visus
/
toolchain-gcc-arm-embedded
Suivre 1
Ajouter aux favoris 0
Bifurcation 0
Code Tickets 0 Demandes d'ajout 0 Versions 0 Wiki Activité
Vous ne pouvez pas sélectionner plus de 25 sujets Les noms de sujets doivent commencer par une lettre ou un nombre, peuvent contenir des tirets ('-') et peuvent comporter jusqu'à 35 caractères.
2 Révisions
1 Branche
Branche: main
Branches Tags
${ item.name }
Créer la branche ${ searchTerm }
de 'main'
${ noResults }
toolchain-gcc-arm-embedded/share/doc/gcc-arm-none-eabi/info/gccint.info

56863 lines
2.6MB
Brut Lien permanent Annotations Historique 

  1. This is gccint.info, produced by makeinfo version 6.5 from gccint.texi.

  2. 

  3. Copyright (C) 1988-2020 Free Software Foundation, Inc.

  4. 

  5.  Permission is granted to copy, distribute and/or modify this document

  6. under the terms of the GNU Free Documentation License, Version 1.3 or

  7. any later version published by the Free Software Foundation; with the

  8. Invariant Sections being "Funding Free Software", the Front-Cover Texts

  9. being (a) (see below), and with the Back-Cover Texts being (b) (see

  10. below).  A copy of the license is included in the section entitled "GNU

  11. Free Documentation License".

  12. 

  13.  (a) The FSF's Front-Cover Text is:

  14. 

  15.  A GNU Manual

  16. 

  17.  (b) The FSF's Back-Cover Text is:

  18. 

  19.  You have freedom to copy and modify this GNU Manual, like GNU software.

  20. Copies published by the Free Software Foundation raise funds for GNU

  21. development.

  22. INFO-DIR-SECTION Software development

  23. START-INFO-DIR-ENTRY

  24. * gccint: (gccint).            Internals of the GNU Compiler Collection.

  25. END-INFO-DIR-ENTRY

  26. 

  27.  This file documents the internals of the GNU compilers.

  28. 

  29.  Copyright (C) 1988-2020 Free Software Foundation, Inc.

  30. 

  31.  Permission is granted to copy, distribute and/or modify this document

  32. under the terms of the GNU Free Documentation License, Version 1.3 or

  33. any later version published by the Free Software Foundation; with the

  34. Invariant Sections being "Funding Free Software", the Front-Cover Texts

  35. being (a) (see below), and with the Back-Cover Texts being (b) (see

  36. below).  A copy of the license is included in the section entitled "GNU

  37. Free Documentation License".

  38. 

  39.  (a) The FSF's Front-Cover Text is:

  40. 

  41.  A GNU Manual

  42. 

  43.  (b) The FSF's Back-Cover Text is:

  44. 

  45.  You have freedom to copy and modify this GNU Manual, like GNU software.

  46. Copies published by the Free Software Foundation raise funds for GNU

  47. development.

  48. 

  49. 

  50. File: gccint.info,  Node: Top,  Next: Contributing

  51. 

  52. Introduction

  53. ************

  54. 

  55. This manual documents the internals of the GNU compilers, including how

  56. to port them to new targets and some information about how to write

  57. front ends for new languages.  It corresponds to the compilers (GNU Arm

  58. Embedded Toolchain 10-2020-q4-major) version 10.2.1.  The use of the GNU

  59. compilers is documented in a separate manual.  *Note Introduction:

  60. (gcc)Top.

  61. 

  62.  This manual is mainly a reference manual rather than a tutorial.  It

  63. discusses how to contribute to GCC (*note Contributing::), the

  64. characteristics of the machines supported by GCC as hosts and targets

  65. (*note Portability::), how GCC relates to the ABIs on such systems

  66. (*note Interface::), and the characteristics of the languages for which

  67. GCC front ends are written (*note Languages::).  It then describes the

  68. GCC source tree structure and build system, some of the interfaces to

  69. GCC front ends, and how support for a target system is implemented in

  70. GCC.

  71. 

  72.  Additional tutorial information is linked to from

  73. <http://gcc.gnu.org/readings.html>.

  74. 

  75. * Menu:

  76. 

  77. * Contributing::    How to contribute to testing and developing GCC.

  78. * Portability::     Goals of GCC's portability features.

  79. * Interface::       Function-call interface of GCC output.

  80. * Libgcc::          Low-level runtime library used by GCC.

  81. * Languages::       Languages for which GCC front ends are written.

  82. * Source Tree::     GCC source tree structure and build system.

  83. * Testsuites::      GCC testsuites.

  84. * Options::         Option specification files.

  85. * Passes::          Order of passes, what they do, and what each file is for.

  86. * poly_int::        Representation of runtime sizes and offsets.

  87. * GENERIC::         Language-independent representation generated by Front Ends

  88. * GIMPLE::          Tuple representation used by Tree SSA optimizers

  89. * Tree SSA::        Analysis and optimization of GIMPLE

  90. * RTL::             Machine-dependent low-level intermediate representation.

  91. * Control Flow::    Maintaining and manipulating the control flow graph.

  92. * Loop Analysis and Representation:: Analysis and representation of loops

  93. * Machine Desc::    How to write machine description instruction patterns.

  94. * Target Macros::   How to write the machine description C macros and functions.

  95. * Host Config::     Writing the 'xm-MACHINE.h' file.

  96. * Fragments::       Writing the 't-TARGET' and 'x-HOST' files.

  97. * Collect2::        How 'collect2' works; how it finds 'ld'.

  98. * Header Dirs::     Understanding the standard header file directories.

  99. * Type Information:: GCC's memory management; generating type information.

  100. * Plugins::         Extending the compiler with plugins.

  101. * LTO::             Using Link-Time Optimization.

  102. 

  103. * Match and Simplify:: How to write expression simplification patterns for GIMPLE and GENERIC

  104. * Static Analyzer:: Working with the static analyzer.

  105. * User Experience Guidelines:: Guidelines for implementing diagnostics and options.

  106. * Funding::         How to help assure funding for free software.

  107. * GNU Project::     The GNU Project and GNU/Linux.

  108. 

  109. * Copying::         GNU General Public License says

  110.                     how you can copy and share GCC.

  111. * GNU Free Documentation License:: How you can copy and share this manual.

  112. * Contributors::    People who have contributed to GCC.

  113. 

  114. * Option Index::    Index to command line options.

  115. * Concept Index::   Index of concepts and symbol names.

  116. 

  117. 

  118. File: gccint.info,  Node: Contributing,  Next: Portability,  Up: Top

  119. 

  120. 1 Contributing to GCC Development

  121. *********************************

  122. 

  123. If you would like to help pretest GCC releases to assure they work well,

  124. current development sources are available via Git (see

  125. <http://gcc.gnu.org/git.html>).  Source and binary snapshots are also

  126. available for FTP; see <http://gcc.gnu.org/snapshots.html>.

  127. 

  128.  If you would like to work on improvements to GCC, please read the

  129. advice at these URLs:

  130. 

  131.      <http://gcc.gnu.org/contribute.html>

  132.      <http://gcc.gnu.org/contributewhy.html>

  133. 

  134. for information on how to make useful contributions and avoid

  135. duplication of effort.  Suggested projects are listed at

  136. <http://gcc.gnu.org/projects/>.

  137. 

  138. 

  139. File: gccint.info,  Node: Portability,  Next: Interface,  Prev: Contributing,  Up: Top

  140. 

  141. 2 GCC and Portability

  142. *********************

  143. 

  144. GCC itself aims to be portable to any machine where 'int' is at least a

  145. 32-bit type.  It aims to target machines with a flat (non-segmented)

  146. byte addressed data address space (the code address space can be

  147. separate).  Target ABIs may have 8, 16, 32 or 64-bit 'int' type.  'char'

  148. can be wider than 8 bits.

  149. 

  150.  GCC gets most of the information about the target machine from a

  151. machine description which gives an algebraic formula for each of the

  152. machine's instructions.  This is a very clean way to describe the

  153. target.  But when the compiler needs information that is difficult to

  154. express in this fashion, ad-hoc parameters have been defined for machine

  155. descriptions.  The purpose of portability is to reduce the total work

  156. needed on the compiler; it was not of interest for its own sake.

  157. 

  158.  GCC does not contain machine dependent code, but it does contain code

  159. that depends on machine parameters such as endianness (whether the most

  160. significant byte has the highest or lowest address of the bytes in a

  161. word) and the availability of autoincrement addressing.  In the

  162. RTL-generation pass, it is often necessary to have multiple strategies

  163. for generating code for a particular kind of syntax tree, strategies

  164. that are usable for different combinations of parameters.  Often, not

  165. all possible cases have been addressed, but only the common ones or only

  166. the ones that have been encountered.  As a result, a new target may

  167. require additional strategies.  You will know if this happens because

  168. the compiler will call 'abort'.  Fortunately, the new strategies can be

  169. added in a machine-independent fashion, and will affect only the target

  170. machines that need them.

  171. 

  172. 

  173. File: gccint.info,  Node: Interface,  Next: Libgcc,  Prev: Portability,  Up: Top

  174. 

  175. 3 Interfacing to GCC Output

  176. ***************************

  177. 

  178. GCC is normally configured to use the same function calling convention

  179. normally in use on the target system.  This is done with the

  180. machine-description macros described (*note Target Macros::).

  181. 

  182.  However, returning of structure and union values is done differently on

  183. some target machines.  As a result, functions compiled with PCC

  184. returning such types cannot be called from code compiled with GCC, and

  185. vice versa.  This does not cause trouble often because few Unix library

  186. routines return structures or unions.

  187. 

  188.  GCC code returns structures and unions that are 1, 2, 4 or 8 bytes long

  189. in the same registers used for 'int' or 'double' return values.  (GCC

  190. typically allocates variables of such types in registers also.)

  191. Structures and unions of other sizes are returned by storing them into

  192. an address passed by the caller (usually in a register).  The target

  193. hook 'TARGET_STRUCT_VALUE_RTX' tells GCC where to pass this address.

  194. 

  195.  By contrast, PCC on most target machines returns structures and unions

  196. of any size by copying the data into an area of static storage, and then

  197. returning the address of that storage as if it were a pointer value.

  198. The caller must copy the data from that memory area to the place where

  199. the value is wanted.  This is slower than the method used by GCC, and

  200. fails to be reentrant.

  201. 

  202.  On some target machines, such as RISC machines and the 80386, the

  203. standard system convention is to pass to the subroutine the address of

  204. where to return the value.  On these machines, GCC has been configured

  205. to be compatible with the standard compiler, when this method is used.

  206. It may not be compatible for structures of 1, 2, 4 or 8 bytes.

  207. 

  208.  GCC uses the system's standard convention for passing arguments.  On

  209. some machines, the first few arguments are passed in registers; in

  210. others, all are passed on the stack.  It would be possible to use

  211. registers for argument passing on any machine, and this would probably

  212. result in a significant speedup.  But the result would be complete

  213. incompatibility with code that follows the standard convention.  So this

  214. change is practical only if you are switching to GCC as the sole C

  215. compiler for the system.  We may implement register argument passing on

  216. certain machines once we have a complete GNU system so that we can

  217. compile the libraries with GCC.

  218. 

  219.  On some machines (particularly the SPARC), certain types of arguments

  220. are passed "by invisible reference".  This means that the value is

  221. stored in memory, and the address of the memory location is passed to

  222. the subroutine.

  223. 

  224.  If you use 'longjmp', beware of automatic variables.  ISO C says that

  225. automatic variables that are not declared 'volatile' have undefined

  226. values after a 'longjmp'.  And this is all GCC promises to do, because

  227. it is very difficult to restore register variables correctly, and one of

  228. GCC's features is that it can put variables in registers without your

  229. asking it to.

  230. 

  231. 

  232. File: gccint.info,  Node: Libgcc,  Next: Languages,  Prev: Interface,  Up: Top

  233. 

  234. 4 The GCC low-level runtime library

  235. ***********************************

  236. 

  237. GCC provides a low-level runtime library, 'libgcc.a' or 'libgcc_s.so.1'

  238. on some platforms.  GCC generates calls to routines in this library

  239. automatically, whenever it needs to perform some operation that is too

  240. complicated to emit inline code for.

  241. 

  242.  Most of the routines in 'libgcc' handle arithmetic operations that the

  243. target processor cannot perform directly.  This includes integer

  244. multiply and divide on some machines, and all floating-point and

  245. fixed-point operations on other machines.  'libgcc' also includes

  246. routines for exception handling, and a handful of miscellaneous

  247. operations.

  248. 

  249.  Some of these routines can be defined in mostly machine-independent C.

  250. Others must be hand-written in assembly language for each processor that

  251. needs them.

  252. 

  253.  GCC will also generate calls to C library routines, such as 'memcpy'

  254. and 'memset', in some cases.  The set of routines that GCC may possibly

  255. use is documented in *note (gcc)Other Builtins::.

  256. 

  257.  These routines take arguments and return values of a specific machine

  258. mode, not a specific C type.  *Note Machine Modes::, for an explanation

  259. of this concept.  For illustrative purposes, in this chapter the

  260. floating point type 'float' is assumed to correspond to 'SFmode';

  261. 'double' to 'DFmode'; and 'long double' to both 'TFmode' and 'XFmode'.

  262. Similarly, the integer types 'int' and 'unsigned int' correspond to

  263. 'SImode'; 'long' and 'unsigned long' to 'DImode'; and 'long long' and

  264. 'unsigned long long' to 'TImode'.

  265. 

  266. * Menu:

  267. 

  268. * Integer library routines::

  269. * Soft float library routines::

  270. * Decimal float library routines::

  271. * Fixed-point fractional library routines::

  272. * Exception handling routines::

  273. * Miscellaneous routines::

  274. 

  275. 

  276. File: gccint.info,  Node: Integer library routines,  Next: Soft float library routines,  Up: Libgcc

  277. 

  278. 4.1 Routines for integer arithmetic

  279. ===================================

  280. 

  281. The integer arithmetic routines are used on platforms that don't provide

  282. hardware support for arithmetic operations on some modes.

  283. 

  284. 4.1.1 Arithmetic functions

  285. --------------------------

  286. 

  287.  -- Runtime Function: int __ashlsi3 (int A, int B)

  288.  -- Runtime Function: long __ashldi3 (long A, int B)

  289.  -- Runtime Function: long long __ashlti3 (long long A, int B)

  290.      These functions return the result of shifting A left by B bits.

  291. 

  292.  -- Runtime Function: int __ashrsi3 (int A, int B)

  293.  -- Runtime Function: long __ashrdi3 (long A, int B)

  294.  -- Runtime Function: long long __ashrti3 (long long A, int B)

  295.      These functions return the result of arithmetically shifting A

  296.      right by B bits.

  297. 

  298.  -- Runtime Function: int __divsi3 (int A, int B)

  299.  -- Runtime Function: long __divdi3 (long A, long B)

  300.  -- Runtime Function: long long __divti3 (long long A, long long B)

  301.      These functions return the quotient of the signed division of A and

  302.      B.

  303. 

  304.  -- Runtime Function: int __lshrsi3 (int A, int B)

  305.  -- Runtime Function: long __lshrdi3 (long A, int B)

  306.  -- Runtime Function: long long __lshrti3 (long long A, int B)

  307.      These functions return the result of logically shifting A right by

  308.      B bits.

  309. 

  310.  -- Runtime Function: int __modsi3 (int A, int B)

  311.  -- Runtime Function: long __moddi3 (long A, long B)

  312.  -- Runtime Function: long long __modti3 (long long A, long long B)

  313.      These functions return the remainder of the signed division of A

  314.      and B.

  315. 

  316.  -- Runtime Function: int __mulsi3 (int A, int B)

  317.  -- Runtime Function: long __muldi3 (long A, long B)

  318.  -- Runtime Function: long long __multi3 (long long A, long long B)

  319.      These functions return the product of A and B.

  320. 

  321.  -- Runtime Function: long __negdi2 (long A)

  322.  -- Runtime Function: long long __negti2 (long long A)

  323.      These functions return the negation of A.

  324. 

  325.  -- Runtime Function: unsigned int __udivsi3 (unsigned int A, unsigned

  326.           int B)

  327.  -- Runtime Function: unsigned long __udivdi3 (unsigned long A, unsigned

  328.           long B)

  329.  -- Runtime Function: unsigned long long __udivti3 (unsigned long long

  330.           A, unsigned long long B)

  331.      These functions return the quotient of the unsigned division of A

  332.      and B.

  333. 

  334.  -- Runtime Function: unsigned long __udivmoddi4 (unsigned long A,

  335.           unsigned long B, unsigned long *C)

  336.  -- Runtime Function: unsigned long long __udivmodti4 (unsigned long

  337.           long A, unsigned long long B, unsigned long long *C)

  338.      These functions calculate both the quotient and remainder of the

  339.      unsigned division of A and B.  The return value is the quotient,

  340.      and the remainder is placed in variable pointed to by C.

  341. 

  342.  -- Runtime Function: unsigned int __umodsi3 (unsigned int A, unsigned

  343.           int B)

  344.  -- Runtime Function: unsigned long __umoddi3 (unsigned long A, unsigned

  345.           long B)

  346.  -- Runtime Function: unsigned long long __umodti3 (unsigned long long

  347.           A, unsigned long long B)

  348.      These functions return the remainder of the unsigned division of A

  349.      and B.

  350. 

  351. 4.1.2 Comparison functions

  352. --------------------------

  353. 

  354. The following functions implement integral comparisons.  These functions

  355. implement a low-level compare, upon which the higher level comparison

  356. operators (such as less than and greater than or equal to) can be

  357. constructed.  The returned values lie in the range zero to two, to allow

  358. the high-level operators to be implemented by testing the returned

  359. result using either signed or unsigned comparison.

  360. 

  361.  -- Runtime Function: int __cmpdi2 (long A, long B)

  362.  -- Runtime Function: int __cmpti2 (long long A, long long B)

  363.      These functions perform a signed comparison of A and B.  If A is

  364.      less than B, they return 0; if A is greater than B, they return 2;

  365.      and if A and B are equal they return 1.

  366. 

  367.  -- Runtime Function: int __ucmpdi2 (unsigned long A, unsigned long B)

  368.  -- Runtime Function: int __ucmpti2 (unsigned long long A, unsigned long

  369.           long B)

  370.      These functions perform an unsigned comparison of A and B.  If A is

  371.      less than B, they return 0; if A is greater than B, they return 2;

  372.      and if A and B are equal they return 1.

  373. 

  374. 4.1.3 Trapping arithmetic functions

  375. -----------------------------------

  376. 

  377. The following functions implement trapping arithmetic.  These functions

  378. call the libc function 'abort' upon signed arithmetic overflow.

  379. 

  380.  -- Runtime Function: int __absvsi2 (int A)

  381.  -- Runtime Function: long __absvdi2 (long A)

  382.      These functions return the absolute value of A.

  383. 

  384.  -- Runtime Function: int __addvsi3 (int A, int B)

  385.  -- Runtime Function: long __addvdi3 (long A, long B)

  386.      These functions return the sum of A and B; that is 'A + B'.

  387. 

  388.  -- Runtime Function: int __mulvsi3 (int A, int B)

  389.  -- Runtime Function: long __mulvdi3 (long A, long B)

  390.      The functions return the product of A and B; that is 'A * B'.

  391. 

  392.  -- Runtime Function: int __negvsi2 (int A)

  393.  -- Runtime Function: long __negvdi2 (long A)

  394.      These functions return the negation of A; that is '-A'.

  395. 

  396.  -- Runtime Function: int __subvsi3 (int A, int B)

  397.  -- Runtime Function: long __subvdi3 (long A, long B)

  398.      These functions return the difference between B and A; that is 'A -

  399.      B'.

  400. 

  401. 4.1.4 Bit operations

  402. --------------------

  403. 

  404.  -- Runtime Function: int __clzsi2 (unsigned int A)

  405.  -- Runtime Function: int __clzdi2 (unsigned long A)

  406.  -- Runtime Function: int __clzti2 (unsigned long long A)

  407.      These functions return the number of leading 0-bits in A, starting

  408.      at the most significant bit position.  If A is zero, the result is

  409.      undefined.

  410. 

  411.  -- Runtime Function: int __ctzsi2 (unsigned int A)

  412.  -- Runtime Function: int __ctzdi2 (unsigned long A)

  413.  -- Runtime Function: int __ctzti2 (unsigned long long A)

  414.      These functions return the number of trailing 0-bits in A, starting

  415.      at the least significant bit position.  If A is zero, the result is

  416.      undefined.

  417. 

  418.  -- Runtime Function: int __ffsdi2 (unsigned long A)

  419.  -- Runtime Function: int __ffsti2 (unsigned long long A)

  420.      These functions return the index of the least significant 1-bit in

  421.      A, or the value zero if A is zero.  The least significant bit is

  422.      index one.

  423. 

  424.  -- Runtime Function: int __paritysi2 (unsigned int A)

  425.  -- Runtime Function: int __paritydi2 (unsigned long A)

  426.  -- Runtime Function: int __parityti2 (unsigned long long A)

  427.      These functions return the value zero if the number of bits set in

  428.      A is even, and the value one otherwise.

  429. 

  430.  -- Runtime Function: int __popcountsi2 (unsigned int A)

  431.  -- Runtime Function: int __popcountdi2 (unsigned long A)

  432.  -- Runtime Function: int __popcountti2 (unsigned long long A)

  433.      These functions return the number of bits set in A.

  434. 

  435.  -- Runtime Function: int32_t __bswapsi2 (int32_t A)

  436.  -- Runtime Function: int64_t __bswapdi2 (int64_t A)

  437.      These functions return the A byteswapped.

  438. 

  439. 

  440. File: gccint.info,  Node: Soft float library routines,  Next: Decimal float library routines,  Prev: Integer library routines,  Up: Libgcc

  441. 

  442. 4.2 Routines for floating point emulation

  443. =========================================

  444. 

  445. The software floating point library is used on machines which do not

  446. have hardware support for floating point.  It is also used whenever

  447. '-msoft-float' is used to disable generation of floating point

  448. instructions.  (Not all targets support this switch.)

  449. 

  450.  For compatibility with other compilers, the floating point emulation

  451. routines can be renamed with the 'DECLARE_LIBRARY_RENAMES' macro (*note

  452. Library Calls::).  In this section, the default names are used.

  453. 

  454.  Presently the library does not support 'XFmode', which is used for

  455. 'long double' on some architectures.

  456. 

  457. 4.2.1 Arithmetic functions

  458. --------------------------

  459. 

  460.  -- Runtime Function: float __addsf3 (float A, float B)

  461.  -- Runtime Function: double __adddf3 (double A, double B)

  462.  -- Runtime Function: long double __addtf3 (long double A, long double

  463.           B)

  464.  -- Runtime Function: long double __addxf3 (long double A, long double

  465.           B)

  466.      These functions return the sum of A and B.

  467. 

  468.  -- Runtime Function: float __subsf3 (float A, float B)

  469.  -- Runtime Function: double __subdf3 (double A, double B)

  470.  -- Runtime Function: long double __subtf3 (long double A, long double

  471.           B)

  472.  -- Runtime Function: long double __subxf3 (long double A, long double

  473.           B)

  474.      These functions return the difference between B and A; that is,

  475.      A - B.

  476. 

  477.  -- Runtime Function: float __mulsf3 (float A, float B)

  478.  -- Runtime Function: double __muldf3 (double A, double B)

  479.  -- Runtime Function: long double __multf3 (long double A, long double

  480.           B)

  481.  -- Runtime Function: long double __mulxf3 (long double A, long double

  482.           B)

  483.      These functions return the product of A and B.

  484. 

  485.  -- Runtime Function: float __divsf3 (float A, float B)

  486.  -- Runtime Function: double __divdf3 (double A, double B)

  487.  -- Runtime Function: long double __divtf3 (long double A, long double

  488.           B)

  489.  -- Runtime Function: long double __divxf3 (long double A, long double

  490.           B)

  491.      These functions return the quotient of A and B; that is, A / B.

  492. 

  493.  -- Runtime Function: float __negsf2 (float A)

  494.  -- Runtime Function: double __negdf2 (double A)

  495.  -- Runtime Function: long double __negtf2 (long double A)

  496.  -- Runtime Function: long double __negxf2 (long double A)

  497.      These functions return the negation of A.  They simply flip the

  498.      sign bit, so they can produce negative zero and negative NaN.

  499. 

  500. 4.2.2 Conversion functions

  501. --------------------------

  502. 

  503.  -- Runtime Function: double __extendsfdf2 (float A)

  504.  -- Runtime Function: long double __extendsftf2 (float A)

  505.  -- Runtime Function: long double __extendsfxf2 (float A)

  506.  -- Runtime Function: long double __extenddftf2 (double A)

  507.  -- Runtime Function: long double __extenddfxf2 (double A)

  508.      These functions extend A to the wider mode of their return type.

  509. 

  510.  -- Runtime Function: double __truncxfdf2 (long double A)

  511.  -- Runtime Function: double __trunctfdf2 (long double A)

  512.  -- Runtime Function: float __truncxfsf2 (long double A)

  513.  -- Runtime Function: float __trunctfsf2 (long double A)

  514.  -- Runtime Function: float __truncdfsf2 (double A)

  515.      These functions truncate A to the narrower mode of their return

  516.      type, rounding toward zero.

  517. 

  518.  -- Runtime Function: int __fixsfsi (float A)

  519.  -- Runtime Function: int __fixdfsi (double A)

  520.  -- Runtime Function: int __fixtfsi (long double A)

  521.  -- Runtime Function: int __fixxfsi (long double A)

  522.      These functions convert A to a signed integer, rounding toward

  523.      zero.

  524. 

  525.  -- Runtime Function: long __fixsfdi (float A)

  526.  -- Runtime Function: long __fixdfdi (double A)

  527.  -- Runtime Function: long __fixtfdi (long double A)

  528.  -- Runtime Function: long __fixxfdi (long double A)

  529.      These functions convert A to a signed long, rounding toward zero.

  530. 

  531.  -- Runtime Function: long long __fixsfti (float A)

  532.  -- Runtime Function: long long __fixdfti (double A)

  533.  -- Runtime Function: long long __fixtfti (long double A)

  534.  -- Runtime Function: long long __fixxfti (long double A)

  535.      These functions convert A to a signed long long, rounding toward

  536.      zero.

  537. 

  538.  -- Runtime Function: unsigned int __fixunssfsi (float A)

  539.  -- Runtime Function: unsigned int __fixunsdfsi (double A)

  540.  -- Runtime Function: unsigned int __fixunstfsi (long double A)

  541.  -- Runtime Function: unsigned int __fixunsxfsi (long double A)

  542.      These functions convert A to an unsigned integer, rounding toward

  543.      zero.  Negative values all become zero.

  544. 

  545.  -- Runtime Function: unsigned long __fixunssfdi (float A)

  546.  -- Runtime Function: unsigned long __fixunsdfdi (double A)

  547.  -- Runtime Function: unsigned long __fixunstfdi (long double A)

  548.  -- Runtime Function: unsigned long __fixunsxfdi (long double A)

  549.      These functions convert A to an unsigned long, rounding toward

  550.      zero.  Negative values all become zero.

  551. 

  552.  -- Runtime Function: unsigned long long __fixunssfti (float A)

  553.  -- Runtime Function: unsigned long long __fixunsdfti (double A)

  554.  -- Runtime Function: unsigned long long __fixunstfti (long double A)

  555.  -- Runtime Function: unsigned long long __fixunsxfti (long double A)

  556.      These functions convert A to an unsigned long long, rounding toward

  557.      zero.  Negative values all become zero.

  558. 

  559.  -- Runtime Function: float __floatsisf (int I)

  560.  -- Runtime Function: double __floatsidf (int I)

  561.  -- Runtime Function: long double __floatsitf (int I)

  562.  -- Runtime Function: long double __floatsixf (int I)

  563.      These functions convert I, a signed integer, to floating point.

  564. 

  565.  -- Runtime Function: float __floatdisf (long I)

  566.  -- Runtime Function: double __floatdidf (long I)

  567.  -- Runtime Function: long double __floatditf (long I)

  568.  -- Runtime Function: long double __floatdixf (long I)

  569.      These functions convert I, a signed long, to floating point.

  570. 

  571.  -- Runtime Function: float __floattisf (long long I)

  572.  -- Runtime Function: double __floattidf (long long I)

  573.  -- Runtime Function: long double __floattitf (long long I)

  574.  -- Runtime Function: long double __floattixf (long long I)

  575.      These functions convert I, a signed long long, to floating point.

  576. 

  577.  -- Runtime Function: float __floatunsisf (unsigned int I)

  578.  -- Runtime Function: double __floatunsidf (unsigned int I)

  579.  -- Runtime Function: long double __floatunsitf (unsigned int I)

  580.  -- Runtime Function: long double __floatunsixf (unsigned int I)

  581.      These functions convert I, an unsigned integer, to floating point.

  582. 

  583.  -- Runtime Function: float __floatundisf (unsigned long I)

  584.  -- Runtime Function: double __floatundidf (unsigned long I)

  585.  -- Runtime Function: long double __floatunditf (unsigned long I)

  586.  -- Runtime Function: long double __floatundixf (unsigned long I)

  587.      These functions convert I, an unsigned long, to floating point.

  588. 

  589.  -- Runtime Function: float __floatuntisf (unsigned long long I)

  590.  -- Runtime Function: double __floatuntidf (unsigned long long I)

  591.  -- Runtime Function: long double __floatuntitf (unsigned long long I)

  592.  -- Runtime Function: long double __floatuntixf (unsigned long long I)

  593.      These functions convert I, an unsigned long long, to floating

  594.      point.

  595. 

  596. 4.2.3 Comparison functions

  597. --------------------------

  598. 

  599. There are two sets of basic comparison functions.

  600. 

  601.  -- Runtime Function: int __cmpsf2 (float A, float B)

  602.  -- Runtime Function: int __cmpdf2 (double A, double B)

  603.  -- Runtime Function: int __cmptf2 (long double A, long double B)

  604.      These functions calculate a <=> b.  That is, if A is less than B,

  605.      they return -1; if A is greater than B, they return 1; and if A and

  606.      B are equal they return 0.  If either argument is NaN they return

  607.      1, but you should not rely on this; if NaN is a possibility, use

  608.      one of the higher-level comparison functions.

  609. 

  610.  -- Runtime Function: int __unordsf2 (float A, float B)

  611.  -- Runtime Function: int __unorddf2 (double A, double B)

  612.  -- Runtime Function: int __unordtf2 (long double A, long double B)

  613.      These functions return a nonzero value if either argument is NaN,

  614.      otherwise 0.

  615. 

  616.  There is also a complete group of higher level functions which

  617. correspond directly to comparison operators.  They implement the ISO C

  618. semantics for floating-point comparisons, taking NaN into account.  Pay

  619. careful attention to the return values defined for each set.  Under the

  620. hood, all of these routines are implemented as

  621. 

  622.        if (__unordXf2 (a, b))

  623.          return E;

  624.        return __cmpXf2 (a, b);

  625. 

  626. where E is a constant chosen to give the proper behavior for NaN.  Thus,

  627. the meaning of the return value is different for each set.  Do not rely

  628. on this implementation; only the semantics documented below are

  629. guaranteed.

  630. 

  631.  -- Runtime Function: int __eqsf2 (float A, float B)

  632.  -- Runtime Function: int __eqdf2 (double A, double B)

  633.  -- Runtime Function: int __eqtf2 (long double A, long double B)

  634.      These functions return zero if neither argument is NaN, and A and B

  635.      are equal.

  636. 

  637.  -- Runtime Function: int __nesf2 (float A, float B)

  638.  -- Runtime Function: int __nedf2 (double A, double B)

  639.  -- Runtime Function: int __netf2 (long double A, long double B)

  640.      These functions return a nonzero value if either argument is NaN,

  641.      or if A and B are unequal.

  642. 

  643.  -- Runtime Function: int __gesf2 (float A, float B)

  644.  -- Runtime Function: int __gedf2 (double A, double B)

  645.  -- Runtime Function: int __getf2 (long double A, long double B)

  646.      These functions return a value greater than or equal to zero if

  647.      neither argument is NaN, and A is greater than or equal to B.

  648. 

  649.  -- Runtime Function: int __ltsf2 (float A, float B)

  650.  -- Runtime Function: int __ltdf2 (double A, double B)

  651.  -- Runtime Function: int __lttf2 (long double A, long double B)

  652.      These functions return a value less than zero if neither argument

  653.      is NaN, and A is strictly less than B.

  654. 

  655.  -- Runtime Function: int __lesf2 (float A, float B)

  656.  -- Runtime Function: int __ledf2 (double A, double B)

  657.  -- Runtime Function: int __letf2 (long double A, long double B)

  658.      These functions return a value less than or equal to zero if

  659.      neither argument is NaN, and A is less than or equal to B.

  660. 

  661.  -- Runtime Function: int __gtsf2 (float A, float B)

  662.  -- Runtime Function: int __gtdf2 (double A, double B)

  663.  -- Runtime Function: int __gttf2 (long double A, long double B)

  664.      These functions return a value greater than zero if neither

  665.      argument is NaN, and A is strictly greater than B.

  666. 

  667. 4.2.4 Other floating-point functions

  668. ------------------------------------

  669. 

  670.  -- Runtime Function: float __powisf2 (float A, int B)

  671.  -- Runtime Function: double __powidf2 (double A, int B)

  672.  -- Runtime Function: long double __powitf2 (long double A, int B)

  673.  -- Runtime Function: long double __powixf2 (long double A, int B)

  674.      These functions convert raise A to the power B.

  675. 

  676.  -- Runtime Function: complex float __mulsc3 (float A, float B, float C,

  677.           float D)

  678.  -- Runtime Function: complex double __muldc3 (double A, double B,

  679.           double C, double D)

  680.  -- Runtime Function: complex long double __multc3 (long double A, long

  681.           double B, long double C, long double D)

  682.  -- Runtime Function: complex long double __mulxc3 (long double A, long

  683.           double B, long double C, long double D)

  684.      These functions return the product of A + iB and C + iD, following

  685.      the rules of C99 Annex G.

  686. 

  687.  -- Runtime Function: complex float __divsc3 (float A, float B, float C,

  688.           float D)

  689.  -- Runtime Function: complex double __divdc3 (double A, double B,

  690.           double C, double D)

  691.  -- Runtime Function: complex long double __divtc3 (long double A, long

  692.           double B, long double C, long double D)

  693.  -- Runtime Function: complex long double __divxc3 (long double A, long

  694.           double B, long double C, long double D)

  695.      These functions return the quotient of A + iB and C + iD (i.e., (A

  696.      + iB) / (C + iD)), following the rules of C99 Annex G.

  697. 

  698. 

  699. File: gccint.info,  Node: Decimal float library routines,  Next: Fixed-point fractional library routines,  Prev: Soft float library routines,  Up: Libgcc

  700. 

  701. 4.3 Routines for decimal floating point emulation

  702. =================================================

  703. 

  704. The software decimal floating point library implements IEEE 754-2008

  705. decimal floating point arithmetic and is only activated on selected

  706. targets.

  707. 

  708.  The software decimal floating point library supports either DPD

  709. (Densely Packed Decimal) or BID (Binary Integer Decimal) encoding as

  710. selected at configure time.

  711. 

  712. 4.3.1 Arithmetic functions

  713. --------------------------

  714. 

  715.  -- Runtime Function: _Decimal32 __dpd_addsd3 (_Decimal32 A, _Decimal32

  716.           B)

  717.  -- Runtime Function: _Decimal32 __bid_addsd3 (_Decimal32 A, _Decimal32

  718.           B)

  719.  -- Runtime Function: _Decimal64 __dpd_adddd3 (_Decimal64 A, _Decimal64

  720.           B)

  721.  -- Runtime Function: _Decimal64 __bid_adddd3 (_Decimal64 A, _Decimal64

  722.           B)

  723.  -- Runtime Function: _Decimal128 __dpd_addtd3 (_Decimal128 A,

  724.           _Decimal128 B)

  725.  -- Runtime Function: _Decimal128 __bid_addtd3 (_Decimal128 A,

  726.           _Decimal128 B)

  727.      These functions return the sum of A and B.

  728. 

  729.  -- Runtime Function: _Decimal32 __dpd_subsd3 (_Decimal32 A, _Decimal32

  730.           B)

  731.  -- Runtime Function: _Decimal32 __bid_subsd3 (_Decimal32 A, _Decimal32

  732.           B)

  733.  -- Runtime Function: _Decimal64 __dpd_subdd3 (_Decimal64 A, _Decimal64

  734.           B)

  735.  -- Runtime Function: _Decimal64 __bid_subdd3 (_Decimal64 A, _Decimal64

  736.           B)

  737.  -- Runtime Function: _Decimal128 __dpd_subtd3 (_Decimal128 A,

  738.           _Decimal128 B)

  739.  -- Runtime Function: _Decimal128 __bid_subtd3 (_Decimal128 A,

  740.           _Decimal128 B)

  741.      These functions return the difference between B and A; that is,

  742.      A - B.

  743. 

  744.  -- Runtime Function: _Decimal32 __dpd_mulsd3 (_Decimal32 A, _Decimal32

  745.           B)

  746.  -- Runtime Function: _Decimal32 __bid_mulsd3 (_Decimal32 A, _Decimal32

  747.           B)

  748.  -- Runtime Function: _Decimal64 __dpd_muldd3 (_Decimal64 A, _Decimal64

  749.           B)

  750.  -- Runtime Function: _Decimal64 __bid_muldd3 (_Decimal64 A, _Decimal64

  751.           B)

  752.  -- Runtime Function: _Decimal128 __dpd_multd3 (_Decimal128 A,

  753.           _Decimal128 B)

  754.  -- Runtime Function: _Decimal128 __bid_multd3 (_Decimal128 A,

  755.           _Decimal128 B)

  756.      These functions return the product of A and B.

  757. 

  758.  -- Runtime Function: _Decimal32 __dpd_divsd3 (_Decimal32 A, _Decimal32

  759.           B)

  760.  -- Runtime Function: _Decimal32 __bid_divsd3 (_Decimal32 A, _Decimal32

  761.           B)

  762.  -- Runtime Function: _Decimal64 __dpd_divdd3 (_Decimal64 A, _Decimal64

  763.           B)

  764.  -- Runtime Function: _Decimal64 __bid_divdd3 (_Decimal64 A, _Decimal64

  765.           B)

  766.  -- Runtime Function: _Decimal128 __dpd_divtd3 (_Decimal128 A,

  767.           _Decimal128 B)

  768.  -- Runtime Function: _Decimal128 __bid_divtd3 (_Decimal128 A,

  769.           _Decimal128 B)

  770.      These functions return the quotient of A and B; that is, A / B.

  771. 

  772.  -- Runtime Function: _Decimal32 __dpd_negsd2 (_Decimal32 A)

  773.  -- Runtime Function: _Decimal32 __bid_negsd2 (_Decimal32 A)

  774.  -- Runtime Function: _Decimal64 __dpd_negdd2 (_Decimal64 A)

  775.  -- Runtime Function: _Decimal64 __bid_negdd2 (_Decimal64 A)

  776.  -- Runtime Function: _Decimal128 __dpd_negtd2 (_Decimal128 A)

  777.  -- Runtime Function: _Decimal128 __bid_negtd2 (_Decimal128 A)

  778.      These functions return the negation of A.  They simply flip the

  779.      sign bit, so they can produce negative zero and negative NaN.

  780. 

  781. 4.3.2 Conversion functions

  782. --------------------------

  783. 

  784.  -- Runtime Function: _Decimal64 __dpd_extendsddd2 (_Decimal32 A)

  785.  -- Runtime Function: _Decimal64 __bid_extendsddd2 (_Decimal32 A)

  786.  -- Runtime Function: _Decimal128 __dpd_extendsdtd2 (_Decimal32 A)

  787.  -- Runtime Function: _Decimal128 __bid_extendsdtd2 (_Decimal32 A)

  788.  -- Runtime Function: _Decimal128 __dpd_extendddtd2 (_Decimal64 A)

  789.  -- Runtime Function: _Decimal128 __bid_extendddtd2 (_Decimal64 A)

  790.  -- Runtime Function: _Decimal32 __dpd_truncddsd2 (_Decimal64 A)

  791.  -- Runtime Function: _Decimal32 __bid_truncddsd2 (_Decimal64 A)

  792.  -- Runtime Function: _Decimal32 __dpd_trunctdsd2 (_Decimal128 A)

  793.  -- Runtime Function: _Decimal32 __bid_trunctdsd2 (_Decimal128 A)

  794.  -- Runtime Function: _Decimal64 __dpd_trunctddd2 (_Decimal128 A)

  795.  -- Runtime Function: _Decimal64 __bid_trunctddd2 (_Decimal128 A)

  796.      These functions convert the value A from one decimal floating type

  797.      to another.

  798. 

  799.  -- Runtime Function: _Decimal64 __dpd_extendsfdd (float A)

  800.  -- Runtime Function: _Decimal64 __bid_extendsfdd (float A)

  801.  -- Runtime Function: _Decimal128 __dpd_extendsftd (float A)

  802.  -- Runtime Function: _Decimal128 __bid_extendsftd (float A)

  803.  -- Runtime Function: _Decimal128 __dpd_extenddftd (double A)

  804.  -- Runtime Function: _Decimal128 __bid_extenddftd (double A)

  805.  -- Runtime Function: _Decimal128 __dpd_extendxftd (long double A)

  806.  -- Runtime Function: _Decimal128 __bid_extendxftd (long double A)

  807.  -- Runtime Function: _Decimal32 __dpd_truncdfsd (double A)

  808.  -- Runtime Function: _Decimal32 __bid_truncdfsd (double A)

  809.  -- Runtime Function: _Decimal32 __dpd_truncxfsd (long double A)

  810.  -- Runtime Function: _Decimal32 __bid_truncxfsd (long double A)

  811.  -- Runtime Function: _Decimal32 __dpd_trunctfsd (long double A)

  812.  -- Runtime Function: _Decimal32 __bid_trunctfsd (long double A)

  813.  -- Runtime Function: _Decimal64 __dpd_truncxfdd (long double A)

  814.  -- Runtime Function: _Decimal64 __bid_truncxfdd (long double A)

  815.  -- Runtime Function: _Decimal64 __dpd_trunctfdd (long double A)

  816.  -- Runtime Function: _Decimal64 __bid_trunctfdd (long double A)

  817.      These functions convert the value of A from a binary floating type

  818.      to a decimal floating type of a different size.

  819. 

  820.  -- Runtime Function: float __dpd_truncddsf (_Decimal64 A)

  821.  -- Runtime Function: float __bid_truncddsf (_Decimal64 A)

  822.  -- Runtime Function: float __dpd_trunctdsf (_Decimal128 A)

  823.  -- Runtime Function: float __bid_trunctdsf (_Decimal128 A)

  824.  -- Runtime Function: double __dpd_extendsddf (_Decimal32 A)

  825.  -- Runtime Function: double __bid_extendsddf (_Decimal32 A)

  826.  -- Runtime Function: double __dpd_trunctddf (_Decimal128 A)

  827.  -- Runtime Function: double __bid_trunctddf (_Decimal128 A)

  828.  -- Runtime Function: long double __dpd_extendsdxf (_Decimal32 A)

  829.  -- Runtime Function: long double __bid_extendsdxf (_Decimal32 A)

  830.  -- Runtime Function: long double __dpd_extendddxf (_Decimal64 A)

  831.  -- Runtime Function: long double __bid_extendddxf (_Decimal64 A)

  832.  -- Runtime Function: long double __dpd_trunctdxf (_Decimal128 A)

  833.  -- Runtime Function: long double __bid_trunctdxf (_Decimal128 A)

  834.  -- Runtime Function: long double __dpd_extendsdtf (_Decimal32 A)

  835.  -- Runtime Function: long double __bid_extendsdtf (_Decimal32 A)

  836.  -- Runtime Function: long double __dpd_extendddtf (_Decimal64 A)

  837.  -- Runtime Function: long double __bid_extendddtf (_Decimal64 A)

  838.      These functions convert the value of A from a decimal floating type

  839.      to a binary floating type of a different size.

  840. 

  841.  -- Runtime Function: _Decimal32 __dpd_extendsfsd (float A)

  842.  -- Runtime Function: _Decimal32 __bid_extendsfsd (float A)

  843.  -- Runtime Function: _Decimal64 __dpd_extenddfdd (double A)

  844.  -- Runtime Function: _Decimal64 __bid_extenddfdd (double A)

  845.  -- Runtime Function: _Decimal128 __dpd_extendtftd (long double A)

  846.  -- Runtime Function: _Decimal128 __bid_extendtftd (long double A)

  847.  -- Runtime Function: float __dpd_truncsdsf (_Decimal32 A)

  848.  -- Runtime Function: float __bid_truncsdsf (_Decimal32 A)

  849.  -- Runtime Function: double __dpd_truncdddf (_Decimal64 A)

  850.  -- Runtime Function: double __bid_truncdddf (_Decimal64 A)

  851.  -- Runtime Function: long double __dpd_trunctdtf (_Decimal128 A)

  852.  -- Runtime Function: long double __bid_trunctdtf (_Decimal128 A)

  853.      These functions convert the value of A between decimal and binary

  854.      floating types of the same size.

  855. 

  856.  -- Runtime Function: int __dpd_fixsdsi (_Decimal32 A)

  857.  -- Runtime Function: int __bid_fixsdsi (_Decimal32 A)

  858.  -- Runtime Function: int __dpd_fixddsi (_Decimal64 A)

  859.  -- Runtime Function: int __bid_fixddsi (_Decimal64 A)

  860.  -- Runtime Function: int __dpd_fixtdsi (_Decimal128 A)

  861.  -- Runtime Function: int __bid_fixtdsi (_Decimal128 A)

  862.      These functions convert A to a signed integer.

  863. 

  864.  -- Runtime Function: long __dpd_fixsddi (_Decimal32 A)

  865.  -- Runtime Function: long __bid_fixsddi (_Decimal32 A)

  866.  -- Runtime Function: long __dpd_fixdddi (_Decimal64 A)

  867.  -- Runtime Function: long __bid_fixdddi (_Decimal64 A)

  868.  -- Runtime Function: long __dpd_fixtddi (_Decimal128 A)

  869.  -- Runtime Function: long __bid_fixtddi (_Decimal128 A)

  870.      These functions convert A to a signed long.

  871. 

  872.  -- Runtime Function: unsigned int __dpd_fixunssdsi (_Decimal32 A)

  873.  -- Runtime Function: unsigned int __bid_fixunssdsi (_Decimal32 A)

  874.  -- Runtime Function: unsigned int __dpd_fixunsddsi (_Decimal64 A)

  875.  -- Runtime Function: unsigned int __bid_fixunsddsi (_Decimal64 A)

  876.  -- Runtime Function: unsigned int __dpd_fixunstdsi (_Decimal128 A)

  877.  -- Runtime Function: unsigned int __bid_fixunstdsi (_Decimal128 A)

  878.      These functions convert A to an unsigned integer.  Negative values

  879.      all become zero.

  880. 

  881.  -- Runtime Function: unsigned long __dpd_fixunssddi (_Decimal32 A)

  882.  -- Runtime Function: unsigned long __bid_fixunssddi (_Decimal32 A)

  883.  -- Runtime Function: unsigned long __dpd_fixunsdddi (_Decimal64 A)

  884.  -- Runtime Function: unsigned long __bid_fixunsdddi (_Decimal64 A)

  885.  -- Runtime Function: unsigned long __dpd_fixunstddi (_Decimal128 A)

  886.  -- Runtime Function: unsigned long __bid_fixunstddi (_Decimal128 A)

  887.      These functions convert A to an unsigned long.  Negative values all

  888.      become zero.

  889. 

  890.  -- Runtime Function: _Decimal32 __dpd_floatsisd (int I)

  891.  -- Runtime Function: _Decimal32 __bid_floatsisd (int I)

  892.  -- Runtime Function: _Decimal64 __dpd_floatsidd (int I)

  893.  -- Runtime Function: _Decimal64 __bid_floatsidd (int I)

  894.  -- Runtime Function: _Decimal128 __dpd_floatsitd (int I)

  895.  -- Runtime Function: _Decimal128 __bid_floatsitd (int I)

  896.      These functions convert I, a signed integer, to decimal floating

  897.      point.

  898. 

  899.  -- Runtime Function: _Decimal32 __dpd_floatdisd (long I)

  900.  -- Runtime Function: _Decimal32 __bid_floatdisd (long I)

  901.  -- Runtime Function: _Decimal64 __dpd_floatdidd (long I)

  902.  -- Runtime Function: _Decimal64 __bid_floatdidd (long I)

  903.  -- Runtime Function: _Decimal128 __dpd_floatditd (long I)

  904.  -- Runtime Function: _Decimal128 __bid_floatditd (long I)

  905.      These functions convert I, a signed long, to decimal floating

  906.      point.

  907. 

  908.  -- Runtime Function: _Decimal32 __dpd_floatunssisd (unsigned int I)

  909.  -- Runtime Function: _Decimal32 __bid_floatunssisd (unsigned int I)

  910.  -- Runtime Function: _Decimal64 __dpd_floatunssidd (unsigned int I)

  911.  -- Runtime Function: _Decimal64 __bid_floatunssidd (unsigned int I)

  912.  -- Runtime Function: _Decimal128 __dpd_floatunssitd (unsigned int I)

  913.  -- Runtime Function: _Decimal128 __bid_floatunssitd (unsigned int I)

  914.      These functions convert I, an unsigned integer, to decimal floating

  915.      point.

  916. 

  917.  -- Runtime Function: _Decimal32 __dpd_floatunsdisd (unsigned long I)

  918.  -- Runtime Function: _Decimal32 __bid_floatunsdisd (unsigned long I)

  919.  -- Runtime Function: _Decimal64 __dpd_floatunsdidd (unsigned long I)

  920.  -- Runtime Function: _Decimal64 __bid_floatunsdidd (unsigned long I)

  921.  -- Runtime Function: _Decimal128 __dpd_floatunsditd (unsigned long I)

  922.  -- Runtime Function: _Decimal128 __bid_floatunsditd (unsigned long I)

  923.      These functions convert I, an unsigned long, to decimal floating

  924.      point.

  925. 

  926. 4.3.3 Comparison functions

  927. --------------------------

  928. 

  929.  -- Runtime Function: int __dpd_unordsd2 (_Decimal32 A, _Decimal32 B)

  930.  -- Runtime Function: int __bid_unordsd2 (_Decimal32 A, _Decimal32 B)

  931.  -- Runtime Function: int __dpd_unorddd2 (_Decimal64 A, _Decimal64 B)

  932.  -- Runtime Function: int __bid_unorddd2 (_Decimal64 A, _Decimal64 B)

  933.  -- Runtime Function: int __dpd_unordtd2 (_Decimal128 A, _Decimal128 B)

  934.  -- Runtime Function: int __bid_unordtd2 (_Decimal128 A, _Decimal128 B)

  935.      These functions return a nonzero value if either argument is NaN,

  936.      otherwise 0.

  937. 

  938.  There is also a complete group of higher level functions which

  939. correspond directly to comparison operators.  They implement the ISO C

  940. semantics for floating-point comparisons, taking NaN into account.  Pay

  941. careful attention to the return values defined for each set.  Under the

  942. hood, all of these routines are implemented as

  943. 

  944.        if (__bid_unordXd2 (a, b))

  945.          return E;

  946.        return __bid_cmpXd2 (a, b);

  947. 

  948. where E is a constant chosen to give the proper behavior for NaN.  Thus,

  949. the meaning of the return value is different for each set.  Do not rely

  950. on this implementation; only the semantics documented below are

  951. guaranteed.

  952. 

  953.  -- Runtime Function: int __dpd_eqsd2 (_Decimal32 A, _Decimal32 B)

  954.  -- Runtime Function: int __bid_eqsd2 (_Decimal32 A, _Decimal32 B)

  955.  -- Runtime Function: int __dpd_eqdd2 (_Decimal64 A, _Decimal64 B)

  956.  -- Runtime Function: int __bid_eqdd2 (_Decimal64 A, _Decimal64 B)

  957.  -- Runtime Function: int __dpd_eqtd2 (_Decimal128 A, _Decimal128 B)

  958.  -- Runtime Function: int __bid_eqtd2 (_Decimal128 A, _Decimal128 B)

  959.      These functions return zero if neither argument is NaN, and A and B

  960.      are equal.

  961. 

  962.  -- Runtime Function: int __dpd_nesd2 (_Decimal32 A, _Decimal32 B)

  963.  -- Runtime Function: int __bid_nesd2 (_Decimal32 A, _Decimal32 B)

  964.  -- Runtime Function: int __dpd_nedd2 (_Decimal64 A, _Decimal64 B)

  965.  -- Runtime Function: int __bid_nedd2 (_Decimal64 A, _Decimal64 B)

  966.  -- Runtime Function: int __dpd_netd2 (_Decimal128 A, _Decimal128 B)

  967.  -- Runtime Function: int __bid_netd2 (_Decimal128 A, _Decimal128 B)

  968.      These functions return a nonzero value if either argument is NaN,

  969.      or if A and B are unequal.

  970. 

  971.  -- Runtime Function: int __dpd_gesd2 (_Decimal32 A, _Decimal32 B)

  972.  -- Runtime Function: int __bid_gesd2 (_Decimal32 A, _Decimal32 B)

  973.  -- Runtime Function: int __dpd_gedd2 (_Decimal64 A, _Decimal64 B)

  974.  -- Runtime Function: int __bid_gedd2 (_Decimal64 A, _Decimal64 B)

  975.  -- Runtime Function: int __dpd_getd2 (_Decimal128 A, _Decimal128 B)

  976.  -- Runtime Function: int __bid_getd2 (_Decimal128 A, _Decimal128 B)

  977.      These functions return a value greater than or equal to zero if

  978.      neither argument is NaN, and A is greater than or equal to B.

  979. 

  980.  -- Runtime Function: int __dpd_ltsd2 (_Decimal32 A, _Decimal32 B)

  981.  -- Runtime Function: int __bid_ltsd2 (_Decimal32 A, _Decimal32 B)

  982.  -- Runtime Function: int __dpd_ltdd2 (_Decimal64 A, _Decimal64 B)

  983.  -- Runtime Function: int __bid_ltdd2 (_Decimal64 A, _Decimal64 B)

  984.  -- Runtime Function: int __dpd_lttd2 (_Decimal128 A, _Decimal128 B)

  985.  -- Runtime Function: int __bid_lttd2 (_Decimal128 A, _Decimal128 B)

  986.      These functions return a value less than zero if neither argument

  987.      is NaN, and A is strictly less than B.

  988. 

  989.  -- Runtime Function: int __dpd_lesd2 (_Decimal32 A, _Decimal32 B)

  990.  -- Runtime Function: int __bid_lesd2 (_Decimal32 A, _Decimal32 B)

  991.  -- Runtime Function: int __dpd_ledd2 (_Decimal64 A, _Decimal64 B)

  992.  -- Runtime Function: int __bid_ledd2 (_Decimal64 A, _Decimal64 B)

  993.  -- Runtime Function: int __dpd_letd2 (_Decimal128 A, _Decimal128 B)

  994.  -- Runtime Function: int __bid_letd2 (_Decimal128 A, _Decimal128 B)

  995.      These functions return a value less than or equal to zero if

  996.      neither argument is NaN, and A is less than or equal to B.

  997. 

  998.  -- Runtime Function: int __dpd_gtsd2 (_Decimal32 A, _Decimal32 B)

  999.  -- Runtime Function: int __bid_gtsd2 (_Decimal32 A, _Decimal32 B)

  1000.  -- Runtime Function: int __dpd_gtdd2 (_Decimal64 A, _Decimal64 B)

  1001.  -- Runtime Function: int __bid_gtdd2 (_Decimal64 A, _Decimal64 B)

  1002.  -- Runtime Function: int __dpd_gttd2 (_Decimal128 A, _Decimal128 B)

  1003.  -- Runtime Function: int __bid_gttd2 (_Decimal128 A, _Decimal128 B)

  1004.      These functions return a value greater than zero if neither

  1005.      argument is NaN, and A is strictly greater than B.

  1006. 

  1007. 

  1008. File: gccint.info,  Node: Fixed-point fractional library routines,  Next: Exception handling routines,  Prev: Decimal float library routines,  Up: Libgcc

  1009. 

  1010. 4.4 Routines for fixed-point fractional emulation

  1011. =================================================

  1012. 

  1013. The software fixed-point library implements fixed-point fractional

  1014. arithmetic, and is only activated on selected targets.

  1015. 

  1016.  For ease of comprehension 'fract' is an alias for the '_Fract' type,

  1017. 'accum' an alias for '_Accum', and 'sat' an alias for '_Sat'.

  1018. 

  1019.  For illustrative purposes, in this section the fixed-point fractional

  1020. type 'short fract' is assumed to correspond to machine mode 'QQmode';

  1021. 'unsigned short fract' to 'UQQmode'; 'fract' to 'HQmode';

  1022. 'unsigned fract' to 'UHQmode'; 'long fract' to 'SQmode';

  1023. 'unsigned long fract' to 'USQmode'; 'long long fract' to 'DQmode'; and

  1024. 'unsigned long long fract' to 'UDQmode'.  Similarly the fixed-point

  1025. accumulator type 'short accum' corresponds to 'HAmode';

  1026. 'unsigned short accum' to 'UHAmode'; 'accum' to 'SAmode';

  1027. 'unsigned accum' to 'USAmode'; 'long accum' to 'DAmode';

  1028. 'unsigned long accum' to 'UDAmode'; 'long long accum' to 'TAmode'; and

  1029. 'unsigned long long accum' to 'UTAmode'.

  1030. 

  1031. 4.4.1 Arithmetic functions

  1032. --------------------------

  1033. 

  1034.  -- Runtime Function: short fract __addqq3 (short fract A, short fract

  1035.           B)

  1036.  -- Runtime Function: fract __addhq3 (fract A, fract B)

  1037.  -- Runtime Function: long fract __addsq3 (long fract A, long fract B)

  1038.  -- Runtime Function: long long fract __adddq3 (long long fract A, long

  1039.           long fract B)

  1040.  -- Runtime Function: unsigned short fract __adduqq3 (unsigned short

  1041.           fract A, unsigned short fract B)

  1042.  -- Runtime Function: unsigned fract __adduhq3 (unsigned fract A,

  1043.           unsigned fract B)

  1044.  -- Runtime Function: unsigned long fract __addusq3 (unsigned long fract

  1045.           A, unsigned long fract B)

  1046.  -- Runtime Function: unsigned long long fract __addudq3 (unsigned long

  1047.           long fract A, unsigned long long fract B)

  1048.  -- Runtime Function: short accum __addha3 (short accum A, short accum

  1049.           B)

  1050.  -- Runtime Function: accum __addsa3 (accum A, accum B)

  1051.  -- Runtime Function: long accum __addda3 (long accum A, long accum B)

  1052.  -- Runtime Function: long long accum __addta3 (long long accum A, long

  1053.           long accum B)

  1054.  -- Runtime Function: unsigned short accum __adduha3 (unsigned short

  1055.           accum A, unsigned short accum B)

  1056.  -- Runtime Function: unsigned accum __addusa3 (unsigned accum A,

  1057.           unsigned accum B)

  1058.  -- Runtime Function: unsigned long accum __adduda3 (unsigned long accum

  1059.           A, unsigned long accum B)

  1060.  -- Runtime Function: unsigned long long accum __adduta3 (unsigned long

  1061.           long accum A, unsigned long long accum B)

  1062.      These functions return the sum of A and B.

  1063. 

  1064.  -- Runtime Function: short fract __ssaddqq3 (short fract A, short fract

  1065.           B)

  1066.  -- Runtime Function: fract __ssaddhq3 (fract A, fract B)

  1067.  -- Runtime Function: long fract __ssaddsq3 (long fract A, long fract B)

  1068.  -- Runtime Function: long long fract __ssadddq3 (long long fract A,

  1069.           long long fract B)

  1070.  -- Runtime Function: short accum __ssaddha3 (short accum A, short accum

  1071.           B)

  1072.  -- Runtime Function: accum __ssaddsa3 (accum A, accum B)

  1073.  -- Runtime Function: long accum __ssaddda3 (long accum A, long accum B)

  1074.  -- Runtime Function: long long accum __ssaddta3 (long long accum A,

  1075.           long long accum B)

  1076.      These functions return the sum of A and B with signed saturation.

  1077. 

  1078.  -- Runtime Function: unsigned short fract __usadduqq3 (unsigned short

  1079.           fract A, unsigned short fract B)

  1080.  -- Runtime Function: unsigned fract __usadduhq3 (unsigned fract A,

  1081.           unsigned fract B)

  1082.  -- Runtime Function: unsigned long fract __usaddusq3 (unsigned long

  1083.           fract A, unsigned long fract B)

  1084.  -- Runtime Function: unsigned long long fract __usaddudq3 (unsigned

  1085.           long long fract A, unsigned long long fract B)

  1086.  -- Runtime Function: unsigned short accum __usadduha3 (unsigned short

  1087.           accum A, unsigned short accum B)

  1088.  -- Runtime Function: unsigned accum __usaddusa3 (unsigned accum A,

  1089.           unsigned accum B)

  1090.  -- Runtime Function: unsigned long accum __usadduda3 (unsigned long

  1091.           accum A, unsigned long accum B)

  1092.  -- Runtime Function: unsigned long long accum __usadduta3 (unsigned

  1093.           long long accum A, unsigned long long accum B)

  1094.      These functions return the sum of A and B with unsigned saturation.

  1095. 

  1096.  -- Runtime Function: short fract __subqq3 (short fract A, short fract

  1097.           B)

  1098.  -- Runtime Function: fract __subhq3 (fract A, fract B)

  1099.  -- Runtime Function: long fract __subsq3 (long fract A, long fract B)

  1100.  -- Runtime Function: long long fract __subdq3 (long long fract A, long

  1101.           long fract B)

  1102.  -- Runtime Function: unsigned short fract __subuqq3 (unsigned short

  1103.           fract A, unsigned short fract B)

  1104.  -- Runtime Function: unsigned fract __subuhq3 (unsigned fract A,

  1105.           unsigned fract B)

  1106.  -- Runtime Function: unsigned long fract __subusq3 (unsigned long fract

  1107.           A, unsigned long fract B)

  1108.  -- Runtime Function: unsigned long long fract __subudq3 (unsigned long

  1109.           long fract A, unsigned long long fract B)

  1110.  -- Runtime Function: short accum __subha3 (short accum A, short accum

  1111.           B)

  1112.  -- Runtime Function: accum __subsa3 (accum A, accum B)

  1113.  -- Runtime Function: long accum __subda3 (long accum A, long accum B)

  1114.  -- Runtime Function: long long accum __subta3 (long long accum A, long

  1115.           long accum B)

  1116.  -- Runtime Function: unsigned short accum __subuha3 (unsigned short

  1117.           accum A, unsigned short accum B)

  1118.  -- Runtime Function: unsigned accum __subusa3 (unsigned accum A,

  1119.           unsigned accum B)

  1120.  -- Runtime Function: unsigned long accum __subuda3 (unsigned long accum

  1121.           A, unsigned long accum B)

  1122.  -- Runtime Function: unsigned long long accum __subuta3 (unsigned long

  1123.           long accum A, unsigned long long accum B)

  1124.      These functions return the difference of A and B; that is, 'A - B'.

  1125. 

  1126.  -- Runtime Function: short fract __sssubqq3 (short fract A, short fract

  1127.           B)

  1128.  -- Runtime Function: fract __sssubhq3 (fract A, fract B)

  1129.  -- Runtime Function: long fract __sssubsq3 (long fract A, long fract B)

  1130.  -- Runtime Function: long long fract __sssubdq3 (long long fract A,

  1131.           long long fract B)

  1132.  -- Runtime Function: short accum __sssubha3 (short accum A, short accum

  1133.           B)

  1134.  -- Runtime Function: accum __sssubsa3 (accum A, accum B)

  1135.  -- Runtime Function: long accum __sssubda3 (long accum A, long accum B)

  1136.  -- Runtime Function: long long accum __sssubta3 (long long accum A,

  1137.           long long accum B)

  1138.      These functions return the difference of A and B with signed

  1139.      saturation; that is, 'A - B'.

  1140. 

  1141.  -- Runtime Function: unsigned short fract __ussubuqq3 (unsigned short

  1142.           fract A, unsigned short fract B)

  1143.  -- Runtime Function: unsigned fract __ussubuhq3 (unsigned fract A,

  1144.           unsigned fract B)

  1145.  -- Runtime Function: unsigned long fract __ussubusq3 (unsigned long

  1146.           fract A, unsigned long fract B)

  1147.  -- Runtime Function: unsigned long long fract __ussubudq3 (unsigned

  1148.           long long fract A, unsigned long long fract B)

  1149.  -- Runtime Function: unsigned short accum __ussubuha3 (unsigned short

  1150.           accum A, unsigned short accum B)

  1151.  -- Runtime Function: unsigned accum __ussubusa3 (unsigned accum A,

  1152.           unsigned accum B)

  1153.  -- Runtime Function: unsigned long accum __ussubuda3 (unsigned long

  1154.           accum A, unsigned long accum B)

  1155.  -- Runtime Function: unsigned long long accum __ussubuta3 (unsigned

  1156.           long long accum A, unsigned long long accum B)

  1157.      These functions return the difference of A and B with unsigned

  1158.      saturation; that is, 'A - B'.

  1159. 

  1160.  -- Runtime Function: short fract __mulqq3 (short fract A, short fract

  1161.           B)

  1162.  -- Runtime Function: fract __mulhq3 (fract A, fract B)

  1163.  -- Runtime Function: long fract __mulsq3 (long fract A, long fract B)

  1164.  -- Runtime Function: long long fract __muldq3 (long long fract A, long

  1165.           long fract B)

  1166.  -- Runtime Function: unsigned short fract __muluqq3 (unsigned short

  1167.           fract A, unsigned short fract B)

  1168.  -- Runtime Function: unsigned fract __muluhq3 (unsigned fract A,

  1169.           unsigned fract B)

  1170.  -- Runtime Function: unsigned long fract __mulusq3 (unsigned long fract

  1171.           A, unsigned long fract B)

  1172.  -- Runtime Function: unsigned long long fract __muludq3 (unsigned long

  1173.           long fract A, unsigned long long fract B)

  1174.  -- Runtime Function: short accum __mulha3 (short accum A, short accum

  1175.           B)

  1176.  -- Runtime Function: accum __mulsa3 (accum A, accum B)

  1177.  -- Runtime Function: long accum __mulda3 (long accum A, long accum B)

  1178.  -- Runtime Function: long long accum __multa3 (long long accum A, long

  1179.           long accum B)

  1180.  -- Runtime Function: unsigned short accum __muluha3 (unsigned short

  1181.           accum A, unsigned short accum B)

  1182.  -- Runtime Function: unsigned accum __mulusa3 (unsigned accum A,

  1183.           unsigned accum B)

  1184.  -- Runtime Function: unsigned long accum __muluda3 (unsigned long accum

  1185.           A, unsigned long accum B)

  1186.  -- Runtime Function: unsigned long long accum __muluta3 (unsigned long

  1187.           long accum A, unsigned long long accum B)

  1188.      These functions return the product of A and B.

  1189. 

  1190.  -- Runtime Function: short fract __ssmulqq3 (short fract A, short fract

  1191.           B)

  1192.  -- Runtime Function: fract __ssmulhq3 (fract A, fract B)

  1193.  -- Runtime Function: long fract __ssmulsq3 (long fract A, long fract B)

  1194.  -- Runtime Function: long long fract __ssmuldq3 (long long fract A,

  1195.           long long fract B)

  1196.  -- Runtime Function: short accum __ssmulha3 (short accum A, short accum

  1197.           B)

  1198.  -- Runtime Function: accum __ssmulsa3 (accum A, accum B)

  1199.  -- Runtime Function: long accum __ssmulda3 (long accum A, long accum B)

  1200.  -- Runtime Function: long long accum __ssmulta3 (long long accum A,

  1201.           long long accum B)

  1202.      These functions return the product of A and B with signed

  1203.      saturation.

  1204. 

  1205.  -- Runtime Function: unsigned short fract __usmuluqq3 (unsigned short

  1206.           fract A, unsigned short fract B)

  1207.  -- Runtime Function: unsigned fract __usmuluhq3 (unsigned fract A,

  1208.           unsigned fract B)

  1209.  -- Runtime Function: unsigned long fract __usmulusq3 (unsigned long

  1210.           fract A, unsigned long fract B)

  1211.  -- Runtime Function: unsigned long long fract __usmuludq3 (unsigned

  1212.           long long fract A, unsigned long long fract B)

  1213.  -- Runtime Function: unsigned short accum __usmuluha3 (unsigned short

  1214.           accum A, unsigned short accum B)

  1215.  -- Runtime Function: unsigned accum __usmulusa3 (unsigned accum A,

  1216.           unsigned accum B)

  1217.  -- Runtime Function: unsigned long accum __usmuluda3 (unsigned long

  1218.           accum A, unsigned long accum B)

  1219.  -- Runtime Function: unsigned long long accum __usmuluta3 (unsigned

  1220.           long long accum A, unsigned long long accum B)

  1221.      These functions return the product of A and B with unsigned

  1222.      saturation.

  1223. 

  1224.  -- Runtime Function: short fract __divqq3 (short fract A, short fract

  1225.           B)

  1226.  -- Runtime Function: fract __divhq3 (fract A, fract B)

  1227.  -- Runtime Function: long fract __divsq3 (long fract A, long fract B)

  1228.  -- Runtime Function: long long fract __divdq3 (long long fract A, long

  1229.           long fract B)

  1230.  -- Runtime Function: short accum __divha3 (short accum A, short accum

  1231.           B)

  1232.  -- Runtime Function: accum __divsa3 (accum A, accum B)

  1233.  -- Runtime Function: long accum __divda3 (long accum A, long accum B)

  1234.  -- Runtime Function: long long accum __divta3 (long long accum A, long

  1235.           long accum B)

  1236.      These functions return the quotient of the signed division of A and

  1237.      B.

  1238. 

  1239.  -- Runtime Function: unsigned short fract __udivuqq3 (unsigned short

  1240.           fract A, unsigned short fract B)

  1241.  -- Runtime Function: unsigned fract __udivuhq3 (unsigned fract A,

  1242.           unsigned fract B)

  1243.  -- Runtime Function: unsigned long fract __udivusq3 (unsigned long

  1244.           fract A, unsigned long fract B)

  1245.  -- Runtime Function: unsigned long long fract __udivudq3 (unsigned long

  1246.           long fract A, unsigned long long fract B)

  1247.  -- Runtime Function: unsigned short accum __udivuha3 (unsigned short

  1248.           accum A, unsigned short accum B)

  1249.  -- Runtime Function: unsigned accum __udivusa3 (unsigned accum A,

  1250.           unsigned accum B)

  1251.  -- Runtime Function: unsigned long accum __udivuda3 (unsigned long

  1252.           accum A, unsigned long accum B)

  1253.  -- Runtime Function: unsigned long long accum __udivuta3 (unsigned long

  1254.           long accum A, unsigned long long accum B)

  1255.      These functions return the quotient of the unsigned division of A

  1256.      and B.

  1257. 

  1258.  -- Runtime Function: short fract __ssdivqq3 (short fract A, short fract

  1259.           B)

  1260.  -- Runtime Function: fract __ssdivhq3 (fract A, fract B)

  1261.  -- Runtime Function: long fract __ssdivsq3 (long fract A, long fract B)

  1262.  -- Runtime Function: long long fract __ssdivdq3 (long long fract A,

  1263.           long long fract B)

  1264.  -- Runtime Function: short accum __ssdivha3 (short accum A, short accum

  1265.           B)

  1266.  -- Runtime Function: accum __ssdivsa3 (accum A, accum B)

  1267.  -- Runtime Function: long accum __ssdivda3 (long accum A, long accum B)

  1268.  -- Runtime Function: long long accum __ssdivta3 (long long accum A,

  1269.           long long accum B)

  1270.      These functions return the quotient of the signed division of A and

  1271.      B with signed saturation.

  1272. 

  1273.  -- Runtime Function: unsigned short fract __usdivuqq3 (unsigned short

  1274.           fract A, unsigned short fract B)

  1275.  -- Runtime Function: unsigned fract __usdivuhq3 (unsigned fract A,

  1276.           unsigned fract B)

  1277.  -- Runtime Function: unsigned long fract __usdivusq3 (unsigned long

  1278.           fract A, unsigned long fract B)

  1279.  -- Runtime Function: unsigned long long fract __usdivudq3 (unsigned

  1280.           long long fract A, unsigned long long fract B)

  1281.  -- Runtime Function: unsigned short accum __usdivuha3 (unsigned short

  1282.           accum A, unsigned short accum B)

  1283.  -- Runtime Function: unsigned accum __usdivusa3 (unsigned accum A,

  1284.           unsigned accum B)

  1285.  -- Runtime Function: unsigned long accum __usdivuda3 (unsigned long

  1286.           accum A, unsigned long accum B)

  1287.  -- Runtime Function: unsigned long long accum __usdivuta3 (unsigned

  1288.           long long accum A, unsigned long long accum B)

  1289.      These functions return the quotient of the unsigned division of A

  1290.      and B with unsigned saturation.

  1291. 

  1292.  -- Runtime Function: short fract __negqq2 (short fract A)

  1293.  -- Runtime Function: fract __neghq2 (fract A)

  1294.  -- Runtime Function: long fract __negsq2 (long fract A)

  1295.  -- Runtime Function: long long fract __negdq2 (long long fract A)

  1296.  -- Runtime Function: unsigned short fract __neguqq2 (unsigned short

  1297.           fract A)

  1298.  -- Runtime Function: unsigned fract __neguhq2 (unsigned fract A)

  1299.  -- Runtime Function: unsigned long fract __negusq2 (unsigned long fract

  1300.           A)

  1301.  -- Runtime Function: unsigned long long fract __negudq2 (unsigned long

  1302.           long fract A)

  1303.  -- Runtime Function: short accum __negha2 (short accum A)

  1304.  -- Runtime Function: accum __negsa2 (accum A)

  1305.  -- Runtime Function: long accum __negda2 (long accum A)

  1306.  -- Runtime Function: long long accum __negta2 (long long accum A)

  1307.  -- Runtime Function: unsigned short accum __neguha2 (unsigned short

  1308.           accum A)

  1309.  -- Runtime Function: unsigned accum __negusa2 (unsigned accum A)

  1310.  -- Runtime Function: unsigned long accum __neguda2 (unsigned long accum

  1311.           A)

  1312.  -- Runtime Function: unsigned long long accum __neguta2 (unsigned long

  1313.           long accum A)

  1314.      These functions return the negation of A.

  1315. 

  1316.  -- Runtime Function: short fract __ssnegqq2 (short fract A)

  1317.  -- Runtime Function: fract __ssneghq2 (fract A)

  1318.  -- Runtime Function: long fract __ssnegsq2 (long fract A)

  1319.  -- Runtime Function: long long fract __ssnegdq2 (long long fract A)

  1320.  -- Runtime Function: short accum __ssnegha2 (short accum A)

  1321.  -- Runtime Function: accum __ssnegsa2 (accum A)

  1322.  -- Runtime Function: long accum __ssnegda2 (long accum A)

  1323.  -- Runtime Function: long long accum __ssnegta2 (long long accum A)

  1324.      These functions return the negation of A with signed saturation.

  1325. 

  1326.  -- Runtime Function: unsigned short fract __usneguqq2 (unsigned short

  1327.           fract A)

  1328.  -- Runtime Function: unsigned fract __usneguhq2 (unsigned fract A)

  1329.  -- Runtime Function: unsigned long fract __usnegusq2 (unsigned long

  1330.           fract A)

  1331.  -- Runtime Function: unsigned long long fract __usnegudq2 (unsigned

  1332.           long long fract A)

  1333.  -- Runtime Function: unsigned short accum __usneguha2 (unsigned short

  1334.           accum A)

  1335.  -- Runtime Function: unsigned accum __usnegusa2 (unsigned accum A)

  1336.  -- Runtime Function: unsigned long accum __usneguda2 (unsigned long

  1337.           accum A)

  1338.  -- Runtime Function: unsigned long long accum __usneguta2 (unsigned

  1339.           long long accum A)

  1340.      These functions return the negation of A with unsigned saturation.

  1341. 

  1342.  -- Runtime Function: short fract __ashlqq3 (short fract A, int B)

  1343.  -- Runtime Function: fract __ashlhq3 (fract A, int B)

  1344.  -- Runtime Function: long fract __ashlsq3 (long fract A, int B)

  1345.  -- Runtime Function: long long fract __ashldq3 (long long fract A, int

  1346.           B)

  1347.  -- Runtime Function: unsigned short fract __ashluqq3 (unsigned short

  1348.           fract A, int B)

  1349.  -- Runtime Function: unsigned fract __ashluhq3 (unsigned fract A, int

  1350.           B)

  1351.  -- Runtime Function: unsigned long fract __ashlusq3 (unsigned long

  1352.           fract A, int B)

  1353.  -- Runtime Function: unsigned long long fract __ashludq3 (unsigned long

  1354.           long fract A, int B)

  1355.  -- Runtime Function: short accum __ashlha3 (short accum A, int B)

  1356.  -- Runtime Function: accum __ashlsa3 (accum A, int B)

  1357.  -- Runtime Function: long accum __ashlda3 (long accum A, int B)

  1358.  -- Runtime Function: long long accum __ashlta3 (long long accum A, int

  1359.           B)

  1360.  -- Runtime Function: unsigned short accum __ashluha3 (unsigned short

  1361.           accum A, int B)

  1362.  -- Runtime Function: unsigned accum __ashlusa3 (unsigned accum A, int

  1363.           B)

  1364.  -- Runtime Function: unsigned long accum __ashluda3 (unsigned long

  1365.           accum A, int B)

  1366.  -- Runtime Function: unsigned long long accum __ashluta3 (unsigned long

  1367.           long accum A, int B)

  1368.      These functions return the result of shifting A left by B bits.

  1369. 

  1370.  -- Runtime Function: short fract __ashrqq3 (short fract A, int B)

  1371.  -- Runtime Function: fract __ashrhq3 (fract A, int B)

  1372.  -- Runtime Function: long fract __ashrsq3 (long fract A, int B)

  1373.  -- Runtime Function: long long fract __ashrdq3 (long long fract A, int

  1374.           B)

  1375.  -- Runtime Function: short accum __ashrha3 (short accum A, int B)

  1376.  -- Runtime Function: accum __ashrsa3 (accum A, int B)

  1377.  -- Runtime Function: long accum __ashrda3 (long accum A, int B)

  1378.  -- Runtime Function: long long accum __ashrta3 (long long accum A, int

  1379.           B)

  1380.      These functions return the result of arithmetically shifting A

  1381.      right by B bits.

  1382. 

  1383.  -- Runtime Function: unsigned short fract __lshruqq3 (unsigned short

  1384.           fract A, int B)

  1385.  -- Runtime Function: unsigned fract __lshruhq3 (unsigned fract A, int

  1386.           B)

  1387.  -- Runtime Function: unsigned long fract __lshrusq3 (unsigned long

  1388.           fract A, int B)

  1389.  -- Runtime Function: unsigned long long fract __lshrudq3 (unsigned long

  1390.           long fract A, int B)

  1391.  -- Runtime Function: unsigned short accum __lshruha3 (unsigned short

  1392.           accum A, int B)

  1393.  -- Runtime Function: unsigned accum __lshrusa3 (unsigned accum A, int

  1394.           B)

  1395.  -- Runtime Function: unsigned long accum __lshruda3 (unsigned long

  1396.           accum A, int B)

  1397.  -- Runtime Function: unsigned long long accum __lshruta3 (unsigned long

  1398.           long accum A, int B)

  1399.      These functions return the result of logically shifting A right by

  1400.      B bits.

  1401. 

  1402.  -- Runtime Function: fract __ssashlhq3 (fract A, int B)

  1403.  -- Runtime Function: long fract __ssashlsq3 (long fract A, int B)

  1404.  -- Runtime Function: long long fract __ssashldq3 (long long fract A,

  1405.           int B)

  1406.  -- Runtime Function: short accum __ssashlha3 (short accum A, int B)

  1407.  -- Runtime Function: accum __ssashlsa3 (accum A, int B)

  1408.  -- Runtime Function: long accum __ssashlda3 (long accum A, int B)

  1409.  -- Runtime Function: long long accum __ssashlta3 (long long accum A,

  1410.           int B)

  1411.      These functions return the result of shifting A left by B bits with

  1412.      signed saturation.

  1413. 

  1414.  -- Runtime Function: unsigned short fract __usashluqq3 (unsigned short

  1415.           fract A, int B)

  1416.  -- Runtime Function: unsigned fract __usashluhq3 (unsigned fract A, int

  1417.           B)

  1418.  -- Runtime Function: unsigned long fract __usashlusq3 (unsigned long

  1419.           fract A, int B)

  1420.  -- Runtime Function: unsigned long long fract __usashludq3 (unsigned

  1421.           long long fract A, int B)

  1422.  -- Runtime Function: unsigned short accum __usashluha3 (unsigned short

  1423.           accum A, int B)

  1424.  -- Runtime Function: unsigned accum __usashlusa3 (unsigned accum A, int

  1425.           B)

  1426.  -- Runtime Function: unsigned long accum __usashluda3 (unsigned long

  1427.           accum A, int B)

  1428.  -- Runtime Function: unsigned long long accum __usashluta3 (unsigned

  1429.           long long accum A, int B)

  1430.      These functions return the result of shifting A left by B bits with

  1431.      unsigned saturation.

  1432. 

  1433. 4.4.2 Comparison functions

  1434. --------------------------

  1435. 

  1436. The following functions implement fixed-point comparisons.  These

  1437. functions implement a low-level compare, upon which the higher level

  1438. comparison operators (such as less than and greater than or equal to)

  1439. can be constructed.  The returned values lie in the range zero to two,

  1440. to allow the high-level operators to be implemented by testing the

  1441. returned result using either signed or unsigned comparison.

  1442. 

  1443.  -- Runtime Function: int __cmpqq2 (short fract A, short fract B)

  1444.  -- Runtime Function: int __cmphq2 (fract A, fract B)

  1445.  -- Runtime Function: int __cmpsq2 (long fract A, long fract B)

  1446.  -- Runtime Function: int __cmpdq2 (long long fract A, long long fract

  1447.           B)

  1448.  -- Runtime Function: int __cmpuqq2 (unsigned short fract A, unsigned

  1449.           short fract B)

  1450.  -- Runtime Function: int __cmpuhq2 (unsigned fract A, unsigned fract B)

  1451.  -- Runtime Function: int __cmpusq2 (unsigned long fract A, unsigned

  1452.           long fract B)

  1453.  -- Runtime Function: int __cmpudq2 (unsigned long long fract A,

  1454.           unsigned long long fract B)

  1455.  -- Runtime Function: int __cmpha2 (short accum A, short accum B)

  1456.  -- Runtime Function: int __cmpsa2 (accum A, accum B)

  1457.  -- Runtime Function: int __cmpda2 (long accum A, long accum B)

  1458.  -- Runtime Function: int __cmpta2 (long long accum A, long long accum

  1459.           B)

  1460.  -- Runtime Function: int __cmpuha2 (unsigned short accum A, unsigned

  1461.           short accum B)

  1462.  -- Runtime Function: int __cmpusa2 (unsigned accum A, unsigned accum B)

  1463.  -- Runtime Function: int __cmpuda2 (unsigned long accum A, unsigned

  1464.           long accum B)

  1465.  -- Runtime Function: int __cmputa2 (unsigned long long accum A,

  1466.           unsigned long long accum B)

  1467.      These functions perform a signed or unsigned comparison of A and B

  1468.      (depending on the selected machine mode).  If A is less than B,

  1469.      they return 0; if A is greater than B, they return 2; and if A and

  1470.      B are equal they return 1.

  1471. 

  1472. 4.4.3 Conversion functions

  1473. --------------------------

  1474. 

  1475.  -- Runtime Function: fract __fractqqhq2 (short fract A)

  1476.  -- Runtime Function: long fract __fractqqsq2 (short fract A)

  1477.  -- Runtime Function: long long fract __fractqqdq2 (short fract A)

  1478.  -- Runtime Function: short accum __fractqqha (short fract A)

  1479.  -- Runtime Function: accum __fractqqsa (short fract A)

  1480.  -- Runtime Function: long accum __fractqqda (short fract A)

  1481.  -- Runtime Function: long long accum __fractqqta (short fract A)

  1482.  -- Runtime Function: unsigned short fract __fractqquqq (short fract A)

  1483.  -- Runtime Function: unsigned fract __fractqquhq (short fract A)

  1484.  -- Runtime Function: unsigned long fract __fractqqusq (short fract A)

  1485.  -- Runtime Function: unsigned long long fract __fractqqudq (short fract

  1486.           A)

  1487.  -- Runtime Function: unsigned short accum __fractqquha (short fract A)

  1488.  -- Runtime Function: unsigned accum __fractqqusa (short fract A)

  1489.  -- Runtime Function: unsigned long accum __fractqquda (short fract A)

  1490.  -- Runtime Function: unsigned long long accum __fractqquta (short fract

  1491.           A)

  1492.  -- Runtime Function: signed char __fractqqqi (short fract A)

  1493.  -- Runtime Function: short __fractqqhi (short fract A)

  1494.  -- Runtime Function: int __fractqqsi (short fract A)

  1495.  -- Runtime Function: long __fractqqdi (short fract A)

  1496.  -- Runtime Function: long long __fractqqti (short fract A)

  1497.  -- Runtime Function: float __fractqqsf (short fract A)

  1498.  -- Runtime Function: double __fractqqdf (short fract A)

  1499.  -- Runtime Function: short fract __fracthqqq2 (fract A)

  1500.  -- Runtime Function: long fract __fracthqsq2 (fract A)

  1501.  -- Runtime Function: long long fract __fracthqdq2 (fract A)

  1502.  -- Runtime Function: short accum __fracthqha (fract A)

  1503.  -- Runtime Function: accum __fracthqsa (fract A)

  1504.  -- Runtime Function: long accum __fracthqda (fract A)

  1505.  -- Runtime Function: long long accum __fracthqta (fract A)

  1506.  -- Runtime Function: unsigned short fract __fracthquqq (fract A)

  1507.  -- Runtime Function: unsigned fract __fracthquhq (fract A)

  1508.  -- Runtime Function: unsigned long fract __fracthqusq (fract A)

  1509.  -- Runtime Function: unsigned long long fract __fracthqudq (fract A)

  1510.  -- Runtime Function: unsigned short accum __fracthquha (fract A)

  1511.  -- Runtime Function: unsigned accum __fracthqusa (fract A)

  1512.  -- Runtime Function: unsigned long accum __fracthquda (fract A)

  1513.  -- Runtime Function: unsigned long long accum __fracthquta (fract A)

  1514.  -- Runtime Function: signed char __fracthqqi (fract A)

  1515.  -- Runtime Function: short __fracthqhi (fract A)

  1516.  -- Runtime Function: int __fracthqsi (fract A)

  1517.  -- Runtime Function: long __fracthqdi (fract A)

  1518.  -- Runtime Function: long long __fracthqti (fract A)

  1519.  -- Runtime Function: float __fracthqsf (fract A)

  1520.  -- Runtime Function: double __fracthqdf (fract A)

  1521.  -- Runtime Function: short fract __fractsqqq2 (long fract A)

  1522.  -- Runtime Function: fract __fractsqhq2 (long fract A)

  1523.  -- Runtime Function: long long fract __fractsqdq2 (long fract A)

  1524.  -- Runtime Function: short accum __fractsqha (long fract A)

  1525.  -- Runtime Function: accum __fractsqsa (long fract A)

  1526.  -- Runtime Function: long accum __fractsqda (long fract A)

  1527.  -- Runtime Function: long long accum __fractsqta (long fract A)

  1528.  -- Runtime Function: unsigned short fract __fractsquqq (long fract A)

  1529.  -- Runtime Function: unsigned fract __fractsquhq (long fract A)

  1530.  -- Runtime Function: unsigned long fract __fractsqusq (long fract A)

  1531.  -- Runtime Function: unsigned long long fract __fractsqudq (long fract

  1532.           A)

  1533.  -- Runtime Function: unsigned short accum __fractsquha (long fract A)

  1534.  -- Runtime Function: unsigned accum __fractsqusa (long fract A)

  1535.  -- Runtime Function: unsigned long accum __fractsquda (long fract A)

  1536.  -- Runtime Function: unsigned long long accum __fractsquta (long fract

  1537.           A)

  1538.  -- Runtime Function: signed char __fractsqqi (long fract A)

  1539.  -- Runtime Function: short __fractsqhi (long fract A)

  1540.  -- Runtime Function: int __fractsqsi (long fract A)

  1541.  -- Runtime Function: long __fractsqdi (long fract A)

  1542.  -- Runtime Function: long long __fractsqti (long fract A)

  1543.  -- Runtime Function: float __fractsqsf (long fract A)

  1544.  -- Runtime Function: double __fractsqdf (long fract A)

  1545.  -- Runtime Function: short fract __fractdqqq2 (long long fract A)

  1546.  -- Runtime Function: fract __fractdqhq2 (long long fract A)

  1547.  -- Runtime Function: long fract __fractdqsq2 (long long fract A)

  1548.  -- Runtime Function: short accum __fractdqha (long long fract A)

  1549.  -- Runtime Function: accum __fractdqsa (long long fract A)

  1550.  -- Runtime Function: long accum __fractdqda (long long fract A)

  1551.  -- Runtime Function: long long accum __fractdqta (long long fract A)

  1552.  -- Runtime Function: unsigned short fract __fractdquqq (long long fract

  1553.           A)

  1554.  -- Runtime Function: unsigned fract __fractdquhq (long long fract A)

  1555.  -- Runtime Function: unsigned long fract __fractdqusq (long long fract

  1556.           A)

  1557.  -- Runtime Function: unsigned long long fract __fractdqudq (long long

  1558.           fract A)

  1559.  -- Runtime Function: unsigned short accum __fractdquha (long long fract

  1560.           A)

  1561.  -- Runtime Function: unsigned accum __fractdqusa (long long fract A)

  1562.  -- Runtime Function: unsigned long accum __fractdquda (long long fract

  1563.           A)

  1564.  -- Runtime Function: unsigned long long accum __fractdquta (long long

  1565.           fract A)

  1566.  -- Runtime Function: signed char __fractdqqi (long long fract A)

  1567.  -- Runtime Function: short __fractdqhi (long long fract A)

  1568.  -- Runtime Function: int __fractdqsi (long long fract A)

  1569.  -- Runtime Function: long __fractdqdi (long long fract A)

  1570.  -- Runtime Function: long long __fractdqti (long long fract A)

  1571.  -- Runtime Function: float __fractdqsf (long long fract A)

  1572.  -- Runtime Function: double __fractdqdf (long long fract A)

  1573.  -- Runtime Function: short fract __fracthaqq (short accum A)

  1574.  -- Runtime Function: fract __fracthahq (short accum A)

  1575.  -- Runtime Function: long fract __fracthasq (short accum A)

  1576.  -- Runtime Function: long long fract __fracthadq (short accum A)

  1577.  -- Runtime Function: accum __fracthasa2 (short accum A)

  1578.  -- Runtime Function: long accum __fracthada2 (short accum A)

  1579.  -- Runtime Function: long long accum __fracthata2 (short accum A)

  1580.  -- Runtime Function: unsigned short fract __fracthauqq (short accum A)

  1581.  -- Runtime Function: unsigned fract __fracthauhq (short accum A)

  1582.  -- Runtime Function: unsigned long fract __fracthausq (short accum A)

  1583.  -- Runtime Function: unsigned long long fract __fracthaudq (short accum

  1584.           A)

  1585.  -- Runtime Function: unsigned short accum __fracthauha (short accum A)

  1586.  -- Runtime Function: unsigned accum __fracthausa (short accum A)

  1587.  -- Runtime Function: unsigned long accum __fracthauda (short accum A)

  1588.  -- Runtime Function: unsigned long long accum __fracthauta (short accum

  1589.           A)

  1590.  -- Runtime Function: signed char __fracthaqi (short accum A)

  1591.  -- Runtime Function: short __fracthahi (short accum A)

  1592.  -- Runtime Function: int __fracthasi (short accum A)

  1593.  -- Runtime Function: long __fracthadi (short accum A)

  1594.  -- Runtime Function: long long __fracthati (short accum A)

  1595.  -- Runtime Function: float __fracthasf (short accum A)

  1596.  -- Runtime Function: double __fracthadf (short accum A)

  1597.  -- Runtime Function: short fract __fractsaqq (accum A)

  1598.  -- Runtime Function: fract __fractsahq (accum A)

  1599.  -- Runtime Function: long fract __fractsasq (accum A)

  1600.  -- Runtime Function: long long fract __fractsadq (accum A)

  1601.  -- Runtime Function: short accum __fractsaha2 (accum A)

  1602.  -- Runtime Function: long accum __fractsada2 (accum A)

  1603.  -- Runtime Function: long long accum __fractsata2 (accum A)

  1604.  -- Runtime Function: unsigned short fract __fractsauqq (accum A)

  1605.  -- Runtime Function: unsigned fract __fractsauhq (accum A)

  1606.  -- Runtime Function: unsigned long fract __fractsausq (accum A)

  1607.  -- Runtime Function: unsigned long long fract __fractsaudq (accum A)

  1608.  -- Runtime Function: unsigned short accum __fractsauha (accum A)

  1609.  -- Runtime Function: unsigned accum __fractsausa (accum A)

  1610.  -- Runtime Function: unsigned long accum __fractsauda (accum A)

  1611.  -- Runtime Function: unsigned long long accum __fractsauta (accum A)

  1612.  -- Runtime Function: signed char __fractsaqi (accum A)

  1613.  -- Runtime Function: short __fractsahi (accum A)

  1614.  -- Runtime Function: int __fractsasi (accum A)

  1615.  -- Runtime Function: long __fractsadi (accum A)

  1616.  -- Runtime Function: long long __fractsati (accum A)

  1617.  -- Runtime Function: float __fractsasf (accum A)

  1618.  -- Runtime Function: double __fractsadf (accum A)

  1619.  -- Runtime Function: short fract __fractdaqq (long accum A)

  1620.  -- Runtime Function: fract __fractdahq (long accum A)

  1621.  -- Runtime Function: long fract __fractdasq (long accum A)

  1622.  -- Runtime Function: long long fract __fractdadq (long accum A)

  1623.  -- Runtime Function: short accum __fractdaha2 (long accum A)

  1624.  -- Runtime Function: accum __fractdasa2 (long accum A)

  1625.  -- Runtime Function: long long accum __fractdata2 (long accum A)

  1626.  -- Runtime Function: unsigned short fract __fractdauqq (long accum A)

  1627.  -- Runtime Function: unsigned fract __fractdauhq (long accum A)

  1628.  -- Runtime Function: unsigned long fract __fractdausq (long accum A)

  1629.  -- Runtime Function: unsigned long long fract __fractdaudq (long accum

  1630.           A)

  1631.  -- Runtime Function: unsigned short accum __fractdauha (long accum A)

  1632.  -- Runtime Function: unsigned accum __fractdausa (long accum A)

  1633.  -- Runtime Function: unsigned long accum __fractdauda (long accum A)

  1634.  -- Runtime Function: unsigned long long accum __fractdauta (long accum

  1635.           A)

  1636.  -- Runtime Function: signed char __fractdaqi (long accum A)

  1637.  -- Runtime Function: short __fractdahi (long accum A)

  1638.  -- Runtime Function: int __fractdasi (long accum A)

  1639.  -- Runtime Function: long __fractdadi (long accum A)

  1640.  -- Runtime Function: long long __fractdati (long accum A)

  1641.  -- Runtime Function: float __fractdasf (long accum A)

  1642.  -- Runtime Function: double __fractdadf (long accum A)

  1643.  -- Runtime Function: short fract __fracttaqq (long long accum A)

  1644.  -- Runtime Function: fract __fracttahq (long long accum A)

  1645.  -- Runtime Function: long fract __fracttasq (long long accum A)

  1646.  -- Runtime Function: long long fract __fracttadq (long long accum A)

  1647.  -- Runtime Function: short accum __fracttaha2 (long long accum A)

  1648.  -- Runtime Function: accum __fracttasa2 (long long accum A)

  1649.  -- Runtime Function: long accum __fracttada2 (long long accum A)

  1650.  -- Runtime Function: unsigned short fract __fracttauqq (long long accum

  1651.           A)

  1652.  -- Runtime Function: unsigned fract __fracttauhq (long long accum A)

  1653.  -- Runtime Function: unsigned long fract __fracttausq (long long accum

  1654.           A)

  1655.  -- Runtime Function: unsigned long long fract __fracttaudq (long long

  1656.           accum A)

  1657.  -- Runtime Function: unsigned short accum __fracttauha (long long accum

  1658.           A)

  1659.  -- Runtime Function: unsigned accum __fracttausa (long long accum A)

  1660.  -- Runtime Function: unsigned long accum __fracttauda (long long accum

  1661.           A)

  1662.  -- Runtime Function: unsigned long long accum __fracttauta (long long

  1663.           accum A)

  1664.  -- Runtime Function: signed char __fracttaqi (long long accum A)

  1665.  -- Runtime Function: short __fracttahi (long long accum A)

  1666.  -- Runtime Function: int __fracttasi (long long accum A)

  1667.  -- Runtime Function: long __fracttadi (long long accum A)

  1668.  -- Runtime Function: long long __fracttati (long long accum A)

  1669.  -- Runtime Function: float __fracttasf (long long accum A)

  1670.  -- Runtime Function: double __fracttadf (long long accum A)

  1671.  -- Runtime Function: short fract __fractuqqqq (unsigned short fract A)

  1672.  -- Runtime Function: fract __fractuqqhq (unsigned short fract A)

  1673.  -- Runtime Function: long fract __fractuqqsq (unsigned short fract A)

  1674.  -- Runtime Function: long long fract __fractuqqdq (unsigned short fract

  1675.           A)

  1676.  -- Runtime Function: short accum __fractuqqha (unsigned short fract A)

  1677.  -- Runtime Function: accum __fractuqqsa (unsigned short fract A)

  1678.  -- Runtime Function: long accum __fractuqqda (unsigned short fract A)

  1679.  -- Runtime Function: long long accum __fractuqqta (unsigned short fract

  1680.           A)

  1681.  -- Runtime Function: unsigned fract __fractuqquhq2 (unsigned short

  1682.           fract A)

  1683.  -- Runtime Function: unsigned long fract __fractuqqusq2 (unsigned short

  1684.           fract A)

  1685.  -- Runtime Function: unsigned long long fract __fractuqqudq2 (unsigned

  1686.           short fract A)

  1687.  -- Runtime Function: unsigned short accum __fractuqquha (unsigned short

  1688.           fract A)

  1689.  -- Runtime Function: unsigned accum __fractuqqusa (unsigned short fract

  1690.           A)

  1691.  -- Runtime Function: unsigned long accum __fractuqquda (unsigned short

  1692.           fract A)

  1693.  -- Runtime Function: unsigned long long accum __fractuqquta (unsigned

  1694.           short fract A)

  1695.  -- Runtime Function: signed char __fractuqqqi (unsigned short fract A)

  1696.  -- Runtime Function: short __fractuqqhi (unsigned short fract A)

  1697.  -- Runtime Function: int __fractuqqsi (unsigned short fract A)

  1698.  -- Runtime Function: long __fractuqqdi (unsigned short fract A)

  1699.  -- Runtime Function: long long __fractuqqti (unsigned short fract A)

  1700.  -- Runtime Function: float __fractuqqsf (unsigned short fract A)

  1701.  -- Runtime Function: double __fractuqqdf (unsigned short fract A)

  1702.  -- Runtime Function: short fract __fractuhqqq (unsigned fract A)

  1703.  -- Runtime Function: fract __fractuhqhq (unsigned fract A)

  1704.  -- Runtime Function: long fract __fractuhqsq (unsigned fract A)

  1705.  -- Runtime Function: long long fract __fractuhqdq (unsigned fract A)

  1706.  -- Runtime Function: short accum __fractuhqha (unsigned fract A)

  1707.  -- Runtime Function: accum __fractuhqsa (unsigned fract A)

  1708.  -- Runtime Function: long accum __fractuhqda (unsigned fract A)

  1709.  -- Runtime Function: long long accum __fractuhqta (unsigned fract A)

  1710.  -- Runtime Function: unsigned short fract __fractuhquqq2 (unsigned

  1711.           fract A)

  1712.  -- Runtime Function: unsigned long fract __fractuhqusq2 (unsigned fract

  1713.           A)

  1714.  -- Runtime Function: unsigned long long fract __fractuhqudq2 (unsigned

  1715.           fract A)

  1716.  -- Runtime Function: unsigned short accum __fractuhquha (unsigned fract

  1717.           A)

  1718.  -- Runtime Function: unsigned accum __fractuhqusa (unsigned fract A)

  1719.  -- Runtime Function: unsigned long accum __fractuhquda (unsigned fract

  1720.           A)

  1721.  -- Runtime Function: unsigned long long accum __fractuhquta (unsigned

  1722.           fract A)

  1723.  -- Runtime Function: signed char __fractuhqqi (unsigned fract A)

  1724.  -- Runtime Function: short __fractuhqhi (unsigned fract A)

  1725.  -- Runtime Function: int __fractuhqsi (unsigned fract A)

  1726.  -- Runtime Function: long __fractuhqdi (unsigned fract A)

  1727.  -- Runtime Function: long long __fractuhqti (unsigned fract A)

  1728.  -- Runtime Function: float __fractuhqsf (unsigned fract A)

  1729.  -- Runtime Function: double __fractuhqdf (unsigned fract A)

  1730.  -- Runtime Function: short fract __fractusqqq (unsigned long fract A)

  1731.  -- Runtime Function: fract __fractusqhq (unsigned long fract A)

  1732.  -- Runtime Function: long fract __fractusqsq (unsigned long fract A)

  1733.  -- Runtime Function: long long fract __fractusqdq (unsigned long fract

  1734.           A)

  1735.  -- Runtime Function: short accum __fractusqha (unsigned long fract A)

  1736.  -- Runtime Function: accum __fractusqsa (unsigned long fract A)

  1737.  -- Runtime Function: long accum __fractusqda (unsigned long fract A)

  1738.  -- Runtime Function: long long accum __fractusqta (unsigned long fract

  1739.           A)

  1740.  -- Runtime Function: unsigned short fract __fractusquqq2 (unsigned long

  1741.           fract A)

  1742.  -- Runtime Function: unsigned fract __fractusquhq2 (unsigned long fract

  1743.           A)

  1744.  -- Runtime Function: unsigned long long fract __fractusqudq2 (unsigned

  1745.           long fract A)

  1746.  -- Runtime Function: unsigned short accum __fractusquha (unsigned long

  1747.           fract A)

  1748.  -- Runtime Function: unsigned accum __fractusqusa (unsigned long fract

  1749.           A)

  1750.  -- Runtime Function: unsigned long accum __fractusquda (unsigned long

  1751.           fract A)

  1752.  -- Runtime Function: unsigned long long accum __fractusquta (unsigned

  1753.           long fract A)

  1754.  -- Runtime Function: signed char __fractusqqi (unsigned long fract A)

  1755.  -- Runtime Function: short __fractusqhi (unsigned long fract A)

  1756.  -- Runtime Function: int __fractusqsi (unsigned long fract A)

  1757.  -- Runtime Function: long __fractusqdi (unsigned long fract A)

  1758.  -- Runtime Function: long long __fractusqti (unsigned long fract A)

  1759.  -- Runtime Function: float __fractusqsf (unsigned long fract A)

  1760.  -- Runtime Function: double __fractusqdf (unsigned long fract A)

  1761.  -- Runtime Function: short fract __fractudqqq (unsigned long long fract

  1762.           A)

  1763.  -- Runtime Function: fract __fractudqhq (unsigned long long fract A)

  1764.  -- Runtime Function: long fract __fractudqsq (unsigned long long fract

  1765.           A)

  1766.  -- Runtime Function: long long fract __fractudqdq (unsigned long long

  1767.           fract A)

  1768.  -- Runtime Function: short accum __fractudqha (unsigned long long fract

  1769.           A)

  1770.  -- Runtime Function: accum __fractudqsa (unsigned long long fract A)

  1771.  -- Runtime Function: long accum __fractudqda (unsigned long long fract

  1772.           A)

  1773.  -- Runtime Function: long long accum __fractudqta (unsigned long long

  1774.           fract A)

  1775.  -- Runtime Function: unsigned short fract __fractudquqq2 (unsigned long

  1776.           long fract A)

  1777.  -- Runtime Function: unsigned fract __fractudquhq2 (unsigned long long

  1778.           fract A)

  1779.  -- Runtime Function: unsigned long fract __fractudqusq2 (unsigned long

  1780.           long fract A)

  1781.  -- Runtime Function: unsigned short accum __fractudquha (unsigned long

  1782.           long fract A)

  1783.  -- Runtime Function: unsigned accum __fractudqusa (unsigned long long

  1784.           fract A)

  1785.  -- Runtime Function: unsigned long accum __fractudquda (unsigned long

  1786.           long fract A)

  1787.  -- Runtime Function: unsigned long long accum __fractudquta (unsigned

  1788.           long long fract A)

  1789.  -- Runtime Function: signed char __fractudqqi (unsigned long long fract

  1790.           A)

  1791.  -- Runtime Function: short __fractudqhi (unsigned long long fract A)

  1792.  -- Runtime Function: int __fractudqsi (unsigned long long fract A)

  1793.  -- Runtime Function: long __fractudqdi (unsigned long long fract A)

  1794.  -- Runtime Function: long long __fractudqti (unsigned long long fract

  1795.           A)

  1796.  -- Runtime Function: float __fractudqsf (unsigned long long fract A)

  1797.  -- Runtime Function: double __fractudqdf (unsigned long long fract A)

  1798.  -- Runtime Function: short fract __fractuhaqq (unsigned short accum A)

  1799.  -- Runtime Function: fract __fractuhahq (unsigned short accum A)

  1800.  -- Runtime Function: long fract __fractuhasq (unsigned short accum A)

  1801.  -- Runtime Function: long long fract __fractuhadq (unsigned short accum

  1802.           A)

  1803.  -- Runtime Function: short accum __fractuhaha (unsigned short accum A)

  1804.  -- Runtime Function: accum __fractuhasa (unsigned short accum A)

  1805.  -- Runtime Function: long accum __fractuhada (unsigned short accum A)

  1806.  -- Runtime Function: long long accum __fractuhata (unsigned short accum

  1807.           A)

  1808.  -- Runtime Function: unsigned short fract __fractuhauqq (unsigned short

  1809.           accum A)

  1810.  -- Runtime Function: unsigned fract __fractuhauhq (unsigned short accum

  1811.           A)

  1812.  -- Runtime Function: unsigned long fract __fractuhausq (unsigned short

  1813.           accum A)

  1814.  -- Runtime Function: unsigned long long fract __fractuhaudq (unsigned

  1815.           short accum A)

  1816.  -- Runtime Function: unsigned accum __fractuhausa2 (unsigned short

  1817.           accum A)

  1818.  -- Runtime Function: unsigned long accum __fractuhauda2 (unsigned short

  1819.           accum A)

  1820.  -- Runtime Function: unsigned long long accum __fractuhauta2 (unsigned

  1821.           short accum A)

  1822.  -- Runtime Function: signed char __fractuhaqi (unsigned short accum A)

  1823.  -- Runtime Function: short __fractuhahi (unsigned short accum A)

  1824.  -- Runtime Function: int __fractuhasi (unsigned short accum A)

  1825.  -- Runtime Function: long __fractuhadi (unsigned short accum A)

  1826.  -- Runtime Function: long long __fractuhati (unsigned short accum A)

  1827.  -- Runtime Function: float __fractuhasf (unsigned short accum A)

  1828.  -- Runtime Function: double __fractuhadf (unsigned short accum A)

  1829.  -- Runtime Function: short fract __fractusaqq (unsigned accum A)

  1830.  -- Runtime Function: fract __fractusahq (unsigned accum A)

  1831.  -- Runtime Function: long fract __fractusasq (unsigned accum A)

  1832.  -- Runtime Function: long long fract __fractusadq (unsigned accum A)

  1833.  -- Runtime Function: short accum __fractusaha (unsigned accum A)

  1834.  -- Runtime Function: accum __fractusasa (unsigned accum A)

  1835.  -- Runtime Function: long accum __fractusada (unsigned accum A)

  1836.  -- Runtime Function: long long accum __fractusata (unsigned accum A)

  1837.  -- Runtime Function: unsigned short fract __fractusauqq (unsigned accum

  1838.           A)

  1839.  -- Runtime Function: unsigned fract __fractusauhq (unsigned accum A)

  1840.  -- Runtime Function: unsigned long fract __fractusausq (unsigned accum

  1841.           A)

  1842.  -- Runtime Function: unsigned long long fract __fractusaudq (unsigned

  1843.           accum A)

  1844.  -- Runtime Function: unsigned short accum __fractusauha2 (unsigned

  1845.           accum A)

  1846.  -- Runtime Function: unsigned long accum __fractusauda2 (unsigned accum

  1847.           A)

  1848.  -- Runtime Function: unsigned long long accum __fractusauta2 (unsigned

  1849.           accum A)

  1850.  -- Runtime Function: signed char __fractusaqi (unsigned accum A)

  1851.  -- Runtime Function: short __fractusahi (unsigned accum A)

  1852.  -- Runtime Function: int __fractusasi (unsigned accum A)

  1853.  -- Runtime Function: long __fractusadi (unsigned accum A)

  1854.  -- Runtime Function: long long __fractusati (unsigned accum A)

  1855.  -- Runtime Function: float __fractusasf (unsigned accum A)

  1856.  -- Runtime Function: double __fractusadf (unsigned accum A)

  1857.  -- Runtime Function: short fract __fractudaqq (unsigned long accum A)

  1858.  -- Runtime Function: fract __fractudahq (unsigned long accum A)

  1859.  -- Runtime Function: long fract __fractudasq (unsigned long accum A)

  1860.  -- Runtime Function: long long fract __fractudadq (unsigned long accum

  1861.           A)

  1862.  -- Runtime Function: short accum __fractudaha (unsigned long accum A)

  1863.  -- Runtime Function: accum __fractudasa (unsigned long accum A)

  1864.  -- Runtime Function: long accum __fractudada (unsigned long accum A)

  1865.  -- Runtime Function: long long accum __fractudata (unsigned long accum

  1866.           A)

  1867.  -- Runtime Function: unsigned short fract __fractudauqq (unsigned long

  1868.           accum A)

  1869.  -- Runtime Function: unsigned fract __fractudauhq (unsigned long accum

  1870.           A)

  1871.  -- Runtime Function: unsigned long fract __fractudausq (unsigned long

  1872.           accum A)

  1873.  -- Runtime Function: unsigned long long fract __fractudaudq (unsigned

  1874.           long accum A)

  1875.  -- Runtime Function: unsigned short accum __fractudauha2 (unsigned long

  1876.           accum A)

  1877.  -- Runtime Function: unsigned accum __fractudausa2 (unsigned long accum

  1878.           A)

  1879.  -- Runtime Function: unsigned long long accum __fractudauta2 (unsigned

  1880.           long accum A)

  1881.  -- Runtime Function: signed char __fractudaqi (unsigned long accum A)

  1882.  -- Runtime Function: short __fractudahi (unsigned long accum A)

  1883.  -- Runtime Function: int __fractudasi (unsigned long accum A)

  1884.  -- Runtime Function: long __fractudadi (unsigned long accum A)

  1885.  -- Runtime Function: long long __fractudati (unsigned long accum A)

  1886.  -- Runtime Function: float __fractudasf (unsigned long accum A)

  1887.  -- Runtime Function: double __fractudadf (unsigned long accum A)

  1888.  -- Runtime Function: short fract __fractutaqq (unsigned long long accum

  1889.           A)

  1890.  -- Runtime Function: fract __fractutahq (unsigned long long accum A)

  1891.  -- Runtime Function: long fract __fractutasq (unsigned long long accum

  1892.           A)

  1893.  -- Runtime Function: long long fract __fractutadq (unsigned long long

  1894.           accum A)

  1895.  -- Runtime Function: short accum __fractutaha (unsigned long long accum

  1896.           A)

  1897.  -- Runtime Function: accum __fractutasa (unsigned long long accum A)

  1898.  -- Runtime Function: long accum __fractutada (unsigned long long accum

  1899.           A)

  1900.  -- Runtime Function: long long accum __fractutata (unsigned long long

  1901.           accum A)

  1902.  -- Runtime Function: unsigned short fract __fractutauqq (unsigned long

  1903.           long accum A)

  1904.  -- Runtime Function: unsigned fract __fractutauhq (unsigned long long

  1905.           accum A)

  1906.  -- Runtime Function: unsigned long fract __fractutausq (unsigned long

  1907.           long accum A)

  1908.  -- Runtime Function: unsigned long long fract __fractutaudq (unsigned

  1909.           long long accum A)

  1910.  -- Runtime Function: unsigned short accum __fractutauha2 (unsigned long

  1911.           long accum A)

  1912.  -- Runtime Function: unsigned accum __fractutausa2 (unsigned long long

  1913.           accum A)

  1914.  -- Runtime Function: unsigned long accum __fractutauda2 (unsigned long

  1915.           long accum A)

  1916.  -- Runtime Function: signed char __fractutaqi (unsigned long long accum

  1917.           A)

  1918.  -- Runtime Function: short __fractutahi (unsigned long long accum A)

  1919.  -- Runtime Function: int __fractutasi (unsigned long long accum A)

  1920.  -- Runtime Function: long __fractutadi (unsigned long long accum A)

  1921.  -- Runtime Function: long long __fractutati (unsigned long long accum

  1922.           A)

  1923.  -- Runtime Function: float __fractutasf (unsigned long long accum A)

  1924.  -- Runtime Function: double __fractutadf (unsigned long long accum A)

  1925.  -- Runtime Function: short fract __fractqiqq (signed char A)

  1926.  -- Runtime Function: fract __fractqihq (signed char A)

  1927.  -- Runtime Function: long fract __fractqisq (signed char A)

  1928.  -- Runtime Function: long long fract __fractqidq (signed char A)

  1929.  -- Runtime Function: short accum __fractqiha (signed char A)

  1930.  -- Runtime Function: accum __fractqisa (signed char A)

  1931.  -- Runtime Function: long accum __fractqida (signed char A)

  1932.  -- Runtime Function: long long accum __fractqita (signed char A)

  1933.  -- Runtime Function: unsigned short fract __fractqiuqq (signed char A)

  1934.  -- Runtime Function: unsigned fract __fractqiuhq (signed char A)

  1935.  -- Runtime Function: unsigned long fract __fractqiusq (signed char A)

  1936.  -- Runtime Function: unsigned long long fract __fractqiudq (signed char

  1937.           A)

  1938.  -- Runtime Function: unsigned short accum __fractqiuha (signed char A)

  1939.  -- Runtime Function: unsigned accum __fractqiusa (signed char A)

  1940.  -- Runtime Function: unsigned long accum __fractqiuda (signed char A)

  1941.  -- Runtime Function: unsigned long long accum __fractqiuta (signed char

  1942.           A)

  1943.  -- Runtime Function: short fract __fracthiqq (short A)

  1944.  -- Runtime Function: fract __fracthihq (short A)

  1945.  -- Runtime Function: long fract __fracthisq (short A)

  1946.  -- Runtime Function: long long fract __fracthidq (short A)

  1947.  -- Runtime Function: short accum __fracthiha (short A)

  1948.  -- Runtime Function: accum __fracthisa (short A)

  1949.  -- Runtime Function: long accum __fracthida (short A)

  1950.  -- Runtime Function: long long accum __fracthita (short A)

  1951.  -- Runtime Function: unsigned short fract __fracthiuqq (short A)

  1952.  -- Runtime Function: unsigned fract __fracthiuhq (short A)

  1953.  -- Runtime Function: unsigned long fract __fracthiusq (short A)

  1954.  -- Runtime Function: unsigned long long fract __fracthiudq (short A)

  1955.  -- Runtime Function: unsigned short accum __fracthiuha (short A)

  1956.  -- Runtime Function: unsigned accum __fracthiusa (short A)

  1957.  -- Runtime Function: unsigned long accum __fracthiuda (short A)

  1958.  -- Runtime Function: unsigned long long accum __fracthiuta (short A)

  1959.  -- Runtime Function: short fract __fractsiqq (int A)

  1960.  -- Runtime Function: fract __fractsihq (int A)

  1961.  -- Runtime Function: long fract __fractsisq (int A)

  1962.  -- Runtime Function: long long fract __fractsidq (int A)

  1963.  -- Runtime Function: short accum __fractsiha (int A)

  1964.  -- Runtime Function: accum __fractsisa (int A)

  1965.  -- Runtime Function: long accum __fractsida (int A)

  1966.  -- Runtime Function: long long accum __fractsita (int A)

  1967.  -- Runtime Function: unsigned short fract __fractsiuqq (int A)

  1968.  -- Runtime Function: unsigned fract __fractsiuhq (int A)

  1969.  -- Runtime Function: unsigned long fract __fractsiusq (int A)

  1970.  -- Runtime Function: unsigned long long fract __fractsiudq (int A)

  1971.  -- Runtime Function: unsigned short accum __fractsiuha (int A)

  1972.  -- Runtime Function: unsigned accum __fractsiusa (int A)

  1973.  -- Runtime Function: unsigned long accum __fractsiuda (int A)

  1974.  -- Runtime Function: unsigned long long accum __fractsiuta (int A)

  1975.  -- Runtime Function: short fract __fractdiqq (long A)

  1976.  -- Runtime Function: fract __fractdihq (long A)

  1977.  -- Runtime Function: long fract __fractdisq (long A)

  1978.  -- Runtime Function: long long fract __fractdidq (long A)

  1979.  -- Runtime Function: short accum __fractdiha (long A)

  1980.  -- Runtime Function: accum __fractdisa (long A)

  1981.  -- Runtime Function: long accum __fractdida (long A)

  1982.  -- Runtime Function: long long accum __fractdita (long A)

  1983.  -- Runtime Function: unsigned short fract __fractdiuqq (long A)

  1984.  -- Runtime Function: unsigned fract __fractdiuhq (long A)

  1985.  -- Runtime Function: unsigned long fract __fractdiusq (long A)

  1986.  -- Runtime Function: unsigned long long fract __fractdiudq (long A)

  1987.  -- Runtime Function: unsigned short accum __fractdiuha (long A)

  1988.  -- Runtime Function: unsigned accum __fractdiusa (long A)

  1989.  -- Runtime Function: unsigned long accum __fractdiuda (long A)

  1990.  -- Runtime Function: unsigned long long accum __fractdiuta (long A)

  1991.  -- Runtime Function: short fract __fracttiqq (long long A)

  1992.  -- Runtime Function: fract __fracttihq (long long A)

  1993.  -- Runtime Function: long fract __fracttisq (long long A)

  1994.  -- Runtime Function: long long fract __fracttidq (long long A)

  1995.  -- Runtime Function: short accum __fracttiha (long long A)

  1996.  -- Runtime Function: accum __fracttisa (long long A)

  1997.  -- Runtime Function: long accum __fracttida (long long A)

  1998.  -- Runtime Function: long long accum __fracttita (long long A)

  1999.  -- Runtime Function: unsigned short fract __fracttiuqq (long long A)

  2000.  -- Runtime Function: unsigned fract __fracttiuhq (long long A)

  2001.  -- Runtime Function: unsigned long fract __fracttiusq (long long A)

  2002.  -- Runtime Function: unsigned long long fract __fracttiudq (long long

  2003.           A)

  2004.  -- Runtime Function: unsigned short accum __fracttiuha (long long A)

  2005.  -- Runtime Function: unsigned accum __fracttiusa (long long A)

  2006.  -- Runtime Function: unsigned long accum __fracttiuda (long long A)

  2007.  -- Runtime Function: unsigned long long accum __fracttiuta (long long

  2008.           A)

  2009.  -- Runtime Function: short fract __fractsfqq (float A)

  2010.  -- Runtime Function: fract __fractsfhq (float A)

  2011.  -- Runtime Function: long fract __fractsfsq (float A)

  2012.  -- Runtime Function: long long fract __fractsfdq (float A)

  2013.  -- Runtime Function: short accum __fractsfha (float A)

  2014.  -- Runtime Function: accum __fractsfsa (float A)

  2015.  -- Runtime Function: long accum __fractsfda (float A)

  2016.  -- Runtime Function: long long accum __fractsfta (float A)

  2017.  -- Runtime Function: unsigned short fract __fractsfuqq (float A)

  2018.  -- Runtime Function: unsigned fract __fractsfuhq (float A)

  2019.  -- Runtime Function: unsigned long fract __fractsfusq (float A)

  2020.  -- Runtime Function: unsigned long long fract __fractsfudq (float A)

  2021.  -- Runtime Function: unsigned short accum __fractsfuha (float A)

  2022.  -- Runtime Function: unsigned accum __fractsfusa (float A)

  2023.  -- Runtime Function: unsigned long accum __fractsfuda (float A)

  2024.  -- Runtime Function: unsigned long long accum __fractsfuta (float A)

  2025.  -- Runtime Function: short fract __fractdfqq (double A)

  2026.  -- Runtime Function: fract __fractdfhq (double A)

  2027.  -- Runtime Function: long fract __fractdfsq (double A)

  2028.  -- Runtime Function: long long fract __fractdfdq (double A)

  2029.  -- Runtime Function: short accum __fractdfha (double A)

  2030.  -- Runtime Function: accum __fractdfsa (double A)

  2031.  -- Runtime Function: long accum __fractdfda (double A)

  2032.  -- Runtime Function: long long accum __fractdfta (double A)

  2033.  -- Runtime Function: unsigned short fract __fractdfuqq (double A)

  2034.  -- Runtime Function: unsigned fract __fractdfuhq (double A)

  2035.  -- Runtime Function: unsigned long fract __fractdfusq (double A)

  2036.  -- Runtime Function: unsigned long long fract __fractdfudq (double A)

  2037.  -- Runtime Function: unsigned short accum __fractdfuha (double A)

  2038.  -- Runtime Function: unsigned accum __fractdfusa (double A)

  2039.  -- Runtime Function: unsigned long accum __fractdfuda (double A)

  2040.  -- Runtime Function: unsigned long long accum __fractdfuta (double A)

  2041.      These functions convert from fractional and signed non-fractionals

  2042.      to fractionals and signed non-fractionals, without saturation.

  2043. 

  2044.  -- Runtime Function: fract __satfractqqhq2 (short fract A)

  2045.  -- Runtime Function: long fract __satfractqqsq2 (short fract A)

  2046.  -- Runtime Function: long long fract __satfractqqdq2 (short fract A)

  2047.  -- Runtime Function: short accum __satfractqqha (short fract A)

  2048.  -- Runtime Function: accum __satfractqqsa (short fract A)

  2049.  -- Runtime Function: long accum __satfractqqda (short fract A)

  2050.  -- Runtime Function: long long accum __satfractqqta (short fract A)

  2051.  -- Runtime Function: unsigned short fract __satfractqquqq (short fract

  2052.           A)

  2053.  -- Runtime Function: unsigned fract __satfractqquhq (short fract A)

  2054.  -- Runtime Function: unsigned long fract __satfractqqusq (short fract

  2055.           A)

  2056.  -- Runtime Function: unsigned long long fract __satfractqqudq (short

  2057.           fract A)

  2058.  -- Runtime Function: unsigned short accum __satfractqquha (short fract

  2059.           A)

  2060.  -- Runtime Function: unsigned accum __satfractqqusa (short fract A)

  2061.  -- Runtime Function: unsigned long accum __satfractqquda (short fract

  2062.           A)

  2063.  -- Runtime Function: unsigned long long accum __satfractqquta (short

  2064.           fract A)

  2065.  -- Runtime Function: short fract __satfracthqqq2 (fract A)

  2066.  -- Runtime Function: long fract __satfracthqsq2 (fract A)

  2067.  -- Runtime Function: long long fract __satfracthqdq2 (fract A)

  2068.  -- Runtime Function: short accum __satfracthqha (fract A)

  2069.  -- Runtime Function: accum __satfracthqsa (fract A)

  2070.  -- Runtime Function: long accum __satfracthqda (fract A)

  2071.  -- Runtime Function: long long accum __satfracthqta (fract A)

  2072.  -- Runtime Function: unsigned short fract __satfracthquqq (fract A)

  2073.  -- Runtime Function: unsigned fract __satfracthquhq (fract A)

  2074.  -- Runtime Function: unsigned long fract __satfracthqusq (fract A)

  2075.  -- Runtime Function: unsigned long long fract __satfracthqudq (fract A)

  2076.  -- Runtime Function: unsigned short accum __satfracthquha (fract A)

  2077.  -- Runtime Function: unsigned accum __satfracthqusa (fract A)

  2078.  -- Runtime Function: unsigned long accum __satfracthquda (fract A)

  2079.  -- Runtime Function: unsigned long long accum __satfracthquta (fract A)

  2080.  -- Runtime Function: short fract __satfractsqqq2 (long fract A)

  2081.  -- Runtime Function: fract __satfractsqhq2 (long fract A)

  2082.  -- Runtime Function: long long fract __satfractsqdq2 (long fract A)

  2083.  -- Runtime Function: short accum __satfractsqha (long fract A)

  2084.  -- Runtime Function: accum __satfractsqsa (long fract A)

  2085.  -- Runtime Function: long accum __satfractsqda (long fract A)

  2086.  -- Runtime Function: long long accum __satfractsqta (long fract A)

  2087.  -- Runtime Function: unsigned short fract __satfractsquqq (long fract

  2088.           A)

  2089.  -- Runtime Function: unsigned fract __satfractsquhq (long fract A)

  2090.  -- Runtime Function: unsigned long fract __satfractsqusq (long fract A)

  2091.  -- Runtime Function: unsigned long long fract __satfractsqudq (long

  2092.           fract A)

  2093.  -- Runtime Function: unsigned short accum __satfractsquha (long fract

  2094.           A)

  2095.  -- Runtime Function: unsigned accum __satfractsqusa (long fract A)

  2096.  -- Runtime Function: unsigned long accum __satfractsquda (long fract A)

  2097.  -- Runtime Function: unsigned long long accum __satfractsquta (long

  2098.           fract A)

  2099.  -- Runtime Function: short fract __satfractdqqq2 (long long fract A)

  2100.  -- Runtime Function: fract __satfractdqhq2 (long long fract A)

  2101.  -- Runtime Function: long fract __satfractdqsq2 (long long fract A)

  2102.  -- Runtime Function: short accum __satfractdqha (long long fract A)

  2103.  -- Runtime Function: accum __satfractdqsa (long long fract A)

  2104.  -- Runtime Function: long accum __satfractdqda (long long fract A)

  2105.  -- Runtime Function: long long accum __satfractdqta (long long fract A)

  2106.  -- Runtime Function: unsigned short fract __satfractdquqq (long long

  2107.           fract A)

  2108.  -- Runtime Function: unsigned fract __satfractdquhq (long long fract A)

  2109.  -- Runtime Function: unsigned long fract __satfractdqusq (long long

  2110.           fract A)

  2111.  -- Runtime Function: unsigned long long fract __satfractdqudq (long

  2112.           long fract A)

  2113.  -- Runtime Function: unsigned short accum __satfractdquha (long long

  2114.           fract A)

  2115.  -- Runtime Function: unsigned accum __satfractdqusa (long long fract A)

  2116.  -- Runtime Function: unsigned long accum __satfractdquda (long long

  2117.           fract A)

  2118.  -- Runtime Function: unsigned long long accum __satfractdquta (long

  2119.           long fract A)

  2120.  -- Runtime Function: short fract __satfracthaqq (short accum A)

  2121.  -- Runtime Function: fract __satfracthahq (short accum A)

  2122.  -- Runtime Function: long fract __satfracthasq (short accum A)

  2123.  -- Runtime Function: long long fract __satfracthadq (short accum A)

  2124.  -- Runtime Function: accum __satfracthasa2 (short accum A)

  2125.  -- Runtime Function: long accum __satfracthada2 (short accum A)

  2126.  -- Runtime Function: long long accum __satfracthata2 (short accum A)

  2127.  -- Runtime Function: unsigned short fract __satfracthauqq (short accum

  2128.           A)

  2129.  -- Runtime Function: unsigned fract __satfracthauhq (short accum A)

  2130.  -- Runtime Function: unsigned long fract __satfracthausq (short accum

  2131.           A)

  2132.  -- Runtime Function: unsigned long long fract __satfracthaudq (short

  2133.           accum A)

  2134.  -- Runtime Function: unsigned short accum __satfracthauha (short accum

  2135.           A)

  2136.  -- Runtime Function: unsigned accum __satfracthausa (short accum A)

  2137.  -- Runtime Function: unsigned long accum __satfracthauda (short accum

  2138.           A)

  2139.  -- Runtime Function: unsigned long long accum __satfracthauta (short

  2140.           accum A)

  2141.  -- Runtime Function: short fract __satfractsaqq (accum A)

  2142.  -- Runtime Function: fract __satfractsahq (accum A)

  2143.  -- Runtime Function: long fract __satfractsasq (accum A)

  2144.  -- Runtime Function: long long fract __satfractsadq (accum A)

  2145.  -- Runtime Function: short accum __satfractsaha2 (accum A)

  2146.  -- Runtime Function: long accum __satfractsada2 (accum A)

  2147.  -- Runtime Function: long long accum __satfractsata2 (accum A)

  2148.  -- Runtime Function: unsigned short fract __satfractsauqq (accum A)

  2149.  -- Runtime Function: unsigned fract __satfractsauhq (accum A)

  2150.  -- Runtime Function: unsigned long fract __satfractsausq (accum A)

  2151.  -- Runtime Function: unsigned long long fract __satfractsaudq (accum A)

  2152.  -- Runtime Function: unsigned short accum __satfractsauha (accum A)

  2153.  -- Runtime Function: unsigned accum __satfractsausa (accum A)

  2154.  -- Runtime Function: unsigned long accum __satfractsauda (accum A)

  2155.  -- Runtime Function: unsigned long long accum __satfractsauta (accum A)

  2156.  -- Runtime Function: short fract __satfractdaqq (long accum A)

  2157.  -- Runtime Function: fract __satfractdahq (long accum A)

  2158.  -- Runtime Function: long fract __satfractdasq (long accum A)

  2159.  -- Runtime Function: long long fract __satfractdadq (long accum A)

  2160.  -- Runtime Function: short accum __satfractdaha2 (long accum A)

  2161.  -- Runtime Function: accum __satfractdasa2 (long accum A)

  2162.  -- Runtime Function: long long accum __satfractdata2 (long accum A)

  2163.  -- Runtime Function: unsigned short fract __satfractdauqq (long accum

  2164.           A)

  2165.  -- Runtime Function: unsigned fract __satfractdauhq (long accum A)

  2166.  -- Runtime Function: unsigned long fract __satfractdausq (long accum A)

  2167.  -- Runtime Function: unsigned long long fract __satfractdaudq (long

  2168.           accum A)

  2169.  -- Runtime Function: unsigned short accum __satfractdauha (long accum

  2170.           A)

  2171.  -- Runtime Function: unsigned accum __satfractdausa (long accum A)

  2172.  -- Runtime Function: unsigned long accum __satfractdauda (long accum A)

  2173.  -- Runtime Function: unsigned long long accum __satfractdauta (long

  2174.           accum A)

  2175.  -- Runtime Function: short fract __satfracttaqq (long long accum A)

  2176.  -- Runtime Function: fract __satfracttahq (long long accum A)

  2177.  -- Runtime Function: long fract __satfracttasq (long long accum A)

  2178.  -- Runtime Function: long long fract __satfracttadq (long long accum A)

  2179.  -- Runtime Function: short accum __satfracttaha2 (long long accum A)

  2180.  -- Runtime Function: accum __satfracttasa2 (long long accum A)

  2181.  -- Runtime Function: long accum __satfracttada2 (long long accum A)

  2182.  -- Runtime Function: unsigned short fract __satfracttauqq (long long

  2183.           accum A)

  2184.  -- Runtime Function: unsigned fract __satfracttauhq (long long accum A)

  2185.  -- Runtime Function: unsigned long fract __satfracttausq (long long

  2186.           accum A)

  2187.  -- Runtime Function: unsigned long long fract __satfracttaudq (long

  2188.           long accum A)

  2189.  -- Runtime Function: unsigned short accum __satfracttauha (long long

  2190.           accum A)

  2191.  -- Runtime Function: unsigned accum __satfracttausa (long long accum A)

  2192.  -- Runtime Function: unsigned long accum __satfracttauda (long long

  2193.           accum A)

  2194.  -- Runtime Function: unsigned long long accum __satfracttauta (long

  2195.           long accum A)

  2196.  -- Runtime Function: short fract __satfractuqqqq (unsigned short fract

  2197.           A)

  2198.  -- Runtime Function: fract __satfractuqqhq (unsigned short fract A)

  2199.  -- Runtime Function: long fract __satfractuqqsq (unsigned short fract

  2200.           A)

  2201.  -- Runtime Function: long long fract __satfractuqqdq (unsigned short

  2202.           fract A)

  2203.  -- Runtime Function: short accum __satfractuqqha (unsigned short fract

  2204.           A)

  2205.  -- Runtime Function: accum __satfractuqqsa (unsigned short fract A)

  2206.  -- Runtime Function: long accum __satfractuqqda (unsigned short fract

  2207.           A)

  2208.  -- Runtime Function: long long accum __satfractuqqta (unsigned short

  2209.           fract A)

  2210.  -- Runtime Function: unsigned fract __satfractuqquhq2 (unsigned short

  2211.           fract A)

  2212.  -- Runtime Function: unsigned long fract __satfractuqqusq2 (unsigned

  2213.           short fract A)

  2214.  -- Runtime Function: unsigned long long fract __satfractuqqudq2

  2215.           (unsigned short fract A)

  2216.  -- Runtime Function: unsigned short accum __satfractuqquha (unsigned

  2217.           short fract A)

  2218.  -- Runtime Function: unsigned accum __satfractuqqusa (unsigned short

  2219.           fract A)

  2220.  -- Runtime Function: unsigned long accum __satfractuqquda (unsigned

  2221.           short fract A)

  2222.  -- Runtime Function: unsigned long long accum __satfractuqquta

  2223.           (unsigned short fract A)

  2224.  -- Runtime Function: short fract __satfractuhqqq (unsigned fract A)

  2225.  -- Runtime Function: fract __satfractuhqhq (unsigned fract A)

  2226.  -- Runtime Function: long fract __satfractuhqsq (unsigned fract A)

  2227.  -- Runtime Function: long long fract __satfractuhqdq (unsigned fract A)

  2228.  -- Runtime Function: short accum __satfractuhqha (unsigned fract A)

  2229.  -- Runtime Function: accum __satfractuhqsa (unsigned fract A)

  2230.  -- Runtime Function: long accum __satfractuhqda (unsigned fract A)

  2231.  -- Runtime Function: long long accum __satfractuhqta (unsigned fract A)

  2232.  -- Runtime Function: unsigned short fract __satfractuhquqq2 (unsigned

  2233.           fract A)

  2234.  -- Runtime Function: unsigned long fract __satfractuhqusq2 (unsigned

  2235.           fract A)

  2236.  -- Runtime Function: unsigned long long fract __satfractuhqudq2

  2237.           (unsigned fract A)

  2238.  -- Runtime Function: unsigned short accum __satfractuhquha (unsigned

  2239.           fract A)

  2240.  -- Runtime Function: unsigned accum __satfractuhqusa (unsigned fract A)

  2241.  -- Runtime Function: unsigned long accum __satfractuhquda (unsigned

  2242.           fract A)

  2243.  -- Runtime Function: unsigned long long accum __satfractuhquta

  2244.           (unsigned fract A)

  2245.  -- Runtime Function: short fract __satfractusqqq (unsigned long fract

  2246.           A)

  2247.  -- Runtime Function: fract __satfractusqhq (unsigned long fract A)

  2248.  -- Runtime Function: long fract __satfractusqsq (unsigned long fract A)

  2249.  -- Runtime Function: long long fract __satfractusqdq (unsigned long

  2250.           fract A)

  2251.  -- Runtime Function: short accum __satfractusqha (unsigned long fract

  2252.           A)

  2253.  -- Runtime Function: accum __satfractusqsa (unsigned long fract A)

  2254.  -- Runtime Function: long accum __satfractusqda (unsigned long fract A)

  2255.  -- Runtime Function: long long accum __satfractusqta (unsigned long

  2256.           fract A)

  2257.  -- Runtime Function: unsigned short fract __satfractusquqq2 (unsigned

  2258.           long fract A)

  2259.  -- Runtime Function: unsigned fract __satfractusquhq2 (unsigned long

  2260.           fract A)

  2261.  -- Runtime Function: unsigned long long fract __satfractusqudq2

  2262.           (unsigned long fract A)

  2263.  -- Runtime Function: unsigned short accum __satfractusquha (unsigned

  2264.           long fract A)

  2265.  -- Runtime Function: unsigned accum __satfractusqusa (unsigned long

  2266.           fract A)

  2267.  -- Runtime Function: unsigned long accum __satfractusquda (unsigned

  2268.           long fract A)

  2269.  -- Runtime Function: unsigned long long accum __satfractusquta

  2270.           (unsigned long fract A)

  2271.  -- Runtime Function: short fract __satfractudqqq (unsigned long long

  2272.           fract A)

  2273.  -- Runtime Function: fract __satfractudqhq (unsigned long long fract A)

  2274.  -- Runtime Function: long fract __satfractudqsq (unsigned long long

  2275.           fract A)

  2276.  -- Runtime Function: long long fract __satfractudqdq (unsigned long

  2277.           long fract A)

  2278.  -- Runtime Function: short accum __satfractudqha (unsigned long long

  2279.           fract A)

  2280.  -- Runtime Function: accum __satfractudqsa (unsigned long long fract A)

  2281.  -- Runtime Function: long accum __satfractudqda (unsigned long long

  2282.           fract A)

  2283.  -- Runtime Function: long long accum __satfractudqta (unsigned long

  2284.           long fract A)

  2285.  -- Runtime Function: unsigned short fract __satfractudquqq2 (unsigned

  2286.           long long fract A)

  2287.  -- Runtime Function: unsigned fract __satfractudquhq2 (unsigned long

  2288.           long fract A)

  2289.  -- Runtime Function: unsigned long fract __satfractudqusq2 (unsigned

  2290.           long long fract A)

  2291.  -- Runtime Function: unsigned short accum __satfractudquha (unsigned

  2292.           long long fract A)

  2293.  -- Runtime Function: unsigned accum __satfractudqusa (unsigned long

  2294.           long fract A)

  2295.  -- Runtime Function: unsigned long accum __satfractudquda (unsigned

  2296.           long long fract A)

  2297.  -- Runtime Function: unsigned long long accum __satfractudquta

  2298.           (unsigned long long fract A)

  2299.  -- Runtime Function: short fract __satfractuhaqq (unsigned short accum

  2300.           A)

  2301.  -- Runtime Function: fract __satfractuhahq (unsigned short accum A)

  2302.  -- Runtime Function: long fract __satfractuhasq (unsigned short accum

  2303.           A)

  2304.  -- Runtime Function: long long fract __satfractuhadq (unsigned short

  2305.           accum A)

  2306.  -- Runtime Function: short accum __satfractuhaha (unsigned short accum

  2307.           A)

  2308.  -- Runtime Function: accum __satfractuhasa (unsigned short accum A)

  2309.  -- Runtime Function: long accum __satfractuhada (unsigned short accum

  2310.           A)

  2311.  -- Runtime Function: long long accum __satfractuhata (unsigned short

  2312.           accum A)

  2313.  -- Runtime Function: unsigned short fract __satfractuhauqq (unsigned

  2314.           short accum A)

  2315.  -- Runtime Function: unsigned fract __satfractuhauhq (unsigned short

  2316.           accum A)

  2317.  -- Runtime Function: unsigned long fract __satfractuhausq (unsigned

  2318.           short accum A)

  2319.  -- Runtime Function: unsigned long long fract __satfractuhaudq

  2320.           (unsigned short accum A)

  2321.  -- Runtime Function: unsigned accum __satfractuhausa2 (unsigned short

  2322.           accum A)

  2323.  -- Runtime Function: unsigned long accum __satfractuhauda2 (unsigned

  2324.           short accum A)

  2325.  -- Runtime Function: unsigned long long accum __satfractuhauta2

  2326.           (unsigned short accum A)

  2327.  -- Runtime Function: short fract __satfractusaqq (unsigned accum A)

  2328.  -- Runtime Function: fract __satfractusahq (unsigned accum A)

  2329.  -- Runtime Function: long fract __satfractusasq (unsigned accum A)

  2330.  -- Runtime Function: long long fract __satfractusadq (unsigned accum A)

  2331.  -- Runtime Function: short accum __satfractusaha (unsigned accum A)

  2332.  -- Runtime Function: accum __satfractusasa (unsigned accum A)

  2333.  -- Runtime Function: long accum __satfractusada (unsigned accum A)

  2334.  -- Runtime Function: long long accum __satfractusata (unsigned accum A)

  2335.  -- Runtime Function: unsigned short fract __satfractusauqq (unsigned

  2336.           accum A)

  2337.  -- Runtime Function: unsigned fract __satfractusauhq (unsigned accum A)

  2338.  -- Runtime Function: unsigned long fract __satfractusausq (unsigned

  2339.           accum A)

  2340.  -- Runtime Function: unsigned long long fract __satfractusaudq

  2341.           (unsigned accum A)

  2342.  -- Runtime Function: unsigned short accum __satfractusauha2 (unsigned

  2343.           accum A)

  2344.  -- Runtime Function: unsigned long accum __satfractusauda2 (unsigned

  2345.           accum A)

  2346.  -- Runtime Function: unsigned long long accum __satfractusauta2

  2347.           (unsigned accum A)

  2348.  -- Runtime Function: short fract __satfractudaqq (unsigned long accum

  2349.           A)

  2350.  -- Runtime Function: fract __satfractudahq (unsigned long accum A)

  2351.  -- Runtime Function: long fract __satfractudasq (unsigned long accum A)

  2352.  -- Runtime Function: long long fract __satfractudadq (unsigned long

  2353.           accum A)

  2354.  -- Runtime Function: short accum __satfractudaha (unsigned long accum

  2355.           A)

  2356.  -- Runtime Function: accum __satfractudasa (unsigned long accum A)

  2357.  -- Runtime Function: long accum __satfractudada (unsigned long accum A)

  2358.  -- Runtime Function: long long accum __satfractudata (unsigned long

  2359.           accum A)

  2360.  -- Runtime Function: unsigned short fract __satfractudauqq (unsigned

  2361.           long accum A)

  2362.  -- Runtime Function: unsigned fract __satfractudauhq (unsigned long

  2363.           accum A)

  2364.  -- Runtime Function: unsigned long fract __satfractudausq (unsigned

  2365.           long accum A)

  2366.  -- Runtime Function: unsigned long long fract __satfractudaudq

  2367.           (unsigned long accum A)

  2368.  -- Runtime Function: unsigned short accum __satfractudauha2 (unsigned

  2369.           long accum A)

  2370.  -- Runtime Function: unsigned accum __satfractudausa2 (unsigned long

  2371.           accum A)

  2372.  -- Runtime Function: unsigned long long accum __satfractudauta2

  2373.           (unsigned long accum A)

  2374.  -- Runtime Function: short fract __satfractutaqq (unsigned long long

  2375.           accum A)

  2376.  -- Runtime Function: fract __satfractutahq (unsigned long long accum A)

  2377.  -- Runtime Function: long fract __satfractutasq (unsigned long long

  2378.           accum A)

  2379.  -- Runtime Function: long long fract __satfractutadq (unsigned long

  2380.           long accum A)

  2381.  -- Runtime Function: short accum __satfractutaha (unsigned long long

  2382.           accum A)

  2383.  -- Runtime Function: accum __satfractutasa (unsigned long long accum A)

  2384.  -- Runtime Function: long accum __satfractutada (unsigned long long

  2385.           accum A)

  2386.  -- Runtime Function: long long accum __satfractutata (unsigned long

  2387.           long accum A)

  2388.  -- Runtime Function: unsigned short fract __satfractutauqq (unsigned

  2389.           long long accum A)

  2390.  -- Runtime Function: unsigned fract __satfractutauhq (unsigned long

  2391.           long accum A)

  2392.  -- Runtime Function: unsigned long fract __satfractutausq (unsigned

  2393.           long long accum A)

  2394.  -- Runtime Function: unsigned long long fract __satfractutaudq

  2395.           (unsigned long long accum A)

  2396.  -- Runtime Function: unsigned short accum __satfractutauha2 (unsigned

  2397.           long long accum A)

  2398.  -- Runtime Function: unsigned accum __satfractutausa2 (unsigned long

  2399.           long accum A)

  2400.  -- Runtime Function: unsigned long accum __satfractutauda2 (unsigned

  2401.           long long accum A)

  2402.  -- Runtime Function: short fract __satfractqiqq (signed char A)

  2403.  -- Runtime Function: fract __satfractqihq (signed char A)

  2404.  -- Runtime Function: long fract __satfractqisq (signed char A)

  2405.  -- Runtime Function: long long fract __satfractqidq (signed char A)

  2406.  -- Runtime Function: short accum __satfractqiha (signed char A)

  2407.  -- Runtime Function: accum __satfractqisa (signed char A)

  2408.  -- Runtime Function: long accum __satfractqida (signed char A)

  2409.  -- Runtime Function: long long accum __satfractqita (signed char A)

  2410.  -- Runtime Function: unsigned short fract __satfractqiuqq (signed char

  2411.           A)

  2412.  -- Runtime Function: unsigned fract __satfractqiuhq (signed char A)

  2413.  -- Runtime Function: unsigned long fract __satfractqiusq (signed char

  2414.           A)

  2415.  -- Runtime Function: unsigned long long fract __satfractqiudq (signed

  2416.           char A)

  2417.  -- Runtime Function: unsigned short accum __satfractqiuha (signed char

  2418.           A)

  2419.  -- Runtime Function: unsigned accum __satfractqiusa (signed char A)

  2420.  -- Runtime Function: unsigned long accum __satfractqiuda (signed char

  2421.           A)

  2422.  -- Runtime Function: unsigned long long accum __satfractqiuta (signed

  2423.           char A)

  2424.  -- Runtime Function: short fract __satfracthiqq (short A)

  2425.  -- Runtime Function: fract __satfracthihq (short A)

  2426.  -- Runtime Function: long fract __satfracthisq (short A)

  2427.  -- Runtime Function: long long fract __satfracthidq (short A)

  2428.  -- Runtime Function: short accum __satfracthiha (short A)

  2429.  -- Runtime Function: accum __satfracthisa (short A)

  2430.  -- Runtime Function: long accum __satfracthida (short A)

  2431.  -- Runtime Function: long long accum __satfracthita (short A)

  2432.  -- Runtime Function: unsigned short fract __satfracthiuqq (short A)

  2433.  -- Runtime Function: unsigned fract __satfracthiuhq (short A)

  2434.  -- Runtime Function: unsigned long fract __satfracthiusq (short A)

  2435.  -- Runtime Function: unsigned long long fract __satfracthiudq (short A)

  2436.  -- Runtime Function: unsigned short accum __satfracthiuha (short A)

  2437.  -- Runtime Function: unsigned accum __satfracthiusa (short A)

  2438.  -- Runtime Function: unsigned long accum __satfracthiuda (short A)

  2439.  -- Runtime Function: unsigned long long accum __satfracthiuta (short A)

  2440.  -- Runtime Function: short fract __satfractsiqq (int A)

  2441.  -- Runtime Function: fract __satfractsihq (int A)

  2442.  -- Runtime Function: long fract __satfractsisq (int A)

  2443.  -- Runtime Function: long long fract __satfractsidq (int A)

  2444.  -- Runtime Function: short accum __satfractsiha (int A)

  2445.  -- Runtime Function: accum __satfractsisa (int A)

  2446.  -- Runtime Function: long accum __satfractsida (int A)

  2447.  -- Runtime Function: long long accum __satfractsita (int A)

  2448.  -- Runtime Function: unsigned short fract __satfractsiuqq (int A)

  2449.  -- Runtime Function: unsigned fract __satfractsiuhq (int A)

  2450.  -- Runtime Function: unsigned long fract __satfractsiusq (int A)

  2451.  -- Runtime Function: unsigned long long fract __satfractsiudq (int A)

  2452.  -- Runtime Function: unsigned short accum __satfractsiuha (int A)

  2453.  -- Runtime Function: unsigned accum __satfractsiusa (int A)

  2454.  -- Runtime Function: unsigned long accum __satfractsiuda (int A)

  2455.  -- Runtime Function: unsigned long long accum __satfractsiuta (int A)

  2456.  -- Runtime Function: short fract __satfractdiqq (long A)

  2457.  -- Runtime Function: fract __satfractdihq (long A)

  2458.  -- Runtime Function: long fract __satfractdisq (long A)

  2459.  -- Runtime Function: long long fract __satfractdidq (long A)

  2460.  -- Runtime Function: short accum __satfractdiha (long A)

  2461.  -- Runtime Function: accum __satfractdisa (long A)

  2462.  -- Runtime Function: long accum __satfractdida (long A)

  2463.  -- Runtime Function: long long accum __satfractdita (long A)

  2464.  -- Runtime Function: unsigned short fract __satfractdiuqq (long A)

  2465.  -- Runtime Function: unsigned fract __satfractdiuhq (long A)

  2466.  -- Runtime Function: unsigned long fract __satfractdiusq (long A)

  2467.  -- Runtime Function: unsigned long long fract __satfractdiudq (long A)

  2468.  -- Runtime Function: unsigned short accum __satfractdiuha (long A)

  2469.  -- Runtime Function: unsigned accum __satfractdiusa (long A)

  2470.  -- Runtime Function: unsigned long accum __satfractdiuda (long A)

  2471.  -- Runtime Function: unsigned long long accum __satfractdiuta (long A)

  2472.  -- Runtime Function: short fract __satfracttiqq (long long A)

  2473.  -- Runtime Function: fract __satfracttihq (long long A)

  2474.  -- Runtime Function: long fract __satfracttisq (long long A)

  2475.  -- Runtime Function: long long fract __satfracttidq (long long A)

  2476.  -- Runtime Function: short accum __satfracttiha (long long A)

  2477.  -- Runtime Function: accum __satfracttisa (long long A)

  2478.  -- Runtime Function: long accum __satfracttida (long long A)

  2479.  -- Runtime Function: long long accum __satfracttita (long long A)

  2480.  -- Runtime Function: unsigned short fract __satfracttiuqq (long long A)

  2481.  -- Runtime Function: unsigned fract __satfracttiuhq (long long A)

  2482.  -- Runtime Function: unsigned long fract __satfracttiusq (long long A)

  2483.  -- Runtime Function: unsigned long long fract __satfracttiudq (long

  2484.           long A)

  2485.  -- Runtime Function: unsigned short accum __satfracttiuha (long long A)

  2486.  -- Runtime Function: unsigned accum __satfracttiusa (long long A)

  2487.  -- Runtime Function: unsigned long accum __satfracttiuda (long long A)

  2488.  -- Runtime Function: unsigned long long accum __satfracttiuta (long

  2489.           long A)

  2490.  -- Runtime Function: short fract __satfractsfqq (float A)

  2491.  -- Runtime Function: fract __satfractsfhq (float A)

  2492.  -- Runtime Function: long fract __satfractsfsq (float A)

  2493.  -- Runtime Function: long long fract __satfractsfdq (float A)

  2494.  -- Runtime Function: short accum __satfractsfha (float A)

  2495.  -- Runtime Function: accum __satfractsfsa (float A)

  2496.  -- Runtime Function: long accum __satfractsfda (float A)

  2497.  -- Runtime Function: long long accum __satfractsfta (float A)

  2498.  -- Runtime Function: unsigned short fract __satfractsfuqq (float A)

  2499.  -- Runtime Function: unsigned fract __satfractsfuhq (float A)

  2500.  -- Runtime Function: unsigned long fract __satfractsfusq (float A)

  2501.  -- Runtime Function: unsigned long long fract __satfractsfudq (float A)

  2502.  -- Runtime Function: unsigned short accum __satfractsfuha (float A)

  2503.  -- Runtime Function: unsigned accum __satfractsfusa (float A)

  2504.  -- Runtime Function: unsigned long accum __satfractsfuda (float A)

  2505.  -- Runtime Function: unsigned long long accum __satfractsfuta (float A)

  2506.  -- Runtime Function: short fract __satfractdfqq (double A)

  2507.  -- Runtime Function: fract __satfractdfhq (double A)

  2508.  -- Runtime Function: long fract __satfractdfsq (double A)

  2509.  -- Runtime Function: long long fract __satfractdfdq (double A)

  2510.  -- Runtime Function: short accum __satfractdfha (double A)

  2511.  -- Runtime Function: accum __satfractdfsa (double A)

  2512.  -- Runtime Function: long accum __satfractdfda (double A)

  2513.  -- Runtime Function: long long accum __satfractdfta (double A)

  2514.  -- Runtime Function: unsigned short fract __satfractdfuqq (double A)

  2515.  -- Runtime Function: unsigned fract __satfractdfuhq (double A)

  2516.  -- Runtime Function: unsigned long fract __satfractdfusq (double A)

  2517.  -- Runtime Function: unsigned long long fract __satfractdfudq (double

  2518.           A)

  2519.  -- Runtime Function: unsigned short accum __satfractdfuha (double A)

  2520.  -- Runtime Function: unsigned accum __satfractdfusa (double A)

  2521.  -- Runtime Function: unsigned long accum __satfractdfuda (double A)

  2522.  -- Runtime Function: unsigned long long accum __satfractdfuta (double

  2523.           A)

  2524.      The functions convert from fractional and signed non-fractionals to

  2525.      fractionals, with saturation.

  2526. 

  2527.  -- Runtime Function: unsigned char __fractunsqqqi (short fract A)

  2528.  -- Runtime Function: unsigned short __fractunsqqhi (short fract A)

  2529.  -- Runtime Function: unsigned int __fractunsqqsi (short fract A)

  2530.  -- Runtime Function: unsigned long __fractunsqqdi (short fract A)

  2531.  -- Runtime Function: unsigned long long __fractunsqqti (short fract A)

  2532.  -- Runtime Function: unsigned char __fractunshqqi (fract A)

  2533.  -- Runtime Function: unsigned short __fractunshqhi (fract A)

  2534.  -- Runtime Function: unsigned int __fractunshqsi (fract A)

  2535.  -- Runtime Function: unsigned long __fractunshqdi (fract A)

  2536.  -- Runtime Function: unsigned long long __fractunshqti (fract A)

  2537.  -- Runtime Function: unsigned char __fractunssqqi (long fract A)

  2538.  -- Runtime Function: unsigned short __fractunssqhi (long fract A)

  2539.  -- Runtime Function: unsigned int __fractunssqsi (long fract A)

  2540.  -- Runtime Function: unsigned long __fractunssqdi (long fract A)

  2541.  -- Runtime Function: unsigned long long __fractunssqti (long fract A)

  2542.  -- Runtime Function: unsigned char __fractunsdqqi (long long fract A)

  2543.  -- Runtime Function: unsigned short __fractunsdqhi (long long fract A)

  2544.  -- Runtime Function: unsigned int __fractunsdqsi (long long fract A)

  2545.  -- Runtime Function: unsigned long __fractunsdqdi (long long fract A)

  2546.  -- Runtime Function: unsigned long long __fractunsdqti (long long fract

  2547.           A)

  2548.  -- Runtime Function: unsigned char __fractunshaqi (short accum A)

  2549.  -- Runtime Function: unsigned short __fractunshahi (short accum A)

  2550.  -- Runtime Function: unsigned int __fractunshasi (short accum A)

  2551.  -- Runtime Function: unsigned long __fractunshadi (short accum A)

  2552.  -- Runtime Function: unsigned long long __fractunshati (short accum A)

  2553.  -- Runtime Function: unsigned char __fractunssaqi (accum A)

  2554.  -- Runtime Function: unsigned short __fractunssahi (accum A)

  2555.  -- Runtime Function: unsigned int __fractunssasi (accum A)

  2556.  -- Runtime Function: unsigned long __fractunssadi (accum A)

  2557.  -- Runtime Function: unsigned long long __fractunssati (accum A)

  2558.  -- Runtime Function: unsigned char __fractunsdaqi (long accum A)

  2559.  -- Runtime Function: unsigned short __fractunsdahi (long accum A)

  2560.  -- Runtime Function: unsigned int __fractunsdasi (long accum A)

  2561.  -- Runtime Function: unsigned long __fractunsdadi (long accum A)

  2562.  -- Runtime Function: unsigned long long __fractunsdati (long accum A)

  2563.  -- Runtime Function: unsigned char __fractunstaqi (long long accum A)

  2564.  -- Runtime Function: unsigned short __fractunstahi (long long accum A)

  2565.  -- Runtime Function: unsigned int __fractunstasi (long long accum A)

  2566.  -- Runtime Function: unsigned long __fractunstadi (long long accum A)

  2567.  -- Runtime Function: unsigned long long __fractunstati (long long accum

  2568.           A)

  2569.  -- Runtime Function: unsigned char __fractunsuqqqi (unsigned short

  2570.           fract A)

  2571.  -- Runtime Function: unsigned short __fractunsuqqhi (unsigned short

  2572.           fract A)

  2573.  -- Runtime Function: unsigned int __fractunsuqqsi (unsigned short fract

  2574.           A)

  2575.  -- Runtime Function: unsigned long __fractunsuqqdi (unsigned short

  2576.           fract A)

  2577.  -- Runtime Function: unsigned long long __fractunsuqqti (unsigned short

  2578.           fract A)

  2579.  -- Runtime Function: unsigned char __fractunsuhqqi (unsigned fract A)

  2580.  -- Runtime Function: unsigned short __fractunsuhqhi (unsigned fract A)

  2581.  -- Runtime Function: unsigned int __fractunsuhqsi (unsigned fract A)

  2582.  -- Runtime Function: unsigned long __fractunsuhqdi (unsigned fract A)

  2583.  -- Runtime Function: unsigned long long __fractunsuhqti (unsigned fract

  2584.           A)

  2585.  -- Runtime Function: unsigned char __fractunsusqqi (unsigned long fract

  2586.           A)

  2587.  -- Runtime Function: unsigned short __fractunsusqhi (unsigned long

  2588.           fract A)

  2589.  -- Runtime Function: unsigned int __fractunsusqsi (unsigned long fract

  2590.           A)

  2591.  -- Runtime Function: unsigned long __fractunsusqdi (unsigned long fract

  2592.           A)

  2593.  -- Runtime Function: unsigned long long __fractunsusqti (unsigned long

  2594.           fract A)

  2595.  -- Runtime Function: unsigned char __fractunsudqqi (unsigned long long

  2596.           fract A)

  2597.  -- Runtime Function: unsigned short __fractunsudqhi (unsigned long long

  2598.           fract A)

  2599.  -- Runtime Function: unsigned int __fractunsudqsi (unsigned long long

  2600.           fract A)

  2601.  -- Runtime Function: unsigned long __fractunsudqdi (unsigned long long

  2602.           fract A)

  2603.  -- Runtime Function: unsigned long long __fractunsudqti (unsigned long

  2604.           long fract A)

  2605.  -- Runtime Function: unsigned char __fractunsuhaqi (unsigned short

  2606.           accum A)

  2607.  -- Runtime Function: unsigned short __fractunsuhahi (unsigned short

  2608.           accum A)

  2609.  -- Runtime Function: unsigned int __fractunsuhasi (unsigned short accum

  2610.           A)

  2611.  -- Runtime Function: unsigned long __fractunsuhadi (unsigned short

  2612.           accum A)

  2613.  -- Runtime Function: unsigned long long __fractunsuhati (unsigned short

  2614.           accum A)

  2615.  -- Runtime Function: unsigned char __fractunsusaqi (unsigned accum A)

  2616.  -- Runtime Function: unsigned short __fractunsusahi (unsigned accum A)

  2617.  -- Runtime Function: unsigned int __fractunsusasi (unsigned accum A)

  2618.  -- Runtime Function: unsigned long __fractunsusadi (unsigned accum A)

  2619.  -- Runtime Function: unsigned long long __fractunsusati (unsigned accum

  2620.           A)

  2621.  -- Runtime Function: unsigned char __fractunsudaqi (unsigned long accum

  2622.           A)

  2623.  -- Runtime Function: unsigned short __fractunsudahi (unsigned long

  2624.           accum A)

  2625.  -- Runtime Function: unsigned int __fractunsudasi (unsigned long accum

  2626.           A)

  2627.  -- Runtime Function: unsigned long __fractunsudadi (unsigned long accum

  2628.           A)

  2629.  -- Runtime Function: unsigned long long __fractunsudati (unsigned long

  2630.           accum A)

  2631.  -- Runtime Function: unsigned char __fractunsutaqi (unsigned long long

  2632.           accum A)

  2633.  -- Runtime Function: unsigned short __fractunsutahi (unsigned long long

  2634.           accum A)

  2635.  -- Runtime Function: unsigned int __fractunsutasi (unsigned long long

  2636.           accum A)

  2637.  -- Runtime Function: unsigned long __fractunsutadi (unsigned long long

  2638.           accum A)

  2639.  -- Runtime Function: unsigned long long __fractunsutati (unsigned long

  2640.           long accum A)

  2641.  -- Runtime Function: short fract __fractunsqiqq (unsigned char A)

  2642.  -- Runtime Function: fract __fractunsqihq (unsigned char A)

  2643.  -- Runtime Function: long fract __fractunsqisq (unsigned char A)

  2644.  -- Runtime Function: long long fract __fractunsqidq (unsigned char A)

  2645.  -- Runtime Function: short accum __fractunsqiha (unsigned char A)

  2646.  -- Runtime Function: accum __fractunsqisa (unsigned char A)

  2647.  -- Runtime Function: long accum __fractunsqida (unsigned char A)

  2648.  -- Runtime Function: long long accum __fractunsqita (unsigned char A)

  2649.  -- Runtime Function: unsigned short fract __fractunsqiuqq (unsigned

  2650.           char A)

  2651.  -- Runtime Function: unsigned fract __fractunsqiuhq (unsigned char A)

  2652.  -- Runtime Function: unsigned long fract __fractunsqiusq (unsigned char

  2653.           A)

  2654.  -- Runtime Function: unsigned long long fract __fractunsqiudq (unsigned

  2655.           char A)

  2656.  -- Runtime Function: unsigned short accum __fractunsqiuha (unsigned

  2657.           char A)

  2658.  -- Runtime Function: unsigned accum __fractunsqiusa (unsigned char A)

  2659.  -- Runtime Function: unsigned long accum __fractunsqiuda (unsigned char

  2660.           A)

  2661.  -- Runtime Function: unsigned long long accum __fractunsqiuta (unsigned

  2662.           char A)

  2663.  -- Runtime Function: short fract __fractunshiqq (unsigned short A)

  2664.  -- Runtime Function: fract __fractunshihq (unsigned short A)

  2665.  -- Runtime Function: long fract __fractunshisq (unsigned short A)

  2666.  -- Runtime Function: long long fract __fractunshidq (unsigned short A)

  2667.  -- Runtime Function: short accum __fractunshiha (unsigned short A)

  2668.  -- Runtime Function: accum __fractunshisa (unsigned short A)

  2669.  -- Runtime Function: long accum __fractunshida (unsigned short A)

  2670.  -- Runtime Function: long long accum __fractunshita (unsigned short A)

  2671.  -- Runtime Function: unsigned short fract __fractunshiuqq (unsigned

  2672.           short A)

  2673.  -- Runtime Function: unsigned fract __fractunshiuhq (unsigned short A)

  2674.  -- Runtime Function: unsigned long fract __fractunshiusq (unsigned

  2675.           short A)

  2676.  -- Runtime Function: unsigned long long fract __fractunshiudq (unsigned

  2677.           short A)

  2678.  -- Runtime Function: unsigned short accum __fractunshiuha (unsigned

  2679.           short A)

  2680.  -- Runtime Function: unsigned accum __fractunshiusa (unsigned short A)

  2681.  -- Runtime Function: unsigned long accum __fractunshiuda (unsigned

  2682.           short A)

  2683.  -- Runtime Function: unsigned long long accum __fractunshiuta (unsigned

  2684.           short A)

  2685.  -- Runtime Function: short fract __fractunssiqq (unsigned int A)

  2686.  -- Runtime Function: fract __fractunssihq (unsigned int A)

  2687.  -- Runtime Function: long fract __fractunssisq (unsigned int A)

  2688.  -- Runtime Function: long long fract __fractunssidq (unsigned int A)

  2689.  -- Runtime Function: short accum __fractunssiha (unsigned int A)

  2690.  -- Runtime Function: accum __fractunssisa (unsigned int A)

  2691.  -- Runtime Function: long accum __fractunssida (unsigned int A)

  2692.  -- Runtime Function: long long accum __fractunssita (unsigned int A)

  2693.  -- Runtime Function: unsigned short fract __fractunssiuqq (unsigned int

  2694.           A)

  2695.  -- Runtime Function: unsigned fract __fractunssiuhq (unsigned int A)

  2696.  -- Runtime Function: unsigned long fract __fractunssiusq (unsigned int

  2697.           A)

  2698.  -- Runtime Function: unsigned long long fract __fractunssiudq (unsigned

  2699.           int A)

  2700.  -- Runtime Function: unsigned short accum __fractunssiuha (unsigned int

  2701.           A)

  2702.  -- Runtime Function: unsigned accum __fractunssiusa (unsigned int A)

  2703.  -- Runtime Function: unsigned long accum __fractunssiuda (unsigned int

  2704.           A)

  2705.  -- Runtime Function: unsigned long long accum __fractunssiuta (unsigned

  2706.           int A)

  2707.  -- Runtime Function: short fract __fractunsdiqq (unsigned long A)

  2708.  -- Runtime Function: fract __fractunsdihq (unsigned long A)

  2709.  -- Runtime Function: long fract __fractunsdisq (unsigned long A)

  2710.  -- Runtime Function: long long fract __fractunsdidq (unsigned long A)

  2711.  -- Runtime Function: short accum __fractunsdiha (unsigned long A)

  2712.  -- Runtime Function: accum __fractunsdisa (unsigned long A)

  2713.  -- Runtime Function: long accum __fractunsdida (unsigned long A)

  2714.  -- Runtime Function: long long accum __fractunsdita (unsigned long A)

  2715.  -- Runtime Function: unsigned short fract __fractunsdiuqq (unsigned

  2716.           long A)

  2717.  -- Runtime Function: unsigned fract __fractunsdiuhq (unsigned long A)

  2718.  -- Runtime Function: unsigned long fract __fractunsdiusq (unsigned long

  2719.           A)

  2720.  -- Runtime Function: unsigned long long fract __fractunsdiudq (unsigned

  2721.           long A)

  2722.  -- Runtime Function: unsigned short accum __fractunsdiuha (unsigned

  2723.           long A)

  2724.  -- Runtime Function: unsigned accum __fractunsdiusa (unsigned long A)

  2725.  -- Runtime Function: unsigned long accum __fractunsdiuda (unsigned long

  2726.           A)

  2727.  -- Runtime Function: unsigned long long accum __fractunsdiuta (unsigned

  2728.           long A)

  2729.  -- Runtime Function: short fract __fractunstiqq (unsigned long long A)

  2730.  -- Runtime Function: fract __fractunstihq (unsigned long long A)

  2731.  -- Runtime Function: long fract __fractunstisq (unsigned long long A)

  2732.  -- Runtime Function: long long fract __fractunstidq (unsigned long long

  2733.           A)

  2734.  -- Runtime Function: short accum __fractunstiha (unsigned long long A)

  2735.  -- Runtime Function: accum __fractunstisa (unsigned long long A)

  2736.  -- Runtime Function: long accum __fractunstida (unsigned long long A)

  2737.  -- Runtime Function: long long accum __fractunstita (unsigned long long

  2738.           A)

  2739.  -- Runtime Function: unsigned short fract __fractunstiuqq (unsigned

  2740.           long long A)

  2741.  -- Runtime Function: unsigned fract __fractunstiuhq (unsigned long long

  2742.           A)

  2743.  -- Runtime Function: unsigned long fract __fractunstiusq (unsigned long

  2744.           long A)

  2745.  -- Runtime Function: unsigned long long fract __fractunstiudq (unsigned

  2746.           long long A)

  2747.  -- Runtime Function: unsigned short accum __fractunstiuha (unsigned

  2748.           long long A)

  2749.  -- Runtime Function: unsigned accum __fractunstiusa (unsigned long long

  2750.           A)

  2751.  -- Runtime Function: unsigned long accum __fractunstiuda (unsigned long

  2752.           long A)

  2753.  -- Runtime Function: unsigned long long accum __fractunstiuta (unsigned

  2754.           long long A)

  2755.      These functions convert from fractionals to unsigned

  2756.      non-fractionals; and from unsigned non-fractionals to fractionals,

  2757.      without saturation.

  2758. 

  2759.  -- Runtime Function: short fract __satfractunsqiqq (unsigned char A)

  2760.  -- Runtime Function: fract __satfractunsqihq (unsigned char A)

  2761.  -- Runtime Function: long fract __satfractunsqisq (unsigned char A)

  2762.  -- Runtime Function: long long fract __satfractunsqidq (unsigned char

  2763.           A)

  2764.  -- Runtime Function: short accum __satfractunsqiha (unsigned char A)

  2765.  -- Runtime Function: accum __satfractunsqisa (unsigned char A)

  2766.  -- Runtime Function: long accum __satfractunsqida (unsigned char A)

  2767.  -- Runtime Function: long long accum __satfractunsqita (unsigned char

  2768.           A)

  2769.  -- Runtime Function: unsigned short fract __satfractunsqiuqq (unsigned

  2770.           char A)

  2771.  -- Runtime Function: unsigned fract __satfractunsqiuhq (unsigned char

  2772.           A)

  2773.  -- Runtime Function: unsigned long fract __satfractunsqiusq (unsigned

  2774.           char A)

  2775.  -- Runtime Function: unsigned long long fract __satfractunsqiudq

  2776.           (unsigned char A)

  2777.  -- Runtime Function: unsigned short accum __satfractunsqiuha (unsigned

  2778.           char A)

  2779.  -- Runtime Function: unsigned accum __satfractunsqiusa (unsigned char

  2780.           A)

  2781.  -- Runtime Function: unsigned long accum __satfractunsqiuda (unsigned

  2782.           char A)

  2783.  -- Runtime Function: unsigned long long accum __satfractunsqiuta

  2784.           (unsigned char A)

  2785.  -- Runtime Function: short fract __satfractunshiqq (unsigned short A)

  2786.  -- Runtime Function: fract __satfractunshihq (unsigned short A)

  2787.  -- Runtime Function: long fract __satfractunshisq (unsigned short A)

  2788.  -- Runtime Function: long long fract __satfractunshidq (unsigned short

  2789.           A)

  2790.  -- Runtime Function: short accum __satfractunshiha (unsigned short A)

  2791.  -- Runtime Function: accum __satfractunshisa (unsigned short A)

  2792.  -- Runtime Function: long accum __satfractunshida (unsigned short A)

  2793.  -- Runtime Function: long long accum __satfractunshita (unsigned short

  2794.           A)

  2795.  -- Runtime Function: unsigned short fract __satfractunshiuqq (unsigned

  2796.           short A)

  2797.  -- Runtime Function: unsigned fract __satfractunshiuhq (unsigned short

  2798.           A)

  2799.  -- Runtime Function: unsigned long fract __satfractunshiusq (unsigned

  2800.           short A)

  2801.  -- Runtime Function: unsigned long long fract __satfractunshiudq

  2802.           (unsigned short A)

  2803.  -- Runtime Function: unsigned short accum __satfractunshiuha (unsigned

  2804.           short A)

  2805.  -- Runtime Function: unsigned accum __satfractunshiusa (unsigned short

  2806.           A)

  2807.  -- Runtime Function: unsigned long accum __satfractunshiuda (unsigned

  2808.           short A)

  2809.  -- Runtime Function: unsigned long long accum __satfractunshiuta

  2810.           (unsigned short A)

  2811.  -- Runtime Function: short fract __satfractunssiqq (unsigned int A)

  2812.  -- Runtime Function: fract __satfractunssihq (unsigned int A)

  2813.  -- Runtime Function: long fract __satfractunssisq (unsigned int A)

  2814.  -- Runtime Function: long long fract __satfractunssidq (unsigned int A)

  2815.  -- Runtime Function: short accum __satfractunssiha (unsigned int A)

  2816.  -- Runtime Function: accum __satfractunssisa (unsigned int A)

  2817.  -- Runtime Function: long accum __satfractunssida (unsigned int A)

  2818.  -- Runtime Function: long long accum __satfractunssita (unsigned int A)

  2819.  -- Runtime Function: unsigned short fract __satfractunssiuqq (unsigned

  2820.           int A)

  2821.  -- Runtime Function: unsigned fract __satfractunssiuhq (unsigned int A)

  2822.  -- Runtime Function: unsigned long fract __satfractunssiusq (unsigned

  2823.           int A)

  2824.  -- Runtime Function: unsigned long long fract __satfractunssiudq

  2825.           (unsigned int A)

  2826.  -- Runtime Function: unsigned short accum __satfractunssiuha (unsigned

  2827.           int A)

  2828.  -- Runtime Function: unsigned accum __satfractunssiusa (unsigned int A)

  2829.  -- Runtime Function: unsigned long accum __satfractunssiuda (unsigned

  2830.           int A)

  2831.  -- Runtime Function: unsigned long long accum __satfractunssiuta

  2832.           (unsigned int A)

  2833.  -- Runtime Function: short fract __satfractunsdiqq (unsigned long A)

  2834.  -- Runtime Function: fract __satfractunsdihq (unsigned long A)

  2835.  -- Runtime Function: long fract __satfractunsdisq (unsigned long A)

  2836.  -- Runtime Function: long long fract __satfractunsdidq (unsigned long

  2837.           A)

  2838.  -- Runtime Function: short accum __satfractunsdiha (unsigned long A)

  2839.  -- Runtime Function: accum __satfractunsdisa (unsigned long A)

  2840.  -- Runtime Function: long accum __satfractunsdida (unsigned long A)

  2841.  -- Runtime Function: long long accum __satfractunsdita (unsigned long

  2842.           A)

  2843.  -- Runtime Function: unsigned short fract __satfractunsdiuqq (unsigned

  2844.           long A)

  2845.  -- Runtime Function: unsigned fract __satfractunsdiuhq (unsigned long

  2846.           A)

  2847.  -- Runtime Function: unsigned long fract __satfractunsdiusq (unsigned

  2848.           long A)

  2849.  -- Runtime Function: unsigned long long fract __satfractunsdiudq

  2850.           (unsigned long A)

  2851.  -- Runtime Function: unsigned short accum __satfractunsdiuha (unsigned

  2852.           long A)

  2853.  -- Runtime Function: unsigned accum __satfractunsdiusa (unsigned long

  2854.           A)

  2855.  -- Runtime Function: unsigned long accum __satfractunsdiuda (unsigned

  2856.           long A)

  2857.  -- Runtime Function: unsigned long long accum __satfractunsdiuta

  2858.           (unsigned long A)

  2859.  -- Runtime Function: short fract __satfractunstiqq (unsigned long long

  2860.           A)

  2861.  -- Runtime Function: fract __satfractunstihq (unsigned long long A)

  2862.  -- Runtime Function: long fract __satfractunstisq (unsigned long long

  2863.           A)

  2864.  -- Runtime Function: long long fract __satfractunstidq (unsigned long

  2865.           long A)

  2866.  -- Runtime Function: short accum __satfractunstiha (unsigned long long

  2867.           A)

  2868.  -- Runtime Function: accum __satfractunstisa (unsigned long long A)

  2869.  -- Runtime Function: long accum __satfractunstida (unsigned long long

  2870.           A)

  2871.  -- Runtime Function: long long accum __satfractunstita (unsigned long

  2872.           long A)

  2873.  -- Runtime Function: unsigned short fract __satfractunstiuqq (unsigned

  2874.           long long A)

  2875.  -- Runtime Function: unsigned fract __satfractunstiuhq (unsigned long

  2876.           long A)

  2877.  -- Runtime Function: unsigned long fract __satfractunstiusq (unsigned

  2878.           long long A)

  2879.  -- Runtime Function: unsigned long long fract __satfractunstiudq

  2880.           (unsigned long long A)

  2881.  -- Runtime Function: unsigned short accum __satfractunstiuha (unsigned

  2882.           long long A)

  2883.  -- Runtime Function: unsigned accum __satfractunstiusa (unsigned long

  2884.           long A)

  2885.  -- Runtime Function: unsigned long accum __satfractunstiuda (unsigned

  2886.           long long A)

  2887.  -- Runtime Function: unsigned long long accum __satfractunstiuta

  2888.           (unsigned long long A)

  2889.      These functions convert from unsigned non-fractionals to

  2890.      fractionals, with saturation.

  2891. 

  2892. 

  2893. File: gccint.info,  Node: Exception handling routines,  Next: Miscellaneous routines,  Prev: Fixed-point fractional library routines,  Up: Libgcc

  2894. 

  2895. 4.5 Language-independent routines for exception handling

  2896. ========================================================

  2897. 

  2898. document me!

  2899. 

  2900.        _Unwind_DeleteException

  2901.        _Unwind_Find_FDE

  2902.        _Unwind_ForcedUnwind

  2903.        _Unwind_GetGR

  2904.        _Unwind_GetIP

  2905.        _Unwind_GetLanguageSpecificData

  2906.        _Unwind_GetRegionStart

  2907.        _Unwind_GetTextRelBase

  2908.        _Unwind_GetDataRelBase

  2909.        _Unwind_RaiseException

  2910.        _Unwind_Resume

  2911.        _Unwind_SetGR

  2912.        _Unwind_SetIP

  2913.        _Unwind_FindEnclosingFunction

  2914.        _Unwind_SjLj_Register

  2915.        _Unwind_SjLj_Unregister

  2916.        _Unwind_SjLj_RaiseException

  2917.        _Unwind_SjLj_ForcedUnwind

  2918.        _Unwind_SjLj_Resume

  2919.        __deregister_frame

  2920.        __deregister_frame_info

  2921.        __deregister_frame_info_bases

  2922.        __register_frame

  2923.        __register_frame_info

  2924.        __register_frame_info_bases

  2925.        __register_frame_info_table

  2926.        __register_frame_info_table_bases

  2927.        __register_frame_table

  2928. 

  2929. 

  2930. File: gccint.info,  Node: Miscellaneous routines,  Prev: Exception handling routines,  Up: Libgcc

  2931. 

  2932. 4.6 Miscellaneous runtime library routines

  2933. ==========================================

  2934. 

  2935. 4.6.1 Cache control functions

  2936. -----------------------------

  2937. 

  2938.  -- Runtime Function: void __clear_cache (char *BEG, char *END)

  2939.      This function clears the instruction cache between BEG and END.

  2940. 

  2941. 4.6.2 Split stack functions and variables

  2942. -----------------------------------------

  2943. 

  2944.  -- Runtime Function: void * __splitstack_find (void *SEGMENT_ARG, void

  2945.           *SP, size_t LEN, void **NEXT_SEGMENT, void **NEXT_SP, void

  2946.           **INITIAL_SP)

  2947.      When using '-fsplit-stack', this call may be used to iterate over

  2948.      the stack segments.  It may be called like this:

  2949.             void *next_segment = NULL;

  2950.             void *next_sp = NULL;

  2951.             void *initial_sp = NULL;

  2952.             void *stack;

  2953.             size_t stack_size;

  2954.             while ((stack = __splitstack_find (next_segment, next_sp,

  2955.                                                &stack_size, &next_segment,

  2956.                                                &next_sp, &initial_sp))

  2957.                    != NULL)

  2958.               {

  2959.                 /* Stack segment starts at stack and is

  2960.                    stack_size bytes long.  */

  2961.               }

  2962. 

  2963.      There is no way to iterate over the stack segments of a different

  2964.      thread.  However, what is permitted is for one thread to call this

  2965.      with the SEGMENT_ARG and SP arguments NULL, to pass NEXT_SEGMENT,

  2966.      NEXT_SP, and INITIAL_SP to a different thread, and then to suspend

  2967.      one way or another.  A different thread may run the subsequent

  2968.      '__splitstack_find' iterations.  Of course, this will only work if

  2969.      the first thread is suspended while the second thread is calling

  2970.      '__splitstack_find'.  If not, the second thread could be looking at

  2971.      the stack while it is changing, and anything could happen.

  2972. 

  2973.  -- Variable: __morestack_segments

  2974.  -- Variable: __morestack_current_segment

  2975.  -- Variable: __morestack_initial_sp

  2976.      Internal variables used by the '-fsplit-stack' implementation.

  2977. 

  2978. 

  2979. File: gccint.info,  Node: Languages,  Next: Source Tree,  Prev: Libgcc,  Up: Top

  2980. 

  2981. 5 Language Front Ends in GCC

  2982. ****************************

  2983. 

  2984. The interface to front ends for languages in GCC, and in particular the

  2985. 'tree' structure (*note GENERIC::), was initially designed for C, and

  2986. many aspects of it are still somewhat biased towards C and C-like

  2987. languages.  It is, however, reasonably well suited to other procedural

  2988. languages, and front ends for many such languages have been written for

  2989. GCC.

  2990. 

  2991.  Writing a compiler as a front end for GCC, rather than compiling

  2992. directly to assembler or generating C code which is then compiled by

  2993. GCC, has several advantages:

  2994. 

  2995.    * GCC front ends benefit from the support for many different target

  2996.      machines already present in GCC.

  2997.    * GCC front ends benefit from all the optimizations in GCC.  Some of

  2998.      these, such as alias analysis, may work better when GCC is

  2999.      compiling directly from source code then when it is compiling from

  3000.      generated C code.

  3001.    * Better debugging information is generated when compiling directly

  3002.      from source code than when going via intermediate generated C code.

  3003. 

  3004.  Because of the advantages of writing a compiler as a GCC front end, GCC

  3005. front ends have also been created for languages very different from

  3006. those for which GCC was designed, such as the declarative

  3007. logic/functional language Mercury.  For these reasons, it may also be

  3008. useful to implement compilers created for specialized purposes (for

  3009. example, as part of a research project) as GCC front ends.

  3010. 

  3011. 

  3012. File: gccint.info,  Node: Source Tree,  Next: Testsuites,  Prev: Languages,  Up: Top

  3013. 

  3014. 6 Source Tree Structure and Build System

  3015. ****************************************

  3016. 

  3017. This chapter describes the structure of the GCC source tree, and how GCC

  3018. is built.  The user documentation for building and installing GCC is in

  3019. a separate manual (<http://gcc.gnu.org/install/>), with which it is

  3020. presumed that you are familiar.

  3021. 

  3022. * Menu:

  3023. 

  3024. * Configure Terms:: Configuration terminology and history.

  3025. * Top Level::       The top level source directory.

  3026. * gcc Directory::   The 'gcc' subdirectory.

  3027. 

  3028. 

  3029. File: gccint.info,  Node: Configure Terms,  Next: Top Level,  Up: Source Tree

  3030. 

  3031. 6.1 Configure Terms and History

  3032. ===============================

  3033. 

  3034. The configure and build process has a long and colorful history, and can

  3035. be confusing to anyone who doesn't know why things are the way they are.

  3036. While there are other documents which describe the configuration process

  3037. in detail, here are a few things that everyone working on GCC should

  3038. know.

  3039. 

  3040.  There are three system names that the build knows about: the machine

  3041. you are building on ("build"), the machine that you are building for

  3042. ("host"), and the machine that GCC will produce code for ("target").

  3043. When you configure GCC, you specify these with '--build=', '--host=',

  3044. and '--target='.

  3045. 

  3046.  Specifying the host without specifying the build should be avoided, as

  3047. 'configure' may (and once did) assume that the host you specify is also

  3048. the build, which may not be true.

  3049. 

  3050.  If build, host, and target are all the same, this is called a "native".

  3051. If build and host are the same but target is different, this is called a

  3052. "cross".  If build, host, and target are all different this is called a

  3053. "canadian" (for obscure reasons dealing with Canada's political party

  3054. and the background of the person working on the build at that time).  If

  3055. host and target are the same, but build is different, you are using a

  3056. cross-compiler to build a native for a different system.  Some people

  3057. call this a "host-x-host", "crossed native", or "cross-built native".

  3058. If build and target are the same, but host is different, you are using a

  3059. cross compiler to build a cross compiler that produces code for the

  3060. machine you're building on.  This is rare, so there is no common way of

  3061. describing it.  There is a proposal to call this a "crossback".

  3062. 

  3063.  If build and host are the same, the GCC you are building will also be

  3064. used to build the target libraries (like 'libstdc++').  If build and

  3065. host are different, you must have already built and installed a cross

  3066. compiler that will be used to build the target libraries (if you

  3067. configured with '--target=foo-bar', this compiler will be called

  3068. 'foo-bar-gcc').

  3069. 

  3070.  In the case of target libraries, the machine you're building for is the

  3071. machine you specified with '--target'.  So, build is the machine you're

  3072. building on (no change there), host is the machine you're building for

  3073. (the target libraries are built for the target, so host is the target

  3074. you specified), and target doesn't apply (because you're not building a

  3075. compiler, you're building libraries).  The configure/make process will

  3076. adjust these variables as needed.  It also sets '$with_cross_host' to

  3077. the original '--host' value in case you need it.

  3078. 

  3079.  The 'libiberty' support library is built up to three times: once for

  3080. the host, once for the target (even if they are the same), and once for

  3081. the build if build and host are different.  This allows it to be used by

  3082. all programs which are generated in the course of the build process.

  3083. 

  3084. 

  3085. File: gccint.info,  Node: Top Level,  Next: gcc Directory,  Prev: Configure Terms,  Up: Source Tree

  3086. 

  3087. 6.2 Top Level Source Directory

  3088. ==============================

  3089. 

  3090. The top level source directory in a GCC distribution contains several

  3091. files and directories that are shared with other software distributions

  3092. such as that of GNU Binutils.  It also contains several subdirectories

  3093. that contain parts of GCC and its runtime libraries:

  3094. 

  3095. 'boehm-gc'

  3096.      The Boehm conservative garbage collector, optionally used as part

  3097.      of the ObjC runtime library when configured with

  3098.      '--enable-objc-gc'.

  3099. 

  3100. 'config'

  3101.      Autoconf macros and Makefile fragments used throughout the tree.

  3102. 

  3103. 'contrib'

  3104.      Contributed scripts that may be found useful in conjunction with

  3105.      GCC.  One of these, 'contrib/texi2pod.pl', is used to generate man

  3106.      pages from Texinfo manuals as part of the GCC build process.

  3107. 

  3108. 'fixincludes'

  3109.      The support for fixing system headers to work with GCC.  See

  3110.      'fixincludes/README' for more information.  The headers fixed by

  3111.      this mechanism are installed in 'LIBSUBDIR/include-fixed'.  Along

  3112.      with those headers, 'README-fixinc' is also installed, as

  3113.      'LIBSUBDIR/include-fixed/README'.

  3114. 

  3115. 'gcc'

  3116.      The main sources of GCC itself (except for runtime libraries),

  3117.      including optimizers, support for different target architectures,

  3118.      language front ends, and testsuites.  *Note The 'gcc' Subdirectory:

  3119.      gcc Directory, for details.

  3120. 

  3121. 'gnattools'

  3122.      Support tools for GNAT.

  3123. 

  3124. 'include'

  3125.      Headers for the 'libiberty' library.

  3126. 

  3127. 'intl'

  3128.      GNU 'libintl', from GNU 'gettext', for systems which do not include

  3129.      it in 'libc'.

  3130. 

  3131. 'libada'

  3132.      The Ada runtime library.

  3133. 

  3134. 'libatomic'

  3135.      The runtime support library for atomic operations (e.g. for

  3136.      '__sync' and '__atomic').

  3137. 

  3138. 'libcpp'

  3139.      The C preprocessor library.

  3140. 

  3141. 'libdecnumber'

  3142.      The Decimal Float support library.

  3143. 

  3144. 'libffi'

  3145.      The 'libffi' library, used as part of the Go runtime library.

  3146. 

  3147. 'libgcc'

  3148.      The GCC runtime library.

  3149. 

  3150. 'libgfortran'

  3151.      The Fortran runtime library.

  3152. 

  3153. 'libgo'

  3154.      The Go runtime library.  The bulk of this library is mirrored from

  3155.      the master Go repository (https://github.com/golang/go).

  3156. 

  3157. 'libgomp'

  3158.      The GNU Offloading and Multi Processing Runtime Library.

  3159. 

  3160. 'libiberty'

  3161.      The 'libiberty' library, used for portability and for some

  3162.      generally useful data structures and algorithms.  *Note

  3163.      Introduction: (libiberty)Top, for more information about this

  3164.      library.

  3165. 

  3166. 'libitm'

  3167.      The runtime support library for transactional memory.

  3168. 

  3169. 'libobjc'

  3170.      The Objective-C and Objective-C++ runtime library.

  3171. 

  3172. 'libquadmath'

  3173.      The runtime support library for quad-precision math operations.

  3174. 

  3175. 'libphobos'

  3176.      The D standard and runtime library.  The bulk of this library is

  3177.      mirrored from the master D repositories (https://github.com/dlang).

  3178. 

  3179. 'libssp'

  3180.      The Stack protector runtime library.

  3181. 

  3182. 'libstdc++-v3'

  3183.      The C++ runtime library.

  3184. 

  3185. 'lto-plugin'

  3186.      Plugin used by the linker if link-time optimizations are enabled.

  3187. 

  3188. 'maintainer-scripts'

  3189.      Scripts used by the 'gccadmin' account on 'gcc.gnu.org'.

  3190. 

  3191. 'zlib'

  3192.      The 'zlib' compression library, used for compressing and

  3193.      uncompressing GCC's intermediate language in LTO object files.

  3194. 

  3195.  The build system in the top level directory, including how recursion

  3196. into subdirectories works and how building runtime libraries for

  3197. multilibs is handled, is documented in a separate manual, included with

  3198. GNU Binutils.  *Note GNU configure and build system: (configure)Top, for

  3199. details.

  3200. 

  3201. 

  3202. File: gccint.info,  Node: gcc Directory,  Prev: Top Level,  Up: Source Tree

  3203. 

  3204. 6.3 The 'gcc' Subdirectory

  3205. ==========================

  3206. 

  3207. The 'gcc' directory contains many files that are part of the C sources

  3208. of GCC, other files used as part of the configuration and build process,

  3209. and subdirectories including documentation and a testsuite.  The files

  3210. that are sources of GCC are documented in a separate chapter.  *Note

  3211. Passes and Files of the Compiler: Passes.

  3212. 

  3213. * Menu:

  3214. 

  3215. * Subdirectories:: Subdirectories of 'gcc'.

  3216. * Configuration::  The configuration process, and the files it uses.

  3217. * Build::          The build system in the 'gcc' directory.

  3218. * Makefile::       Targets in 'gcc/Makefile'.

  3219. * Library Files::  Library source files and headers under 'gcc/'.

  3220. * Headers::        Headers installed by GCC.

  3221. * Documentation::  Building documentation in GCC.

  3222. * Front End::      Anatomy of a language front end.

  3223. * Back End::       Anatomy of a target back end.

  3224. 

  3225. 

  3226. File: gccint.info,  Node: Subdirectories,  Next: Configuration,  Up: gcc Directory

  3227. 

  3228. 6.3.1 Subdirectories of 'gcc'

  3229. -----------------------------

  3230. 

  3231. The 'gcc' directory contains the following subdirectories:

  3232. 

  3233. 'LANGUAGE'

  3234.      Subdirectories for various languages.  Directories containing a

  3235.      file 'config-lang.in' are language subdirectories.  The contents of

  3236.      the subdirectories 'c' (for C), 'cp' (for C++), 'objc' (for

  3237.      Objective-C), 'objcp' (for Objective-C++), and 'lto' (for LTO) are

  3238.      documented in this manual (*note Passes and Files of the Compiler:

  3239.      Passes.); those for other languages are not.  *Note Anatomy of a

  3240.      Language Front End: Front End, for details of the files in these

  3241.      directories.

  3242. 

  3243. 'common'

  3244.      Source files shared between the compiler drivers (such as 'gcc')

  3245.      and the compilers proper (such as 'cc1').  If an architecture

  3246.      defines target hooks shared between those places, it also has a

  3247.      subdirectory in 'common/config'.  *Note Target Structure::.

  3248. 

  3249. 'config'

  3250.      Configuration files for supported architectures and operating

  3251.      systems.  *Note Anatomy of a Target Back End: Back End, for details

  3252.      of the files in this directory.

  3253. 

  3254. 'doc'

  3255.      Texinfo documentation for GCC, together with automatically

  3256.      generated man pages and support for converting the installation

  3257.      manual to HTML.  *Note Documentation::.

  3258. 

  3259. 'ginclude'

  3260.      System headers installed by GCC, mainly those required by the C

  3261.      standard of freestanding implementations.  *Note Headers Installed

  3262.      by GCC: Headers, for details of when these and other headers are

  3263.      installed.

  3264. 

  3265. 'po'

  3266.      Message catalogs with translations of messages produced by GCC into

  3267.      various languages, 'LANGUAGE.po'.  This directory also contains

  3268.      'gcc.pot', the template for these message catalogues, 'exgettext',

  3269.      a wrapper around 'gettext' to extract the messages from the GCC

  3270.      sources and create 'gcc.pot', which is run by 'make gcc.pot', and

  3271.      'EXCLUDES', a list of files from which messages should not be

  3272.      extracted.

  3273. 

  3274. 'testsuite'

  3275.      The GCC testsuites (except for those for runtime libraries).  *Note

  3276.      Testsuites::.

  3277. 

  3278. 

  3279. File: gccint.info,  Node: Configuration,  Next: Build,  Prev: Subdirectories,  Up: gcc Directory

  3280. 

  3281. 6.3.2 Configuration in the 'gcc' Directory

  3282. ------------------------------------------

  3283. 

  3284. The 'gcc' directory is configured with an Autoconf-generated script

  3285. 'configure'.  The 'configure' script is generated from 'configure.ac'

  3286. and 'aclocal.m4'.  From the files 'configure.ac' and 'acconfig.h',

  3287. Autoheader generates the file 'config.in'.  The file 'cstamp-h.in' is

  3288. used as a timestamp.

  3289. 

  3290. * Menu:

  3291. 

  3292. * Config Fragments::     Scripts used by 'configure'.

  3293. * System Config::        The 'config.build', 'config.host', and

  3294.                          'config.gcc' files.

  3295. * Configuration Files::  Files created by running 'configure'.

  3296. 

  3297. 

  3298. File: gccint.info,  Node: Config Fragments,  Next: System Config,  Up: Configuration

  3299. 

  3300. 6.3.2.1 Scripts Used by 'configure'

  3301. ...................................

  3302. 

  3303. 'configure' uses some other scripts to help in its work:

  3304. 

  3305.    * The standard GNU 'config.sub' and 'config.guess' files, kept in the

  3306.      top level directory, are used.

  3307. 

  3308.    * The file 'config.gcc' is used to handle configuration specific to

  3309.      the particular target machine.  The file 'config.build' is used to

  3310.      handle configuration specific to the particular build machine.  The

  3311.      file 'config.host' is used to handle configuration specific to the

  3312.      particular host machine.  (In general, these should only be used

  3313.      for features that cannot reasonably be tested in Autoconf feature

  3314.      tests.)  *Note The 'config.build'; 'config.host'; and 'config.gcc'

  3315.      Files: System Config, for details of the contents of these files.

  3316. 

  3317.    * Each language subdirectory has a file 'LANGUAGE/config-lang.in'

  3318.      that is used for front-end-specific configuration.  *Note The Front

  3319.      End 'config-lang.in' File: Front End Config, for details of this

  3320.      file.

  3321. 

  3322.    * A helper script 'configure.frag' is used as part of creating the

  3323.      output of 'configure'.

  3324. 

  3325. 

  3326. File: gccint.info,  Node: System Config,  Next: Configuration Files,  Prev: Config Fragments,  Up: Configuration

  3327. 

  3328. 6.3.2.2 The 'config.build'; 'config.host'; and 'config.gcc' Files

  3329. .................................................................

  3330. 

  3331. The 'config.build' file contains specific rules for particular systems

  3332. which GCC is built on.  This should be used as rarely as possible, as

  3333. the behavior of the build system can always be detected by autoconf.

  3334. 

  3335.  The 'config.host' file contains specific rules for particular systems

  3336. which GCC will run on.  This is rarely needed.

  3337. 

  3338.  The 'config.gcc' file contains specific rules for particular systems

  3339. which GCC will generate code for.  This is usually needed.

  3340. 

  3341.  Each file has a list of the shell variables it sets, with descriptions,

  3342. at the top of the file.

  3343. 

  3344.  FIXME: document the contents of these files, and what variables should

  3345. be set to control build, host and target configuration.

  3346. 

  3347. 

  3348. File: gccint.info,  Node: Configuration Files,  Prev: System Config,  Up: Configuration

  3349. 

  3350. 6.3.2.3 Files Created by 'configure'

  3351. ....................................

  3352. 

  3353. Here we spell out what files will be set up by 'configure' in the 'gcc'

  3354. directory.  Some other files are created as temporary files in the

  3355. configuration process, and are not used in the subsequent build; these

  3356. are not documented.

  3357. 

  3358.    * 'Makefile' is constructed from 'Makefile.in', together with the

  3359.      host and target fragments (*note Makefile Fragments: Fragments.)

  3360.      't-TARGET' and 'x-HOST' from 'config', if any, and language

  3361.      Makefile fragments 'LANGUAGE/Make-lang.in'.

  3362.    * 'auto-host.h' contains information about the host machine

  3363.      determined by 'configure'.  If the host machine is different from

  3364.      the build machine, then 'auto-build.h' is also created, containing

  3365.      such information about the build machine.

  3366.    * 'config.status' is a script that may be run to recreate the current

  3367.      configuration.

  3368.    * 'configargs.h' is a header containing details of the arguments

  3369.      passed to 'configure' to configure GCC, and of the thread model

  3370.      used.

  3371.    * 'cstamp-h' is used as a timestamp.

  3372.    * If a language 'config-lang.in' file (*note The Front End

  3373.      'config-lang.in' File: Front End Config.) sets 'outputs', then the

  3374.      files listed in 'outputs' there are also generated.

  3375. 

  3376.  The following configuration headers are created from the Makefile,

  3377. using 'mkconfig.sh', rather than directly by 'configure'.  'config.h',

  3378. 'bconfig.h' and 'tconfig.h' all contain the 'xm-MACHINE.h' header, if

  3379. any, appropriate to the host, build and target machines respectively,

  3380. the configuration headers for the target, and some definitions; for the

  3381. host and build machines, these include the autoconfigured headers

  3382. generated by 'configure'.  The other configuration headers are

  3383. determined by 'config.gcc'.  They also contain the typedefs for 'rtx',

  3384. 'rtvec' and 'tree'.

  3385. 

  3386.    * 'config.h', for use in programs that run on the host machine.

  3387.    * 'bconfig.h', for use in programs that run on the build machine.

  3388.    * 'tconfig.h', for use in programs and libraries for the target

  3389.      machine.

  3390.    * 'tm_p.h', which includes the header 'MACHINE-protos.h' that

  3391.      contains prototypes for functions in the target 'MACHINE.c' file.

  3392.      The 'MACHINE-protos.h' header is included after the 'rtl.h' and/or

  3393.      'tree.h' would have been included.  The 'tm_p.h' also includes the

  3394.      header 'tm-preds.h' which is generated by 'genpreds' program during

  3395.      the build to define the declarations and inline functions for the

  3396.      predicate functions.

  3397. 

  3398. 

  3399. File: gccint.info,  Node: Build,  Next: Makefile,  Prev: Configuration,  Up: gcc Directory

  3400. 

  3401. 6.3.3 Build System in the 'gcc' Directory

  3402. -----------------------------------------

  3403. 

  3404. FIXME: describe the build system, including what is built in what

  3405. stages.  Also list the various source files that are used in the build

  3406. process but aren't source files of GCC itself and so aren't documented

  3407. below (*note Passes::).

  3408. 

  3409. 

  3410. File: gccint.info,  Node: Makefile,  Next: Library Files,  Prev: Build,  Up: gcc Directory

  3411. 

  3412. 6.3.4 Makefile Targets

  3413. ----------------------

  3414. 

  3415. These targets are available from the 'gcc' directory:

  3416. 

  3417. 'all'

  3418.      This is the default target.  Depending on what your

  3419.      build/host/target configuration is, it coordinates all the things

  3420.      that need to be built.

  3421. 

  3422. 'doc'

  3423.      Produce info-formatted documentation and man pages.  Essentially it

  3424.      calls 'make man' and 'make info'.

  3425. 

  3426. 'dvi'

  3427.      Produce DVI-formatted documentation.

  3428. 

  3429. 'pdf'

  3430.      Produce PDF-formatted documentation.

  3431. 

  3432. 'html'

  3433.      Produce HTML-formatted documentation.

  3434. 

  3435. 'man'

  3436.      Generate man pages.

  3437. 

  3438. 'info'

  3439.      Generate info-formatted pages.

  3440. 

  3441. 'mostlyclean'

  3442.      Delete the files made while building the compiler.

  3443. 

  3444. 'clean'

  3445.      That, and all the other files built by 'make all'.

  3446. 

  3447. 'distclean'

  3448.      That, and all the files created by 'configure'.

  3449. 

  3450. 'maintainer-clean'

  3451.      Distclean plus any file that can be generated from other files.

  3452.      Note that additional tools may be required beyond what is normally

  3453.      needed to build GCC.

  3454. 

  3455. 'srcextra'

  3456.      Generates files in the source directory that are not

  3457.      version-controlled but should go into a release tarball.

  3458. 

  3459. 'srcinfo'

  3460. 'srcman'

  3461.      Copies the info-formatted and manpage documentation into the source

  3462.      directory usually for the purpose of generating a release tarball.

  3463. 

  3464. 'install'

  3465.      Installs GCC.

  3466. 

  3467. 'uninstall'

  3468.      Deletes installed files, though this is not supported.

  3469. 

  3470. 'check'

  3471.      Run the testsuite.  This creates a 'testsuite' subdirectory that

  3472.      has various '.sum' and '.log' files containing the results of the

  3473.      testing.  You can run subsets with, for example, 'make check-gcc'.

  3474.      You can specify specific tests by setting 'RUNTESTFLAGS' to be the

  3475.      name of the '.exp' file, optionally followed by (for some tests) an

  3476.      equals and a file wildcard, like:

  3477. 

  3478.           make check-gcc RUNTESTFLAGS="execute.exp=19980413-*"

  3479. 

  3480.      Note that running the testsuite may require additional tools be

  3481.      installed, such as Tcl or DejaGnu.

  3482. 

  3483.  The toplevel tree from which you start GCC compilation is not the GCC

  3484. directory, but rather a complex Makefile that coordinates the various

  3485. steps of the build, including bootstrapping the compiler and using the

  3486. new compiler to build target libraries.

  3487. 

  3488.  When GCC is configured for a native configuration, the default action

  3489. for 'make' is to do a full three-stage bootstrap.  This means that GCC

  3490. is built three times--once with the native compiler, once with the

  3491. native-built compiler it just built, and once with the compiler it built

  3492. the second time.  In theory, the last two should produce the same

  3493. results, which 'make compare' can check.  Each stage is configured

  3494. separately and compiled into a separate directory, to minimize problems

  3495. due to ABI incompatibilities between the native compiler and GCC.

  3496. 

  3497.  If you do a change, rebuilding will also start from the first stage and

  3498. "bubble" up the change through the three stages.  Each stage is taken

  3499. from its build directory (if it had been built previously), rebuilt, and

  3500. copied to its subdirectory.  This will allow you to, for example,

  3501. continue a bootstrap after fixing a bug which causes the stage2 build to

  3502. crash.  It does not provide as good coverage of the compiler as

  3503. bootstrapping from scratch, but it ensures that the new code is

  3504. syntactically correct (e.g., that you did not use GCC extensions by

  3505. mistake), and avoids spurious bootstrap comparison failures(1).

  3506. 

  3507.  Other targets available from the top level include:

  3508. 

  3509. 'bootstrap-lean'

  3510.      Like 'bootstrap', except that the various stages are removed once

  3511.      they're no longer needed.  This saves disk space.

  3512. 

  3513. 'bootstrap2'

  3514. 'bootstrap2-lean'

  3515.      Performs only the first two stages of bootstrap.  Unlike a

  3516.      three-stage bootstrap, this does not perform a comparison to test

  3517.      that the compiler is running properly.  Note that the disk space

  3518.      required by a "lean" bootstrap is approximately independent of the

  3519.      number of stages.

  3520. 

  3521. 'stageN-bubble (N = 1...4, profile, feedback)'

  3522.      Rebuild all the stages up to N, with the appropriate flags,

  3523.      "bubbling" the changes as described above.

  3524. 

  3525. 'all-stageN (N = 1...4, profile, feedback)'

  3526.      Assuming that stage N has already been built, rebuild it with the

  3527.      appropriate flags.  This is rarely needed.

  3528. 

  3529. 'cleanstrap'

  3530.      Remove everything ('make clean') and rebuilds ('make bootstrap').

  3531. 

  3532. 'compare'

  3533.      Compares the results of stages 2 and 3.  This ensures that the

  3534.      compiler is running properly, since it should produce the same

  3535.      object files regardless of how it itself was compiled.

  3536. 

  3537. 'profiledbootstrap'

  3538.      Builds a compiler with profiling feedback information.  In this

  3539.      case, the second and third stages are named 'profile' and

  3540.      'feedback', respectively.  For more information, see the

  3541.      installation instructions.

  3542. 

  3543. 'restrap'

  3544.      Restart a bootstrap, so that everything that was not built with the

  3545.      system compiler is rebuilt.

  3546. 

  3547. 'stageN-start (N = 1...4, profile, feedback)'

  3548.      For each package that is bootstrapped, rename directories so that,

  3549.      for example, 'gcc' points to the stageN GCC, compiled with the

  3550.      stageN-1 GCC(2).

  3551. 

  3552.      You will invoke this target if you need to test or debug the stageN

  3553.      GCC.  If you only need to execute GCC (but you need not run 'make'

  3554.      either to rebuild it or to run test suites), you should be able to

  3555.      work directly in the 'stageN-gcc' directory.  This makes it easier

  3556.      to debug multiple stages in parallel.

  3557. 

  3558. 'stage'

  3559.      For each package that is bootstrapped, relocate its build directory

  3560.      to indicate its stage.  For example, if the 'gcc' directory points

  3561.      to the stage2 GCC, after invoking this target it will be renamed to

  3562.      'stage2-gcc'.

  3563. 

  3564.  If you wish to use non-default GCC flags when compiling the stage2 and

  3565. stage3 compilers, set 'BOOT_CFLAGS' on the command line when doing

  3566. 'make'.

  3567. 

  3568.  Usually, the first stage only builds the languages that the compiler is

  3569. written in: typically, C and maybe Ada.  If you are debugging a

  3570. miscompilation of a different stage2 front-end (for example, of the

  3571. Fortran front-end), you may want to have front-ends for other languages

  3572. in the first stage as well.  To do so, set 'STAGE1_LANGUAGES' on the

  3573. command line when doing 'make'.

  3574. 

  3575.  For example, in the aforementioned scenario of debugging a Fortran

  3576. front-end miscompilation caused by the stage1 compiler, you may need a

  3577. command like

  3578. 

  3579.      make stage2-bubble STAGE1_LANGUAGES=c,fortran

  3580. 

  3581.  Alternatively, you can use per-language targets to build and test

  3582. languages that are not enabled by default in stage1.  For example, 'make

  3583. f951' will build a Fortran compiler even in the stage1 build directory.

  3584. 

  3585.    ---------- Footnotes ----------

  3586. 

  3587.    (1) Except if the compiler was buggy and miscompiled some of the

  3588. files that were not modified.  In this case, it's best to use 'make

  3589. restrap'.

  3590. 

  3591.    (2) Customarily, the system compiler is also termed the 'stage0' GCC.

  3592. 

  3593. 

  3594. File: gccint.info,  Node: Library Files,  Next: Headers,  Prev: Makefile,  Up: gcc Directory

  3595. 

  3596. 6.3.5 Library Source Files and Headers under the 'gcc' Directory

  3597. ----------------------------------------------------------------

  3598. 

  3599. FIXME: list here, with explanation, all the C source files and headers

  3600. under the 'gcc' directory that aren't built into the GCC executable but

  3601. rather are part of runtime libraries and object files, such as

  3602. 'crtstuff.c' and 'unwind-dw2.c'.  *Note Headers Installed by GCC:

  3603. Headers, for more information about the 'ginclude' directory.

  3604. 

  3605. 

  3606. File: gccint.info,  Node: Headers,  Next: Documentation,  Prev: Library Files,  Up: gcc Directory

  3607. 

  3608. 6.3.6 Headers Installed by GCC

  3609. ------------------------------

  3610. 

  3611. In general, GCC expects the system C library to provide most of the

  3612. headers to be used with it.  However, GCC will fix those headers if

  3613. necessary to make them work with GCC, and will install some headers

  3614. required of freestanding implementations.  These headers are installed

  3615. in 'LIBSUBDIR/include'.  Headers for non-C runtime libraries are also

  3616. installed by GCC; these are not documented here.  (FIXME: document them

  3617. somewhere.)

  3618. 

  3619.  Several of the headers GCC installs are in the 'ginclude' directory.

  3620. These headers, 'iso646.h', 'stdarg.h', 'stdbool.h', and 'stddef.h', are

  3621. installed in 'LIBSUBDIR/include', unless the target Makefile fragment

  3622. (*note Target Fragment::) overrides this by setting 'USER_H'.

  3623. 

  3624.  In addition to these headers and those generated by fixing system

  3625. headers to work with GCC, some other headers may also be installed in

  3626. 'LIBSUBDIR/include'.  'config.gcc' may set 'extra_headers'; this

  3627. specifies additional headers under 'config' to be installed on some

  3628. systems.

  3629. 

  3630.  GCC installs its own version of '<float.h>', from 'ginclude/float.h'.

  3631. This is done to cope with command-line options that change the

  3632. representation of floating point numbers.

  3633. 

  3634.  GCC also installs its own version of '<limits.h>'; this is generated

  3635. from 'glimits.h', together with 'limitx.h' and 'limity.h' if the system

  3636. also has its own version of '<limits.h>'.  (GCC provides its own header

  3637. because it is required of ISO C freestanding implementations, but needs

  3638. to include the system header from its own header as well because other

  3639. standards such as POSIX specify additional values to be defined in

  3640. '<limits.h>'.)  The system's '<limits.h>' header is used via

  3641. 'LIBSUBDIR/include/syslimits.h', which is copied from 'gsyslimits.h' if

  3642. it does not need fixing to work with GCC; if it needs fixing,

  3643. 'syslimits.h' is the fixed copy.

  3644. 

  3645.  GCC can also install '<tgmath.h>'.  It will do this when 'config.gcc'

  3646. sets 'use_gcc_tgmath' to 'yes'.

  3647. 

  3648. 

  3649. File: gccint.info,  Node: Documentation,  Next: Front End,  Prev: Headers,  Up: gcc Directory

  3650. 

  3651. 6.3.7 Building Documentation

  3652. ----------------------------

  3653. 

  3654. The main GCC documentation is in the form of manuals in Texinfo format.

  3655. These are installed in Info format; DVI versions may be generated by

  3656. 'make dvi', PDF versions by 'make pdf', and HTML versions by 'make

  3657. html'.  In addition, some man pages are generated from the Texinfo

  3658. manuals, there are some other text files with miscellaneous

  3659. documentation, and runtime libraries have their own documentation

  3660. outside the 'gcc' directory.  FIXME: document the documentation for

  3661. runtime libraries somewhere.

  3662. 

  3663. * Menu:

  3664. 

  3665. * Texinfo Manuals::      GCC manuals in Texinfo format.

  3666. * Man Page Generation::  Generating man pages from Texinfo manuals.

  3667. * Miscellaneous Docs::   Miscellaneous text files with documentation.

  3668. 

  3669. 

  3670. File: gccint.info,  Node: Texinfo Manuals,  Next: Man Page Generation,  Up: Documentation

  3671. 

  3672. 6.3.7.1 Texinfo Manuals

  3673. .......................

  3674. 

  3675. The manuals for GCC as a whole, and the C and C++ front ends, are in

  3676. files 'doc/*.texi'.  Other front ends have their own manuals in files

  3677. 'LANGUAGE/*.texi'.  Common files 'doc/include/*.texi' are provided which

  3678. may be included in multiple manuals; the following files are in

  3679. 'doc/include':

  3680. 

  3681. 'fdl.texi'

  3682.      The GNU Free Documentation License.

  3683. 'funding.texi'

  3684.      The section "Funding Free Software".

  3685. 'gcc-common.texi'

  3686.      Common definitions for manuals.

  3687. 'gpl_v3.texi'

  3688.      The GNU General Public License.

  3689. 'texinfo.tex'

  3690.      A copy of 'texinfo.tex' known to work with the GCC manuals.

  3691. 

  3692.  DVI-formatted manuals are generated by 'make dvi', which uses

  3693. 'texi2dvi' (via the Makefile macro '$(TEXI2DVI)').  PDF-formatted

  3694. manuals are generated by 'make pdf', which uses 'texi2pdf' (via the

  3695. Makefile macro '$(TEXI2PDF)').  HTML formatted manuals are generated by

  3696. 'make html'.  Info manuals are generated by 'make info' (which is run as

  3697. part of a bootstrap); this generates the manuals in the source

  3698. directory, using 'makeinfo' via the Makefile macro '$(MAKEINFO)', and

  3699. they are included in release distributions.

  3700. 

  3701.  Manuals are also provided on the GCC web site, in both HTML and

  3702. PostScript forms.  This is done via the script

  3703. 'maintainer-scripts/update_web_docs_git'.  Each manual to be provided

  3704. online must be listed in the definition of 'MANUALS' in that file; a

  3705. file 'NAME.texi' must only appear once in the source tree, and the

  3706. output manual must have the same name as the source file.  (However,

  3707. other Texinfo files, included in manuals but not themselves the root

  3708. files of manuals, may have names that appear more than once in the

  3709. source tree.)  The manual file 'NAME.texi' should only include other

  3710. files in its own directory or in 'doc/include'.  HTML manuals will be

  3711. generated by 'makeinfo --html', PostScript manuals by 'texi2dvi' and

  3712. 'dvips', and PDF manuals by 'texi2pdf'.  All Texinfo files that are

  3713. parts of manuals must be version-controlled, even if they are generated

  3714. files, for the generation of online manuals to work.

  3715. 

  3716.  The installation manual, 'doc/install.texi', is also provided on the

  3717. GCC web site.  The HTML version is generated by the script

  3718. 'doc/install.texi2html'.

  3719. 

  3720. 

  3721. File: gccint.info,  Node: Man Page Generation,  Next: Miscellaneous Docs,  Prev: Texinfo Manuals,  Up: Documentation

  3722. 

  3723. 6.3.7.2 Man Page Generation

  3724. ...........................

  3725. 

  3726. Because of user demand, in addition to full Texinfo manuals, man pages

  3727. are provided which contain extracts from those manuals.  These man pages

  3728. are generated from the Texinfo manuals using 'contrib/texi2pod.pl' and

  3729. 'pod2man'.  (The man page for 'g++', 'cp/g++.1', just contains a '.so'

  3730. reference to 'gcc.1', but all the other man pages are generated from

  3731. Texinfo manuals.)

  3732. 

  3733.  Because many systems may not have the necessary tools installed to

  3734. generate the man pages, they are only generated if the 'configure'

  3735. script detects that recent enough tools are installed, and the Makefiles

  3736. allow generating man pages to fail without aborting the build.  Man

  3737. pages are also included in release distributions.  They are generated in

  3738. the source directory.

  3739. 

  3740.  Magic comments in Texinfo files starting '@c man' control what parts of

  3741. a Texinfo file go into a man page.  Only a subset of Texinfo is

  3742. supported by 'texi2pod.pl', and it may be necessary to add support for

  3743. more Texinfo features to this script when generating new man pages.  To

  3744. improve the man page output, some special Texinfo macros are provided in

  3745. 'doc/include/gcc-common.texi' which 'texi2pod.pl' understands:

  3746. 

  3747. '@gcctabopt'

  3748.      Use in the form '@table @gcctabopt' for tables of options, where

  3749.      for printed output the effect of '@code' is better than that of

  3750.      '@option' but for man page output a different effect is wanted.

  3751. '@gccoptlist'

  3752.      Use for summary lists of options in manuals.

  3753. '@gol'

  3754.      Use at the end of each line inside '@gccoptlist'.  This is

  3755.      necessary to avoid problems with differences in how the

  3756.      '@gccoptlist' macro is handled by different Texinfo formatters.

  3757. 

  3758.  FIXME: describe the 'texi2pod.pl' input language and magic comments in

  3759. more detail.

  3760. 

  3761. 

  3762. File: gccint.info,  Node: Miscellaneous Docs,  Prev: Man Page Generation,  Up: Documentation

  3763. 

  3764. 6.3.7.3 Miscellaneous Documentation

  3765. ...................................

  3766. 

  3767. In addition to the formal documentation that is installed by GCC, there

  3768. are several other text files in the 'gcc' subdirectory with

  3769. miscellaneous documentation:

  3770. 

  3771. 'ABOUT-GCC-NLS'

  3772.      Notes on GCC's Native Language Support.  FIXME: this should be part

  3773.      of this manual rather than a separate file.

  3774. 'ABOUT-NLS'

  3775.      Notes on the Free Translation Project.

  3776. 'COPYING'

  3777. 'COPYING3'

  3778.      The GNU General Public License, Versions 2 and 3.

  3779. 'COPYING.LIB'

  3780. 'COPYING3.LIB'

  3781.      The GNU Lesser General Public License, Versions 2.1 and 3.

  3782. '*ChangeLog*'

  3783. '*/ChangeLog*'

  3784.      Change log files for various parts of GCC.

  3785. 'LANGUAGES'

  3786.      Details of a few changes to the GCC front-end interface.  FIXME:

  3787.      the information in this file should be part of general

  3788.      documentation of the front-end interface in this manual.

  3789. 'ONEWS'

  3790.      Information about new features in old versions of GCC.  (For recent

  3791.      versions, the information is on the GCC web site.)

  3792. 'README.Portability'

  3793.      Information about portability issues when writing code in GCC.

  3794.      FIXME: why isn't this part of this manual or of the GCC Coding

  3795.      Conventions?

  3796. 

  3797.  FIXME: document such files in subdirectories, at least 'config', 'c',

  3798. 'cp', 'objc', 'testsuite'.

  3799. 

  3800. 

  3801. File: gccint.info,  Node: Front End,  Next: Back End,  Prev: Documentation,  Up: gcc Directory

  3802. 

  3803. 6.3.8 Anatomy of a Language Front End

  3804. -------------------------------------

  3805. 

  3806. A front end for a language in GCC has the following parts:

  3807. 

  3808.    * A directory 'LANGUAGE' under 'gcc' containing source files for that

  3809.      front end.  *Note The Front End 'LANGUAGE' Directory: Front End

  3810.      Directory, for details.

  3811.    * A mention of the language in the list of supported languages in

  3812.      'gcc/doc/install.texi'.

  3813.    * A mention of the name under which the language's runtime library is

  3814.      recognized by '--enable-shared=PACKAGE' in the documentation of

  3815.      that option in 'gcc/doc/install.texi'.

  3816.    * A mention of any special prerequisites for building the front end

  3817.      in the documentation of prerequisites in 'gcc/doc/install.texi'.

  3818.    * Details of contributors to that front end in

  3819.      'gcc/doc/contrib.texi'.  If the details are in that front end's own

  3820.      manual then there should be a link to that manual's list in

  3821.      'contrib.texi'.

  3822.    * Information about support for that language in

  3823.      'gcc/doc/frontends.texi'.

  3824.    * Information about standards for that language, and the front end's

  3825.      support for them, in 'gcc/doc/standards.texi'.  This may be a link

  3826.      to such information in the front end's own manual.

  3827.    * Details of source file suffixes for that language and '-x LANG'

  3828.      options supported, in 'gcc/doc/invoke.texi'.

  3829.    * Entries in 'default_compilers' in 'gcc.c' for source file suffixes

  3830.      for that language.

  3831.    * Preferably testsuites, which may be under 'gcc/testsuite' or

  3832.      runtime library directories.  FIXME: document somewhere how to

  3833.      write testsuite harnesses.

  3834.    * Probably a runtime library for the language, outside the 'gcc'

  3835.      directory.  FIXME: document this further.

  3836.    * Details of the directories of any runtime libraries in

  3837.      'gcc/doc/sourcebuild.texi'.

  3838.    * Check targets in 'Makefile.def' for the top-level 'Makefile' to

  3839.      check just the compiler or the compiler and runtime library for the

  3840.      language.

  3841. 

  3842.  If the front end is added to the official GCC source repository, the

  3843. following are also necessary:

  3844. 

  3845.    * At least one Bugzilla component for bugs in that front end and

  3846.      runtime libraries.  This category needs to be added to the Bugzilla

  3847.      database.

  3848.    * Normally, one or more maintainers of that front end listed in

  3849.      'MAINTAINERS'.

  3850.    * Mentions on the GCC web site in 'index.html' and 'frontends.html',

  3851.      with any relevant links on 'readings.html'.  (Front ends that are

  3852.      not an official part of GCC may also be listed on 'frontends.html',

  3853.      with relevant links.)

  3854.    * A news item on 'index.html', and possibly an announcement on the

  3855.      <gcc-announce@gcc.gnu.org> mailing list.

  3856.    * The front end's manuals should be mentioned in

  3857.      'maintainer-scripts/update_web_docs_git' (*note Texinfo Manuals::)

  3858.      and the online manuals should be linked to from

  3859.      'onlinedocs/index.html'.

  3860.    * Any old releases or CVS repositories of the front end, before its

  3861.      inclusion in GCC, should be made available on the GCC web site at

  3862.      <https://gcc.gnu.org/pub/gcc/old-releases/>.

  3863.    * The release and snapshot script 'maintainer-scripts/gcc_release'

  3864.      should be updated to generate appropriate tarballs for this front

  3865.      end.

  3866.    * If this front end includes its own version files that include the

  3867.      current date, 'maintainer-scripts/update_version' should be updated

  3868.      accordingly.

  3869. 

  3870. * Menu:

  3871. 

  3872. * Front End Directory::  The front end 'LANGUAGE' directory.

  3873. * Front End Config::     The front end 'config-lang.in' file.

  3874. * Front End Makefile::   The front end 'Make-lang.in' file.

  3875. 

  3876. 

  3877. File: gccint.info,  Node: Front End Directory,  Next: Front End Config,  Up: Front End

  3878. 

  3879. 6.3.8.1 The Front End 'LANGUAGE' Directory

  3880. ..........................................

  3881. 

  3882. A front end 'LANGUAGE' directory contains the source files of that front

  3883. end (but not of any runtime libraries, which should be outside the 'gcc'

  3884. directory).  This includes documentation, and possibly some subsidiary

  3885. programs built alongside the front end.  Certain files are special and

  3886. other parts of the compiler depend on their names:

  3887. 

  3888. 'config-lang.in'

  3889.      This file is required in all language subdirectories.  *Note The

  3890.      Front End 'config-lang.in' File: Front End Config, for details of

  3891.      its contents

  3892. 'Make-lang.in'

  3893.      This file is required in all language subdirectories.  *Note The

  3894.      Front End 'Make-lang.in' File: Front End Makefile, for details of

  3895.      its contents.

  3896. 'lang.opt'

  3897.      This file registers the set of switches that the front end accepts

  3898.      on the command line, and their '--help' text.  *Note Options::.

  3899. 'lang-specs.h'

  3900.      This file provides entries for 'default_compilers' in 'gcc.c' which

  3901.      override the default of giving an error that a compiler for that

  3902.      language is not installed.

  3903. 'LANGUAGE-tree.def'

  3904.      This file, which need not exist, defines any language-specific tree

  3905.      codes.

  3906. 

  3907. 

  3908. File: gccint.info,  Node: Front End Config,  Next: Front End Makefile,  Prev: Front End Directory,  Up: Front End

  3909. 

  3910. 6.3.8.2 The Front End 'config-lang.in' File

  3911. ...........................................

  3912. 

  3913. Each language subdirectory contains a 'config-lang.in' file.  This file

  3914. is a shell script that may define some variables describing the

  3915. language:

  3916. 

  3917. 'language'

  3918.      This definition must be present, and gives the name of the language

  3919.      for some purposes such as arguments to '--enable-languages'.

  3920. 'lang_requires'

  3921.      If defined, this variable lists (space-separated) language front

  3922.      ends other than C that this front end requires to be enabled (with

  3923.      the names given being their 'language' settings).  For example, the

  3924.      Obj-C++ front end depends on the C++ and ObjC front ends, so sets

  3925.      'lang_requires="objc c++"'.

  3926. 'subdir_requires'

  3927.      If defined, this variable lists (space-separated) front end

  3928.      directories other than C that this front end requires to be

  3929.      present.  For example, the Objective-C++ front end uses source

  3930.      files from the C++ and Objective-C front ends, so sets

  3931.      'subdir_requires="cp objc"'.

  3932. 'target_libs'

  3933.      If defined, this variable lists (space-separated) targets in the

  3934.      top level 'Makefile' to build the runtime libraries for this

  3935.      language, such as 'target-libobjc'.

  3936. 'lang_dirs'

  3937.      If defined, this variable lists (space-separated) top level

  3938.      directories (parallel to 'gcc'), apart from the runtime libraries,

  3939.      that should not be configured if this front end is not built.

  3940. 'build_by_default'

  3941.      If defined to 'no', this language front end is not built unless

  3942.      enabled in a '--enable-languages' argument.  Otherwise, front ends

  3943.      are built by default, subject to any special logic in

  3944.      'configure.ac' (as is present to disable the Ada front end if the

  3945.      Ada compiler is not already installed).

  3946. 'boot_language'

  3947.      If defined to 'yes', this front end is built in stage1 of the

  3948.      bootstrap.  This is only relevant to front ends written in their

  3949.      own languages.

  3950. 'compilers'

  3951.      If defined, a space-separated list of compiler executables that

  3952.      will be run by the driver.  The names here will each end with

  3953.      '\$(exeext)'.

  3954. 'outputs'

  3955.      If defined, a space-separated list of files that should be

  3956.      generated by 'configure' substituting values in them.  This

  3957.      mechanism can be used to create a file 'LANGUAGE/Makefile' from

  3958.      'LANGUAGE/Makefile.in', but this is deprecated, building everything

  3959.      from the single 'gcc/Makefile' is preferred.

  3960. 'gtfiles'

  3961.      If defined, a space-separated list of files that should be scanned

  3962.      by 'gengtype.c' to generate the garbage collection tables and

  3963.      routines for this language.  This excludes the files that are

  3964.      common to all front ends.  *Note Type Information::.

  3965. 

  3966. 

  3967. File: gccint.info,  Node: Front End Makefile,  Prev: Front End Config,  Up: Front End

  3968. 

  3969. 6.3.8.3 The Front End 'Make-lang.in' File

  3970. .........................................

  3971. 

  3972. Each language subdirectory contains a 'Make-lang.in' file.  It contains

  3973. targets 'LANG.HOOK' (where 'LANG' is the setting of 'language' in

  3974. 'config-lang.in') for the following values of 'HOOK', and any other

  3975. Makefile rules required to build those targets (which may if necessary

  3976. use other Makefiles specified in 'outputs' in 'config-lang.in', although

  3977. this is deprecated).  It also adds any testsuite targets that can use

  3978. the standard rule in 'gcc/Makefile.in' to the variable 'lang_checks'.

  3979. 

  3980. 'all.cross'

  3981. 'start.encap'

  3982. 'rest.encap'

  3983.      FIXME: exactly what goes in each of these targets?

  3984. 'tags'

  3985.      Build an 'etags' 'TAGS' file in the language subdirectory in the

  3986.      source tree.

  3987. 'info'

  3988.      Build info documentation for the front end, in the build directory.

  3989.      This target is only called by 'make bootstrap' if a suitable

  3990.      version of 'makeinfo' is available, so does not need to check for

  3991.      this, and should fail if an error occurs.

  3992. 'dvi'

  3993.      Build DVI documentation for the front end, in the build directory.

  3994.      This should be done using '$(TEXI2DVI)', with appropriate '-I'

  3995.      arguments pointing to directories of included files.

  3996. 'pdf'

  3997.      Build PDF documentation for the front end, in the build directory.

  3998.      This should be done using '$(TEXI2PDF)', with appropriate '-I'

  3999.      arguments pointing to directories of included files.

  4000. 'html'

  4001.      Build HTML documentation for the front end, in the build directory.

  4002. 'man'

  4003.      Build generated man pages for the front end from Texinfo manuals

  4004.      (*note Man Page Generation::), in the build directory.  This target

  4005.      is only called if the necessary tools are available, but should

  4006.      ignore errors so as not to stop the build if errors occur; man

  4007.      pages are optional and the tools involved may be installed in a

  4008.      broken way.

  4009. 'install-common'

  4010.      Install everything that is part of the front end, apart from the

  4011.      compiler executables listed in 'compilers' in 'config-lang.in'.

  4012. 'install-info'

  4013.      Install info documentation for the front end, if it is present in

  4014.      the source directory.  This target should have dependencies on info

  4015.      files that should be installed.

  4016. 'install-man'

  4017.      Install man pages for the front end.  This target should ignore

  4018.      errors.

  4019. 'install-plugin'

  4020.      Install headers needed for plugins.

  4021. 'srcextra'

  4022.      Copies its dependencies into the source directory.  This generally

  4023.      should be used for generated files such as Bison output files which

  4024.      are not version-controlled, but should be included in any release

  4025.      tarballs.  This target will be executed during a bootstrap if

  4026.      '--enable-generated-files-in-srcdir' was specified as a 'configure'

  4027.      option.

  4028. 'srcinfo'

  4029. 'srcman'

  4030.      Copies its dependencies into the source directory.  These targets

  4031.      will be executed during a bootstrap if

  4032.      '--enable-generated-files-in-srcdir' was specified as a 'configure'

  4033.      option.

  4034. 'uninstall'

  4035.      Uninstall files installed by installing the compiler.  This is

  4036.      currently documented not to be supported, so the hook need not do

  4037.      anything.

  4038. 'mostlyclean'

  4039. 'clean'

  4040. 'distclean'

  4041. 'maintainer-clean'

  4042.      The language parts of the standard GNU '*clean' targets.  *Note

  4043.      Standard Targets for Users: (standards)Standard Targets, for

  4044.      details of the standard targets.  For GCC, 'maintainer-clean'

  4045.      should delete all generated files in the source directory that are

  4046.      not version-controlled, but should not delete anything that is.

  4047. 

  4048.  'Make-lang.in' must also define a variable 'LANG_OBJS' to a list of

  4049. host object files that are used by that language.

  4050. 

  4051. 

  4052. File: gccint.info,  Node: Back End,  Prev: Front End,  Up: gcc Directory

  4053. 

  4054. 6.3.9 Anatomy of a Target Back End

  4055. ----------------------------------

  4056. 

  4057. A back end for a target architecture in GCC has the following parts:

  4058. 

  4059.    * A directory 'MACHINE' under 'gcc/config', containing a machine

  4060.      description 'MACHINE.md' file (*note Machine Descriptions: Machine

  4061.      Desc.), header files 'MACHINE.h' and 'MACHINE-protos.h' and a

  4062.      source file 'MACHINE.c' (*note Target Description Macros and

  4063.      Functions: Target Macros.), possibly a target Makefile fragment

  4064.      't-MACHINE' (*note The Target Makefile Fragment: Target Fragment.),

  4065.      and maybe some other files.  The names of these files may be

  4066.      changed from the defaults given by explicit specifications in

  4067.      'config.gcc'.

  4068.    * If necessary, a file 'MACHINE-modes.def' in the 'MACHINE'

  4069.      directory, containing additional machine modes to represent

  4070.      condition codes.  *Note Condition Code::, for further details.

  4071.    * An optional 'MACHINE.opt' file in the 'MACHINE' directory,

  4072.      containing a list of target-specific options.  You can also add

  4073.      other option files using the 'extra_options' variable in

  4074.      'config.gcc'.  *Note Options::.

  4075.    * Entries in 'config.gcc' (*note The 'config.gcc' File: System

  4076.      Config.) for the systems with this target architecture.

  4077.    * Documentation in 'gcc/doc/invoke.texi' for any command-line options

  4078.      supported by this target (*note Run-time Target Specification:

  4079.      Run-time Target.).  This means both entries in the summary table of

  4080.      options and details of the individual options.

  4081.    * Documentation in 'gcc/doc/extend.texi' for any target-specific

  4082.      attributes supported (*note Defining target-specific uses of

  4083.      '__attribute__': Target Attributes.), including where the same

  4084.      attribute is already supported on some targets, which are

  4085.      enumerated in the manual.

  4086.    * Documentation in 'gcc/doc/extend.texi' for any target-specific

  4087.      pragmas supported.

  4088.    * Documentation in 'gcc/doc/extend.texi' of any target-specific

  4089.      built-in functions supported.

  4090.    * Documentation in 'gcc/doc/extend.texi' of any target-specific

  4091.      format checking styles supported.

  4092.    * Documentation in 'gcc/doc/md.texi' of any target-specific

  4093.      constraint letters (*note Constraints for Particular Machines:

  4094.      Machine Constraints.).

  4095.    * A note in 'gcc/doc/contrib.texi' under the person or people who

  4096.      contributed the target support.

  4097.    * Entries in 'gcc/doc/install.texi' for all target triplets supported

  4098.      with this target architecture, giving details of any special notes

  4099.      about installation for this target, or saying that there are no

  4100.      special notes if there are none.

  4101.    * Possibly other support outside the 'gcc' directory for runtime

  4102.      libraries.  FIXME: reference docs for this.  The 'libstdc++'

  4103.      porting manual needs to be installed as info for this to work, or

  4104.      to be a chapter of this manual.

  4105. 

  4106.  The 'MACHINE.h' header is included very early in GCC's standard

  4107. sequence of header files, while 'MACHINE-protos.h' is included late in

  4108. the sequence.  Thus 'MACHINE-protos.h' can include declarations

  4109. referencing types that are not defined when 'MACHINE.h' is included,

  4110. specifically including those from 'rtl.h' and 'tree.h'.  Since both RTL

  4111. and tree types may not be available in every context where

  4112. 'MACHINE-protos.h' is included, in this file you should guard

  4113. declarations using these types inside appropriate '#ifdef RTX_CODE' or

  4114. '#ifdef TREE_CODE' conditional code segments.

  4115. 

  4116.  If the backend uses shared data structures that require 'GTY' markers

  4117. for garbage collection (*note Type Information::), you must declare

  4118. those in 'MACHINE.h' rather than 'MACHINE-protos.h'.  Any definitions

  4119. required for building libgcc must also go in 'MACHINE.h'.

  4120. 

  4121.  GCC uses the macro 'IN_TARGET_CODE' to distinguish between

  4122. machine-specific '.c' and '.cc' files and machine-independent '.c' and

  4123. '.cc' files.  Machine-specific files should use the directive:

  4124. 

  4125.      #define IN_TARGET_CODE 1

  4126. 

  4127.  before including 'config.h'.

  4128. 

  4129.  If the back end is added to the official GCC source repository, the

  4130. following are also necessary:

  4131. 

  4132.    * An entry for the target architecture in 'readings.html' on the GCC

  4133.      web site, with any relevant links.

  4134.    * Details of the properties of the back end and target architecture

  4135.      in 'backends.html' on the GCC web site.

  4136.    * A news item about the contribution of support for that target

  4137.      architecture, in 'index.html' on the GCC web site.

  4138.    * Normally, one or more maintainers of that target listed in

  4139.      'MAINTAINERS'.  Some existing architectures may be unmaintained,

  4140.      but it would be unusual to add support for a target that does not

  4141.      have a maintainer when support is added.

  4142.    * Target triplets covering all 'config.gcc' stanzas for the target,

  4143.      in the list in 'contrib/config-list.mk'.

  4144. 

  4145. 

  4146. File: gccint.info,  Node: Testsuites,  Next: Options,  Prev: Source Tree,  Up: Top

  4147. 

  4148. 7 Testsuites

  4149. ************

  4150. 

  4151. GCC contains several testsuites to help maintain compiler quality.  Most

  4152. of the runtime libraries and language front ends in GCC have testsuites.

  4153. Currently only the C language testsuites are documented here; FIXME:

  4154. document the others.

  4155. 

  4156. * Menu:

  4157. 

  4158. * Test Idioms::     Idioms used in testsuite code.

  4159. * Test Directives:: Directives used within DejaGnu tests.

  4160. * Ada Tests::       The Ada language testsuites.

  4161. * C Tests::         The C language testsuites.

  4162. * LTO Testing::     Support for testing link-time optimizations.

  4163. * gcov Testing::    Support for testing gcov.

  4164. * profopt Testing:: Support for testing profile-directed optimizations.

  4165. * compat Testing::  Support for testing binary compatibility.

  4166. * Torture Tests::   Support for torture testing using multiple options.

  4167. * GIMPLE Tests::    Support for testing GIMPLE passes.

  4168. * RTL Tests::       Support for testing RTL passes.

  4169. 

  4170. 

  4171. File: gccint.info,  Node: Test Idioms,  Next: Test Directives,  Up: Testsuites

  4172. 

  4173. 7.1 Idioms Used in Testsuite Code

  4174. =================================

  4175. 

  4176. In general, C testcases have a trailing '-N.c', starting with '-1.c', in

  4177. case other testcases with similar names are added later.  If the test is

  4178. a test of some well-defined feature, it should have a name referring to

  4179. that feature such as 'FEATURE-1.c'.  If it does not test a well-defined

  4180. feature but just happens to exercise a bug somewhere in the compiler,

  4181. and a bug report has been filed for this bug in the GCC bug database,

  4182. 'prBUG-NUMBER-1.c' is the appropriate form of name.  Otherwise (for

  4183. miscellaneous bugs not filed in the GCC bug database), and previously

  4184. more generally, test cases are named after the date on which they were

  4185. added.  This allows people to tell at a glance whether a test failure is

  4186. because of a recently found bug that has not yet been fixed, or whether

  4187. it may be a regression, but does not give any other information about

  4188. the bug or where discussion of it may be found.  Some other language

  4189. testsuites follow similar conventions.

  4190. 

  4191.  In the 'gcc.dg' testsuite, it is often necessary to test that an error

  4192. is indeed a hard error and not just a warning--for example, where it is

  4193. a constraint violation in the C standard, which must become an error

  4194. with '-pedantic-errors'.  The following idiom, where the first line

  4195. shown is line LINE of the file and the line that generates the error, is

  4196. used for this:

  4197. 

  4198.      /* { dg-bogus "warning" "warning in place of error" } */

  4199.      /* { dg-error "REGEXP" "MESSAGE" { target *-*-* } LINE } */

  4200. 

  4201.  It may be necessary to check that an expression is an integer constant

  4202. expression and has a certain value.  To check that 'E' has value 'V', an

  4203. idiom similar to the following is used:

  4204. 

  4205.      char x[((E) == (V) ? 1 : -1)];

  4206. 

  4207.  In 'gcc.dg' tests, '__typeof__' is sometimes used to make assertions

  4208. about the types of expressions.  See, for example,

  4209. 'gcc.dg/c99-condexpr-1.c'.  The more subtle uses depend on the exact

  4210. rules for the types of conditional expressions in the C standard; see,

  4211. for example, 'gcc.dg/c99-intconst-1.c'.

  4212. 

  4213.  It is useful to be able to test that optimizations are being made

  4214. properly.  This cannot be done in all cases, but it can be done where

  4215. the optimization will lead to code being optimized away (for example,

  4216. where flow analysis or alias analysis should show that certain code

  4217. cannot be called) or to functions not being called because they have

  4218. been expanded as built-in functions.  Such tests go in

  4219. 'gcc.c-torture/execute'.  Where code should be optimized away, a call to

  4220. a nonexistent function such as 'link_failure ()' may be inserted; a

  4221. definition

  4222. 

  4223.      #ifndef __OPTIMIZE__

  4224.      void

  4225.      link_failure (void)

  4226.      {

  4227.        abort ();

  4228.      }

  4229.      #endif

  4230. 

  4231. will also be needed so that linking still succeeds when the test is run

  4232. without optimization.  When all calls to a built-in function should have

  4233. been optimized and no calls to the non-built-in version of the function

  4234. should remain, that function may be defined as 'static' to call 'abort

  4235. ()' (although redeclaring a function as static may not work on all

  4236. targets).

  4237. 

  4238.  All testcases must be portable.  Target-specific testcases must have

  4239. appropriate code to avoid causing failures on unsupported systems;

  4240. unfortunately, the mechanisms for this differ by directory.

  4241. 

  4242.  FIXME: discuss non-C testsuites here.

  4243. 

  4244. 

  4245. File: gccint.info,  Node: Test Directives,  Next: Ada Tests,  Prev: Test Idioms,  Up: Testsuites

  4246. 

  4247. 7.2 Directives used within DejaGnu tests

  4248. ========================================

  4249. 

  4250. * Menu:

  4251. 

  4252. * Directives::  Syntax and descriptions of test directives.

  4253. * Selectors:: Selecting targets to which a test applies.

  4254. * Effective-Target Keywords:: Keywords describing target attributes.

  4255. * Add Options:: Features for 'dg-add-options'

  4256. * Require Support:: Variants of 'dg-require-SUPPORT'

  4257. * Final Actions:: Commands for use in 'dg-final'

  4258. 

  4259. 

  4260. File: gccint.info,  Node: Directives,  Next: Selectors,  Up: Test Directives

  4261. 

  4262. 7.2.1 Syntax and Descriptions of test directives

  4263. ------------------------------------------------

  4264. 

  4265. Test directives appear within comments in a test source file and begin

  4266. with 'dg-'.  Some of these are defined within DejaGnu and others are

  4267. local to the GCC testsuite.

  4268. 

  4269.  The order in which test directives appear in a test can be important:

  4270. directives local to GCC sometimes override information used by the

  4271. DejaGnu directives, which know nothing about the GCC directives, so the

  4272. DejaGnu directives must precede GCC directives.

  4273. 

  4274.  Several test directives include selectors (*note Selectors::) which are

  4275. usually preceded by the keyword 'target' or 'xfail'.

  4276. 

  4277. 7.2.1.1 Specify how to build the test

  4278. .....................................

  4279. 

  4280. '{ dg-do DO-WHAT-KEYWORD [{ target/xfail SELECTOR }] }'

  4281.      DO-WHAT-KEYWORD specifies how the test is compiled and whether it

  4282.      is executed.  It is one of:

  4283. 

  4284.      'preprocess'

  4285.           Compile with '-E' to run only the preprocessor.

  4286.      'compile'

  4287.           Compile with '-S' to produce an assembly code file.

  4288.      'assemble'

  4289.           Compile with '-c' to produce a relocatable object file.

  4290.      'link'

  4291.           Compile, assemble, and link to produce an executable file.

  4292.      'run'

  4293.           Produce and run an executable file, which is expected to

  4294.           return an exit code of 0.

  4295. 

  4296.      The default is 'compile'.  That can be overridden for a set of

  4297.      tests by redefining 'dg-do-what-default' within the '.exp' file for

  4298.      those tests.

  4299. 

  4300.      If the directive includes the optional '{ target SELECTOR }' then

  4301.      the test is skipped unless the target system matches the SELECTOR.

  4302. 

  4303.      If DO-WHAT-KEYWORD is 'run' and the directive includes the optional

  4304.      '{ xfail SELECTOR }' and the selector is met then the test is

  4305.      expected to fail.  The 'xfail' clause is ignored for other values

  4306.      of DO-WHAT-KEYWORD; those tests can use directive 'dg-xfail-if'.

  4307. 

  4308. 7.2.1.2 Specify additional compiler options

  4309. ...........................................

  4310. 

  4311. '{ dg-options OPTIONS [{ target SELECTOR }] }'

  4312.      This DejaGnu directive provides a list of compiler options, to be

  4313.      used if the target system matches SELECTOR, that replace the

  4314.      default options used for this set of tests.

  4315. 

  4316. '{ dg-add-options FEATURE ... }'

  4317.      Add any compiler options that are needed to access certain

  4318.      features.  This directive does nothing on targets that enable the

  4319.      features by default, or that don't provide them at all.  It must

  4320.      come after all 'dg-options' directives.  For supported values of

  4321.      FEATURE see *note Add Options::.

  4322. 

  4323. '{ dg-additional-options OPTIONS [{ target SELECTOR }] }'

  4324.      This directive provides a list of compiler options, to be used if

  4325.      the target system matches SELECTOR, that are added to the default

  4326.      options used for this set of tests.

  4327. 

  4328. 7.2.1.3 Modify the test timeout value

  4329. .....................................

  4330. 

  4331. The normal timeout limit, in seconds, is found by searching the

  4332. following in order:

  4333. 

  4334.    * the value defined by an earlier 'dg-timeout' directive in the test

  4335. 

  4336.    * variable TOOL_TIMEOUT defined by the set of tests

  4337. 

  4338.    * GCC,TIMEOUT set in the target board

  4339. 

  4340.    * 300

  4341. 

  4342. '{ dg-timeout N [{target SELECTOR }] }'

  4343.      Set the time limit for the compilation and for the execution of the

  4344.      test to the specified number of seconds.

  4345. 

  4346. '{ dg-timeout-factor X [{ target SELECTOR }] }'

  4347.      Multiply the normal time limit for compilation and execution of the

  4348.      test by the specified floating-point factor.

  4349. 

  4350. 7.2.1.4 Skip a test for some targets

  4351. ....................................

  4352. 

  4353. '{ dg-skip-if COMMENT { SELECTOR } [{ INCLUDE-OPTS } [{ EXCLUDE-OPTS }]] }'

  4354.      Arguments INCLUDE-OPTS and EXCLUDE-OPTS are lists in which each

  4355.      element is a string of zero or more GCC options.  Skip the test if

  4356.      all of the following conditions are met:

  4357.         * the test system is included in SELECTOR

  4358. 

  4359.         * for at least one of the option strings in INCLUDE-OPTS, every

  4360.           option from that string is in the set of options with which

  4361.           the test would be compiled; use '"*"' for an INCLUDE-OPTS list

  4362.           that matches any options; that is the default if INCLUDE-OPTS

  4363.           is not specified

  4364. 

  4365.         * for each of the option strings in EXCLUDE-OPTS, at least one

  4366.           option from that string is not in the set of options with

  4367.           which the test would be compiled; use '""' for an empty

  4368.           EXCLUDE-OPTS list; that is the default if EXCLUDE-OPTS is not

  4369.           specified

  4370. 

  4371.      For example, to skip a test if option '-Os' is present:

  4372. 

  4373.           /* { dg-skip-if "" { *-*-* }  { "-Os" } { "" } } */

  4374. 

  4375.      To skip a test if both options '-O2' and '-g' are present:

  4376. 

  4377.           /* { dg-skip-if "" { *-*-* }  { "-O2 -g" } { "" } } */

  4378. 

  4379.      To skip a test if either '-O2' or '-O3' is present:

  4380. 

  4381.           /* { dg-skip-if "" { *-*-* }  { "-O2" "-O3" } { "" } } */

  4382. 

  4383.      To skip a test unless option '-Os' is present:

  4384. 

  4385.           /* { dg-skip-if "" { *-*-* }  { "*" } { "-Os" } } */

  4386. 

  4387.      To skip a test if either '-O2' or '-O3' is used with '-g' but not

  4388.      if '-fpic' is also present:

  4389. 

  4390.           /* { dg-skip-if "" { *-*-* }  { "-O2 -g" "-O3 -g" } { "-fpic" } } */

  4391. 

  4392. '{ dg-require-effective-target KEYWORD [{ SELECTOR }] }'

  4393.      Skip the test if the test target, including current multilib flags,

  4394.      is not covered by the effective-target keyword.  If the directive

  4395.      includes the optional '{ SELECTOR }' then the effective-target test

  4396.      is only performed if the target system matches the SELECTOR.  This

  4397.      directive must appear after any 'dg-do' directive in the test and

  4398.      before any 'dg-additional-sources' directive.  *Note

  4399.      Effective-Target Keywords::.

  4400. 

  4401. '{ dg-require-SUPPORT args }'

  4402.      Skip the test if the target does not provide the required support.

  4403.      These directives must appear after any 'dg-do' directive in the

  4404.      test and before any 'dg-additional-sources' directive.  They

  4405.      require at least one argument, which can be an empty string if the

  4406.      specific procedure does not examine the argument.  *Note Require

  4407.      Support::, for a complete list of these directives.

  4408. 

  4409. 7.2.1.5 Expect a test to fail for some targets

  4410. ..............................................

  4411. 

  4412. '{ dg-xfail-if COMMENT { SELECTOR } [{ INCLUDE-OPTS } [{ EXCLUDE-OPTS }]] }'

  4413.      Expect the test to fail if the conditions (which are the same as

  4414.      for 'dg-skip-if') are met.  This does not affect the execute step.

  4415. 

  4416. '{ dg-xfail-run-if COMMENT { SELECTOR } [{ INCLUDE-OPTS } [{ EXCLUDE-OPTS }]] }'

  4417.      Expect the execute step of a test to fail if the conditions (which

  4418.      are the same as for 'dg-skip-if') are met.

  4419. 

  4420. 7.2.1.6 Expect the test executable to fail

  4421. ..........................................

  4422. 

  4423. '{ dg-shouldfail COMMENT [{ SELECTOR } [{ INCLUDE-OPTS } [{ EXCLUDE-OPTS }]]] }'

  4424.      Expect the test executable to return a nonzero exit status if the

  4425.      conditions (which are the same as for 'dg-skip-if') are met.

  4426. 

  4427. 7.2.1.7 Verify compiler messages

  4428. ................................

  4429. 

  4430. Where LINE is an accepted argument for these commands, a value of '0'

  4431. can be used if there is no line associated with the message.

  4432. 

  4433. '{ dg-error REGEXP [COMMENT [{ target/xfail SELECTOR } [LINE] ]] }'

  4434.      This DejaGnu directive appears on a source line that is expected to

  4435.      get an error message, or else specifies the source line associated

  4436.      with the message.  If there is no message for that line or if the

  4437.      text of that message is not matched by REGEXP then the check fails

  4438.      and COMMENT is included in the 'FAIL' message.  The check does not

  4439.      look for the string 'error' unless it is part of REGEXP.

  4440. 

  4441. '{ dg-warning REGEXP [COMMENT [{ target/xfail SELECTOR } [LINE] ]] }'

  4442.      This DejaGnu directive appears on a source line that is expected to

  4443.      get a warning message, or else specifies the source line associated

  4444.      with the message.  If there is no message for that line or if the

  4445.      text of that message is not matched by REGEXP then the check fails

  4446.      and COMMENT is included in the 'FAIL' message.  The check does not

  4447.      look for the string 'warning' unless it is part of REGEXP.

  4448. 

  4449. '{ dg-message REGEXP [COMMENT [{ target/xfail SELECTOR } [LINE] ]] }'

  4450.      The line is expected to get a message other than an error or

  4451.      warning.  If there is no message for that line or if the text of

  4452.      that message is not matched by REGEXP then the check fails and

  4453.      COMMENT is included in the 'FAIL' message.

  4454. 

  4455. '{ dg-bogus REGEXP [COMMENT [{ target/xfail SELECTOR } [LINE] ]] }'

  4456.      This DejaGnu directive appears on a source line that should not get

  4457.      a message matching REGEXP, or else specifies the source line

  4458.      associated with the bogus message.  It is usually used with 'xfail'

  4459.      to indicate that the message is a known problem for a particular

  4460.      set of targets.

  4461. 

  4462. '{ dg-line LINENUMVAR }'

  4463.      This DejaGnu directive sets the variable LINENUMVAR to the line

  4464.      number of the source line.  The variable LINENUMVAR can then be

  4465.      used in subsequent 'dg-error', 'dg-warning', 'dg-message' and

  4466.      'dg-bogus' directives.  For example:

  4467. 

  4468.           int a;   /* { dg-line first_def_a } */

  4469.           float a; /* { dg-error "conflicting types of" } */

  4470.           /* { dg-message "previous declaration of" "" { target *-*-* } first_def_a } */

  4471. 

  4472. '{ dg-excess-errors COMMENT [{ target/xfail SELECTOR }] }'

  4473.      This DejaGnu directive indicates that the test is expected to fail

  4474.      due to compiler messages that are not handled by 'dg-error',

  4475.      'dg-warning' or 'dg-bogus'.  For this directive 'xfail' has the

  4476.      same effect as 'target'.

  4477. 

  4478. '{ dg-prune-output REGEXP }'

  4479.      Prune messages matching REGEXP from the test output.

  4480. 

  4481. 7.2.1.8 Verify output of the test executable

  4482. ............................................

  4483. 

  4484. '{ dg-output REGEXP [{ target/xfail SELECTOR }] }'

  4485.      This DejaGnu directive compares REGEXP to the combined output that

  4486.      the test executable writes to 'stdout' and 'stderr'.

  4487. 

  4488. 7.2.1.9 Specify environment variables for a test

  4489. ................................................

  4490. 

  4491. '{ dg-set-compiler-env-var VAR_NAME "VAR_VALUE" }'

  4492.      Specify that the environment variable VAR_NAME needs to be set to

  4493.      VAR_VALUE before invoking the compiler on the test file.

  4494. 

  4495. '{ dg-set-target-env-var VAR_NAME "VAR_VALUE" }'

  4496.      Specify that the environment variable VAR_NAME needs to be set to

  4497.      VAR_VALUE before execution of the program created by the test.

  4498. 

  4499. 7.2.1.10 Specify additional files for a test

  4500. ............................................

  4501. 

  4502. '{ dg-additional-files "FILELIST" }'

  4503.      Specify additional files, other than source files, that must be

  4504.      copied to the system where the compiler runs.

  4505. 

  4506. '{ dg-additional-sources "FILELIST" }'

  4507.      Specify additional source files to appear in the compile line

  4508.      following the main test file.

  4509. 

  4510. 7.2.1.11 Add checks at the end of a test

  4511. ........................................

  4512. 

  4513. '{ dg-final { LOCAL-DIRECTIVE } }'

  4514.      This DejaGnu directive is placed within a comment anywhere in the

  4515.      source file and is processed after the test has been compiled and

  4516.      run.  Multiple 'dg-final' commands are processed in the order in

  4517.      which they appear in the source file.  *Note Final Actions::, for a

  4518.      list of directives that can be used within 'dg-final'.

  4519. 

  4520. 

  4521. File: gccint.info,  Node: Selectors,  Next: Effective-Target Keywords,  Prev: Directives,  Up: Test Directives

  4522. 

  4523. 7.2.2 Selecting targets to which a test applies

  4524. -----------------------------------------------

  4525. 

  4526. Several test directives include SELECTORs to limit the targets for which

  4527. a test is run or to declare that a test is expected to fail on

  4528. particular targets.

  4529. 

  4530.  A selector is:

  4531.    * one or more target triplets, possibly including wildcard

  4532.      characters; use '*-*-*' to match any target

  4533.    * a single effective-target keyword (*note Effective-Target

  4534.      Keywords::)

  4535.    * a logical expression

  4536. 

  4537.  Depending on the context, the selector specifies whether a test is

  4538. skipped and reported as unsupported or is expected to fail.  A context

  4539. that allows either 'target' or 'xfail' also allows '{ target SELECTOR1

  4540. xfail SELECTOR2 }' to skip the test for targets that don't match

  4541. SELECTOR1 and the test to fail for targets that match SELECTOR2.

  4542. 

  4543.  A selector expression appears within curly braces and uses a single

  4544. logical operator: one of '!', '&&', or '||'.  An operand is another

  4545. selector expression, an effective-target keyword, a single target

  4546. triplet, or a list of target triplets within quotes or curly braces.

  4547. For example:

  4548. 

  4549.      { target { ! "hppa*-*-* ia64*-*-*" } }

  4550.      { target { powerpc*-*-* && lp64 } }

  4551.      { xfail { lp64 || vect_no_align } }

  4552. 

  4553. 

  4554. File: gccint.info,  Node: Effective-Target Keywords,  Next: Add Options,  Prev: Selectors,  Up: Test Directives

  4555. 

  4556. 7.2.3 Keywords describing target attributes

  4557. -------------------------------------------

  4558. 

  4559. Effective-target keywords identify sets of targets that support

  4560. particular functionality.  They are used to limit tests to be run only

  4561. for particular targets, or to specify that particular sets of targets

  4562. are expected to fail some tests.

  4563. 

  4564.  Effective-target keywords are defined in 'lib/target-supports.exp' in

  4565. the GCC testsuite, with the exception of those that are documented as

  4566. being local to a particular test directory.

  4567. 

  4568.  The 'effective target' takes into account all of the compiler options

  4569. with which the test will be compiled, including the multilib options.

  4570. By convention, keywords ending in '_nocache' can also include options

  4571. specified for the particular test in an earlier 'dg-options' or

  4572. 'dg-add-options' directive.

  4573. 

  4574. 7.2.3.1 Endianness

  4575. ..................

  4576. 

  4577. 'be'

  4578.      Target uses big-endian memory order for multi-byte and multi-word

  4579.      data.

  4580. 

  4581. 'le'

  4582.      Target uses little-endian memory order for multi-byte and

  4583.      multi-word data.

  4584. 

  4585. 7.2.3.2 Data type sizes

  4586. .......................

  4587. 

  4588. 'ilp32'

  4589.      Target has 32-bit 'int', 'long', and pointers.

  4590. 

  4591. 'lp64'

  4592.      Target has 32-bit 'int', 64-bit 'long' and pointers.

  4593. 

  4594. 'llp64'

  4595.      Target has 32-bit 'int' and 'long', 64-bit 'long long' and

  4596.      pointers.

  4597. 

  4598. 'double64'

  4599.      Target has 64-bit 'double'.

  4600. 

  4601. 'double64plus'

  4602.      Target has 'double' that is 64 bits or longer.

  4603. 

  4604. 'longdouble128'

  4605.      Target has 128-bit 'long double'.

  4606. 

  4607. 'int32plus'

  4608.      Target has 'int' that is at 32 bits or longer.

  4609. 

  4610. 'int16'

  4611.      Target has 'int' that is 16 bits or shorter.

  4612. 

  4613. 'longlong64'

  4614.      Target has 64-bit 'long long'.

  4615. 

  4616. 'long_neq_int'

  4617.      Target has 'int' and 'long' with different sizes.

  4618. 

  4619. 'int_eq_float'

  4620.      Target has 'int' and 'float' with the same size.

  4621. 

  4622. 'ptr_eq_long'

  4623.      Target has pointers ('void *') and 'long' with the same size.

  4624. 

  4625. 'large_double'

  4626.      Target supports 'double' that is longer than 'float'.

  4627. 

  4628. 'large_long_double'

  4629.      Target supports 'long double' that is longer than 'double'.

  4630. 

  4631. 'ptr32plus'

  4632.      Target has pointers that are 32 bits or longer.

  4633. 

  4634. 'size20plus'

  4635.      Target has a 20-bit or larger address space, so at least supports

  4636.      16-bit array and structure sizes.

  4637. 

  4638. 'size32plus'

  4639.      Target has a 32-bit or larger address space, so at least supports

  4640.      24-bit array and structure sizes.

  4641. 

  4642. '4byte_wchar_t'

  4643.      Target has 'wchar_t' that is at least 4 bytes.

  4644. 

  4645. 'floatN'

  4646.      Target has the '_FloatN' type.

  4647. 

  4648. 'floatNx'

  4649.      Target has the '_FloatNx' type.

  4650. 

  4651. 'floatN_runtime'

  4652.      Target has the '_FloatN' type, including runtime support for any

  4653.      options added with 'dg-add-options'.

  4654. 

  4655. 'floatNx_runtime'

  4656.      Target has the '_FloatNx' type, including runtime support for any

  4657.      options added with 'dg-add-options'.

  4658. 

  4659. 'floatn_nx_runtime'

  4660.      Target has runtime support for any options added with

  4661.      'dg-add-options' for any '_FloatN' or '_FloatNx' type.

  4662. 

  4663. 'inf'

  4664.      Target supports floating point infinite ('inf') for type 'double'.

  4665. 

  4666. 7.2.3.3 Fortran-specific attributes

  4667. ...................................

  4668. 

  4669. 'fortran_integer_16'

  4670.      Target supports Fortran 'integer' that is 16 bytes or longer.

  4671. 

  4672. 'fortran_real_10'

  4673.      Target supports Fortran 'real' that is 10 bytes or longer.

  4674. 

  4675. 'fortran_real_16'

  4676.      Target supports Fortran 'real' that is 16 bytes or longer.

  4677. 

  4678. 'fortran_large_int'

  4679.      Target supports Fortran 'integer' kinds larger than 'integer(8)'.

  4680. 

  4681. 'fortran_large_real'

  4682.      Target supports Fortran 'real' kinds larger than 'real(8)'.

  4683. 

  4684. 7.2.3.4 Vector-specific attributes

  4685. ..................................

  4686. 

  4687. 'vect_align_stack_vars'

  4688.      The target's ABI allows stack variables to be aligned to the

  4689.      preferred vector alignment.

  4690. 

  4691. 'vect_avg_qi'

  4692.      Target supports both signed and unsigned averaging operations on

  4693.      vectors of bytes.

  4694. 

  4695. 'vect_mulhrs_hi'

  4696.      Target supports both signed and unsigned

  4697.      multiply-high-with-round-and-scale operations on vectors of

  4698.      half-words.

  4699. 

  4700. 'vect_sdiv_pow2_si'

  4701.      Target supports signed division by constant power-of-2 operations

  4702.      on vectors of 4-byte integers.

  4703. 

  4704. 'vect_condition'

  4705.      Target supports vector conditional operations.

  4706. 

  4707. 'vect_cond_mixed'

  4708.      Target supports vector conditional operations where comparison

  4709.      operands have different type from the value operands.

  4710. 

  4711. 'vect_double'

  4712.      Target supports hardware vectors of 'double'.

  4713. 

  4714. 'vect_double_cond_arith'

  4715.      Target supports conditional addition, subtraction, multiplication,

  4716.      division, minimum and maximum on vectors of 'double', via the

  4717.      'cond_' optabs.

  4718. 

  4719. 'vect_element_align_preferred'

  4720.      The target's preferred vector alignment is the same as the element

  4721.      alignment.

  4722. 

  4723. 'vect_float'

  4724.      Target supports hardware vectors of 'float' when

  4725.      '-funsafe-math-optimizations' is in effect.

  4726. 

  4727. 'vect_float_strict'

  4728.      Target supports hardware vectors of 'float' when

  4729.      '-funsafe-math-optimizations' is not in effect.  This implies

  4730.      'vect_float'.

  4731. 

  4732. 'vect_int'

  4733.      Target supports hardware vectors of 'int'.

  4734. 

  4735. 'vect_long'

  4736.      Target supports hardware vectors of 'long'.

  4737. 

  4738. 'vect_long_long'

  4739.      Target supports hardware vectors of 'long long'.

  4740. 

  4741. 'vect_check_ptrs'

  4742.      Target supports the 'check_raw_ptrs' and 'check_war_ptrs' optabs on

  4743.      vectors.

  4744. 

  4745. 'vect_fully_masked'

  4746.      Target supports fully-masked (also known as fully-predicated)

  4747.      loops, so that vector loops can handle partial as well as full

  4748.      vectors.

  4749. 

  4750. 'vect_masked_store'

  4751.      Target supports vector masked stores.

  4752. 

  4753. 'vect_scatter_store'

  4754.      Target supports vector scatter stores.

  4755. 

  4756. 'vect_aligned_arrays'

  4757.      Target aligns arrays to vector alignment boundary.

  4758. 

  4759. 'vect_hw_misalign'

  4760.      Target supports a vector misalign access.

  4761. 

  4762. 'vect_no_align'

  4763.      Target does not support a vector alignment mechanism.

  4764. 

  4765. 'vect_peeling_profitable'

  4766.      Target might require to peel loops for alignment purposes.

  4767. 

  4768. 'vect_no_int_min_max'

  4769.      Target does not support a vector min and max instruction on 'int'.

  4770. 

  4771. 'vect_no_int_add'

  4772.      Target does not support a vector add instruction on 'int'.

  4773. 

  4774. 'vect_no_bitwise'

  4775.      Target does not support vector bitwise instructions.

  4776. 

  4777. 'vect_bool_cmp'

  4778.      Target supports comparison of 'bool' vectors for at least one

  4779.      vector length.

  4780. 

  4781. 'vect_char_add'

  4782.      Target supports addition of 'char' vectors for at least one vector

  4783.      length.

  4784. 

  4785. 'vect_char_mult'

  4786.      Target supports 'vector char' multiplication.

  4787. 

  4788. 'vect_short_mult'

  4789.      Target supports 'vector short' multiplication.

  4790. 

  4791. 'vect_int_mult'

  4792.      Target supports 'vector int' multiplication.

  4793. 

  4794. 'vect_long_mult'

  4795.      Target supports 64 bit 'vector long' multiplication.

  4796. 

  4797. 'vect_extract_even_odd'

  4798.      Target supports vector even/odd element extraction.

  4799. 

  4800. 'vect_extract_even_odd_wide'

  4801.      Target supports vector even/odd element extraction of vectors with

  4802.      elements 'SImode' or larger.

  4803. 

  4804. 'vect_interleave'

  4805.      Target supports vector interleaving.

  4806. 

  4807. 'vect_strided'

  4808.      Target supports vector interleaving and extract even/odd.

  4809. 

  4810. 'vect_strided_wide'

  4811.      Target supports vector interleaving and extract even/odd for wide

  4812.      element types.

  4813. 

  4814. 'vect_perm'

  4815.      Target supports vector permutation.

  4816. 

  4817. 'vect_perm_byte'

  4818.      Target supports permutation of vectors with 8-bit elements.

  4819. 

  4820. 'vect_perm_short'

  4821.      Target supports permutation of vectors with 16-bit elements.

  4822. 

  4823. 'vect_perm3_byte'

  4824.      Target supports permutation of vectors with 8-bit elements, and for

  4825.      the default vector length it is possible to permute:

  4826.           { a0, a1, a2, b0, b1, b2, ... }

  4827.      to:

  4828.           { a0, a0, a0, b0, b0, b0, ... }

  4829.           { a1, a1, a1, b1, b1, b1, ... }

  4830.           { a2, a2, a2, b2, b2, b2, ... }

  4831.      using only two-vector permutes, regardless of how long the sequence

  4832.      is.

  4833. 

  4834. 'vect_perm3_int'

  4835.      Like 'vect_perm3_byte', but for 32-bit elements.

  4836. 

  4837. 'vect_perm3_short'

  4838.      Like 'vect_perm3_byte', but for 16-bit elements.

  4839. 

  4840. 'vect_shift'

  4841.      Target supports a hardware vector shift operation.

  4842. 

  4843. 'vect_unaligned_possible'

  4844.      Target prefers vectors to have an alignment greater than element

  4845.      alignment, but also allows unaligned vector accesses in some

  4846.      circumstances.

  4847. 

  4848. 'vect_variable_length'

  4849.      Target has variable-length vectors.

  4850. 

  4851. 'vect_widen_sum_hi_to_si'

  4852.      Target supports a vector widening summation of 'short' operands

  4853.      into 'int' results, or can promote (unpack) from 'short' to 'int'.

  4854. 

  4855. 'vect_widen_sum_qi_to_hi'

  4856.      Target supports a vector widening summation of 'char' operands into

  4857.      'short' results, or can promote (unpack) from 'char' to 'short'.

  4858. 

  4859. 'vect_widen_sum_qi_to_si'

  4860.      Target supports a vector widening summation of 'char' operands into

  4861.      'int' results.

  4862. 

  4863. 'vect_widen_mult_qi_to_hi'

  4864.      Target supports a vector widening multiplication of 'char' operands

  4865.      into 'short' results, or can promote (unpack) from 'char' to

  4866.      'short' and perform non-widening multiplication of 'short'.

  4867. 

  4868. 'vect_widen_mult_hi_to_si'

  4869.      Target supports a vector widening multiplication of 'short'

  4870.      operands into 'int' results, or can promote (unpack) from 'short'

  4871.      to 'int' and perform non-widening multiplication of 'int'.

  4872. 

  4873. 'vect_widen_mult_si_to_di_pattern'

  4874.      Target supports a vector widening multiplication of 'int' operands

  4875.      into 'long' results.

  4876. 

  4877. 'vect_sdot_qi'

  4878.      Target supports a vector dot-product of 'signed char'.

  4879. 

  4880. 'vect_udot_qi'

  4881.      Target supports a vector dot-product of 'unsigned char'.

  4882. 

  4883. 'vect_sdot_hi'

  4884.      Target supports a vector dot-product of 'signed short'.

  4885. 

  4886. 'vect_udot_hi'

  4887.      Target supports a vector dot-product of 'unsigned short'.

  4888. 

  4889. 'vect_pack_trunc'

  4890.      Target supports a vector demotion (packing) of 'short' to 'char'

  4891.      and from 'int' to 'short' using modulo arithmetic.

  4892. 

  4893. 'vect_unpack'

  4894.      Target supports a vector promotion (unpacking) of 'char' to 'short'

  4895.      and from 'char' to 'int'.

  4896. 

  4897. 'vect_intfloat_cvt'

  4898.      Target supports conversion from 'signed int' to 'float'.

  4899. 

  4900. 'vect_uintfloat_cvt'

  4901.      Target supports conversion from 'unsigned int' to 'float'.

  4902. 

  4903. 'vect_floatint_cvt'

  4904.      Target supports conversion from 'float' to 'signed int'.

  4905. 

  4906. 'vect_floatuint_cvt'

  4907.      Target supports conversion from 'float' to 'unsigned int'.

  4908. 

  4909. 'vect_intdouble_cvt'

  4910.      Target supports conversion from 'signed int' to 'double'.

  4911. 

  4912. 'vect_doubleint_cvt'

  4913.      Target supports conversion from 'double' to 'signed int'.

  4914. 

  4915. 'vect_max_reduc'

  4916.      Target supports max reduction for vectors.

  4917. 

  4918. 'vect_sizes_16B_8B'

  4919.      Target supports 16- and 8-bytes vectors.

  4920. 

  4921. 'vect_sizes_32B_16B'

  4922.      Target supports 32- and 16-bytes vectors.

  4923. 

  4924. 'vect_logical_reduc'

  4925.      Target supports AND, IOR and XOR reduction on vectors.

  4926. 

  4927. 'vect_fold_extract_last'

  4928.      Target supports the 'fold_extract_last' optab.

  4929. 

  4930. 7.2.3.5 Thread Local Storage attributes

  4931. .......................................

  4932. 

  4933. 'tls'

  4934.      Target supports thread-local storage.

  4935. 

  4936. 'tls_native'

  4937.      Target supports native (rather than emulated) thread-local storage.

  4938. 

  4939. 'tls_runtime'

  4940.      Test system supports executing TLS executables.

  4941. 

  4942. 7.2.3.6 Decimal floating point attributes

  4943. .........................................

  4944. 

  4945. 'dfp'

  4946.      Targets supports compiling decimal floating point extension to C.

  4947. 

  4948. 'dfp_nocache'

  4949.      Including the options used to compile this particular test, the

  4950.      target supports compiling decimal floating point extension to C.

  4951. 

  4952. 'dfprt'

  4953.      Test system can execute decimal floating point tests.

  4954. 

  4955. 'dfprt_nocache'

  4956.      Including the options used to compile this particular test, the

  4957.      test system can execute decimal floating point tests.

  4958. 

  4959. 'hard_dfp'

  4960.      Target generates decimal floating point instructions with current

  4961.      options.

  4962. 

  4963. 7.2.3.7 ARM-specific attributes

  4964. ...............................

  4965. 

  4966. 'arm32'

  4967.      ARM target generates 32-bit code.

  4968. 

  4969. 'arm_little_endian'

  4970.      ARM target that generates little-endian code.

  4971. 

  4972. 'arm_eabi'

  4973.      ARM target adheres to the ABI for the ARM Architecture.

  4974. 

  4975. 'arm_fp_ok'

  4976.      ARM target defines '__ARM_FP' using '-mfloat-abi=softfp' or

  4977.      equivalent options.  Some multilibs may be incompatible with these

  4978.      options.

  4979. 

  4980. 'arm_fp_dp_ok'

  4981.      ARM target defines '__ARM_FP' with double-precision support using

  4982.      '-mfloat-abi=softfp' or equivalent options.  Some multilibs may be

  4983.      incompatible with these options.

  4984. 

  4985. 'arm_hf_eabi'

  4986.      ARM target adheres to the VFP and Advanced SIMD Register Arguments

  4987.      variant of the ABI for the ARM Architecture (as selected with

  4988.      '-mfloat-abi=hard').

  4989. 

  4990. 'arm_softfloat'

  4991.      ARM target uses the soft-float ABI with no floating-point

  4992.      instructions used whatsoever (as selected with '-mfloat-abi=soft').

  4993. 

  4994. 'arm_hard_vfp_ok'

  4995.      ARM target supports '-mfpu=vfp -mfloat-abi=hard'.  Some multilibs

  4996.      may be incompatible with these options.

  4997. 

  4998. 'arm_iwmmxt_ok'

  4999.      ARM target supports '-mcpu=iwmmxt'.  Some multilibs may be

  5000.      incompatible with this option.

  5001. 

  5002. 'arm_neon'

  5003.      ARM target supports generating NEON instructions.

  5004. 

  5005. 'arm_tune_string_ops_prefer_neon'

  5006.      Test CPU tune supports inlining string operations with NEON

  5007.      instructions.

  5008. 

  5009. 'arm_neon_hw'

  5010.      Test system supports executing NEON instructions.

  5011. 

  5012. 'arm_neonv2_hw'

  5013.      Test system supports executing NEON v2 instructions.

  5014. 

  5015. 'arm_neon_ok'

  5016.      ARM Target supports '-mfpu=neon -mfloat-abi=softfp' or compatible

  5017.      options.  Some multilibs may be incompatible with these options.

  5018. 

  5019. 'arm_neon_ok_no_float_abi'

  5020.      ARM Target supports NEON with '-mfpu=neon', but without any

  5021.      -mfloat-abi= option.  Some multilibs may be incompatible with this

  5022.      option.

  5023. 

  5024. 'arm_neonv2_ok'

  5025.      ARM Target supports '-mfpu=neon-vfpv4 -mfloat-abi=softfp' or

  5026.      compatible options.  Some multilibs may be incompatible with these

  5027.      options.

  5028. 

  5029. 'arm_fp16_ok'

  5030.      Target supports options to generate VFP half-precision

  5031.      floating-point instructions.  Some multilibs may be incompatible

  5032.      with these options.  This test is valid for ARM only.

  5033. 

  5034. 'arm_fp16_hw'

  5035.      Target supports executing VFP half-precision floating-point

  5036.      instructions.  This test is valid for ARM only.

  5037. 

  5038. 'arm_neon_fp16_ok'

  5039.      ARM Target supports '-mfpu=neon-fp16 -mfloat-abi=softfp' or

  5040.      compatible options, including '-mfp16-format=ieee' if necessary to

  5041.      obtain the '__fp16' type.  Some multilibs may be incompatible with

  5042.      these options.

  5043. 

  5044. 'arm_neon_fp16_hw'

  5045.      Test system supports executing Neon half-precision float

  5046.      instructions.  (Implies previous.)

  5047. 

  5048. 'arm_fp16_alternative_ok'

  5049.      ARM target supports the ARM FP16 alternative format.  Some

  5050.      multilibs may be incompatible with the options needed.

  5051. 

  5052. 'arm_fp16_none_ok'

  5053.      ARM target supports specifying none as the ARM FP16 format.

  5054. 

  5055. 'arm_thumb1_ok'

  5056.      ARM target generates Thumb-1 code for '-mthumb'.

  5057. 

  5058. 'arm_thumb2_ok'

  5059.      ARM target generates Thumb-2 code for '-mthumb'.

  5060. 

  5061. 'arm_nothumb'

  5062.      ARM target that is not using Thumb.

  5063. 

  5064. 'arm_vfp_ok'

  5065.      ARM target supports '-mfpu=vfp -mfloat-abi=softfp'.  Some multilibs

  5066.      may be incompatible with these options.

  5067. 

  5068. 'arm_vfp3_ok'

  5069.      ARM target supports '-mfpu=vfp3 -mfloat-abi=softfp'.  Some

  5070.      multilibs may be incompatible with these options.

  5071. 

  5072. 'arm_arch_v8a_hard_ok'

  5073.      The compiler is targeting 'arm*-*-*' and can compile and assemble

  5074.      code using the options '-march=armv8-a -mfpu=neon-fp-armv8

  5075.      -mfloat-abi=hard'.  This is not enough to guarantee that linking

  5076.      works.

  5077. 

  5078. 'arm_arch_v8a_hard_multilib'

  5079.      The compiler is targeting 'arm*-*-*' and can build programs using

  5080.      the options '-march=armv8-a -mfpu=neon-fp-armv8 -mfloat-abi=hard'.

  5081.      The target can also run the resulting binaries.

  5082. 

  5083. 'arm_v8_vfp_ok'

  5084.      ARM target supports '-mfpu=fp-armv8 -mfloat-abi=softfp'.  Some

  5085.      multilibs may be incompatible with these options.

  5086. 

  5087. 'arm_v8_neon_ok'

  5088.      ARM target supports '-mfpu=neon-fp-armv8 -mfloat-abi=softfp'.  Some

  5089.      multilibs may be incompatible with these options.

  5090. 

  5091. 'arm_v8_1a_neon_ok'

  5092.      ARM target supports options to generate ARMv8.1-A Adv.SIMD

  5093.      instructions.  Some multilibs may be incompatible with these

  5094.      options.

  5095. 

  5096. 'arm_v8_1a_neon_hw'

  5097.      ARM target supports executing ARMv8.1-A Adv.SIMD instructions.

  5098.      Some multilibs may be incompatible with the options needed.

  5099.      Implies arm_v8_1a_neon_ok.

  5100. 

  5101. 'arm_acq_rel'

  5102.      ARM target supports acquire-release instructions.

  5103. 

  5104. 'arm_v8_2a_fp16_scalar_ok'

  5105.      ARM target supports options to generate instructions for ARMv8.2-A

  5106.      and scalar instructions from the FP16 extension.  Some multilibs

  5107.      may be incompatible with these options.

  5108. 

  5109. 'arm_v8_2a_fp16_scalar_hw'

  5110.      ARM target supports executing instructions for ARMv8.2-A and scalar

  5111.      instructions from the FP16 extension.  Some multilibs may be

  5112.      incompatible with these options.  Implies arm_v8_2a_fp16_neon_ok.

  5113. 

  5114. 'arm_v8_2a_fp16_neon_ok'

  5115.      ARM target supports options to generate instructions from ARMv8.2-A

  5116.      with the FP16 extension.  Some multilibs may be incompatible with

  5117.      these options.  Implies arm_v8_2a_fp16_scalar_ok.

  5118. 

  5119. 'arm_v8_2a_fp16_neon_hw'

  5120.      ARM target supports executing instructions from ARMv8.2-A with the

  5121.      FP16 extension.  Some multilibs may be incompatible with these

  5122.      options.  Implies arm_v8_2a_fp16_neon_ok and

  5123.      arm_v8_2a_fp16_scalar_hw.

  5124. 

  5125. 'arm_v8_2a_dotprod_neon_ok'

  5126.      ARM target supports options to generate instructions from ARMv8.2-A

  5127.      with the Dot Product extension.  Some multilibs may be incompatible

  5128.      with these options.

  5129. 

  5130. 'arm_v8_2a_dotprod_neon_hw'

  5131.      ARM target supports executing instructions from ARMv8.2-A with the

  5132.      Dot Product extension.  Some multilibs may be incompatible with

  5133.      these options.  Implies arm_v8_2a_dotprod_neon_ok.

  5134. 

  5135. 'arm_fp16fml_neon_ok'

  5136.      ARM target supports extensions to generate the 'VFMAL' and 'VFMLS'

  5137.      half-precision floating-point instructions available from ARMv8.2-A

  5138.      and onwards.  Some multilibs may be incompatible with these

  5139.      options.

  5140. 

  5141. 'arm_v8_2a_bf16_neon_ok'

  5142.      ARM target supports options to generate instructions from ARMv8.2-A

  5143.      with the BFloat16 extension (bf16).  Some multilibs may be

  5144.      incompatible with these options.

  5145. 

  5146. 'arm_v8_2a_i8mm_ok'

  5147.      ARM target supports options to generate instructions from ARMv8.2-A

  5148.      with the 8-Bit Integer Matrix Multiply extension (i8mm).  Some

  5149.      multilibs may be incompatible with these options.

  5150. 

  5151. 'arm_v8_1m_mve_ok'

  5152.      ARM target supports options to generate instructions from ARMv8.1-M

  5153.      with the M-Profile Vector Extension (MVE). Some multilibs may be

  5154.      incompatible with these options.

  5155. 

  5156. 'arm_v8_1m_mve_fp_ok'

  5157.      ARM target supports options to generate instructions from ARMv8.1-M

  5158.      with the Half-precision floating-point instructions (HP),

  5159.      Floating-point Extension (FP) along with M-Profile Vector Extension

  5160.      (MVE). Some multilibs may be incompatible with these options.

  5161. 

  5162. 'arm_mve_hw'

  5163.      Test system supports executing MVE instructions.

  5164. 

  5165. 'arm_v8m_main_cde'

  5166.      ARM target supports options to generate instructions from ARMv8-M

  5167.      with the Custom Datapath Extension (CDE). Some multilibs may be

  5168.      incompatible with these options.

  5169. 

  5170. 'arm_v8m_main_cde_fp'

  5171.      ARM target supports options to generate instructions from ARMv8-M

  5172.      with the Custom Datapath Extension (CDE) and floating-point (VFP).

  5173.      Some multilibs may be incompatible with these options.

  5174. 

  5175. 'arm_v8_1m_main_cde_mve'

  5176.      ARM target supports options to generate instructions from ARMv8.1-M

  5177.      with the Custom Datapath Extension (CDE) and M-Profile Vector

  5178.      Extension (MVE). Some multilibs may be incompatible with these

  5179.      options.

  5180. 

  5181. 'arm_prefer_ldrd_strd'

  5182.      ARM target prefers 'LDRD' and 'STRD' instructions over 'LDM' and

  5183.      'STM' instructions.

  5184. 

  5185. 'arm_thumb1_movt_ok'

  5186.      ARM target generates Thumb-1 code for '-mthumb' with 'MOVW' and

  5187.      'MOVT' instructions available.

  5188. 

  5189. 'arm_thumb1_cbz_ok'

  5190.      ARM target generates Thumb-1 code for '-mthumb' with 'CBZ' and

  5191.      'CBNZ' instructions available.

  5192. 

  5193. 'arm_divmod_simode'

  5194.      ARM target for which divmod transform is disabled, if it supports

  5195.      hardware div instruction.

  5196. 

  5197. 'arm_cmse_ok'

  5198.      ARM target supports ARMv8-M Security Extensions, enabled by the

  5199.      '-mcmse' option.

  5200. 

  5201. 'arm_coproc1_ok'

  5202.      ARM target supports the following coprocessor instructions: 'CDP',

  5203.      'LDC', 'STC', 'MCR' and 'MRC'.

  5204. 

  5205. 'arm_coproc2_ok'

  5206.      ARM target supports all the coprocessor instructions also listed as

  5207.      supported in *note arm_coproc1_ok:: in addition to the following:

  5208.      'CDP2', 'LDC2', 'LDC2l', 'STC2', 'STC2l', 'MCR2' and 'MRC2'.

  5209. 

  5210. 'arm_coproc3_ok'

  5211.      ARM target supports all the coprocessor instructions also listed as

  5212.      supported in *note arm_coproc2_ok:: in addition the following:

  5213.      'MCRR' and 'MRRC'.

  5214. 

  5215. 'arm_coproc4_ok'

  5216.      ARM target supports all the coprocessor instructions also listed as

  5217.      supported in *note arm_coproc3_ok:: in addition the following:

  5218.      'MCRR2' and 'MRRC2'.

  5219. 

  5220. 'arm_simd32_ok'

  5221.      ARM Target supports options suitable for accessing the SIMD32

  5222.      intrinsics from 'arm_acle.h'.  Some multilibs may be incompatible

  5223.      with these options.

  5224. 

  5225. 'arm_qbit_ok'

  5226.      ARM Target supports options suitable for accessing the Q-bit

  5227.      manipulation intrinsics from 'arm_acle.h'.  Some multilibs may be

  5228.      incompatible with these options.

  5229. 

  5230. 'arm_softfp_ok'

  5231.      ARM target supports the '-mfloat-abi=softfp' option.

  5232. 

  5233. 'arm_hard_ok'

  5234.      ARM target supports the '-mfloat-abi=hard' option.

  5235. 

  5236. 7.2.3.8 AArch64-specific attributes

  5237. ...................................

  5238. 

  5239. 'aarch64_asm_<ext>_ok'

  5240.      AArch64 assembler supports the architecture extension 'ext' via the

  5241.      '.arch_extension' pseudo-op.

  5242. 'aarch64_tiny'

  5243.      AArch64 target which generates instruction sequences for tiny

  5244.      memory model.

  5245. 'aarch64_small'

  5246.      AArch64 target which generates instruction sequences for small

  5247.      memory model.

  5248. 'aarch64_large'

  5249.      AArch64 target which generates instruction sequences for large

  5250.      memory model.

  5251. 'aarch64_little_endian'

  5252.      AArch64 target which generates instruction sequences for little

  5253.      endian.

  5254. 'aarch64_big_endian'

  5255.      AArch64 target which generates instruction sequences for big

  5256.      endian.

  5257. 'aarch64_small_fpic'

  5258.      Binutils installed on test system supports relocation types

  5259.      required by -fpic for AArch64 small memory model.

  5260. 'aarch64_sve_hw'

  5261.      AArch64 target that is able to generate and execute SVE code

  5262.      (regardless of whether it does so by default).

  5263. 'aarch64_sve128_hw'

  5264. 'aarch64_sve256_hw'

  5265. 'aarch64_sve512_hw'

  5266. 'aarch64_sve1024_hw'

  5267. 'aarch64_sve2048_hw'

  5268.      Like 'aarch64_sve_hw', but also test for an exact hardware vector

  5269.      length.

  5270. 

  5271. 'aarch64_fjcvtzs_hw'

  5272.      AArch64 target that is able to generate and execute armv8.3-a

  5273.      FJCVTZS instruction.

  5274. 

  5275. 7.2.3.9 MIPS-specific attributes

  5276. ................................

  5277. 

  5278. 'mips64'

  5279.      MIPS target supports 64-bit instructions.

  5280. 

  5281. 'nomips16'

  5282.      MIPS target does not produce MIPS16 code.

  5283. 

  5284. 'mips16_attribute'

  5285.      MIPS target can generate MIPS16 code.

  5286. 

  5287. 'mips_loongson'

  5288.      MIPS target is a Loongson-2E or -2F target using an ABI that

  5289.      supports the Loongson vector modes.

  5290. 

  5291. 'mips_msa'

  5292.      MIPS target supports '-mmsa', MIPS SIMD Architecture (MSA).

  5293. 

  5294. 'mips_newabi_large_long_double'

  5295.      MIPS target supports 'long double' larger than 'double' when using

  5296.      the new ABI.

  5297. 

  5298. 'mpaired_single'

  5299.      MIPS target supports '-mpaired-single'.

  5300. 

  5301. 7.2.3.10 PowerPC-specific attributes

  5302. ....................................

  5303. 

  5304. 'dfp_hw'

  5305.      PowerPC target supports executing hardware DFP instructions.

  5306. 

  5307. 'p8vector_hw'

  5308.      PowerPC target supports executing VSX instructions (ISA 2.07).

  5309. 

  5310. 'powerpc64'

  5311.      Test system supports executing 64-bit instructions.

  5312. 

  5313. 'powerpc_altivec'

  5314.      PowerPC target supports AltiVec.

  5315. 

  5316. 'powerpc_altivec_ok'

  5317.      PowerPC target supports '-maltivec'.

  5318. 

  5319. 'powerpc_eabi_ok'

  5320.      PowerPC target supports '-meabi'.

  5321. 

  5322. 'powerpc_elfv2'

  5323.      PowerPC target supports '-mabi=elfv2'.

  5324. 

  5325. 'powerpc_fprs'

  5326.      PowerPC target supports floating-point registers.

  5327. 

  5328. 'powerpc_hard_double'

  5329.      PowerPC target supports hardware double-precision floating-point.

  5330. 

  5331. 'powerpc_htm_ok'

  5332.      PowerPC target supports '-mhtm'

  5333. 

  5334. 'powerpc_p8vector_ok'

  5335.      PowerPC target supports '-mpower8-vector'

  5336. 

  5337. 'powerpc_popcntb_ok'

  5338.      PowerPC target supports the 'popcntb' instruction, indicating that

  5339.      this target supports '-mcpu=power5'.

  5340. 

  5341. 'powerpc_ppu_ok'

  5342.      PowerPC target supports '-mcpu=cell'.

  5343. 

  5344. 'powerpc_spe'

  5345.      PowerPC target supports PowerPC SPE.

  5346. 

  5347. 'powerpc_spe_nocache'

  5348.      Including the options used to compile this particular test, the

  5349.      PowerPC target supports PowerPC SPE.

  5350. 

  5351. 'powerpc_spu'

  5352.      PowerPC target supports PowerPC SPU.

  5353. 

  5354. 'powerpc_vsx_ok'

  5355.      PowerPC target supports '-mvsx'.

  5356. 

  5357. 'powerpc_405_nocache'

  5358.      Including the options used to compile this particular test, the

  5359.      PowerPC target supports PowerPC 405.

  5360. 

  5361. 'ppc_recip_hw'

  5362.      PowerPC target supports executing reciprocal estimate instructions.

  5363. 

  5364. 'vmx_hw'

  5365.      PowerPC target supports executing AltiVec instructions.

  5366. 

  5367. 'vsx_hw'

  5368.      PowerPC target supports executing VSX instructions (ISA 2.06).

  5369. 

  5370. 7.2.3.11 Other hardware attributes

  5371. ..................................

  5372. 

  5373. 'autoincdec'

  5374.      Target supports autoincrement/decrement addressing.

  5375. 

  5376. 'avx'

  5377.      Target supports compiling 'avx' instructions.

  5378. 

  5379. 'avx_runtime'

  5380.      Target supports the execution of 'avx' instructions.

  5381. 

  5382. 'avx2'

  5383.      Target supports compiling 'avx2' instructions.

  5384. 

  5385. 'avx2_runtime'

  5386.      Target supports the execution of 'avx2' instructions.

  5387. 

  5388. 'avx512f'

  5389.      Target supports compiling 'avx512f' instructions.

  5390. 

  5391. 'avx512f_runtime'

  5392.      Target supports the execution of 'avx512f' instructions.

  5393. 

  5394. 'avx512vp2intersect'

  5395.      Target supports the execution of 'avx512vp2intersect' instructions.

  5396. 

  5397. 'cell_hw'

  5398.      Test system can execute AltiVec and Cell PPU instructions.

  5399. 

  5400. 'coldfire_fpu'

  5401.      Target uses a ColdFire FPU.

  5402. 

  5403. 'divmod'

  5404.      Target supporting hardware divmod insn or divmod libcall.

  5405. 

  5406. 'divmod_simode'

  5407.      Target supporting hardware divmod insn or divmod libcall for

  5408.      SImode.

  5409. 

  5410. 'hard_float'

  5411.      Target supports FPU instructions.

  5412. 

  5413. 'non_strict_align'

  5414.      Target does not require strict alignment.

  5415. 

  5416. 'pie_copyreloc'

  5417.      The x86-64 target linker supports PIE with copy reloc.

  5418. 

  5419. 'rdrand'

  5420.      Target supports x86 'rdrand' instruction.

  5421. 

  5422. 'sqrt_insn'

  5423.      Target has a square root instruction that the compiler can

  5424.      generate.

  5425. 

  5426. 'sse'

  5427.      Target supports compiling 'sse' instructions.

  5428. 

  5429. 'sse_runtime'

  5430.      Target supports the execution of 'sse' instructions.

  5431. 

  5432. 'sse2'

  5433.      Target supports compiling 'sse2' instructions.

  5434. 

  5435. 'sse2_runtime'

  5436.      Target supports the execution of 'sse2' instructions.

  5437. 

  5438. 'sync_char_short'

  5439.      Target supports atomic operations on 'char' and 'short'.

  5440. 

  5441. 'sync_int_long'

  5442.      Target supports atomic operations on 'int' and 'long'.

  5443. 

  5444. 'ultrasparc_hw'

  5445.      Test environment appears to run executables on a simulator that

  5446.      accepts only 'EM_SPARC' executables and chokes on 'EM_SPARC32PLUS'

  5447.      or 'EM_SPARCV9' executables.

  5448. 

  5449. 'vect_cmdline_needed'

  5450.      Target requires a command line argument to enable a SIMD

  5451.      instruction set.

  5452. 

  5453. 'xorsign'

  5454.      Target supports the xorsign optab expansion.

  5455. 

  5456. 7.2.3.12 Environment attributes

  5457. ...............................

  5458. 

  5459. 'c'

  5460.      The language for the compiler under test is C.

  5461. 

  5462. 'c++'

  5463.      The language for the compiler under test is C++.

  5464. 

  5465. 'c99_runtime'

  5466.      Target provides a full C99 runtime.

  5467. 

  5468. 'correct_iso_cpp_string_wchar_protos'

  5469.      Target 'string.h' and 'wchar.h' headers provide C++ required

  5470.      overloads for 'strchr' etc.  functions.

  5471. 

  5472. 'd_runtime'

  5473.      Target provides the D runtime.

  5474. 

  5475. 'd_runtime_has_std_library'

  5476.      Target provides the D standard library (Phobos).

  5477. 

  5478. 'dummy_wcsftime'

  5479.      Target uses a dummy 'wcsftime' function that always returns zero.

  5480. 

  5481. 'fd_truncate'

  5482.      Target can truncate a file from a file descriptor, as used by

  5483.      'libgfortran/io/unix.c:fd_truncate'; i.e. 'ftruncate' or 'chsize'.

  5484. 

  5485. 'fenv'

  5486.      Target provides 'fenv.h' include file.

  5487. 

  5488. 'fenv_exceptions'

  5489.      Target supports 'fenv.h' with all the standard IEEE exceptions and

  5490.      floating-point exceptions are raised by arithmetic operations.

  5491. 

  5492. 'fileio'

  5493.      Target offers such file I/O library functions as 'fopen', 'fclose',

  5494.      'tmpnam', and 'remove'.  This is a link-time requirement for the

  5495.      presence of the functions in the library; even if they fail at

  5496.      runtime, the requirement is still regarded as satisfied.

  5497. 

  5498. 'freestanding'

  5499.      Target is 'freestanding' as defined in section 4 of the C99

  5500.      standard.  Effectively, it is a target which supports no extra

  5501.      headers or libraries other than what is considered essential.

  5502. 

  5503. 'gettimeofday'

  5504.      Target supports 'gettimeofday'.

  5505. 

  5506. 'init_priority'

  5507.      Target supports constructors with initialization priority

  5508.      arguments.

  5509. 

  5510. 'inttypes_types'

  5511.      Target has the basic signed and unsigned types in 'inttypes.h'.

  5512.      This is for tests that GCC's notions of these types agree with

  5513.      those in the header, as some systems have only 'inttypes.h'.

  5514. 

  5515. 'lax_strtofp'

  5516.      Target might have errors of a few ULP in string to floating-point

  5517.      conversion functions and overflow is not always detected correctly

  5518.      by those functions.

  5519. 

  5520. 'mempcpy'

  5521.      Target provides 'mempcpy' function.

  5522. 

  5523. 'mmap'

  5524.      Target supports 'mmap'.

  5525. 

  5526. 'newlib'

  5527.      Target supports Newlib.

  5528. 

  5529. 'newlib_nano_io'

  5530.      GCC was configured with '--enable-newlib-nano-formatted-io', which

  5531.      reduces the code size of Newlib formatted I/O functions.

  5532. 

  5533. 'pow10'

  5534.      Target provides 'pow10' function.

  5535. 

  5536. 'pthread'

  5537.      Target can compile using 'pthread.h' with no errors or warnings.

  5538. 

  5539. 'pthread_h'

  5540.      Target has 'pthread.h'.

  5541. 

  5542. 'run_expensive_tests'

  5543.      Expensive testcases (usually those that consume excessive amounts

  5544.      of CPU time) should be run on this target.  This can be enabled by

  5545.      setting the 'GCC_TEST_RUN_EXPENSIVE' environment variable to a

  5546.      non-empty string.

  5547. 

  5548. 'simulator'

  5549.      Test system runs executables on a simulator (i.e. slowly) rather

  5550.      than hardware (i.e. fast).

  5551. 

  5552. 'signal'

  5553.      Target has 'signal.h'.

  5554. 

  5555. 'stabs'

  5556.      Target supports the stabs debugging format.

  5557. 

  5558. 'stdint_types'

  5559.      Target has the basic signed and unsigned C types in 'stdint.h'.

  5560.      This will be obsolete when GCC ensures a working 'stdint.h' for all

  5561.      targets.

  5562. 

  5563. 'stpcpy'

  5564.      Target provides 'stpcpy' function.

  5565. 

  5566. 'trampolines'

  5567.      Target supports trampolines.

  5568. 

  5569. 'uclibc'

  5570.      Target supports uClibc.

  5571. 

  5572. 'unwrapped'

  5573.      Target does not use a status wrapper.

  5574. 

  5575. 'vxworks_kernel'

  5576.      Target is a VxWorks kernel.

  5577. 

  5578. 'vxworks_rtp'

  5579.      Target is a VxWorks RTP.

  5580. 

  5581. 'wchar'

  5582.      Target supports wide characters.

  5583. 

  5584. 7.2.3.13 Other attributes

  5585. .........................

  5586. 

  5587. 'automatic_stack_alignment'

  5588.      Target supports automatic stack alignment.

  5589. 

  5590. 'branch_cost'

  5591.      Target supports '-branch-cost=N'.

  5592. 

  5593. 'cxa_atexit'

  5594.      Target uses '__cxa_atexit'.

  5595. 

  5596. 'default_packed'

  5597.      Target has packed layout of structure members by default.

  5598. 

  5599. 'exceptions'

  5600.      Target supports exceptions.

  5601. 

  5602. 'exceptions_enabled'

  5603.      Target supports exceptions and they are enabled in the current

  5604.      testing configuration.

  5605. 

  5606. 'fgraphite'

  5607.      Target supports Graphite optimizations.

  5608. 

  5609. 'fixed_point'

  5610.      Target supports fixed-point extension to C.

  5611. 

  5612. 'fopenacc'

  5613.      Target supports OpenACC via '-fopenacc'.

  5614. 

  5615. 'fopenmp'

  5616.      Target supports OpenMP via '-fopenmp'.

  5617. 

  5618. 'fpic'

  5619.      Target supports '-fpic' and '-fPIC'.

  5620. 

  5621. 'freorder'

  5622.      Target supports '-freorder-blocks-and-partition'.

  5623. 

  5624. 'fstack_protector'

  5625.      Target supports '-fstack-protector'.

  5626. 

  5627. 'gas'

  5628.      Target uses GNU 'as'.

  5629. 

  5630. 'gc_sections'

  5631.      Target supports '--gc-sections'.

  5632. 

  5633. 'gld'

  5634.      Target uses GNU 'ld'.

  5635. 

  5636. 'keeps_null_pointer_checks'

  5637.      Target keeps null pointer checks, either due to the use of

  5638.      '-fno-delete-null-pointer-checks' or hardwired into the target.

  5639. 

  5640. 'llvm_binutils'

  5641.      Target is using an LLVM assembler and/or linker, instead of GNU

  5642.      Binutils.

  5643. 

  5644. 'lto'

  5645.      Compiler has been configured to support link-time optimization

  5646.      (LTO).

  5647. 

  5648. 'lto_incremental'

  5649.      Compiler and linker support link-time optimization relocatable

  5650.      linking with '-r' and '-flto' options.

  5651. 

  5652. 'naked_functions'

  5653.      Target supports the 'naked' function attribute.

  5654. 

  5655. 'named_sections'

  5656.      Target supports named sections.

  5657. 

  5658. 'natural_alignment_32'

  5659.      Target uses natural alignment (aligned to type size) for types of

  5660.      32 bits or less.

  5661. 

  5662. 'target_natural_alignment_64'

  5663.      Target uses natural alignment (aligned to type size) for types of

  5664.      64 bits or less.

  5665. 

  5666. 'noinit'

  5667.      Target supports the 'noinit' variable attribute.

  5668. 

  5669. 'nonpic'

  5670.      Target does not generate PIC by default.

  5671. 

  5672. 'offload_gcn'

  5673.      Target has been configured for OpenACC/OpenMP offloading on AMD

  5674.      GCN.

  5675. 

  5676. 'pie_enabled'

  5677.      Target generates PIE by default.

  5678. 

  5679. 'pcc_bitfield_type_matters'

  5680.      Target defines 'PCC_BITFIELD_TYPE_MATTERS'.

  5681. 

  5682. 'pe_aligned_commons'

  5683.      Target supports '-mpe-aligned-commons'.

  5684. 

  5685. 'pie'

  5686.      Target supports '-pie', '-fpie' and '-fPIE'.

  5687. 

  5688. 'rdynamic'

  5689.      Target supports '-rdynamic'.

  5690. 

  5691. 'scalar_all_fma'

  5692.      Target supports all four fused multiply-add optabs for both 'float'

  5693.      and 'double'.  These optabs are: 'fma_optab', 'fms_optab',

  5694.      'fnma_optab' and 'fnms_optab'.

  5695. 

  5696. 'section_anchors'

  5697.      Target supports section anchors.

  5698. 

  5699. 'short_enums'

  5700.      Target defaults to short enums.

  5701. 

  5702. 'stack_size'

  5703.      Target has limited stack size.  The stack size limit can be

  5704.      obtained using the STACK_SIZE macro defined by *note

  5705.      'dg-add-options' feature 'stack_size': stack_size_ao.

  5706. 

  5707. 'static'

  5708.      Target supports '-static'.

  5709. 

  5710. 'static_libgfortran'

  5711.      Target supports statically linking 'libgfortran'.

  5712. 

  5713. 'string_merging'

  5714.      Target supports merging string constants at link time.

  5715. 

  5716. 'ucn'

  5717.      Target supports compiling and assembling UCN.

  5718. 

  5719. 'ucn_nocache'

  5720.      Including the options used to compile this particular test, the

  5721.      target supports compiling and assembling UCN.

  5722. 

  5723. 'unaligned_stack'

  5724.      Target does not guarantee that its 'STACK_BOUNDARY' is greater than

  5725.      or equal to the required vector alignment.

  5726. 

  5727. 'vector_alignment_reachable'

  5728.      Vector alignment is reachable for types of 32 bits or less.

  5729. 

  5730. 'vector_alignment_reachable_for_64bit'

  5731.      Vector alignment is reachable for types of 64 bits or less.

  5732. 

  5733. 'wchar_t_char16_t_compatible'

  5734.      Target supports 'wchar_t' that is compatible with 'char16_t'.

  5735. 

  5736. 'wchar_t_char32_t_compatible'

  5737.      Target supports 'wchar_t' that is compatible with 'char32_t'.

  5738. 

  5739. 'comdat_group'

  5740.      Target uses comdat groups.

  5741. 

  5742. 'indirect_calls'

  5743.      Target supports indirect calls, i.e.  calls where the target is not

  5744.      constant.

  5745. 

  5746. 7.2.3.14 Local to tests in 'gcc.target/i386'

  5747. ............................................

  5748. 

  5749. '3dnow'

  5750.      Target supports compiling '3dnow' instructions.

  5751. 

  5752. 'aes'

  5753.      Target supports compiling 'aes' instructions.

  5754. 

  5755. 'fma4'

  5756.      Target supports compiling 'fma4' instructions.

  5757. 

  5758. 'mfentry'

  5759.      Target supports the '-mfentry' option that alters the position of

  5760.      profiling calls such that they precede the prologue.

  5761. 

  5762. 'ms_hook_prologue'

  5763.      Target supports attribute 'ms_hook_prologue'.

  5764. 

  5765. 'pclmul'

  5766.      Target supports compiling 'pclmul' instructions.

  5767. 

  5768. 'sse3'

  5769.      Target supports compiling 'sse3' instructions.

  5770. 

  5771. 'sse4'

  5772.      Target supports compiling 'sse4' instructions.

  5773. 

  5774. 'sse4a'

  5775.      Target supports compiling 'sse4a' instructions.

  5776. 

  5777. 'ssse3'

  5778.      Target supports compiling 'ssse3' instructions.

  5779. 

  5780. 'vaes'

  5781.      Target supports compiling 'vaes' instructions.

  5782. 

  5783. 'vpclmul'

  5784.      Target supports compiling 'vpclmul' instructions.

  5785. 

  5786. 'xop'

  5787.      Target supports compiling 'xop' instructions.

  5788. 

  5789. 7.2.3.15 Local to tests in 'gcc.test-framework'

  5790. ...............................................

  5791. 

  5792. 'no'

  5793.      Always returns 0.

  5794. 

  5795. 'yes'

  5796.      Always returns 1.

  5797. 

  5798. 

  5799. File: gccint.info,  Node: Add Options,  Next: Require Support,  Prev: Effective-Target Keywords,  Up: Test Directives

  5800. 

  5801. 7.2.4 Features for 'dg-add-options'

  5802. -----------------------------------

  5803. 

  5804. The supported values of FEATURE for directive 'dg-add-options' are:

  5805. 

  5806. 'arm_fp'

  5807.      '__ARM_FP' definition.  Only ARM targets support this feature, and

  5808.      only then in certain modes; see the *note arm_fp_ok effective

  5809.      target keyword: arm_fp_ok.

  5810. 

  5811. 'arm_fp_dp'

  5812.      '__ARM_FP' definition with double-precision support.  Only ARM

  5813.      targets support this feature, and only then in certain modes; see

  5814.      the *note arm_fp_dp_ok effective target keyword: arm_fp_dp_ok.

  5815. 

  5816. 'arm_neon'

  5817.      NEON support.  Only ARM targets support this feature, and only then

  5818.      in certain modes; see the *note arm_neon_ok effective target

  5819.      keyword: arm_neon_ok.

  5820. 

  5821. 'arm_fp16'

  5822.      VFP half-precision floating point support.  This does not select

  5823.      the FP16 format; for that, use *note arm_fp16_ieee: arm_fp16_ieee.

  5824.      or *note arm_fp16_alternative: arm_fp16_alternative. instead.  This

  5825.      feature is only supported by ARM targets and then only in certain

  5826.      modes; see the *note arm_fp16_ok effective target keyword:

  5827.      arm_fp16_ok.

  5828. 

  5829. 'arm_fp16_ieee'

  5830.      ARM IEEE 754-2008 format VFP half-precision floating point support.

  5831.      This feature is only supported by ARM targets and then only in

  5832.      certain modes; see the *note arm_fp16_ok effective target keyword:

  5833.      arm_fp16_ok.

  5834. 

  5835. 'arm_fp16_alternative'

  5836.      ARM Alternative format VFP half-precision floating point support.

  5837.      This feature is only supported by ARM targets and then only in

  5838.      certain modes; see the *note arm_fp16_ok effective target keyword:

  5839.      arm_fp16_ok.

  5840. 

  5841. 'arm_neon_fp16'

  5842.      NEON and half-precision floating point support.  Only ARM targets

  5843.      support this feature, and only then in certain modes; see the *note

  5844.      arm_neon_fp16_ok effective target keyword: arm_neon_fp16_ok.

  5845. 

  5846. 'arm_vfp3'

  5847.      arm vfp3 floating point support; see the *note arm_vfp3_ok

  5848.      effective target keyword: arm_vfp3_ok.

  5849. 

  5850. 'arm_arch_v8a_hard'

  5851.      Add options for ARMv8-A and the hard-float variant of the AAPCS, if

  5852.      this is supported by the compiler; see the *note

  5853.      arm_arch_v8a_hard_ok: arm_arch_v8a_hard_ok. effective target

  5854.      keyword.

  5855. 

  5856. 'arm_v8_1a_neon'

  5857.      Add options for ARMv8.1-A with Adv.SIMD support, if this is

  5858.      supported by the target; see the *note arm_v8_1a_neon_ok:

  5859.      arm_v8_1a_neon_ok. effective target keyword.

  5860. 

  5861. 'arm_v8_2a_fp16_scalar'

  5862.      Add options for ARMv8.2-A with scalar FP16 support, if this is

  5863.      supported by the target; see the *note arm_v8_2a_fp16_scalar_ok:

  5864.      arm_v8_2a_fp16_scalar_ok. effective target keyword.

  5865. 

  5866. 'arm_v8_2a_fp16_neon'

  5867.      Add options for ARMv8.2-A with Adv.SIMD FP16 support, if this is

  5868.      supported by the target; see the *note arm_v8_2a_fp16_neon_ok:

  5869.      arm_v8_2a_fp16_neon_ok. effective target keyword.

  5870. 

  5871. 'arm_v8_2a_dotprod_neon'

  5872.      Add options for ARMv8.2-A with Adv.SIMD Dot Product support, if

  5873.      this is supported by the target; see the *note

  5874.      arm_v8_2a_dotprod_neon_ok:: effective target keyword.

  5875. 

  5876. 'arm_fp16fml_neon'

  5877.      Add options to enable generation of the 'VFMAL' and 'VFMSL'

  5878.      instructions, if this is supported by the target; see the *note

  5879.      arm_fp16fml_neon_ok:: effective target keyword.

  5880. 

  5881. 'bind_pic_locally'

  5882.      Add the target-specific flags needed to enable functions to bind

  5883.      locally when using pic/PIC passes in the testsuite.

  5884. 

  5885. 'floatN'

  5886.      Add the target-specific flags needed to use the '_FloatN' type.

  5887. 

  5888. 'floatNx'

  5889.      Add the target-specific flags needed to use the '_FloatNx' type.

  5890. 

  5891. 'ieee'

  5892.      Add the target-specific flags needed to enable full IEEE compliance

  5893.      mode.

  5894. 

  5895. 'mips16_attribute'

  5896.      'mips16' function attributes.  Only MIPS targets support this

  5897.      feature, and only then in certain modes.

  5898. 

  5899. 'stack_size'

  5900.      Add the flags needed to define macro STACK_SIZE and set it to the

  5901.      stack size limit associated with the *note 'stack_size' effective

  5902.      target: stack_size_et.

  5903. 

  5904. 'sqrt_insn'

  5905.      Add the target-specific flags needed to enable hardware square root

  5906.      instructions, if any.

  5907. 

  5908. 'tls'

  5909.      Add the target-specific flags needed to use thread-local storage.

  5910. 

  5911. 

  5912. File: gccint.info,  Node: Require Support,  Next: Final Actions,  Prev: Add Options,  Up: Test Directives

  5913. 

  5914. 7.2.5 Variants of 'dg-require-SUPPORT'

  5915. --------------------------------------

  5916. 

  5917. A few of the 'dg-require' directives take arguments.

  5918. 

  5919. 'dg-require-iconv CODESET'

  5920.      Skip the test if the target does not support iconv.  CODESET is the

  5921.      codeset to convert to.

  5922. 

  5923. 'dg-require-profiling PROFOPT'

  5924.      Skip the test if the target does not support profiling with option

  5925.      PROFOPT.

  5926. 

  5927. 'dg-require-stack-check CHECK'

  5928.      Skip the test if the target does not support the '-fstack-check'

  5929.      option.  If CHECK is '""', support for '-fstack-check' is checked,

  5930.      for '-fstack-check=("CHECK")' otherwise.

  5931. 

  5932. 'dg-require-stack-size SIZE'

  5933.      Skip the test if the target does not support a stack size of SIZE.

  5934. 

  5935. 'dg-require-visibility VIS'

  5936.      Skip the test if the target does not support the 'visibility'

  5937.      attribute.  If VIS is '""', support for 'visibility("hidden")' is

  5938.      checked, for 'visibility("VIS")' otherwise.

  5939. 

  5940.  The original 'dg-require' directives were defined before there was

  5941. support for effective-target keywords.  The directives that do not take

  5942. arguments could be replaced with effective-target keywords.

  5943. 

  5944. 'dg-require-alias ""'

  5945.      Skip the test if the target does not support the 'alias' attribute.

  5946. 

  5947. 'dg-require-ascii-locale ""'

  5948.      Skip the test if the host does not support an ASCII locale.

  5949. 

  5950. 'dg-require-compat-dfp ""'

  5951.      Skip this test unless both compilers in a 'compat' testsuite

  5952.      support decimal floating point.

  5953. 

  5954. 'dg-require-cxa-atexit ""'

  5955.      Skip the test if the target does not support '__cxa_atexit'.  This

  5956.      is equivalent to 'dg-require-effective-target cxa_atexit'.

  5957. 

  5958. 'dg-require-dll ""'

  5959.      Skip the test if the target does not support DLL attributes.

  5960. 

  5961. 'dg-require-dot ""'

  5962.      Skip the test if the host does not have 'dot'.

  5963. 

  5964. 'dg-require-fork ""'

  5965.      Skip the test if the target does not support 'fork'.

  5966. 

  5967. 'dg-require-gc-sections ""'

  5968.      Skip the test if the target's linker does not support the

  5969.      '--gc-sections' flags.  This is equivalent to

  5970.      'dg-require-effective-target gc-sections'.

  5971. 

  5972. 'dg-require-host-local ""'

  5973.      Skip the test if the host is remote, rather than the same as the

  5974.      build system.  Some tests are incompatible with DejaGnu's handling

  5975.      of remote hosts, which involves copying the source file to the host

  5976.      and compiling it with a relative path and "'-o a.out'".

  5977. 

  5978. 'dg-require-mkfifo ""'

  5979.      Skip the test if the target does not support 'mkfifo'.

  5980. 

  5981. 'dg-require-named-sections ""'

  5982.      Skip the test is the target does not support named sections.  This

  5983.      is equivalent to 'dg-require-effective-target named_sections'.

  5984. 

  5985. 'dg-require-weak ""'

  5986.      Skip the test if the target does not support weak symbols.

  5987. 

  5988. 'dg-require-weak-override ""'

  5989.      Skip the test if the target does not support overriding weak

  5990.      symbols.

  5991. 

  5992. 

  5993. File: gccint.info,  Node: Final Actions,  Prev: Require Support,  Up: Test Directives

  5994. 

  5995. 7.2.6 Commands for use in 'dg-final'

  5996. ------------------------------------

  5997. 

  5998. The GCC testsuite defines the following directives to be used within

  5999. 'dg-final'.

  6000. 

  6001. 7.2.6.1 Scan a particular file

  6002. ..............................

  6003. 

  6004. 'scan-file FILENAME REGEXP [{ target/xfail SELECTOR }]'

  6005.      Passes if REGEXP matches text in FILENAME.

  6006. 'scan-file-not FILENAME REGEXP [{ target/xfail SELECTOR }]'

  6007.      Passes if REGEXP does not match text in FILENAME.

  6008. 'scan-module MODULE REGEXP [{ target/xfail SELECTOR }]'

  6009.      Passes if REGEXP matches in Fortran module MODULE.

  6010. 'dg-check-dot FILENAME'

  6011.      Passes if FILENAME is a valid '.dot' file (by running 'dot -Tpng'

  6012.      on it, and verifying the exit code is 0).

  6013. 

  6014. 7.2.6.2 Scan the assembly output

  6015. ................................

  6016. 

  6017. 'scan-assembler REGEX [{ target/xfail SELECTOR }]'

  6018.      Passes if REGEX matches text in the test's assembler output.

  6019. 

  6020. 'scan-assembler-not REGEX [{ target/xfail SELECTOR }]'

  6021.      Passes if REGEX does not match text in the test's assembler output.

  6022. 

  6023. 'scan-assembler-times REGEX NUM [{ target/xfail SELECTOR }]'

  6024.      Passes if REGEX is matched exactly NUM times in the test's

  6025.      assembler output.

  6026. 

  6027. 'scan-assembler-dem REGEX [{ target/xfail SELECTOR }]'

  6028.      Passes if REGEX matches text in the test's demangled assembler

  6029.      output.

  6030. 

  6031. 'scan-assembler-dem-not REGEX [{ target/xfail SELECTOR }]'

  6032.      Passes if REGEX does not match text in the test's demangled

  6033.      assembler output.

  6034. 

  6035. 'scan-hidden SYMBOL [{ target/xfail SELECTOR }]'

  6036.      Passes if SYMBOL is defined as a hidden symbol in the test's

  6037.      assembly output.

  6038. 

  6039. 'scan-not-hidden SYMBOL [{ target/xfail SELECTOR }]'

  6040.      Passes if SYMBOL is not defined as a hidden symbol in the test's

  6041.      assembly output.

  6042. 

  6043. 'check-function-bodies PREFIX TERMINATOR [OPTIONS [{ target/xfail SELECTOR }]]'

  6044.      Looks through the source file for comments that give the expected

  6045.      assembly output for selected functions.  Each line of expected

  6046.      output starts with the prefix string PREFIX and the expected output

  6047.      for a function as a whole is followed by a line that starts with

  6048.      the string TERMINATOR.  Specifying an empty terminator is

  6049.      equivalent to specifying '"*/"'.

  6050. 

  6051.      OPTIONS, if specified, is a list of regular expressions, each of

  6052.      which matches a full command-line option.  A non-empty list

  6053.      prevents the test from running unless all of the given options are

  6054.      present on the command line.  This can help if a source file is

  6055.      compiled both with and without optimization, since it is rarely

  6056.      useful to check the full function body for unoptimized code.

  6057. 

  6058.      The first line of the expected output for a function FN has the

  6059.      form:

  6060. 

  6061.           PREFIX FN:  [{ target/xfail SELECTOR }]

  6062. 

  6063.      Subsequent lines of the expected output also start with PREFIX.  In

  6064.      both cases, whitespace after PREFIX is not significant.

  6065. 

  6066.      The test discards assembly directives such as '.cfi_startproc' and

  6067.      local label definitions such as '.LFB0' from the compiler's

  6068.      assembly output.  It then matches the result against the expected

  6069.      output for a function as a single regular expression.  This means

  6070.      that later lines can use backslashes to refer back to '(...)'

  6071.      captures on earlier lines.  For example:

  6072. 

  6073.           /* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" } } */

  6074.           ...

  6075.           /*

  6076.           ** add_w0_s8_m:

  6077.           **	mov	(z[0-9]+\.b), w0

  6078.           **	add	z0\.b, p0/m, z0\.b, \1

  6079.           **	ret

  6080.           */

  6081.           svint8_t add_w0_s8_m (...) { ... }

  6082.           ...

  6083.           /*

  6084.           ** add_b0_s8_m:

  6085.           **	mov	(z[0-9]+\.b), b0

  6086.           **	add	z1\.b, p0/m, z1\.b, \1

  6087.           **	ret

  6088.           */

  6089.           svint8_t add_b0_s8_m (...) { ... }

  6090. 

  6091.      checks whether the implementations of 'add_w0_s8_m' and

  6092.      'add_b0_s8_m' match the regular expressions given.  The test only

  6093.      runs when '-DCHECK_ASM' is passed on the command line.

  6094. 

  6095.      It is possible to create non-capturing multi-line regular

  6096.      expression groups of the form '(A|B|...)' by putting the '(', '|'

  6097.      and ')' on separate lines (each still using PREFIX).  For example:

  6098. 

  6099.           /*

  6100.           ** cmple_f16_tied:

  6101.           ** (

  6102.           **	fcmge	p0\.h, p0/z, z1\.h, z0\.h

  6103.           ** |

  6104.           **	fcmle	p0\.h, p0/z, z0\.h, z1\.h

  6105.           ** )

  6106.           **	ret

  6107.           */

  6108.           svbool_t cmple_f16_tied (...) { ... }

  6109. 

  6110.      checks whether 'cmple_f16_tied' is implemented by the 'fcmge'

  6111.      instruction followed by 'ret' or by the 'fcmle' instruction

  6112.      followed by 'ret'.  The test is still a single regular rexpression.

  6113. 

  6114.      A line containing just:

  6115. 

  6116.           PREFIX ...

  6117. 

  6118.      stands for zero or more unmatched lines; the whitespace after

  6119.      PREFIX is again not significant.

  6120. 

  6121. 7.2.6.3 Scan optimization dump files

  6122. ....................................

  6123. 

  6124. These commands are available for KIND of 'tree', 'ltrans-tree',

  6125. 'offload-tree', 'rtl', 'offload-rtl', 'ipa', and 'wpa-ipa'.

  6126. 

  6127. 'scan-KIND-dump REGEX SUFFIX [{ target/xfail SELECTOR }]'

  6128.      Passes if REGEX matches text in the dump file with suffix SUFFIX.

  6129. 

  6130. 'scan-KIND-dump-not REGEX SUFFIX [{ target/xfail SELECTOR }]'

  6131.      Passes if REGEX does not match text in the dump file with suffix

  6132.      SUFFIX.

  6133. 

  6134. 'scan-KIND-dump-times REGEX NUM SUFFIX [{ target/xfail SELECTOR }]'

  6135.      Passes if REGEX is found exactly NUM times in the dump file with

  6136.      suffix SUFFIX.

  6137. 

  6138. 'scan-KIND-dump-dem REGEX SUFFIX [{ target/xfail SELECTOR }]'

  6139.      Passes if REGEX matches demangled text in the dump file with suffix

  6140.      SUFFIX.

  6141. 

  6142. 'scan-KIND-dump-dem-not REGEX SUFFIX [{ target/xfail SELECTOR }]'

  6143.      Passes if REGEX does not match demangled text in the dump file with

  6144.      suffix SUFFIX.

  6145. 

  6146. 7.2.6.4 Check for output files

  6147. ..............................

  6148. 

  6149. 'output-exists [{ target/xfail SELECTOR }]'

  6150.      Passes if compiler output file exists.

  6151. 

  6152. 'output-exists-not [{ target/xfail SELECTOR }]'

  6153.      Passes if compiler output file does not exist.

  6154. 

  6155. 'scan-symbol REGEXP [{ target/xfail SELECTOR }]'

  6156.      Passes if the pattern is present in the final executable.

  6157. 

  6158. 'scan-symbol-not REGEXP [{ target/xfail SELECTOR }]'

  6159.      Passes if the pattern is absent from the final executable.

  6160. 

  6161. 7.2.6.5 Checks for 'gcov' tests

  6162. ...............................

  6163. 

  6164. 'run-gcov SOURCEFILE'

  6165.      Check line counts in 'gcov' tests.

  6166. 

  6167. 'run-gcov [branches] [calls] { OPTS SOURCEFILE }'

  6168.      Check branch and/or call counts, in addition to line counts, in

  6169.      'gcov' tests.

  6170. 

  6171. 7.2.6.6 Clean up generated test files

  6172. .....................................

  6173. 

  6174. Usually the test-framework removes files that were generated during

  6175. testing.  If a testcase, for example, uses any dumping mechanism to

  6176. inspect a passes dump file, the testsuite recognized the dump option

  6177. passed to the tool and schedules a final cleanup to remove these files.

  6178. 

  6179.  There are, however, following additional cleanup directives that can be

  6180. used to annotate a testcase "manually".

  6181. 'cleanup-coverage-files'

  6182.      Removes coverage data files generated for this test.

  6183. 

  6184. 'cleanup-modules "LIST-OF-EXTRA-MODULES"'

  6185.      Removes Fortran module files generated for this test, excluding the

  6186.      module names listed in keep-modules.  Cleaning up module files is

  6187.      usually done automatically by the testsuite by looking at the

  6188.      source files and removing the modules after the test has been

  6189.      executed.

  6190.           module MoD1

  6191.           end module MoD1

  6192.           module Mod2

  6193.           end module Mod2

  6194.           module moD3

  6195.           end module moD3

  6196.           module mod4

  6197.           end module mod4

  6198.           ! { dg-final { cleanup-modules "mod1 mod2" } } ! redundant

  6199.           ! { dg-final { keep-modules "mod3 mod4" } }

  6200. 

  6201. 'keep-modules "LIST-OF-MODULES-NOT-TO-DELETE"'

  6202.      Whitespace separated list of module names that should not be

  6203.      deleted by cleanup-modules.  If the list of modules is empty, all

  6204.      modules defined in this file are kept.

  6205.           module maybe_unneeded

  6206.           end module maybe_unneeded

  6207.           module keep1

  6208.           end module keep1

  6209.           module keep2

  6210.           end module keep2

  6211.           ! { dg-final { keep-modules "keep1 keep2" } } ! just keep these two

  6212.           ! { dg-final { keep-modules "" } } ! keep all

  6213. 

  6214. 'dg-keep-saved-temps "LIST-OF-SUFFIXES-NOT-TO-DELETE"'

  6215.      Whitespace separated list of suffixes that should not be deleted

  6216.      automatically in a testcase that uses '-save-temps'.

  6217.           // { dg-options "-save-temps -fpch-preprocess -I." }

  6218.           int main() { return 0; }

  6219.           // { dg-keep-saved-temps ".s" } ! just keep assembler file

  6220.           // { dg-keep-saved-temps ".s" ".i" } ! ... and .i

  6221.           // { dg-keep-saved-temps ".ii" ".o" } ! or just .ii and .o

  6222. 

  6223. 'cleanup-profile-file'

  6224.      Removes profiling files generated for this test.

  6225. 

  6226. 

  6227. File: gccint.info,  Node: Ada Tests,  Next: C Tests,  Prev: Test Directives,  Up: Testsuites

  6228. 

  6229. 7.3 Ada Language Testsuites

  6230. ===========================

  6231. 

  6232. The Ada testsuite includes executable tests from the ACATS testsuite,

  6233. publicly available at <http://www.ada-auth.org/acats.html>.

  6234. 

  6235.  These tests are integrated in the GCC testsuite in the 'ada/acats'

  6236. directory, and enabled automatically when running 'make check', assuming

  6237. the Ada language has been enabled when configuring GCC.

  6238. 

  6239.  You can also run the Ada testsuite independently, using 'make

  6240. check-ada', or run a subset of the tests by specifying which chapter to

  6241. run, e.g.:

  6242. 

  6243.      $ make check-ada CHAPTERS="c3 c9"

  6244. 

  6245.  The tests are organized by directory, each directory corresponding to a

  6246. chapter of the Ada Reference Manual.  So for example, 'c9' corresponds

  6247. to chapter 9, which deals with tasking features of the language.

  6248. 

  6249.  The tests are run using two 'sh' scripts: 'run_acats' and 'run_all.sh'.

  6250. To run the tests using a simulator or a cross target, see the small

  6251. customization section at the top of 'run_all.sh'.

  6252. 

  6253.  These tests are run using the build tree: they can be run without doing

  6254. a 'make install'.

  6255. 

  6256. 

  6257. File: gccint.info,  Node: C Tests,  Next: LTO Testing,  Prev: Ada Tests,  Up: Testsuites

  6258. 

  6259. 7.4 C Language Testsuites

  6260. =========================

  6261. 

  6262. GCC contains the following C language testsuites, in the 'gcc/testsuite'

  6263. directory:

  6264. 

  6265. 'gcc.dg'

  6266.      This contains tests of particular features of the C compiler, using

  6267.      the more modern 'dg' harness.  Correctness tests for various

  6268.      compiler features should go here if possible.

  6269. 

  6270.      Magic comments determine whether the file is preprocessed,

  6271.      compiled, linked or run.  In these tests, error and warning message

  6272.      texts are compared against expected texts or regular expressions

  6273.      given in comments.  These tests are run with the options '-ansi

  6274.      -pedantic' unless other options are given in the test.  Except as

  6275.      noted below they are not run with multiple optimization options.

  6276. 'gcc.dg/compat'

  6277.      This subdirectory contains tests for binary compatibility using

  6278.      'lib/compat.exp', which in turn uses the language-independent

  6279.      support (*note Support for testing binary compatibility: compat

  6280.      Testing.).

  6281. 'gcc.dg/cpp'

  6282.      This subdirectory contains tests of the preprocessor.

  6283. 'gcc.dg/debug'

  6284.      This subdirectory contains tests for debug formats.  Tests in this

  6285.      subdirectory are run for each debug format that the compiler

  6286.      supports.

  6287. 'gcc.dg/format'

  6288.      This subdirectory contains tests of the '-Wformat' format checking.

  6289.      Tests in this directory are run with and without '-DWIDE'.

  6290. 'gcc.dg/noncompile'

  6291.      This subdirectory contains tests of code that should not compile

  6292.      and does not need any special compilation options.  They are run

  6293.      with multiple optimization options, since sometimes invalid code

  6294.      crashes the compiler with optimization.

  6295. 'gcc.dg/special'

  6296.      FIXME: describe this.

  6297. 

  6298. 'gcc.c-torture'

  6299.      This contains particular code fragments which have historically

  6300.      broken easily.  These tests are run with multiple optimization

  6301.      options, so tests for features which only break at some

  6302.      optimization levels belong here.  This also contains tests to check

  6303.      that certain optimizations occur.  It might be worthwhile to

  6304.      separate the correctness tests cleanly from the code quality tests,

  6305.      but it hasn't been done yet.

  6306. 

  6307. 'gcc.c-torture/compat'

  6308.      FIXME: describe this.

  6309. 

  6310.      This directory should probably not be used for new tests.

  6311. 'gcc.c-torture/compile'

  6312.      This testsuite contains test cases that should compile, but do not

  6313.      need to link or run.  These test cases are compiled with several

  6314.      different combinations of optimization options.  All warnings are

  6315.      disabled for these test cases, so this directory is not suitable if

  6316.      you wish to test for the presence or absence of compiler warnings.

  6317.      While special options can be set, and tests disabled on specific

  6318.      platforms, by the use of '.x' files, mostly these test cases should

  6319.      not contain platform dependencies.  FIXME: discuss how defines such

  6320.      as 'STACK_SIZE' are used.

  6321. 'gcc.c-torture/execute'

  6322.      This testsuite contains test cases that should compile, link and

  6323.      run; otherwise the same comments as for 'gcc.c-torture/compile'

  6324.      apply.

  6325. 'gcc.c-torture/execute/ieee'

  6326.      This contains tests which are specific to IEEE floating point.

  6327. 'gcc.c-torture/unsorted'

  6328.      FIXME: describe this.

  6329. 

  6330.      This directory should probably not be used for new tests.

  6331. 'gcc.misc-tests'

  6332.      This directory contains C tests that require special handling.

  6333.      Some of these tests have individual expect files, and others share

  6334.      special-purpose expect files:

  6335. 

  6336.      'bprob*.c'

  6337.           Test '-fbranch-probabilities' using

  6338.           'gcc.misc-tests/bprob.exp', which in turn uses the generic,

  6339.           language-independent framework (*note Support for testing

  6340.           profile-directed optimizations: profopt Testing.).

  6341. 

  6342.      'gcov*.c'

  6343.           Test 'gcov' output using 'gcov.exp', which in turn uses the

  6344.           language-independent support (*note Support for testing gcov:

  6345.           gcov Testing.).

  6346. 

  6347.      'i386-pf-*.c'

  6348.           Test i386-specific support for data prefetch using

  6349.           'i386-prefetch.exp'.

  6350. 

  6351. 'gcc.test-framework'

  6352.      'dg-*.c'

  6353.           Test the testsuite itself using

  6354.           'gcc.test-framework/test-framework.exp'.

  6355. 

  6356.  FIXME: merge in 'testsuite/README.gcc' and discuss the format of test

  6357. cases and magic comments more.

  6358. 

  6359. 

  6360. File: gccint.info,  Node: LTO Testing,  Next: gcov Testing,  Prev: C Tests,  Up: Testsuites

  6361. 

  6362. 7.5 Support for testing link-time optimizations

  6363. ===============================================

  6364. 

  6365. Tests for link-time optimizations usually require multiple source files

  6366. that are compiled separately, perhaps with different sets of options.

  6367. There are several special-purpose test directives used for these tests.

  6368. 

  6369. '{ dg-lto-do DO-WHAT-KEYWORD }'

  6370.      DO-WHAT-KEYWORD specifies how the test is compiled and whether it

  6371.      is executed.  It is one of:

  6372. 

  6373.      'assemble'

  6374.           Compile with '-c' to produce a relocatable object file.

  6375.      'link'

  6376.           Compile, assemble, and link to produce an executable file.

  6377.      'run'

  6378.           Produce and run an executable file, which is expected to

  6379.           return an exit code of 0.

  6380. 

  6381.      The default is 'assemble'.  That can be overridden for a set of

  6382.      tests by redefining 'dg-do-what-default' within the '.exp' file for

  6383.      those tests.

  6384. 

  6385.      Unlike 'dg-do', 'dg-lto-do' does not support an optional 'target'

  6386.      or 'xfail' list.  Use 'dg-skip-if', 'dg-xfail-if', or

  6387.      'dg-xfail-run-if'.

  6388. 

  6389. '{ dg-lto-options { { OPTIONS } [{ OPTIONS }] } [{ target SELECTOR }]}'

  6390.      This directive provides a list of one or more sets of compiler

  6391.      options to override LTO_OPTIONS.  Each test will be compiled and

  6392.      run with each of these sets of options.

  6393. 

  6394. '{ dg-extra-ld-options OPTIONS [{ target SELECTOR }]}'

  6395.      This directive adds OPTIONS to the linker options used.

  6396. 

  6397. '{ dg-suppress-ld-options OPTIONS [{ target SELECTOR }]}'

  6398.      This directive removes OPTIONS from the set of linker options used.

  6399. 

  6400. 

  6401. File: gccint.info,  Node: gcov Testing,  Next: profopt Testing,  Prev: LTO Testing,  Up: Testsuites

  6402. 

  6403. 7.6 Support for testing 'gcov'

  6404. ==============================

  6405. 

  6406. Language-independent support for testing 'gcov', and for checking that

  6407. branch profiling produces expected values, is provided by the expect

  6408. file 'lib/gcov.exp'.  'gcov' tests also rely on procedures in

  6409. 'lib/gcc-dg.exp' to compile and run the test program.  A typical 'gcov'

  6410. test contains the following DejaGnu commands within comments:

  6411. 

  6412.      { dg-options "--coverage" }

  6413.      { dg-do run { target native } }

  6414.      { dg-final { run-gcov sourcefile } }

  6415. 

  6416.  Checks of 'gcov' output can include line counts, branch percentages,

  6417. and call return percentages.  All of these checks are requested via

  6418. commands that appear in comments in the test's source file.  Commands to

  6419. check line counts are processed by default.  Commands to check branch

  6420. percentages and call return percentages are processed if the 'run-gcov'

  6421. command has arguments 'branches' or 'calls', respectively.  For example,

  6422. the following specifies checking both, as well as passing '-b' to

  6423. 'gcov':

  6424. 

  6425.      { dg-final { run-gcov branches calls { -b sourcefile } } }

  6426. 

  6427.  A line count command appears within a comment on the source line that

  6428. is expected to get the specified count and has the form 'count(CNT)'.  A

  6429. test should only check line counts for lines that will get the same

  6430. count for any architecture.

  6431. 

  6432.  Commands to check branch percentages ('branch') and call return

  6433. percentages ('returns') are very similar to each other.  A beginning

  6434. command appears on or before the first of a range of lines that will

  6435. report the percentage, and the ending command follows that range of

  6436. lines.  The beginning command can include a list of percentages, all of

  6437. which are expected to be found within the range.  A range is terminated

  6438. by the next command of the same kind.  A command 'branch(end)' or

  6439. 'returns(end)' marks the end of a range without starting a new one.  For

  6440. example:

  6441. 

  6442.      if (i > 10 && j > i && j < 20)  /* branch(27 50 75) */

  6443.                                      /* branch(end) */

  6444.        foo (i, j);

  6445. 

  6446.  For a call return percentage, the value specified is the percentage of

  6447. calls reported to return.  For a branch percentage, the value is either

  6448. the expected percentage or 100 minus that value, since the direction of

  6449. a branch can differ depending on the target or the optimization level.

  6450. 

  6451.  Not all branches and calls need to be checked.  A test should not check

  6452. for branches that might be optimized away or replaced with predicated

  6453. instructions.  Don't check for calls inserted by the compiler or ones

  6454. that might be inlined or optimized away.

  6455. 

  6456.  A single test can check for combinations of line counts, branch

  6457. percentages, and call return percentages.  The command to check a line

  6458. count must appear on the line that will report that count, but commands

  6459. to check branch percentages and call return percentages can bracket the

  6460. lines that report them.

  6461. 

  6462. 

  6463. File: gccint.info,  Node: profopt Testing,  Next: compat Testing,  Prev: gcov Testing,  Up: Testsuites

  6464. 

  6465. 7.7 Support for testing profile-directed optimizations

  6466. ======================================================

  6467. 

  6468. The file 'profopt.exp' provides language-independent support for

  6469. checking correct execution of a test built with profile-directed

  6470. optimization.  This testing requires that a test program be built and

  6471. executed twice.  The first time it is compiled to generate profile data,

  6472. and the second time it is compiled to use the data that was generated

  6473. during the first execution.  The second execution is to verify that the

  6474. test produces the expected results.

  6475. 

  6476.  To check that the optimization actually generated better code, a test

  6477. can be built and run a third time with normal optimizations to verify

  6478. that the performance is better with the profile-directed optimizations.

  6479. 'profopt.exp' has the beginnings of this kind of support.

  6480. 

  6481.  'profopt.exp' provides generic support for profile-directed

  6482. optimizations.  Each set of tests that uses it provides information

  6483. about a specific optimization:

  6484. 

  6485. 'tool'

  6486.      tool being tested, e.g., 'gcc'

  6487. 

  6488. 'profile_option'

  6489.      options used to generate profile data

  6490. 

  6491. 'feedback_option'

  6492.      options used to optimize using that profile data

  6493. 

  6494. 'prof_ext'

  6495.      suffix of profile data files

  6496. 

  6497. 'PROFOPT_OPTIONS'

  6498.      list of options with which to run each test, similar to the lists

  6499.      for torture tests

  6500. 

  6501. '{ dg-final-generate { LOCAL-DIRECTIVE } }'

  6502.      This directive is similar to 'dg-final', but the LOCAL-DIRECTIVE is

  6503.      run after the generation of profile data.

  6504. 

  6505. '{ dg-final-use { LOCAL-DIRECTIVE } }'

  6506.      The LOCAL-DIRECTIVE is run after the profile data have been used.

  6507. 

  6508. 

  6509. File: gccint.info,  Node: compat Testing,  Next: Torture Tests,  Prev: profopt Testing,  Up: Testsuites

  6510. 

  6511. 7.8 Support for testing binary compatibility

  6512. ============================================

  6513. 

  6514. The file 'compat.exp' provides language-independent support for binary

  6515. compatibility testing.  It supports testing interoperability of two

  6516. compilers that follow the same ABI, or of multiple sets of compiler

  6517. options that should not affect binary compatibility.  It is intended to

  6518. be used for testsuites that complement ABI testsuites.

  6519. 

  6520.  A test supported by this framework has three parts, each in a separate

  6521. source file: a main program and two pieces that interact with each other

  6522. to split up the functionality being tested.

  6523. 

  6524. 'TESTNAME_main.SUFFIX'

  6525.      Contains the main program, which calls a function in file

  6526.      'TESTNAME_x.SUFFIX'.

  6527. 

  6528. 'TESTNAME_x.SUFFIX'

  6529.      Contains at least one call to a function in 'TESTNAME_y.SUFFIX'.

  6530. 

  6531. 'TESTNAME_y.SUFFIX'

  6532.      Shares data with, or gets arguments from, 'TESTNAME_x.SUFFIX'.

  6533. 

  6534.  Within each test, the main program and one functional piece are

  6535. compiled by the GCC under test.  The other piece can be compiled by an

  6536. alternate compiler.  If no alternate compiler is specified, then all

  6537. three source files are all compiled by the GCC under test.  You can

  6538. specify pairs of sets of compiler options.  The first element of such a

  6539. pair specifies options used with the GCC under test, and the second

  6540. element of the pair specifies options used with the alternate compiler.

  6541. Each test is compiled with each pair of options.

  6542. 

  6543.  'compat.exp' defines default pairs of compiler options.  These can be

  6544. overridden by defining the environment variable 'COMPAT_OPTIONS' as:

  6545. 

  6546.      COMPAT_OPTIONS="[list [list {TST1} {ALT1}]

  6547.        ...[list {TSTN} {ALTN}]]"

  6548. 

  6549.  where TSTI and ALTI are lists of options, with TSTI used by the

  6550. compiler under test and ALTI used by the alternate compiler.  For

  6551. example, with '[list [list {-g -O0} {-O3}] [list {-fpic} {-fPIC -O2}]]',

  6552. the test is first built with '-g -O0' by the compiler under test and

  6553. with '-O3' by the alternate compiler.  The test is built a second time

  6554. using '-fpic' by the compiler under test and '-fPIC -O2' by the

  6555. alternate compiler.

  6556. 

  6557.  An alternate compiler is specified by defining an environment variable

  6558. to be the full pathname of an installed compiler; for C define

  6559. 'ALT_CC_UNDER_TEST', and for C++ define 'ALT_CXX_UNDER_TEST'.  These

  6560. will be written to the 'site.exp' file used by DejaGnu.  The default is

  6561. to build each test with the compiler under test using the first of each

  6562. pair of compiler options from 'COMPAT_OPTIONS'.  When

  6563. 'ALT_CC_UNDER_TEST' or 'ALT_CXX_UNDER_TEST' is 'same', each test is

  6564. built using the compiler under test but with combinations of the options

  6565. from 'COMPAT_OPTIONS'.

  6566. 

  6567.  To run only the C++ compatibility suite using the compiler under test

  6568. and another version of GCC using specific compiler options, do the

  6569. following from 'OBJDIR/gcc':

  6570. 

  6571.      rm site.exp

  6572.      make -k \

  6573.        ALT_CXX_UNDER_TEST=${alt_prefix}/bin/g++ \

  6574.        COMPAT_OPTIONS="LISTS AS SHOWN ABOVE" \

  6575.        check-c++ \

  6576.        RUNTESTFLAGS="compat.exp"

  6577. 

  6578.  A test that fails when the source files are compiled with different

  6579. compilers, but passes when the files are compiled with the same

  6580. compiler, demonstrates incompatibility of the generated code or runtime

  6581. support.  A test that fails for the alternate compiler but passes for

  6582. the compiler under test probably tests for a bug that was fixed in the

  6583. compiler under test but is present in the alternate compiler.

  6584. 

  6585.  The binary compatibility tests support a small number of test framework

  6586. commands that appear within comments in a test file.

  6587. 

  6588. 'dg-require-*'

  6589.      These commands can be used in 'TESTNAME_main.SUFFIX' to skip the

  6590.      test if specific support is not available on the target.

  6591. 

  6592. 'dg-options'

  6593.      The specified options are used for compiling this particular source

  6594.      file, appended to the options from 'COMPAT_OPTIONS'.  When this

  6595.      command appears in 'TESTNAME_main.SUFFIX' the options are also used

  6596.      to link the test program.

  6597. 

  6598. 'dg-xfail-if'

  6599.      This command can be used in a secondary source file to specify that

  6600.      compilation is expected to fail for particular options on

  6601.      particular targets.

  6602. 

  6603. 

  6604. File: gccint.info,  Node: Torture Tests,  Next: GIMPLE Tests,  Prev: compat Testing,  Up: Testsuites

  6605. 

  6606. 7.9 Support for torture testing using multiple options

  6607. ======================================================

  6608. 

  6609. Throughout the compiler testsuite there are several directories whose

  6610. tests are run multiple times, each with a different set of options.

  6611. These are known as torture tests.  'lib/torture-options.exp' defines

  6612. procedures to set up these lists:

  6613. 

  6614. 'torture-init'

  6615.      Initialize use of torture lists.

  6616. 'set-torture-options'

  6617.      Set lists of torture options to use for tests with and without

  6618.      loops.  Optionally combine a set of torture options with a set of

  6619.      other options, as is done with Objective-C runtime options.

  6620. 'torture-finish'

  6621.      Finalize use of torture lists.

  6622. 

  6623.  The '.exp' file for a set of tests that use torture options must

  6624. include calls to these three procedures if:

  6625. 

  6626.    * It calls 'gcc-dg-runtest' and overrides DG_TORTURE_OPTIONS.

  6627. 

  6628.    * It calls ${TOOL}'-torture' or ${TOOL}'-torture-execute', where TOOL

  6629.      is 'c', 'fortran', or 'objc'.

  6630. 

  6631.    * It calls 'dg-pch'.

  6632. 

  6633.  It is not necessary for a '.exp' file that calls 'gcc-dg-runtest' to

  6634. call the torture procedures if the tests should use the list in

  6635. DG_TORTURE_OPTIONS defined in 'gcc-dg.exp'.

  6636. 

  6637.  Most uses of torture options can override the default lists by defining

  6638. TORTURE_OPTIONS or add to the default list by defining

  6639. ADDITIONAL_TORTURE_OPTIONS.  Define these in a '.dejagnurc' file or add

  6640. them to the 'site.exp' file; for example

  6641. 

  6642.      set ADDITIONAL_TORTURE_OPTIONS  [list \

  6643.        { -O2 -ftree-loop-linear } \

  6644.        { -O2 -fpeel-loops } ]

  6645. 

  6646. 

  6647. File: gccint.info,  Node: GIMPLE Tests,  Next: RTL Tests,  Prev: Torture Tests,  Up: Testsuites

  6648. 

  6649. 7.10 Support for testing GIMPLE passes

  6650. ======================================

  6651. 

  6652. As of gcc 7, C functions can be tagged with '__GIMPLE' to indicate that

  6653. the function body will be GIMPLE, rather than C. The compiler requires

  6654. the option '-fgimple' to enable this functionality.  For example:

  6655. 

  6656.      /* { dg-do compile } */

  6657.      /* { dg-options "-O -fgimple" } */

  6658. 

  6659.      void __GIMPLE (startwith ("dse2")) foo ()

  6660.      {

  6661.        int a;

  6662. 

  6663.      bb_2:

  6664.        if (a > 4)

  6665.          goto bb_3;

  6666.        else

  6667.          goto bb_4;

  6668. 

  6669.      bb_3:

  6670.        a_2 = 10;

  6671.        goto bb_5;

  6672. 

  6673.      bb_4:

  6674.        a_3 = 20;

  6675. 

  6676.      bb_5:

  6677.        a_1 = __PHI (bb_3: a_2, bb_4: a_3);

  6678.        a_4 = a_1 + 4;

  6679. 

  6680.        return;

  6681.      }

  6682. 

  6683.  The 'startwith' argument indicates at which pass to begin.

  6684. 

  6685.  Use the dump modifier '-gimple' (e.g. '-fdump-tree-all-gimple') to make

  6686. tree dumps more closely follow the format accepted by the GIMPLE parser.

  6687. 

  6688.  Example DejaGnu tests of GIMPLE can be seen in the source tree at

  6689. 'gcc/testsuite/gcc.dg/gimplefe-*.c'.

  6690. 

  6691.  The '__GIMPLE' parser is integrated with the C tokenizer and

  6692. preprocessor, so it should be possible to use macros to build out test

  6693. coverage.

  6694. 

  6695. 

  6696. File: gccint.info,  Node: RTL Tests,  Prev: GIMPLE Tests,  Up: Testsuites

  6697. 

  6698. 7.11 Support for testing RTL passes

  6699. ===================================

  6700. 

  6701. As of gcc 7, C functions can be tagged with '__RTL' to indicate that the

  6702. function body will be RTL, rather than C. For example:

  6703. 

  6704.      double __RTL (startwith ("ira")) test (struct foo *f, const struct bar *b)

  6705.      {

  6706.        (function "test"

  6707.           [...snip; various directives go in here...]

  6708.        ) ;; function "test"

  6709.      }

  6710. 

  6711.  The 'startwith' argument indicates at which pass to begin.

  6712. 

  6713.  The parser expects the RTL body to be in the format emitted by this

  6714. dumping function:

  6715. 

  6716.      DEBUG_FUNCTION void

  6717.      print_rtx_function (FILE *outfile, function *fn, bool compact);

  6718. 

  6719.  when "compact" is true.  So you can capture RTL in the correct format

  6720. from the debugger using:

  6721. 

  6722.      (gdb) print_rtx_function (stderr, cfun, true);

  6723. 

  6724.  and copy and paste the output into the body of the C function.

  6725. 

  6726.  Example DejaGnu tests of RTL can be seen in the source tree under

  6727. 'gcc/testsuite/gcc.dg/rtl'.

  6728. 

  6729.  The '__RTL' parser is not integrated with the C tokenizer or

  6730. preprocessor, and works simply by reading the relevant lines within the

  6731. braces.  In particular, the RTL body must be on separate lines from the

  6732. enclosing braces, and the preprocessor is not usable within it.

  6733. 

  6734. 

  6735. File: gccint.info,  Node: Options,  Next: Passes,  Prev: Testsuites,  Up: Top

  6736. 

  6737. 8 Option specification files

  6738. ****************************

  6739. 

  6740. Most GCC command-line options are described by special option definition

  6741. files, the names of which conventionally end in '.opt'.  This chapter

  6742. describes the format of these files.

  6743. 

  6744. * Menu:

  6745. 

  6746. * Option file format::   The general layout of the files

  6747. * Option properties::    Supported option properties

  6748. 

  6749. 

  6750. File: gccint.info,  Node: Option file format,  Next: Option properties,  Up: Options

  6751. 

  6752. 8.1 Option file format

  6753. ======================

  6754. 

  6755. Option files are a simple list of records in which each field occupies

  6756. its own line and in which the records themselves are separated by blank

  6757. lines.  Comments may appear on their own line anywhere within the file

  6758. and are preceded by semicolons.  Whitespace is allowed before the

  6759. semicolon.

  6760. 

  6761.  The files can contain the following types of record:

  6762. 

  6763.    * A language definition record.  These records have two fields: the

  6764.      string 'Language' and the name of the language.  Once a language

  6765.      has been declared in this way, it can be used as an option

  6766.      property.  *Note Option properties::.

  6767. 

  6768.    * A target specific save record to save additional information.

  6769.      These records have two fields: the string 'TargetSave', and a

  6770.      declaration type to go in the 'cl_target_option' structure.

  6771. 

  6772.    * A variable record to define a variable used to store option

  6773.      information.  These records have two fields: the string 'Variable',

  6774.      and a declaration of the type and name of the variable, optionally

  6775.      with an initializer (but without any trailing ';').  These records

  6776.      may be used for variables used for many options where declaring the

  6777.      initializer in a single option definition record, or duplicating it

  6778.      in many records, would be inappropriate, or for variables set in

  6779.      option handlers rather than referenced by 'Var' properties.

  6780. 

  6781.    * A variable record to define a variable used to store option

  6782.      information.  These records have two fields: the string

  6783.      'TargetVariable', and a declaration of the type and name of the

  6784.      variable, optionally with an initializer (but without any trailing

  6785.      ';').  'TargetVariable' is a combination of 'Variable' and

  6786.      'TargetSave' records in that the variable is defined in the

  6787.      'gcc_options' structure, but these variables are also stored in the

  6788.      'cl_target_option' structure.  The variables are saved in the

  6789.      target save code and restored in the target restore code.

  6790. 

  6791.    * A variable record to record any additional files that the

  6792.      'options.h' file should include.  This is useful to provide

  6793.      enumeration or structure definitions needed for target variables.

  6794.      These records have two fields: the string 'HeaderInclude' and the

  6795.      name of the include file.

  6796. 

  6797.    * A variable record to record any additional files that the

  6798.      'options.c' or 'options-save.c' file should include.  This is

  6799.      useful to provide inline functions needed for target variables

  6800.      and/or '#ifdef' sequences to properly set up the initialization.

  6801.      These records have two fields: the string 'SourceInclude' and the

  6802.      name of the include file.

  6803. 

  6804.    * An enumeration record to define a set of strings that may be used

  6805.      as arguments to an option or options.  These records have three

  6806.      fields: the string 'Enum', a space-separated list of properties and

  6807.      help text used to describe the set of strings in '--help' output.

  6808.      Properties use the same format as option properties; the following

  6809.      are valid:

  6810.      'Name(NAME)'

  6811.           This property is required; NAME must be a name (suitable for

  6812.           use in C identifiers) used to identify the set of strings in

  6813.           'Enum' option properties.

  6814. 

  6815.      'Type(TYPE)'

  6816.           This property is required; TYPE is the C type for variables

  6817.           set by options using this enumeration together with 'Var'.

  6818. 

  6819.      'UnknownError(MESSAGE)'

  6820.           The message MESSAGE will be used as an error message if the

  6821.           argument is invalid; for enumerations without 'UnknownError',

  6822.           a generic error message is used.  MESSAGE should contain a

  6823.           single '%qs' format, which will be used to format the invalid

  6824.           argument.

  6825. 

  6826.    * An enumeration value record to define one of the strings in a set

  6827.      given in an 'Enum' record.  These records have two fields: the

  6828.      string 'EnumValue' and a space-separated list of properties.

  6829.      Properties use the same format as option properties; the following

  6830.      are valid:

  6831.      'Enum(NAME)'

  6832.           This property is required; NAME says which 'Enum' record this

  6833.           'EnumValue' record corresponds to.

  6834. 

  6835.      'String(STRING)'

  6836.           This property is required; STRING is the string option

  6837.           argument being described by this record.

  6838. 

  6839.      'Value(VALUE)'

  6840.           This property is required; it says what value (representable

  6841.           as 'int') should be used for the given string.

  6842. 

  6843.      'Canonical'

  6844.           This property is optional.  If present, it says the present

  6845.           string is the canonical one among all those with the given

  6846.           value.  Other strings yielding that value will be mapped to

  6847.           this one so specs do not need to handle them.

  6848. 

  6849.      'DriverOnly'

  6850.           This property is optional.  If present, the present string

  6851.           will only be accepted by the driver.  This is used for cases

  6852.           such as '-march=native' that are processed by the driver so

  6853.           that 'gcc -v' shows how the options chosen depended on the

  6854.           system on which the compiler was run.

  6855. 

  6856.    * An option definition record.  These records have the following

  6857.      fields:

  6858.        1. the name of the option, with the leading "-" removed

  6859.        2. a space-separated list of option properties (*note Option

  6860.           properties::)

  6861.        3. the help text to use for '--help' (omitted if the second field

  6862.           contains the 'Undocumented' property).

  6863. 

  6864.      By default, all options beginning with "f", "W" or "m" are

  6865.      implicitly assumed to take a "no-" form.  This form should not be

  6866.      listed separately.  If an option beginning with one of these

  6867.      letters does not have a "no-" form, you can use the

  6868.      'RejectNegative' property to reject it.

  6869. 

  6870.      The help text is automatically line-wrapped before being displayed.

  6871.      Normally the name of the option is printed on the left-hand side of

  6872.      the output and the help text is printed on the right.  However, if

  6873.      the help text contains a tab character, the text to the left of the

  6874.      tab is used instead of the option's name and the text to the right

  6875.      of the tab forms the help text.  This allows you to elaborate on

  6876.      what type of argument the option takes.

  6877. 

  6878.    * A target mask record.  These records have one field of the form

  6879.      'Mask(X)'.  The options-processing script will automatically

  6880.      allocate a bit in 'target_flags' (*note Run-time Target::) for each

  6881.      mask name X and set the macro 'MASK_X' to the appropriate bitmask.

  6882.      It will also declare a 'TARGET_X' macro that has the value 1 when

  6883.      bit 'MASK_X' is set and 0 otherwise.

  6884. 

  6885.      They are primarily intended to declare target masks that are not

  6886.      associated with user options, either because these masks represent

  6887.      internal switches or because the options are not available on all

  6888.      configurations and yet the masks always need to be defined.

  6889. 

  6890. 

  6891. File: gccint.info,  Node: Option properties,  Prev: Option file format,  Up: Options

  6892. 

  6893. 8.2 Option properties

  6894. =====================

  6895. 

  6896. The second field of an option record can specify any of the following

  6897. properties.  When an option takes an argument, it is enclosed in

  6898. parentheses following the option property name.  The parser that handles

  6899. option files is quite simplistic, and will be tricked by any nested

  6900. parentheses within the argument text itself; in this case, the entire

  6901. option argument can be wrapped in curly braces within the parentheses to

  6902. demarcate it, e.g.:

  6903. 

  6904.      Condition({defined (USE_CYGWIN_LIBSTDCXX_WRAPPERS)})

  6905. 

  6906. 'Common'

  6907.      The option is available for all languages and targets.

  6908. 

  6909. 'Target'

  6910.      The option is available for all languages but is target-specific.

  6911. 

  6912. 'Driver'

  6913.      The option is handled by the compiler driver using code not shared

  6914.      with the compilers proper ('cc1' etc.).

  6915. 

  6916. 'LANGUAGE'

  6917.      The option is available when compiling for the given language.

  6918. 

  6919.      It is possible to specify several different languages for the same

  6920.      option.  Each LANGUAGE must have been declared by an earlier

  6921.      'Language' record.  *Note Option file format::.

  6922. 

  6923. 'RejectDriver'

  6924.      The option is only handled by the compilers proper ('cc1' etc.) and

  6925.      should not be accepted by the driver.

  6926. 

  6927. 'RejectNegative'

  6928.      The option does not have a "no-" form.  All options beginning with

  6929.      "f", "W" or "m" are assumed to have a "no-" form unless this

  6930.      property is used.

  6931. 

  6932. 'Negative(OTHERNAME)'

  6933.      The option will turn off another option OTHERNAME, which is the

  6934.      option name with the leading "-" removed.  This chain action will

  6935.      propagate through the 'Negative' property of the option to be

  6936.      turned off.  The driver will prune options, removing those that are

  6937.      turned off by some later option.  This pruning is not done for

  6938.      options with 'Joined' or 'JoinedOrMissing' properties, unless the

  6939.      options have either 'RejectNegative' property or the 'Negative'

  6940.      property mentions an option other than itself.

  6941. 

  6942.      As a consequence, if you have a group of mutually-exclusive

  6943.      options, their 'Negative' properties should form a circular chain.

  6944.      For example, if options '-A', '-B' and '-C' are mutually exclusive,

  6945.      their respective 'Negative' properties should be 'Negative(B)',

  6946.      'Negative(C)' and 'Negative(A)'.

  6947. 

  6948. 'Joined'

  6949. 'Separate'

  6950.      The option takes a mandatory argument.  'Joined' indicates that the

  6951.      option and argument can be included in the same 'argv' entry (as

  6952.      with '-mflush-func=NAME', for example).  'Separate' indicates that

  6953.      the option and argument can be separate 'argv' entries (as with

  6954.      '-o').  An option is allowed to have both of these properties.

  6955. 

  6956. 'JoinedOrMissing'

  6957.      The option takes an optional argument.  If the argument is given,

  6958.      it will be part of the same 'argv' entry as the option itself.

  6959. 

  6960.      This property cannot be used alongside 'Joined' or 'Separate'.

  6961. 

  6962. 'MissingArgError(MESSAGE)'

  6963.      For an option marked 'Joined' or 'Separate', the message MESSAGE

  6964.      will be used as an error message if the mandatory argument is

  6965.      missing; for options without 'MissingArgError', a generic error

  6966.      message is used.  MESSAGE should contain a single '%qs' format,

  6967.      which will be used to format the name of the option passed.

  6968. 

  6969. 'Args(N)'

  6970.      For an option marked 'Separate', indicate that it takes N

  6971.      arguments.  The default is 1.

  6972. 

  6973. 'UInteger'

  6974.      The option's argument is a non-negative integer consisting of

  6975.      either decimal or hexadecimal digits interpreted as 'int'.

  6976.      Hexadecimal integers may optionally start with the '0x' or '0X'

  6977.      prefix.  The option parser validates and converts the argument

  6978.      before passing it to the relevant option handler.  'UInteger'

  6979.      should also be used with options like '-falign-loops' where both

  6980.      '-falign-loops' and '-falign-loops'=N are supported to make sure

  6981.      the saved options are given a full integer.  Positive values of the

  6982.      argument in excess of 'INT_MAX' wrap around zero.

  6983. 

  6984. 'Host_Wide_Int'

  6985.      The option's argument is a non-negative integer consisting of

  6986.      either decimal or hexadecimal digits interpreted as the widest

  6987.      integer type on the host.  As with an 'UInteger' argument,

  6988.      hexadecimal integers may optionally start with the '0x' or '0X'

  6989.      prefix.  The option parser validates and converts the argument

  6990.      before passing it to the relevant option handler.  'Host_Wide_Int'

  6991.      should be used with options that need to accept very large values.

  6992.      Positive values of the argument in excess of 'HOST_WIDE_INT_M1U'

  6993.      are assigned 'HOST_WIDE_INT_M1U'.

  6994. 

  6995. 'IntegerRange(N, M)'

  6996.      The options's arguments are integers of type 'int'.  The option's

  6997.      parser validates that the value of an option integer argument is

  6998.      within the closed range [N, M].

  6999. 

  7000. 'ByteSize'

  7001.      A property applicable only to 'UInteger' or 'Host_Wide_Int'

  7002.      arguments.  The option's integer argument is interpreted as if in

  7003.      infinite precision using saturation arithmetic in the corresponding

  7004.      type.  The argument may be followed by a 'byte-size' suffix

  7005.      designating a multiple of bytes such as 'kB' and 'KiB' for kilobyte

  7006.      and kibibyte, respectively, 'MB' and 'MiB' for megabyte and

  7007.      mebibyte, 'GB' and 'GiB' for gigabyte and gigibyte, and so on.

  7008.      'ByteSize' should be used for with options that take a very large

  7009.      argument representing a size in bytes, such as '-Wlarger-than='.

  7010. 

  7011. 'ToLower'

  7012.      The option's argument should be converted to lowercase as part of

  7013.      putting it in canonical form, and before comparing with the strings

  7014.      indicated by any 'Enum' property.

  7015. 

  7016. 'NoDriverArg'

  7017.      For an option marked 'Separate', the option only takes an argument

  7018.      in the compiler proper, not in the driver.  This is for

  7019.      compatibility with existing options that are used both directly and

  7020.      via '-Wp,'; new options should not have this property.

  7021. 

  7022. 'Var(VAR)'

  7023.      The state of this option should be stored in variable VAR (actually

  7024.      a macro for 'global_options.x_VAR').  The way that the state is

  7025.      stored depends on the type of option:

  7026. 

  7027. 'WarnRemoved'

  7028.      The option is removed and every usage of such option will result in

  7029.      a warning.  We use it option backward compatibility.

  7030. 

  7031. 'Var(VAR, SET)'

  7032.      The option controls an integer variable VAR and is active when VAR

  7033.      equals SET.  The option parser will set VAR to SET when the

  7034.      positive form of the option is used and '!SET' when the "no-" form

  7035.      is used.

  7036. 

  7037.      VAR is declared in the same way as for the single-argument form

  7038.      described above.

  7039. 

  7040.         * If the option uses the 'Mask' or 'InverseMask' properties, VAR

  7041.           is the integer variable that contains the mask.

  7042. 

  7043.         * If the option is a normal on/off switch, VAR is an integer

  7044.           variable that is nonzero when the option is enabled.  The

  7045.           options parser will set the variable to 1 when the positive

  7046.           form of the option is used and 0 when the "no-" form is used.

  7047. 

  7048.         * If the option takes an argument and has the 'UInteger'

  7049.           property, VAR is an integer variable that stores the value of

  7050.           the argument.

  7051. 

  7052.         * If the option takes an argument and has the 'Enum' property,

  7053.           VAR is a variable (type given in the 'Type' property of the

  7054.           'Enum' record whose 'Name' property has the same argument as

  7055.           the 'Enum' property of this option) that stores the value of

  7056.           the argument.

  7057. 

  7058.         * If the option has the 'Defer' property, VAR is a pointer to a

  7059.           'VEC(cl_deferred_option,heap)' that stores the option for

  7060.           later processing.  (VAR is declared with type 'void *' and

  7061.           needs to be cast to 'VEC(cl_deferred_option,heap)' before

  7062.           use.)

  7063. 

  7064.         * Otherwise, if the option takes an argument, VAR is a pointer

  7065.           to the argument string.  The pointer will be null if the

  7066.           argument is optional and wasn't given.

  7067. 

  7068.      The option-processing script will usually zero-initialize VAR.  You

  7069.      can modify this behavior using 'Init'.

  7070. 

  7071. 'Init(VALUE)'

  7072.      The variable specified by the 'Var' property should be statically

  7073.      initialized to VALUE.  If more than one option using the same

  7074.      variable specifies 'Init', all must specify the same initializer.

  7075. 

  7076. 'Mask(NAME)'

  7077.      The option is associated with a bit in the 'target_flags' variable

  7078.      (*note Run-time Target::) and is active when that bit is set.  You

  7079.      may also specify 'Var' to select a variable other than

  7080.      'target_flags'.

  7081. 

  7082.      The options-processing script will automatically allocate a unique

  7083.      bit for the option.  If the option is attached to 'target_flags',

  7084.      the script will set the macro 'MASK_NAME' to the appropriate

  7085.      bitmask.  It will also declare a 'TARGET_NAME' macro that has the

  7086.      value 1 when the option is active and 0 otherwise.  If you use

  7087.      'Var' to attach the option to a different variable, the bitmask

  7088.      macro with be called 'OPTION_MASK_NAME'.

  7089. 

  7090. 'InverseMask(OTHERNAME)'

  7091. 'InverseMask(OTHERNAME, THISNAME)'

  7092.      The option is the inverse of another option that has the

  7093.      'Mask(OTHERNAME)' property.  If THISNAME is given, the

  7094.      options-processing script will declare a 'TARGET_THISNAME' macro

  7095.      that is 1 when the option is active and 0 otherwise.

  7096. 

  7097. 'Enum(NAME)'

  7098.      The option's argument is a string from the set of strings

  7099.      associated with the corresponding 'Enum' record.  The string is

  7100.      checked and converted to the integer specified in the corresponding

  7101.      'EnumValue' record before being passed to option handlers.

  7102. 

  7103. 'Defer'

  7104.      The option should be stored in a vector, specified with 'Var', for

  7105.      later processing.

  7106. 

  7107. 'Alias(OPT)'

  7108. 'Alias(OPT, ARG)'

  7109. 'Alias(OPT, POSARG, NEGARG)'

  7110.      The option is an alias for '-OPT' (or the negative form of that

  7111.      option, depending on 'NegativeAlias').  In the first form, any

  7112.      argument passed to the alias is considered to be passed to '-OPT',

  7113.      and '-OPT' is considered to be negated if the alias is used in

  7114.      negated form.  In the second form, the alias may not be negated or

  7115.      have an argument, and POSARG is considered to be passed as an

  7116.      argument to '-OPT'.  In the third form, the alias may not have an

  7117.      argument, if the alias is used in the positive form then POSARG is

  7118.      considered to be passed to '-OPT', and if the alias is used in the

  7119.      negative form then NEGARG is considered to be passed to '-OPT'.

  7120. 

  7121.      Aliases should not specify 'Var' or 'Mask' or 'UInteger'.  Aliases

  7122.      should normally specify the same languages as the target of the

  7123.      alias; the flags on the target will be used to determine any

  7124.      diagnostic for use of an option for the wrong language, while those

  7125.      on the alias will be used to identify what command-line text is the

  7126.      option and what text is any argument to that option.

  7127. 

  7128.      When an 'Alias' definition is used for an option, driver specs do

  7129.      not need to handle it and no 'OPT_' enumeration value is defined

  7130.      for it; only the canonical form of the option will be seen in those

  7131.      places.

  7132. 

  7133. 'NegativeAlias'

  7134.      For an option marked with 'Alias(OPT)', the option is considered to

  7135.      be an alias for the positive form of '-OPT' if negated and for the

  7136.      negative form of '-OPT' if not negated.  'NegativeAlias' may not be

  7137.      used with the forms of 'Alias' taking more than one argument.

  7138. 

  7139. 'Ignore'

  7140.      This option is ignored apart from printing any warning specified

  7141.      using 'Warn'.  The option will not be seen by specs and no 'OPT_'

  7142.      enumeration value is defined for it.

  7143. 

  7144. 'SeparateAlias'

  7145.      For an option marked with 'Joined', 'Separate' and 'Alias', the

  7146.      option only acts as an alias when passed a separate argument; with

  7147.      a joined argument it acts as a normal option, with an 'OPT_'

  7148.      enumeration value.  This is for compatibility with the Java '-d'

  7149.      option and should not be used for new options.

  7150. 

  7151. 'Warn(MESSAGE)'

  7152.      If this option is used, output the warning MESSAGE.  MESSAGE is a

  7153.      format string, either taking a single operand with a '%qs' format

  7154.      which is the option name, or not taking any operands, which is

  7155.      passed to the 'warning' function.  If an alias is marked 'Warn',

  7156.      the target of the alias must not also be marked 'Warn'.

  7157. 

  7158. 'Report'

  7159.      The state of the option should be printed by '-fverbose-asm'.

  7160. 

  7161. 'Warning'

  7162.      This is a warning option and should be shown as such in '--help'

  7163.      output.  This flag does not currently affect anything other than

  7164.      '--help'.

  7165. 

  7166. 'Optimization'

  7167.      This is an optimization option.  It should be shown as such in

  7168.      '--help' output, and any associated variable named using 'Var'

  7169.      should be saved and restored when the optimization level is changed

  7170.      with 'optimize' attributes.

  7171. 

  7172. 'PerFunction'

  7173.      This is an option that can be overridden on a per-function basis.

  7174.      'Optimization' implies 'PerFunction', but options that do not

  7175.      affect executable code generation may use this flag instead, so

  7176.      that the option is not taken into account in ways that might affect

  7177.      executable code generation.

  7178. 

  7179. 'Param'

  7180.      This is an option that is a parameter.

  7181. 

  7182. 'Undocumented'

  7183.      The option is deliberately missing documentation and should not be

  7184.      included in the '--help' output.

  7185. 

  7186. 'Condition(COND)'

  7187.      The option should only be accepted if preprocessor condition COND

  7188.      is true.  Note that any C declarations associated with the option

  7189.      will be present even if COND is false; COND simply controls whether

  7190.      the option is accepted and whether it is printed in the '--help'

  7191.      output.

  7192. 

  7193. 'Save'

  7194.      Build the 'cl_target_option' structure to hold a copy of the

  7195.      option, add the functions 'cl_target_option_save' and

  7196.      'cl_target_option_restore' to save and restore the options.

  7197. 

  7198. 'SetByCombined'

  7199.      The option may also be set by a combined option such as

  7200.      '-ffast-math'.  This causes the 'gcc_options' struct to have a

  7201.      field 'frontend_set_NAME', where 'NAME' is the name of the field

  7202.      holding the value of this option (without the leading 'x_').  This

  7203.      gives the front end a way to indicate that the value has been set

  7204.      explicitly and should not be changed by the combined option.  For

  7205.      example, some front ends use this to prevent '-ffast-math' and

  7206.      '-fno-fast-math' from changing the value of '-fmath-errno' for

  7207.      languages that do not use 'errno'.

  7208. 

  7209. 'EnabledBy(OPT)'

  7210. 'EnabledBy(OPT || OPT2)'

  7211. 'EnabledBy(OPT && OPT2)'

  7212.      If not explicitly set, the option is set to the value of '-OPT';

  7213.      multiple options can be given, separated by '||'.  The third form

  7214.      using '&&' specifies that the option is only set if both OPT and

  7215.      OPT2 are set.  The options OPT and OPT2 must have the 'Common'

  7216.      property; otherwise, use 'LangEnabledBy'.

  7217. 

  7218. 'LangEnabledBy(LANGUAGE, OPT)'

  7219. 'LangEnabledBy(LANGUAGE, OPT, POSARG, NEGARG)'

  7220.      When compiling for the given language, the option is set to the

  7221.      value of '-OPT', if not explicitly set.  OPT can be also a list of

  7222.      '||' separated options.  In the second form, if OPT is used in the

  7223.      positive form then POSARG is considered to be passed to the option,

  7224.      and if OPT is used in the negative form then NEGARG is considered

  7225.      to be passed to the option.  It is possible to specify several

  7226.      different languages.  Each LANGUAGE must have been declared by an

  7227.      earlier 'Language' record.  *Note Option file format::.

  7228. 

  7229. 'NoDWARFRecord'

  7230.      The option is omitted from the producer string written by

  7231.      '-grecord-gcc-switches'.

  7232. 

  7233. 'PchIgnore'

  7234.      Even if this is a target option, this option will not be recorded /

  7235.      compared to determine if a precompiled header file matches.

  7236. 

  7237. 'CPP(VAR)'

  7238.      The state of this option should be kept in sync with the

  7239.      preprocessor option VAR.  If this property is set, then properties

  7240.      'Var' and 'Init' must be set as well.

  7241. 

  7242. 'CppReason(CPP_W_ENUM)'

  7243.      This warning option corresponds to 'cpplib.h' warning reason code

  7244.      CPP_W_ENUM.  This should only be used for warning options of the

  7245.      C-family front-ends.

  7246. 

  7247. 

  7248. File: gccint.info,  Node: Passes,  Next: poly_int,  Prev: Options,  Up: Top

  7249. 

  7250. 9 Passes and Files of the Compiler

  7251. **********************************

  7252. 

  7253. This chapter is dedicated to giving an overview of the optimization and

  7254. code generation passes of the compiler.  In the process, it describes

  7255. some of the language front end interface, though this description is no

  7256. where near complete.

  7257. 

  7258. * Menu:

  7259. 

  7260. * Parsing pass::         The language front end turns text into bits.

  7261. * Gimplification pass::  The bits are turned into something we can optimize.

  7262. * Pass manager::         Sequencing the optimization passes.

  7263. * IPA passes::           Inter-procedural optimizations.

  7264. * Tree SSA passes::      Optimizations on a high-level representation.

  7265. * RTL passes::           Optimizations on a low-level representation.

  7266. * Optimization info::    Dumping optimization information from passes.

  7267. 

  7268. 

  7269. File: gccint.info,  Node: Parsing pass,  Next: Gimplification pass,  Up: Passes

  7270. 

  7271. 9.1 Parsing pass

  7272. ================

  7273. 

  7274. The language front end is invoked only once, via

  7275. 'lang_hooks.parse_file', to parse the entire input.  The language front

  7276. end may use any intermediate language representation deemed appropriate.

  7277. The C front end uses GENERIC trees (*note GENERIC::), plus a double

  7278. handful of language specific tree codes defined in 'c-common.def'.  The

  7279. Fortran front end uses a completely different private representation.

  7280. 

  7281.  At some point the front end must translate the representation used in

  7282. the front end to a representation understood by the language-independent

  7283. portions of the compiler.  Current practice takes one of two forms.  The

  7284. C front end manually invokes the gimplifier (*note GIMPLE::) on each

  7285. function, and uses the gimplifier callbacks to convert the

  7286. language-specific tree nodes directly to GIMPLE before passing the

  7287. function off to be compiled.  The Fortran front end converts from a

  7288. private representation to GENERIC, which is later lowered to GIMPLE when

  7289. the function is compiled.  Which route to choose probably depends on how

  7290. well GENERIC (plus extensions) can be made to match up with the source

  7291. language and necessary parsing data structures.

  7292. 

  7293.  BUG: Gimplification must occur before nested function lowering, and

  7294. nested function lowering must be done by the front end before passing

  7295. the data off to cgraph.

  7296. 

  7297.  TODO: Cgraph should control nested function lowering.  It would only be

  7298. invoked when it is certain that the outer-most function is used.

  7299. 

  7300.  TODO: Cgraph needs a gimplify_function callback.  It should be invoked

  7301. when (1) it is certain that the function is used, (2) warning flags

  7302. specified by the user require some amount of compilation in order to

  7303. honor, (3) the language indicates that semantic analysis is not complete

  7304. until gimplification occurs.  Hum... this sounds overly complicated.

  7305. Perhaps we should just have the front end gimplify always; in most cases

  7306. it's only one function call.

  7307. 

  7308.  The front end needs to pass all function definitions and top level

  7309. declarations off to the middle-end so that they can be compiled and

  7310. emitted to the object file.  For a simple procedural language, it is

  7311. usually most convenient to do this as each top level declaration or

  7312. definition is seen.  There is also a distinction to be made between

  7313. generating functional code and generating complete debug information.

  7314. The only thing that is absolutely required for functional code is that

  7315. function and data _definitions_ be passed to the middle-end.  For

  7316. complete debug information, function, data and type declarations should

  7317. all be passed as well.

  7318. 

  7319.  In any case, the front end needs each complete top-level function or

  7320. data declaration, and each data definition should be passed to

  7321. 'rest_of_decl_compilation'.  Each complete type definition should be

  7322. passed to 'rest_of_type_compilation'.  Each function definition should

  7323. be passed to 'cgraph_finalize_function'.

  7324. 

  7325.  TODO: I know rest_of_compilation currently has all sorts of RTL

  7326. generation semantics.  I plan to move all code generation bits (both

  7327. Tree and RTL) to compile_function.  Should we hide cgraph from the front

  7328. ends and move back to rest_of_compilation as the official interface?

  7329. Possibly we should rename all three interfaces such that the names match

  7330. in some meaningful way and that is more descriptive than "rest_of".

  7331. 

  7332.  The middle-end will, at its option, emit the function and data

  7333. definitions immediately or queue them for later processing.

  7334. 

  7335. 

  7336. File: gccint.info,  Node: Gimplification pass,  Next: Pass manager,  Prev: Parsing pass,  Up: Passes

  7337. 

  7338. 9.2 Gimplification pass

  7339. =======================

  7340. 

  7341. "Gimplification" is a whimsical term for the process of converting the

  7342. intermediate representation of a function into the GIMPLE language

  7343. (*note GIMPLE::).  The term stuck, and so words like "gimplification",

  7344. "gimplify", "gimplifier" and the like are sprinkled throughout this

  7345. section of code.

  7346. 

  7347.  While a front end may certainly choose to generate GIMPLE directly if

  7348. it chooses, this can be a moderately complex process unless the

  7349. intermediate language used by the front end is already fairly simple.

  7350. Usually it is easier to generate GENERIC trees plus extensions and let

  7351. the language-independent gimplifier do most of the work.

  7352. 

  7353.  The main entry point to this pass is 'gimplify_function_tree' located

  7354. in 'gimplify.c'.  From here we process the entire function gimplifying

  7355. each statement in turn.  The main workhorse for this pass is

  7356. 'gimplify_expr'.  Approximately everything passes through here at least

  7357. once, and it is from here that we invoke the 'lang_hooks.gimplify_expr'

  7358. callback.

  7359. 

  7360.  The callback should examine the expression in question and return

  7361. 'GS_UNHANDLED' if the expression is not a language specific construct

  7362. that requires attention.  Otherwise it should alter the expression in

  7363. some way to such that forward progress is made toward producing valid

  7364. GIMPLE.  If the callback is certain that the transformation is complete

  7365. and the expression is valid GIMPLE, it should return 'GS_ALL_DONE'.

  7366. Otherwise it should return 'GS_OK', which will cause the expression to

  7367. be processed again.  If the callback encounters an error during the

  7368. transformation (because the front end is relying on the gimplification

  7369. process to finish semantic checks), it should return 'GS_ERROR'.

  7370. 

  7371. 

  7372. File: gccint.info,  Node: Pass manager,  Next: IPA passes,  Prev: Gimplification pass,  Up: Passes

  7373. 

  7374. 9.3 Pass manager

  7375. ================

  7376. 

  7377. The pass manager is located in 'passes.c', 'tree-optimize.c' and

  7378. 'tree-pass.h'.  It processes passes as described in 'passes.def'.  Its

  7379. job is to run all of the individual passes in the correct order, and

  7380. take care of standard bookkeeping that applies to every pass.

  7381. 

  7382.  The theory of operation is that each pass defines a structure that

  7383. represents everything we need to know about that pass--when it should be

  7384. run, how it should be run, what intermediate language form or

  7385. on-the-side data structures it needs.  We register the pass to be run in

  7386. some particular order, and the pass manager arranges for everything to

  7387. happen in the correct order.

  7388. 

  7389.  The actuality doesn't completely live up to the theory at present.

  7390. Command-line switches and 'timevar_id_t' enumerations must still be

  7391. defined elsewhere.  The pass manager validates constraints but does not

  7392. attempt to (re-)generate data structures or lower intermediate language

  7393. form based on the requirements of the next pass.  Nevertheless, what is

  7394. present is useful, and a far sight better than nothing at all.

  7395. 

  7396.  Each pass should have a unique name.  Each pass may have its own dump

  7397. file (for GCC debugging purposes).  Passes with a name starting with a

  7398. star do not dump anything.  Sometimes passes are supposed to share a

  7399. dump file / option name.  To still give these unique names, you can use

  7400. a prefix that is delimited by a space from the part that is used for the

  7401. dump file / option name.  E.g.  When the pass name is "ud dce", the name

  7402. used for dump file/options is "dce".

  7403. 

  7404.  TODO: describe the global variables set up by the pass manager, and a

  7405. brief description of how a new pass should use it.  I need to look at

  7406. what info RTL passes use first...

  7407. 

  7408. 

  7409. File: gccint.info,  Node: IPA passes,  Next: Tree SSA passes,  Prev: Pass manager,  Up: Passes

  7410. 

  7411. 9.4 Inter-procedural optimization passes

  7412. ========================================

  7413. 

  7414. The inter-procedural optimization (IPA) passes use call graph

  7415. information to perform transformations across function boundaries.  IPA

  7416. is a critical part of link-time optimization (LTO) and whole-program

  7417. (WHOPR) optimization, and these passes are structured with the needs of

  7418. LTO and WHOPR in mind by dividing their operations into stages.  For

  7419. detailed discussion of the LTO/WHOPR IPA pass stages and interfaces, see

  7420. *note IPA::.

  7421. 

  7422.  The following briefly describes the inter-procedural optimization (IPA)

  7423. passes, which are split into small IPA passes, regular IPA passes, and

  7424. late IPA passes, according to the LTO/WHOPR processing model.

  7425. 

  7426. * Menu:

  7427. 

  7428. * Small IPA passes::

  7429. * Regular IPA passes::

  7430. * Late IPA passes::

  7431. 

  7432. 

  7433. File: gccint.info,  Node: Small IPA passes,  Next: Regular IPA passes,  Up: IPA passes

  7434. 

  7435. 9.4.1 Small IPA passes

  7436. ----------------------

  7437. 

  7438. A small IPA pass is a pass derived from 'simple_ipa_opt_pass'.  As

  7439. described in *note IPA::, it does everything at once and defines only

  7440. the _Execute_ stage.  During this stage it accesses and modifies the

  7441. function bodies.  No 'generate_summary', 'read_summary', or

  7442. 'write_summary' hooks are defined.

  7443. 

  7444.    * IPA free lang data

  7445. 

  7446.      This pass frees resources that are used by the front end but are

  7447.      not needed once it is done.  It is located in 'tree.c' and is

  7448.      described by 'pass_ipa_free_lang_data'.

  7449. 

  7450.    * IPA function and variable visibility

  7451. 

  7452.      This is a local function pass handling visibilities of all symbols.

  7453.      This happens before LTO streaming, so '-fwhole-program' should be

  7454.      ignored at this level.  It is located in 'ipa-visibility.c' and is

  7455.      described by 'pass_ipa_function_and_variable_visibility'.

  7456. 

  7457.    * IPA remove symbols

  7458. 

  7459.      This pass performs reachability analysis and reclaims all

  7460.      unreachable nodes.  It is located in 'passes.c' and is described by

  7461.      'pass_ipa_remove_symbols'.

  7462. 

  7463.    * IPA OpenACC

  7464. 

  7465.      This is a pass group for OpenACC processing.  It is located in

  7466.      'tree-ssa-loop.c' and is described by 'pass_ipa_oacc'.

  7467. 

  7468.    * IPA points-to analysis

  7469. 

  7470.      This is a tree-based points-to analysis pass.  The idea behind this

  7471.      analyzer is to generate set constraints from the program, then

  7472.      solve the resulting constraints in order to generate the points-to

  7473.      sets.  It is located in 'tree-ssa-structalias.c' and is described

  7474.      by 'pass_ipa_pta'.

  7475. 

  7476.    * IPA OpenACC kernels

  7477. 

  7478.      This is a pass group for processing OpenACC kernels regions.  It is

  7479.      a subpass of the IPA OpenACC pass group that runs on offloaded

  7480.      functions containing OpenACC kernels loops.  It is located in

  7481.      'tree-ssa-loop.c' and is described by 'pass_ipa_oacc_kernels'.

  7482. 

  7483.    * Target clone

  7484. 

  7485.      This is a pass for parsing functions with multiple target

  7486.      attributes.  It is located in 'multiple_target.c' and is described

  7487.      by 'pass_target_clone'.

  7488. 

  7489.    * IPA auto profile

  7490. 

  7491.      This pass uses AutoFDO profiling data to annotate the control flow

  7492.      graph.  It is located in 'auto-profile.c' and is described by

  7493.      'pass_ipa_auto_profile'.

  7494. 

  7495.    * IPA tree profile

  7496. 

  7497.      This pass does profiling for all functions in the call graph.  It

  7498.      calculates branch probabilities and basic block execution counts.

  7499.      It is located in 'tree-profile.c' and is described by

  7500.      'pass_ipa_tree_profile'.

  7501. 

  7502.    * IPA free function summary

  7503. 

  7504.      This pass is a small IPA pass when argument 'small_p' is true.  It

  7505.      releases inline function summaries and call summaries.  It is

  7506.      located in 'ipa-fnsummary.c' and is described by

  7507.      'pass_ipa_free_free_fn_summary'.

  7508. 

  7509.    * IPA increase alignment

  7510. 

  7511.      This pass increases the alignment of global arrays to improve

  7512.      vectorization.  It is located in 'tree-vectorizer.c' and is

  7513.      described by 'pass_ipa_increase_alignment'.

  7514. 

  7515.    * IPA transactional memory

  7516. 

  7517.      This pass is for transactional memory support.  It is located in

  7518.      'trans-mem.c' and is described by 'pass_ipa_tm'.

  7519. 

  7520.    * IPA lower emulated TLS

  7521. 

  7522.      This pass lowers thread-local storage (TLS) operations to emulation

  7523.      functions provided by libgcc.  It is located in 'tree-emutls.c' and

  7524.      is described by 'pass_ipa_lower_emutls'.

  7525. 

  7526. 

  7527. File: gccint.info,  Node: Regular IPA passes,  Next: Late IPA passes,  Prev: Small IPA passes,  Up: IPA passes

  7528. 

  7529. 9.4.2 Regular IPA passes

  7530. ------------------------

  7531. 

  7532. A regular IPA pass is a pass derived from 'ipa_opt_pass_d' that is

  7533. executed in WHOPR compilation.  Regular IPA passes may have summary

  7534. hooks implemented in any of the LGEN, WPA or LTRANS stages (*note

  7535. IPA::).

  7536. 

  7537.    * IPA whole program visibility

  7538. 

  7539.      This pass performs various optimizations involving symbol

  7540.      visibility with '-fwhole-program', including symbol privatization,

  7541.      discovering local functions, and dismantling comdat groups.  It is

  7542.      located in 'ipa-visibility.c' and is described by

  7543.      'pass_ipa_whole_program_visibility'.

  7544. 

  7545.    * IPA profile

  7546. 

  7547.      The IPA profile pass propagates profiling frequencies across the

  7548.      call graph.  It is located in 'ipa-profile.c' and is described by

  7549.      'pass_ipa_profile'.

  7550. 

  7551.    * IPA identical code folding

  7552. 

  7553.      This is the inter-procedural identical code folding pass.  The goal

  7554.      of this transformation is to discover functions and read-only

  7555.      variables that have exactly the same semantics.  It is located in

  7556.      'ipa-icf.c' and is described by 'pass_ipa_icf'.

  7557. 

  7558.    * IPA devirtualization

  7559. 

  7560.      This pass performs speculative devirtualization based on the type

  7561.      inheritance graph.  When a polymorphic call has only one likely

  7562.      target in the unit, it is turned into a speculative call.  It is

  7563.      located in 'ipa-devirt.c' and is described by 'pass_ipa_devirt'.

  7564. 

  7565.    * IPA constant propagation

  7566. 

  7567.      The goal of this pass is to discover functions that are always

  7568.      invoked with some arguments with the same known constant values and

  7569.      to modify the functions accordingly.  It can also do partial

  7570.      specialization and type-based devirtualization.  It is located in

  7571.      'ipa-cp.c' and is described by 'pass_ipa_cp'.

  7572. 

  7573.    * IPA scalar replacement of aggregates

  7574. 

  7575.      This pass can replace an aggregate parameter with a set of other

  7576.      parameters representing part of the original, turning those passed

  7577.      by reference into new ones which pass the value directly.  It also

  7578.      removes unused function return values and unused function

  7579.      parameters.  This pass is located in 'ipa-sra.c' and is described

  7580.      by 'pass_ipa_sra'.

  7581. 

  7582.    * IPA constructor/destructor merge

  7583. 

  7584.      This pass merges multiple constructors and destructors for static

  7585.      objects into single functions.  It's only run at LTO time unless

  7586.      the target doesn't support constructors and destructors natively.

  7587.      The pass is located in 'ipa.c' and is described by

  7588.      'pass_ipa_cdtor_merge'.

  7589. 

  7590.    * IPA HSA

  7591. 

  7592.      This pass is part of the GCC support for HSA (Heterogeneous System

  7593.      Architecture) accelerators.  It is responsible for creation of HSA

  7594.      clones and emitting HSAIL instructions for them.  It is located in

  7595.      'ipa-hsa.c' and is described by 'pass_ipa_hsa'.

  7596. 

  7597.    * IPA function summary

  7598. 

  7599.      This pass provides function analysis for inter-procedural passes.

  7600.      It collects estimates of function body size, execution time, and

  7601.      frame size for each function.  It also estimates information about

  7602.      function calls: call statement size, time and how often the

  7603.      parameters change for each call.  It is located in

  7604.      'ipa-fnsummary.c' and is described by 'pass_ipa_fn_summary'.

  7605. 

  7606.    * IPA inline

  7607. 

  7608.      The IPA inline pass handles function inlining with whole-program

  7609.      knowledge.  Small functions that are candidates for inlining are

  7610.      ordered in increasing badness, bounded by unit growth parameters.

  7611.      Unreachable functions are removed from the call graph.  Functions

  7612.      called once and not exported from the unit are inlined.  This pass

  7613.      is located in 'ipa-inline.c' and is described by 'pass_ipa_inline'.

  7614. 

  7615.    * IPA pure/const analysis

  7616. 

  7617.      This pass marks functions as being either const ('TREE_READONLY')

  7618.      or pure ('DECL_PURE_P').  The per-function information is produced

  7619.      by 'pure_const_generate_summary', then the global information is

  7620.      computed by performing a transitive closure over the call graph.

  7621.      It is located in 'ipa-pure-const.c' and is described by

  7622.      'pass_ipa_pure_const'.

  7623. 

  7624.    * IPA free function summary

  7625. 

  7626.      This pass is a regular IPA pass when argument 'small_p' is false.

  7627.      It releases inline function summaries and call summaries.  It is

  7628.      located in 'ipa-fnsummary.c' and is described by

  7629.      'pass_ipa_free_fn_summary'.

  7630. 

  7631.    * IPA reference

  7632. 

  7633.      This pass gathers information about how variables whose scope is

  7634.      confined to the compilation unit are used.  It is located in

  7635.      'ipa-reference.c' and is described by 'pass_ipa_reference'.

  7636. 

  7637.    * IPA single use

  7638. 

  7639.      This pass checks whether variables are used by a single function.

  7640.      It is located in 'ipa.c' and is described by 'pass_ipa_single_use'.

  7641. 

  7642.    * IPA comdats

  7643. 

  7644.      This pass looks for static symbols that are used exclusively within

  7645.      one comdat group, and moves them into that comdat group.  It is

  7646.      located in 'ipa-comdats.c' and is described by 'pass_ipa_comdats'.

  7647. 

  7648. 

  7649. File: gccint.info,  Node: Late IPA passes,  Prev: Regular IPA passes,  Up: IPA passes

  7650. 

  7651. 9.4.3 Late IPA passes

  7652. ---------------------

  7653. 

  7654. Late IPA passes are simple IPA passes executed after the regular passes.

  7655. In WHOPR mode the passes are executed after partitioning and thus see

  7656. just parts of the compiled unit.

  7657. 

  7658.    * Materialize all clones

  7659. 

  7660.      Once all functions from compilation unit are in memory, produce all

  7661.      clones and update all calls.  It is located in 'ipa.c' and is

  7662.      described by 'pass_materialize_all_clones'.

  7663. 

  7664.    * IPA points-to analysis

  7665. 

  7666.      Points-to analysis; this is the same as the points-to-analysis pass

  7667.      run with the small IPA passes (*note Small IPA passes::).

  7668. 

  7669.    * OpenMP simd clone

  7670. 

  7671.      This is the OpenMP constructs' SIMD clone pass.  It creates the

  7672.      appropriate SIMD clones for functions tagged as elemental SIMD

  7673.      functions.  It is located in 'omp-simd-clone.c' and is described by

  7674.      'pass_omp_simd_clone'.

  7675. 

  7676. 

  7677. File: gccint.info,  Node: Tree SSA passes,  Next: RTL passes,  Prev: IPA passes,  Up: Passes

  7678. 

  7679. 9.5 Tree SSA passes

  7680. ===================

  7681. 

  7682. The following briefly describes the Tree optimization passes that are

  7683. run after gimplification and what source files they are located in.

  7684. 

  7685.    * Remove useless statements

  7686. 

  7687.      This pass is an extremely simple sweep across the gimple code in

  7688.      which we identify obviously dead code and remove it.  Here we do

  7689.      things like simplify 'if' statements with constant conditions,

  7690.      remove exception handling constructs surrounding code that

  7691.      obviously cannot throw, remove lexical bindings that contain no

  7692.      variables, and other assorted simplistic cleanups.  The idea is to

  7693.      get rid of the obvious stuff quickly rather than wait until later

  7694.      when it's more work to get rid of it.  This pass is located in

  7695.      'tree-cfg.c' and described by 'pass_remove_useless_stmts'.

  7696. 

  7697.    * OpenMP lowering

  7698. 

  7699.      If OpenMP generation ('-fopenmp') is enabled, this pass lowers

  7700.      OpenMP constructs into GIMPLE.

  7701. 

  7702.      Lowering of OpenMP constructs involves creating replacement

  7703.      expressions for local variables that have been mapped using data

  7704.      sharing clauses, exposing the control flow of most synchronization

  7705.      directives and adding region markers to facilitate the creation of

  7706.      the control flow graph.  The pass is located in 'omp-low.c' and is

  7707.      described by 'pass_lower_omp'.

  7708. 

  7709.    * OpenMP expansion

  7710. 

  7711.      If OpenMP generation ('-fopenmp') is enabled, this pass expands

  7712.      parallel regions into their own functions to be invoked by the

  7713.      thread library.  The pass is located in 'omp-low.c' and is

  7714.      described by 'pass_expand_omp'.

  7715. 

  7716.    * Lower control flow

  7717. 

  7718.      This pass flattens 'if' statements ('COND_EXPR') and moves lexical

  7719.      bindings ('BIND_EXPR') out of line.  After this pass, all 'if'

  7720.      statements will have exactly two 'goto' statements in its 'then'

  7721.      and 'else' arms.  Lexical binding information for each statement

  7722.      will be found in 'TREE_BLOCK' rather than being inferred from its

  7723.      position under a 'BIND_EXPR'.  This pass is found in 'gimple-low.c'

  7724.      and is described by 'pass_lower_cf'.

  7725. 

  7726.    * Lower exception handling control flow

  7727. 

  7728.      This pass decomposes high-level exception handling constructs

  7729.      ('TRY_FINALLY_EXPR' and 'TRY_CATCH_EXPR') into a form that

  7730.      explicitly represents the control flow involved.  After this pass,

  7731.      'lookup_stmt_eh_region' will return a non-negative number for any

  7732.      statement that may have EH control flow semantics; examine

  7733.      'tree_can_throw_internal' or 'tree_can_throw_external' for exact

  7734.      semantics.  Exact control flow may be extracted from

  7735.      'foreach_reachable_handler'.  The EH region nesting tree is defined

  7736.      in 'except.h' and built in 'except.c'.  The lowering pass itself is

  7737.      in 'tree-eh.c' and is described by 'pass_lower_eh'.

  7738. 

  7739.    * Build the control flow graph

  7740. 

  7741.      This pass decomposes a function into basic blocks and creates all

  7742.      of the edges that connect them.  It is located in 'tree-cfg.c' and

  7743.      is described by 'pass_build_cfg'.

  7744. 

  7745.    * Find all referenced variables

  7746. 

  7747.      This pass walks the entire function and collects an array of all

  7748.      variables referenced in the function, 'referenced_vars'.  The index

  7749.      at which a variable is found in the array is used as a UID for the

  7750.      variable within this function.  This data is needed by the SSA

  7751.      rewriting routines.  The pass is located in 'tree-dfa.c' and is

  7752.      described by 'pass_referenced_vars'.

  7753. 

  7754.    * Enter static single assignment form

  7755. 

  7756.      This pass rewrites the function such that it is in SSA form.  After

  7757.      this pass, all 'is_gimple_reg' variables will be referenced by

  7758.      'SSA_NAME', and all occurrences of other variables will be

  7759.      annotated with 'VDEFS' and 'VUSES'; PHI nodes will have been

  7760.      inserted as necessary for each basic block.  This pass is located

  7761.      in 'tree-ssa.c' and is described by 'pass_build_ssa'.

  7762. 

  7763.    * Warn for uninitialized variables

  7764. 

  7765.      This pass scans the function for uses of 'SSA_NAME's that are fed

  7766.      by default definition.  For non-parameter variables, such uses are

  7767.      uninitialized.  The pass is run twice, before and after

  7768.      optimization (if turned on).  In the first pass we only warn for

  7769.      uses that are positively uninitialized; in the second pass we warn

  7770.      for uses that are possibly uninitialized.  The pass is located in

  7771.      'tree-ssa.c' and is defined by 'pass_early_warn_uninitialized' and

  7772.      'pass_late_warn_uninitialized'.

  7773. 

  7774.    * Dead code elimination

  7775. 

  7776.      This pass scans the function for statements without side effects

  7777.      whose result is unused.  It does not do memory life analysis, so

  7778.      any value that is stored in memory is considered used.  The pass is

  7779.      run multiple times throughout the optimization process.  It is

  7780.      located in 'tree-ssa-dce.c' and is described by 'pass_dce'.

  7781. 

  7782.    * Dominator optimizations

  7783. 

  7784.      This pass performs trivial dominator-based copy and constant

  7785.      propagation, expression simplification, and jump threading.  It is

  7786.      run multiple times throughout the optimization process.  It is

  7787.      located in 'tree-ssa-dom.c' and is described by 'pass_dominator'.

  7788. 

  7789.    * Forward propagation of single-use variables

  7790. 

  7791.      This pass attempts to remove redundant computation by substituting

  7792.      variables that are used once into the expression that uses them and

  7793.      seeing if the result can be simplified.  It is located in

  7794.      'tree-ssa-forwprop.c' and is described by 'pass_forwprop'.

  7795. 

  7796.    * Copy Renaming

  7797. 

  7798.      This pass attempts to change the name of compiler temporaries

  7799.      involved in copy operations such that SSA->normal can coalesce the

  7800.      copy away.  When compiler temporaries are copies of user variables,

  7801.      it also renames the compiler temporary to the user variable

  7802.      resulting in better use of user symbols.  It is located in

  7803.      'tree-ssa-copyrename.c' and is described by 'pass_copyrename'.

  7804. 

  7805.    * PHI node optimizations

  7806. 

  7807.      This pass recognizes forms of PHI inputs that can be represented as

  7808.      conditional expressions and rewrites them into straight line code.

  7809.      It is located in 'tree-ssa-phiopt.c' and is described by

  7810.      'pass_phiopt'.

  7811. 

  7812.    * May-alias optimization

  7813. 

  7814.      This pass performs a flow sensitive SSA-based points-to analysis.

  7815.      The resulting may-alias, must-alias, and escape analysis

  7816.      information is used to promote variables from in-memory addressable

  7817.      objects to non-aliased variables that can be renamed into SSA form.

  7818.      We also update the 'VDEF'/'VUSE' memory tags for non-renameable

  7819.      aggregates so that we get fewer false kills.  The pass is located

  7820.      in 'tree-ssa-alias.c' and is described by 'pass_may_alias'.

  7821. 

  7822.      Interprocedural points-to information is located in

  7823.      'tree-ssa-structalias.c' and described by 'pass_ipa_pta'.

  7824. 

  7825.    * Profiling

  7826. 

  7827.      This pass instruments the function in order to collect runtime

  7828.      block and value profiling data.  Such data may be fed back into the

  7829.      compiler on a subsequent run so as to allow optimization based on

  7830.      expected execution frequencies.  The pass is located in

  7831.      'tree-profile.c' and is described by 'pass_ipa_tree_profile'.

  7832. 

  7833.    * Static profile estimation

  7834. 

  7835.      This pass implements series of heuristics to guess propababilities

  7836.      of branches.  The resulting predictions are turned into edge

  7837.      profile by propagating branches across the control flow graphs.

  7838.      The pass is located in 'tree-profile.c' and is described by

  7839.      'pass_profile'.

  7840. 

  7841.    * Lower complex arithmetic

  7842. 

  7843.      This pass rewrites complex arithmetic operations into their

  7844.      component scalar arithmetic operations.  The pass is located in

  7845.      'tree-complex.c' and is described by 'pass_lower_complex'.

  7846. 

  7847.    * Scalar replacement of aggregates

  7848. 

  7849.      This pass rewrites suitable non-aliased local aggregate variables

  7850.      into a set of scalar variables.  The resulting scalar variables are

  7851.      rewritten into SSA form, which allows subsequent optimization

  7852.      passes to do a significantly better job with them.  The pass is

  7853.      located in 'tree-sra.c' and is described by 'pass_sra'.

  7854. 

  7855.    * Dead store elimination

  7856. 

  7857.      This pass eliminates stores to memory that are subsequently

  7858.      overwritten by another store, without any intervening loads.  The

  7859.      pass is located in 'tree-ssa-dse.c' and is described by 'pass_dse'.

  7860. 

  7861.    * Tail recursion elimination

  7862. 

  7863.      This pass transforms tail recursion into a loop.  It is located in

  7864.      'tree-tailcall.c' and is described by 'pass_tail_recursion'.

  7865. 

  7866.    * Forward store motion

  7867. 

  7868.      This pass sinks stores and assignments down the flowgraph closer to

  7869.      their use point.  The pass is located in 'tree-ssa-sink.c' and is

  7870.      described by 'pass_sink_code'.

  7871. 

  7872.    * Partial redundancy elimination

  7873. 

  7874.      This pass eliminates partially redundant computations, as well as

  7875.      performing load motion.  The pass is located in 'tree-ssa-pre.c'

  7876.      and is described by 'pass_pre'.

  7877. 

  7878.      Just before partial redundancy elimination, if

  7879.      '-funsafe-math-optimizations' is on, GCC tries to convert divisions

  7880.      to multiplications by the reciprocal.  The pass is located in

  7881.      'tree-ssa-math-opts.c' and is described by 'pass_cse_reciprocal'.

  7882. 

  7883.    * Full redundancy elimination

  7884. 

  7885.      This is a simpler form of PRE that only eliminates redundancies

  7886.      that occur on all paths.  It is located in 'tree-ssa-pre.c' and

  7887.      described by 'pass_fre'.

  7888. 

  7889.    * Loop optimization

  7890. 

  7891.      The main driver of the pass is placed in 'tree-ssa-loop.c' and

  7892.      described by 'pass_loop'.

  7893. 

  7894.      The optimizations performed by this pass are:

  7895. 

  7896.      Loop invariant motion.  This pass moves only invariants that would

  7897.      be hard to handle on RTL level (function calls, operations that

  7898.      expand to nontrivial sequences of insns).  With '-funswitch-loops'

  7899.      it also moves operands of conditions that are invariant out of the

  7900.      loop, so that we can use just trivial invariantness analysis in

  7901.      loop unswitching.  The pass also includes store motion.  The pass

  7902.      is implemented in 'tree-ssa-loop-im.c'.

  7903. 

  7904.      Canonical induction variable creation.  This pass creates a simple

  7905.      counter for number of iterations of the loop and replaces the exit

  7906.      condition of the loop using it, in case when a complicated analysis

  7907.      is necessary to determine the number of iterations.  Later

  7908.      optimizations then may determine the number easily.  The pass is

  7909.      implemented in 'tree-ssa-loop-ivcanon.c'.

  7910. 

  7911.      Induction variable optimizations.  This pass performs standard

  7912.      induction variable optimizations, including strength reduction,

  7913.      induction variable merging and induction variable elimination.  The

  7914.      pass is implemented in 'tree-ssa-loop-ivopts.c'.

  7915. 

  7916.      Loop unswitching.  This pass moves the conditional jumps that are

  7917.      invariant out of the loops.  To achieve this, a duplicate of the

  7918.      loop is created for each possible outcome of conditional jump(s).

  7919.      The pass is implemented in 'tree-ssa-loop-unswitch.c'.

  7920. 

  7921.      Loop splitting.  If a loop contains a conditional statement that is

  7922.      always true for one part of the iteration space and false for the

  7923.      other this pass splits the loop into two, one dealing with one side

  7924.      the other only with the other, thereby removing one inner-loop

  7925.      conditional.  The pass is implemented in 'tree-ssa-loop-split.c'.

  7926. 

  7927.      The optimizations also use various utility functions contained in

  7928.      'tree-ssa-loop-manip.c', 'cfgloop.c', 'cfgloopanal.c' and

  7929.      'cfgloopmanip.c'.

  7930. 

  7931.      Vectorization.  This pass transforms loops to operate on vector

  7932.      types instead of scalar types.  Data parallelism across loop

  7933.      iterations is exploited to group data elements from consecutive

  7934.      iterations into a vector and operate on them in parallel.

  7935.      Depending on available target support the loop is conceptually

  7936.      unrolled by a factor 'VF' (vectorization factor), which is the

  7937.      number of elements operated upon in parallel in each iteration, and

  7938.      the 'VF' copies of each scalar operation are fused to form a vector

  7939.      operation.  Additional loop transformations such as peeling and

  7940.      versioning may take place to align the number of iterations, and to

  7941.      align the memory accesses in the loop.  The pass is implemented in

  7942.      'tree-vectorizer.c' (the main driver), 'tree-vect-loop.c' and

  7943.      'tree-vect-loop-manip.c' (loop specific parts and general loop

  7944.      utilities), 'tree-vect-slp' (loop-aware SLP functionality),

  7945.      'tree-vect-stmts.c' and 'tree-vect-data-refs.c'.  Analysis of data

  7946.      references is in 'tree-data-ref.c'.

  7947. 

  7948.      SLP Vectorization.  This pass performs vectorization of

  7949.      straight-line code.  The pass is implemented in 'tree-vectorizer.c'

  7950.      (the main driver), 'tree-vect-slp.c', 'tree-vect-stmts.c' and

  7951.      'tree-vect-data-refs.c'.

  7952. 

  7953.      Autoparallelization.  This pass splits the loop iteration space to

  7954.      run into several threads.  The pass is implemented in

  7955.      'tree-parloops.c'.

  7956. 

  7957.      Graphite is a loop transformation framework based on the polyhedral

  7958.      model.  Graphite stands for Gimple Represented as Polyhedra.  The

  7959.      internals of this infrastructure are documented in

  7960.      <http://gcc.gnu.org/wiki/Graphite>.  The passes working on this

  7961.      representation are implemented in the various 'graphite-*' files.

  7962. 

  7963.    * Tree level if-conversion for vectorizer

  7964. 

  7965.      This pass applies if-conversion to simple loops to help vectorizer.

  7966.      We identify if convertible loops, if-convert statements and merge

  7967.      basic blocks in one big block.  The idea is to present loop in such

  7968.      form so that vectorizer can have one to one mapping between

  7969.      statements and available vector operations.  This pass is located

  7970.      in 'tree-if-conv.c' and is described by 'pass_if_conversion'.

  7971. 

  7972.    * Conditional constant propagation

  7973. 

  7974.      This pass relaxes a lattice of values in order to identify those

  7975.      that must be constant even in the presence of conditional branches.

  7976.      The pass is located in 'tree-ssa-ccp.c' and is described by

  7977.      'pass_ccp'.

  7978. 

  7979.      A related pass that works on memory loads and stores, and not just

  7980.      register values, is located in 'tree-ssa-ccp.c' and described by

  7981.      'pass_store_ccp'.

  7982. 

  7983.    * Conditional copy propagation

  7984. 

  7985.      This is similar to constant propagation but the lattice of values

  7986.      is the "copy-of" relation.  It eliminates redundant copies from the

  7987.      code.  The pass is located in 'tree-ssa-copy.c' and described by

  7988.      'pass_copy_prop'.

  7989. 

  7990.      A related pass that works on memory copies, and not just register

  7991.      copies, is located in 'tree-ssa-copy.c' and described by

  7992.      'pass_store_copy_prop'.

  7993. 

  7994.    * Value range propagation

  7995. 

  7996.      This transformation is similar to constant propagation but instead

  7997.      of propagating single constant values, it propagates known value

  7998.      ranges.  The implementation is based on Patterson's range

  7999.      propagation algorithm (Accurate Static Branch Prediction by Value

  8000.      Range Propagation, J. R. C. Patterson, PLDI '95).  In contrast to

  8001.      Patterson's algorithm, this implementation does not propagate

  8002.      branch probabilities nor it uses more than a single range per SSA

  8003.      name.  This means that the current implementation cannot be used

  8004.      for branch prediction (though adapting it would not be difficult).

  8005.      The pass is located in 'tree-vrp.c' and is described by 'pass_vrp'.

  8006. 

  8007.    * Folding built-in functions

  8008. 

  8009.      This pass simplifies built-in functions, as applicable, with

  8010.      constant arguments or with inferable string lengths.  It is located

  8011.      in 'tree-ssa-ccp.c' and is described by 'pass_fold_builtins'.

  8012. 

  8013.    * Split critical edges

  8014. 

  8015.      This pass identifies critical edges and inserts empty basic blocks

  8016.      such that the edge is no longer critical.  The pass is located in

  8017.      'tree-cfg.c' and is described by 'pass_split_crit_edges'.

  8018. 

  8019.    * Control dependence dead code elimination

  8020. 

  8021.      This pass is a stronger form of dead code elimination that can

  8022.      eliminate unnecessary control flow statements.  It is located in

  8023.      'tree-ssa-dce.c' and is described by 'pass_cd_dce'.

  8024. 

  8025.    * Tail call elimination

  8026. 

  8027.      This pass identifies function calls that may be rewritten into

  8028.      jumps.  No code transformation is actually applied here, but the

  8029.      data and control flow problem is solved.  The code transformation

  8030.      requires target support, and so is delayed until RTL.  In the

  8031.      meantime 'CALL_EXPR_TAILCALL' is set indicating the possibility.

  8032.      The pass is located in 'tree-tailcall.c' and is described by

  8033.      'pass_tail_calls'.  The RTL transformation is handled by

  8034.      'fixup_tail_calls' in 'calls.c'.

  8035. 

  8036.    * Warn for function return without value

  8037. 

  8038.      For non-void functions, this pass locates return statements that do

  8039.      not specify a value and issues a warning.  Such a statement may

  8040.      have been injected by falling off the end of the function.  This

  8041.      pass is run last so that we have as much time as possible to prove

  8042.      that the statement is not reachable.  It is located in 'tree-cfg.c'

  8043.      and is described by 'pass_warn_function_return'.

  8044. 

  8045.    * Leave static single assignment form

  8046. 

  8047.      This pass rewrites the function such that it is in normal form.  At

  8048.      the same time, we eliminate as many single-use temporaries as

  8049.      possible, so the intermediate language is no longer GIMPLE, but

  8050.      GENERIC.  The pass is located in 'tree-outof-ssa.c' and is

  8051.      described by 'pass_del_ssa'.

  8052. 

  8053.    * Merge PHI nodes that feed into one another

  8054. 

  8055.      This is part of the CFG cleanup passes.  It attempts to join PHI

  8056.      nodes from a forwarder CFG block into another block with PHI nodes.

  8057.      The pass is located in 'tree-cfgcleanup.c' and is described by

  8058.      'pass_merge_phi'.

  8059. 

  8060.    * Return value optimization

  8061. 

  8062.      If a function always returns the same local variable, and that

  8063.      local variable is an aggregate type, then the variable is replaced

  8064.      with the return value for the function (i.e., the function's

  8065.      DECL_RESULT). This is equivalent to the C++ named return value

  8066.      optimization applied to GIMPLE.  The pass is located in

  8067.      'tree-nrv.c' and is described by 'pass_nrv'.

  8068. 

  8069.    * Return slot optimization

  8070. 

  8071.      If a function returns a memory object and is called as 'var =

  8072.      foo()', this pass tries to change the call so that the address of

  8073.      'var' is sent to the caller to avoid an extra memory copy.  This

  8074.      pass is located in 'tree-nrv.c' and is described by

  8075.      'pass_return_slot'.

  8076. 

  8077.    * Optimize calls to '__builtin_object_size'

  8078. 

  8079.      This is a propagation pass similar to CCP that tries to remove

  8080.      calls to '__builtin_object_size' when the size of the object can be

  8081.      computed at compile-time.  This pass is located in

  8082.      'tree-object-size.c' and is described by 'pass_object_sizes'.

  8083. 

  8084.    * Loop invariant motion

  8085. 

  8086.      This pass removes expensive loop-invariant computations out of

  8087.      loops.  The pass is located in 'tree-ssa-loop.c' and described by

  8088.      'pass_lim'.

  8089. 

  8090.    * Loop nest optimizations

  8091. 

  8092.      This is a family of loop transformations that works on loop nests.

  8093.      It includes loop interchange, scaling, skewing and reversal and

  8094.      they are all geared to the optimization of data locality in array

  8095.      traversals and the removal of dependencies that hamper

  8096.      optimizations such as loop parallelization and vectorization.  The

  8097.      pass is located in 'tree-loop-linear.c' and described by

  8098.      'pass_linear_transform'.

  8099. 

  8100.    * Removal of empty loops

  8101. 

  8102.      This pass removes loops with no code in them.  The pass is located

  8103.      in 'tree-ssa-loop-ivcanon.c' and described by 'pass_empty_loop'.

  8104. 

  8105.    * Unrolling of small loops

  8106. 

  8107.      This pass completely unrolls loops with few iterations.  The pass

  8108.      is located in 'tree-ssa-loop-ivcanon.c' and described by

  8109.      'pass_complete_unroll'.

  8110. 

  8111.    * Predictive commoning

  8112. 

  8113.      This pass makes the code reuse the computations from the previous

  8114.      iterations of the loops, especially loads and stores to memory.  It

  8115.      does so by storing the values of these computations to a bank of

  8116.      temporary variables that are rotated at the end of loop.  To avoid

  8117.      the need for this rotation, the loop is then unrolled and the

  8118.      copies of the loop body are rewritten to use the appropriate

  8119.      version of the temporary variable.  This pass is located in

  8120.      'tree-predcom.c' and described by 'pass_predcom'.

  8121. 

  8122.    * Array prefetching

  8123. 

  8124.      This pass issues prefetch instructions for array references inside

  8125.      loops.  The pass is located in 'tree-ssa-loop-prefetch.c' and

  8126.      described by 'pass_loop_prefetch'.

  8127. 

  8128.    * Reassociation

  8129. 

  8130.      This pass rewrites arithmetic expressions to enable optimizations

  8131.      that operate on them, like redundancy elimination and

  8132.      vectorization.  The pass is located in 'tree-ssa-reassoc.c' and

  8133.      described by 'pass_reassoc'.

  8134. 

  8135.    * Optimization of 'stdarg' functions

  8136. 

  8137.      This pass tries to avoid the saving of register arguments into the

  8138.      stack on entry to 'stdarg' functions.  If the function doesn't use

  8139.      any 'va_start' macros, no registers need to be saved.  If

  8140.      'va_start' macros are used, the 'va_list' variables don't escape

  8141.      the function, it is only necessary to save registers that will be

  8142.      used in 'va_arg' macros.  For instance, if 'va_arg' is only used

  8143.      with integral types in the function, floating point registers don't

  8144.      need to be saved.  This pass is located in 'tree-stdarg.c' and

  8145.      described by 'pass_stdarg'.

  8146. 

  8147. 

  8148. File: gccint.info,  Node: RTL passes,  Next: Optimization info,  Prev: Tree SSA passes,  Up: Passes

  8149. 

  8150. 9.6 RTL passes

  8151. ==============

  8152. 

  8153. The following briefly describes the RTL generation and optimization

  8154. passes that are run after the Tree optimization passes.

  8155. 

  8156.    * RTL generation

  8157. 

  8158.      The source files for RTL generation include 'stmt.c', 'calls.c',

  8159.      'expr.c', 'explow.c', 'expmed.c', 'function.c', 'optabs.c' and

  8160.      'emit-rtl.c'.  Also, the file 'insn-emit.c', generated from the

  8161.      machine description by the program 'genemit', is used in this pass.

  8162.      The header file 'expr.h' is used for communication within this

  8163.      pass.

  8164. 

  8165.      The header files 'insn-flags.h' and 'insn-codes.h', generated from

  8166.      the machine description by the programs 'genflags' and 'gencodes',

  8167.      tell this pass which standard names are available for use and which

  8168.      patterns correspond to them.

  8169. 

  8170.    * Generation of exception landing pads

  8171. 

  8172.      This pass generates the glue that handles communication between the

  8173.      exception handling library routines and the exception handlers

  8174.      within the function.  Entry points in the function that are invoked

  8175.      by the exception handling library are called "landing pads".  The

  8176.      code for this pass is located in 'except.c'.

  8177. 

  8178.    * Control flow graph cleanup

  8179. 

  8180.      This pass removes unreachable code, simplifies jumps to next, jumps

  8181.      to jump, jumps across jumps, etc.  The pass is run multiple times.

  8182.      For historical reasons, it is occasionally referred to as the "jump

  8183.      optimization pass".  The bulk of the code for this pass is in

  8184.      'cfgcleanup.c', and there are support routines in 'cfgrtl.c' and

  8185.      'jump.c'.

  8186. 

  8187.    * Forward propagation of single-def values

  8188. 

  8189.      This pass attempts to remove redundant computation by substituting

  8190.      variables that come from a single definition, and seeing if the

  8191.      result can be simplified.  It performs copy propagation and

  8192.      addressing mode selection.  The pass is run twice, with values

  8193.      being propagated into loops only on the second run.  The code is

  8194.      located in 'fwprop.c'.

  8195. 

  8196.    * Common subexpression elimination

  8197. 

  8198.      This pass removes redundant computation within basic blocks, and

  8199.      optimizes addressing modes based on cost.  The pass is run twice.

  8200.      The code for this pass is located in 'cse.c'.

  8201. 

  8202.    * Global common subexpression elimination

  8203. 

  8204.      This pass performs two different types of GCSE depending on whether

  8205.      you are optimizing for size or not (LCM based GCSE tends to

  8206.      increase code size for a gain in speed, while Morel-Renvoise based

  8207.      GCSE does not).  When optimizing for size, GCSE is done using

  8208.      Morel-Renvoise Partial Redundancy Elimination, with the exception

  8209.      that it does not try to move invariants out of loops--that is left

  8210.      to the loop optimization pass.  If MR PRE GCSE is done, code

  8211.      hoisting (aka unification) is also done, as well as load motion.

  8212.      If you are optimizing for speed, LCM (lazy code motion) based GCSE

  8213.      is done.  LCM is based on the work of Knoop, Ruthing, and Steffen.

  8214.      LCM based GCSE also does loop invariant code motion.  We also

  8215.      perform load and store motion when optimizing for speed.

  8216.      Regardless of which type of GCSE is used, the GCSE pass also

  8217.      performs global constant and copy propagation.  The source file for

  8218.      this pass is 'gcse.c', and the LCM routines are in 'lcm.c'.

  8219. 

  8220.    * Loop optimization

  8221. 

  8222.      This pass performs several loop related optimizations.  The source

  8223.      files 'cfgloopanal.c' and 'cfgloopmanip.c' contain generic loop

  8224.      analysis and manipulation code.  Initialization and finalization of

  8225.      loop structures is handled by 'loop-init.c'.  A loop invariant

  8226.      motion pass is implemented in 'loop-invariant.c'.  Basic block

  8227.      level optimizations--unrolling, and peeling loops-- are implemented

  8228.      in 'loop-unroll.c'.  Replacing of the exit condition of loops by

  8229.      special machine-dependent instructions is handled by

  8230.      'loop-doloop.c'.

  8231. 

  8232.    * Jump bypassing

  8233. 

  8234.      This pass is an aggressive form of GCSE that transforms the control

  8235.      flow graph of a function by propagating constants into conditional

  8236.      branch instructions.  The source file for this pass is 'gcse.c'.

  8237. 

  8238.    * If conversion

  8239. 

  8240.      This pass attempts to replace conditional branches and surrounding

  8241.      assignments with arithmetic, boolean value producing comparison

  8242.      instructions, and conditional move instructions.  In the very last

  8243.      invocation after reload/LRA, it will generate predicated

  8244.      instructions when supported by the target.  The code is located in

  8245.      'ifcvt.c'.

  8246. 

  8247.    * Web construction

  8248. 

  8249.      This pass splits independent uses of each pseudo-register.  This

  8250.      can improve effect of the other transformation, such as CSE or

  8251.      register allocation.  The code for this pass is located in 'web.c'.

  8252. 

  8253.    * Instruction combination

  8254. 

  8255.      This pass attempts to combine groups of two or three instructions

  8256.      that are related by data flow into single instructions.  It

  8257.      combines the RTL expressions for the instructions by substitution,

  8258.      simplifies the result using algebra, and then attempts to match the

  8259.      result against the machine description.  The code is located in

  8260.      'combine.c'.

  8261. 

  8262.    * Mode switching optimization

  8263. 

  8264.      This pass looks for instructions that require the processor to be

  8265.      in a specific "mode" and minimizes the number of mode changes

  8266.      required to satisfy all users.  What these modes are, and what they

  8267.      apply to are completely target-specific.  The code for this pass is

  8268.      located in 'mode-switching.c'.

  8269. 

  8270.    * Modulo scheduling

  8271. 

  8272.      This pass looks at innermost loops and reorders their instructions

  8273.      by overlapping different iterations.  Modulo scheduling is

  8274.      performed immediately before instruction scheduling.  The code for

  8275.      this pass is located in 'modulo-sched.c'.

  8276. 

  8277.    * Instruction scheduling

  8278. 

  8279.      This pass looks for instructions whose output will not be available

  8280.      by the time that it is used in subsequent instructions.  Memory

  8281.      loads and floating point instructions often have this behavior on

  8282.      RISC machines.  It re-orders instructions within a basic block to

  8283.      try to separate the definition and use of items that otherwise

  8284.      would cause pipeline stalls.  This pass is performed twice, before

  8285.      and after register allocation.  The code for this pass is located

  8286.      in 'haifa-sched.c', 'sched-deps.c', 'sched-ebb.c', 'sched-rgn.c'

  8287.      and 'sched-vis.c'.

  8288. 

  8289.    * Register allocation

  8290. 

  8291.      These passes make sure that all occurrences of pseudo registers are

  8292.      eliminated, either by allocating them to a hard register, replacing

  8293.      them by an equivalent expression (e.g. a constant) or by placing

  8294.      them on the stack.  This is done in several subpasses:

  8295. 

  8296.         * The integrated register allocator (IRA).  It is called

  8297.           integrated because coalescing, register live range splitting,

  8298.           and hard register preferencing are done on-the-fly during

  8299.           coloring.  It also has better integration with the reload/LRA

  8300.           pass.  Pseudo-registers spilled by the allocator or the

  8301.           reload/LRA have still a chance to get hard-registers if the

  8302.           reload/LRA evicts some pseudo-registers from hard-registers.

  8303.           The allocator helps to choose better pseudos for spilling

  8304.           based on their live ranges and to coalesce stack slots

  8305.           allocated for the spilled pseudo-registers.  IRA is a regional

  8306.           register allocator which is transformed into Chaitin-Briggs

  8307.           allocator if there is one region.  By default, IRA chooses

  8308.           regions using register pressure but the user can force it to

  8309.           use one region or regions corresponding to all loops.

  8310. 

  8311.           Source files of the allocator are 'ira.c', 'ira-build.c',

  8312.           'ira-costs.c', 'ira-conflicts.c', 'ira-color.c', 'ira-emit.c',

  8313.           'ira-lives', plus header files 'ira.h' and 'ira-int.h' used

  8314.           for the communication between the allocator and the rest of

  8315.           the compiler and between the IRA files.

  8316. 

  8317.         * Reloading.  This pass renumbers pseudo registers with the

  8318.           hardware registers numbers they were allocated.  Pseudo

  8319.           registers that did not get hard registers are replaced with

  8320.           stack slots.  Then it finds instructions that are invalid

  8321.           because a value has failed to end up in a register, or has

  8322.           ended up in a register of the wrong kind.  It fixes up these

  8323.           instructions by reloading the problematical values temporarily

  8324.           into registers.  Additional instructions are generated to do

  8325.           the copying.

  8326. 

  8327.           The reload pass also optionally eliminates the frame pointer

  8328.           and inserts instructions to save and restore call-clobbered

  8329.           registers around calls.

  8330. 

  8331.           Source files are 'reload.c' and 'reload1.c', plus the header

  8332.           'reload.h' used for communication between them.

  8333. 

  8334.         * This pass is a modern replacement of the reload pass.  Source

  8335.           files are 'lra.c', 'lra-assign.c', 'lra-coalesce.c',

  8336.           'lra-constraints.c', 'lra-eliminations.c', 'lra-lives.c',

  8337.           'lra-remat.c', 'lra-spills.c', the header 'lra-int.h' used for

  8338.           communication between them, and the header 'lra.h' used for

  8339.           communication between LRA and the rest of compiler.

  8340. 

  8341.           Unlike the reload pass, intermediate LRA decisions are

  8342.           reflected in RTL as much as possible.  This reduces the number

  8343.           of target-dependent macros and hooks, leaving instruction

  8344.           constraints as the primary source of control.

  8345. 

  8346.           LRA is run on targets for which TARGET_LRA_P returns true.

  8347. 

  8348.    * Basic block reordering

  8349. 

  8350.      This pass implements profile guided code positioning.  If profile

  8351.      information is not available, various types of static analysis are

  8352.      performed to make the predictions normally coming from the profile

  8353.      feedback (IE execution frequency, branch probability, etc).  It is

  8354.      implemented in the file 'bb-reorder.c', and the various prediction

  8355.      routines are in 'predict.c'.

  8356. 

  8357.    * Variable tracking

  8358. 

  8359.      This pass computes where the variables are stored at each position

  8360.      in code and generates notes describing the variable locations to

  8361.      RTL code.  The location lists are then generated according to these

  8362.      notes to debug information if the debugging information format

  8363.      supports location lists.  The code is located in 'var-tracking.c'.

  8364. 

  8365.    * Delayed branch scheduling

  8366. 

  8367.      This optional pass attempts to find instructions that can go into

  8368.      the delay slots of other instructions, usually jumps and calls.

  8369.      The code for this pass is located in 'reorg.c'.

  8370. 

  8371.    * Branch shortening

  8372. 

  8373.      On many RISC machines, branch instructions have a limited range.

  8374.      Thus, longer sequences of instructions must be used for long

  8375.      branches.  In this pass, the compiler figures out what how far each

  8376.      instruction will be from each other instruction, and therefore

  8377.      whether the usual instructions, or the longer sequences, must be

  8378.      used for each branch.  The code for this pass is located in

  8379.      'final.c'.

  8380. 

  8381.    * Register-to-stack conversion

  8382. 

  8383.      Conversion from usage of some hard registers to usage of a register

  8384.      stack may be done at this point.  Currently, this is supported only

  8385.      for the floating-point registers of the Intel 80387 coprocessor.

  8386.      The code for this pass is located in 'reg-stack.c'.

  8387. 

  8388.    * Final

  8389. 

  8390.      This pass outputs the assembler code for the function.  The source

  8391.      files are 'final.c' plus 'insn-output.c'; the latter is generated

  8392.      automatically from the machine description by the tool 'genoutput'.

  8393.      The header file 'conditions.h' is used for communication between

  8394.      these files.

  8395. 

  8396.    * Debugging information output

  8397. 

  8398.      This is run after final because it must output the stack slot

  8399.      offsets for pseudo registers that did not get hard registers.

  8400.      Source files are 'dbxout.c' for DBX symbol table format,

  8401.      'dwarfout.c' for DWARF symbol table format, files 'dwarf2out.c' and

  8402.      'dwarf2asm.c' for DWARF2 symbol table format, and 'vmsdbgout.c' for

  8403.      VMS debug symbol table format.

  8404. 

  8405. 

  8406. File: gccint.info,  Node: Optimization info,  Prev: RTL passes,  Up: Passes

  8407. 

  8408. 9.7 Optimization info

  8409. =====================

  8410. 

  8411. This section is describes dump infrastructure which is common to both

  8412. pass dumps as well as optimization dumps.  The goal for this

  8413. infrastructure is to provide both gcc developers and users detailed

  8414. information about various compiler transformations and optimizations.

  8415. 

  8416. * Menu:

  8417. 

  8418. * Dump setup::                         Setup of optimization dumps.

  8419. * Optimization groups::                Groups made up of optimization passes.

  8420. * Dump files and streams::             Dump output file names and streams.

  8421. * Dump output verbosity::              How much information to dump.

  8422. * Dump types::                         Various types of dump functions.

  8423. * Dump examples::                      Sample usage.

  8424. 

  8425. 

  8426. File: gccint.info,  Node: Dump setup,  Next: Optimization groups,  Up: Optimization info

  8427. 

  8428. 9.7.1 Dump setup

  8429. ----------------

  8430. 

  8431. A dump_manager class is defined in 'dumpfile.h'.  Various passes

  8432. register dumping pass-specific information via 'dump_register' in

  8433. 'passes.c'.  During the registration, an optimization pass can select

  8434. its optimization group (*note Optimization groups::).  After that

  8435. optimization information corresponding to the entire group (presumably

  8436. from multiple passes) can be output via command-line switches.  Note

  8437. that if a pass does not fit into any of the pre-defined groups, it can

  8438. select 'OPTGROUP_NONE'.

  8439. 

  8440.  Note that in general, a pass need not know its dump output file name,

  8441. whether certain flags are enabled, etc.  However, for legacy reasons,

  8442. passes could also call 'dump_begin' which returns a stream in case the

  8443. particular pass has optimization dumps enabled.  A pass could call

  8444. 'dump_end' when the dump has ended.  These methods should go away once

  8445. all the passes are converted to use the new dump infrastructure.

  8446. 

  8447.  The recommended way to setup the dump output is via 'dump_start' and

  8448. 'dump_end'.

  8449. 

  8450. 

  8451. File: gccint.info,  Node: Optimization groups,  Next: Dump files and streams,  Prev: Dump setup,  Up: Optimization info

  8452. 

  8453. 9.7.2 Optimization groups

  8454. -------------------------

  8455. 

  8456. The optimization passes are grouped into several categories.  Currently

  8457. defined categories in 'dumpfile.h' are

  8458. 

  8459. 'OPTGROUP_IPA'

  8460.      IPA optimization passes.  Enabled by '-ipa'

  8461. 

  8462. 'OPTGROUP_LOOP'

  8463.      Loop optimization passes.  Enabled by '-loop'.

  8464. 

  8465. 'OPTGROUP_INLINE'

  8466.      Inlining passes.  Enabled by '-inline'.

  8467. 

  8468. 'OPTGROUP_OMP'

  8469.      OMP (Offloading and Multi Processing) passes.  Enabled by '-omp'.

  8470. 

  8471. 'OPTGROUP_VEC'

  8472.      Vectorization passes.  Enabled by '-vec'.

  8473. 

  8474. 'OPTGROUP_OTHER'

  8475.      All other optimization passes which do not fall into one of the

  8476.      above.

  8477. 

  8478. 'OPTGROUP_ALL'

  8479.      All optimization passes.  Enabled by '-optall'.

  8480. 

  8481.  By using groups a user could selectively enable optimization

  8482. information only for a group of passes.  By default, the optimization

  8483. information for all the passes is dumped.

  8484. 

  8485. 

  8486. File: gccint.info,  Node: Dump files and streams,  Next: Dump output verbosity,  Prev: Optimization groups,  Up: Optimization info

  8487. 

  8488. 9.7.3 Dump files and streams

  8489. ----------------------------

  8490. 

  8491. There are two separate output streams available for outputting

  8492. optimization information from passes.  Note that both these streams

  8493. accept 'stderr' and 'stdout' as valid streams and thus it is possible to

  8494. dump output to standard output or error.  This is specially handy for

  8495. outputting all available information in a single file by redirecting

  8496. 'stderr'.

  8497. 

  8498. 'pstream'

  8499.      This stream is for pass-specific dump output.  For example,

  8500.      '-fdump-tree-vect=foo.v' dumps tree vectorization pass output into

  8501.      the given file name 'foo.v'.  If the file name is not provided, the

  8502.      default file name is based on the source file and pass number.

  8503.      Note that one could also use special file names 'stdout' and

  8504.      'stderr' for dumping to standard output and standard error

  8505.      respectively.

  8506. 

  8507. 'alt_stream'

  8508.      This steam is used for printing optimization specific output in

  8509.      response to the '-fopt-info'.  Again a file name can be given.  If

  8510.      the file name is not given, it defaults to 'stderr'.

  8511. 

  8512. 

  8513. File: gccint.info,  Node: Dump output verbosity,  Next: Dump types,  Prev: Dump files and streams,  Up: Optimization info

  8514. 

  8515. 9.7.4 Dump output verbosity

  8516. ---------------------------

  8517. 

  8518. The dump verbosity has the following options

  8519. 

  8520. 'optimized'

  8521.      Print information when an optimization is successfully applied.  It

  8522.      is up to a pass to decide which information is relevant.  For

  8523.      example, the vectorizer passes print the source location of loops

  8524.      which got successfully vectorized.

  8525. 

  8526. 'missed'

  8527.      Print information about missed optimizations.  Individual passes

  8528.      control which information to include in the output.  For example,

  8529. 

  8530.           gcc -O2 -ftree-vectorize -fopt-info-vec-missed

  8531. 

  8532.      will print information about missed optimization opportunities from

  8533.      vectorization passes on stderr.

  8534. 

  8535. 'note'

  8536.      Print verbose information about optimizations, such as certain

  8537.      transformations, more detailed messages about decisions etc.

  8538. 

  8539. 'all'

  8540.      Print detailed optimization information.  This includes OPTIMIZED,

  8541.      MISSED, and NOTE.

  8542. 

  8543. 

  8544. File: gccint.info,  Node: Dump types,  Next: Dump examples,  Prev: Dump output verbosity,  Up: Optimization info

  8545. 

  8546. 9.7.5 Dump types

  8547. ----------------

  8548. 

  8549. 'dump_printf'

  8550. 

  8551.      This is a generic method for doing formatted output.  It takes an

  8552.      additional argument 'dump_kind' which signifies the type of dump.

  8553.      This method outputs information only when the dumps are enabled for

  8554.      this particular 'dump_kind'.  Note that the caller doesn't need to

  8555.      know if the particular dump is enabled or not, or even the file

  8556.      name.  The caller only needs to decide which dump output

  8557.      information is relevant, and under what conditions.  This

  8558.      determines the associated flags.

  8559. 

  8560.      Consider the following example from 'loop-unroll.c' where an

  8561.      informative message about a loop (along with its location) is

  8562.      printed when any of the following flags is enabled

  8563. 

  8564.         - optimization messages

  8565.         - RTL dumps

  8566.         - detailed dumps

  8567. 

  8568.           int report_flags = MSG_OPTIMIZED_LOCATIONS | TDF_RTL | TDF_DETAILS;

  8569.           dump_printf_loc (report_flags, insn,

  8570.                            "loop turned into non-loop; it never loops.\n");

  8571. 

  8572. 'dump_basic_block'

  8573.      Output basic block.

  8574. 'dump_generic_expr'

  8575.      Output generic expression.

  8576. 'dump_gimple_stmt'

  8577.      Output gimple statement.

  8578. 

  8579.      Note that the above methods also have variants prefixed with

  8580.      '_loc', such as 'dump_printf_loc', which are similar except they

  8581.      also output the source location information.  The '_loc' variants

  8582.      take a 'const dump_location_t &'.  This class can be constructed

  8583.      from a 'gimple *' or from a 'rtx_insn *', and so callers can pass a

  8584.      'gimple *' or a 'rtx_insn *' as the '_loc' argument.  The

  8585.      'dump_location_t' constructor will extract the source location from

  8586.      the statement or instruction, along with the profile count, and the

  8587.      location in GCC's own source code (or the plugin) from which the

  8588.      dump call was emitted.  Only the source location is currently used.

  8589.      There is also a 'dump_user_location_t' class, capturing the source

  8590.      location and profile count, but not the dump emission location, so

  8591.      that locations in the user's code can be passed around.  This can

  8592.      also be constructed from a 'gimple *' and from a 'rtx_insn *', and

  8593.      it too can be passed as the '_loc' argument.

  8594. 

  8595. 

  8596. File: gccint.info,  Node: Dump examples,  Prev: Dump types,  Up: Optimization info

  8597. 

  8598. 9.7.6 Dump examples

  8599. -------------------

  8600. 

  8601.      gcc -O3 -fopt-info-missed=missed.all

  8602. 

  8603.  outputs missed optimization report from all the passes into

  8604. 'missed.all'.

  8605. 

  8606.  As another example,

  8607.      gcc -O3 -fopt-info-inline-optimized-missed=inline.txt

  8608. 

  8609.  will output information about missed optimizations as well as optimized

  8610. locations from all the inlining passes into 'inline.txt'.

  8611. 

  8612.  If the FILENAME is provided, then the dumps from all the applicable

  8613. optimizations are concatenated into the 'filename'.  Otherwise the dump

  8614. is output onto 'stderr'.  If OPTIONS is omitted, it defaults to

  8615. 'optimized-optall', which means dump all information about successful

  8616. optimizations from all the passes.  In the following example, the

  8617. optimization information is output on to 'stderr'.

  8618. 

  8619.      gcc -O3 -fopt-info

  8620. 

  8621.  Note that '-fopt-info-vec-missed' behaves the same as

  8622. '-fopt-info-missed-vec'.  The order of the optimization group names and

  8623. message types listed after '-fopt-info' does not matter.

  8624. 

  8625.  As another example, consider

  8626. 

  8627.      gcc -fopt-info-vec-missed=vec.miss -fopt-info-loop-optimized=loop.opt

  8628. 

  8629.  Here the two output file names 'vec.miss' and 'loop.opt' are in

  8630. conflict since only one output file is allowed.  In this case, only the

  8631. first option takes effect and the subsequent options are ignored.  Thus

  8632. only the 'vec.miss' is produced which containts dumps from the

  8633. vectorizer about missed opportunities.

  8634. 

  8635. 

  8636. File: gccint.info,  Node: poly_int,  Next: GENERIC,  Prev: Passes,  Up: Top

  8637. 

  8638. 10 Sizes and offsets as runtime invariants

  8639. ******************************************

  8640. 

  8641. GCC allows the size of a hardware register to be a runtime invariant

  8642. rather than a compile-time constant.  This in turn means that various

  8643. sizes and offsets must also be runtime invariants rather than

  8644. compile-time constants, such as:

  8645. 

  8646.    * the size of a general 'machine_mode' (*note Machine Modes::);

  8647. 

  8648.    * the size of a spill slot;

  8649. 

  8650.    * the offset of something within a stack frame;

  8651. 

  8652.    * the number of elements in a vector;

  8653. 

  8654.    * the size and offset of a 'mem' rtx (*note Regs and Memory::); and

  8655. 

  8656.    * the byte offset in a 'subreg' rtx (*note Regs and Memory::).

  8657. 

  8658.  The motivating example is the Arm SVE ISA, whose vector registers can

  8659. be any multiple of 128 bits between 128 and 2048 inclusive.  The

  8660. compiler normally produces code that works for all SVE register sizes,

  8661. with the actual size only being known at runtime.

  8662. 

  8663.  GCC's main representation of such runtime invariants is the 'poly_int'

  8664. class.  This chapter describes what 'poly_int' does, lists the available

  8665. operations, and gives some general usage guidelines.

  8666. 

  8667. * Menu:

  8668. 

  8669. * Overview of poly_int::

  8670. * Consequences of using poly_int::

  8671. * Comparisons involving poly_int::

  8672. * Arithmetic on poly_ints::

  8673. * Alignment of poly_ints::

  8674. * Computing bounds on poly_ints::

  8675. * Converting poly_ints::

  8676. * Miscellaneous poly_int routines::

  8677. * Guidelines for using poly_int::

  8678. 

  8679. 

  8680. File: gccint.info,  Node: Overview of poly_int,  Next: Consequences of using poly_int,  Up: poly_int

  8681. 

  8682. 10.1 Overview of 'poly_int'

  8683. ===========================

  8684. 

  8685. We define indeterminates X1, ..., XN whose values are only known at

  8686. runtime and use polynomials of the form:

  8687. 

  8688.      C0 + C1 * X1 + ... + CN * XN

  8689. 

  8690.  to represent a size or offset whose value might depend on some of these

  8691. indeterminates.  The coefficients C0, ..., CN are always known at

  8692. compile time, with the C0 term being the "constant" part that does not

  8693. depend on any runtime value.

  8694. 

  8695.  GCC uses the 'poly_int' class to represent these coefficients.  The

  8696. class has two template parameters: the first specifies the number of

  8697. coefficients (N + 1) and the second specifies the type of the

  8698. coefficients.  For example, 'poly_int<2, unsigned short>' represents a

  8699. polynomial with two coefficients (and thus one indeterminate), with each

  8700. coefficient having type 'unsigned short'.  When N is 0, the class

  8701. degenerates to a single compile-time constant C0.

  8702. 

  8703.  The number of coefficients needed for compilation is a fixed property

  8704. of each target and is specified by the configuration macro

  8705. 'NUM_POLY_INT_COEFFS'.  The default value is 1, since most targets do

  8706. not have such runtime invariants.  Targets that need a different value

  8707. should '#define' the macro in their 'CPU-modes.def' file.  *Note Back

  8708. End::.

  8709. 

  8710.  'poly_int' makes the simplifying requirement that each indeterminate

  8711. must be a nonnegative integer.  An indeterminate value of 0 should

  8712. usually represent the minimum possible runtime value, with C0 specifying

  8713. the value in that case.

  8714. 

  8715.  For example, when targetting the Arm SVE ISA, the single indeterminate

  8716. represents the number of 128-bit blocks in a vector _beyond the minimum

  8717. length of 128 bits_.  Thus the number of 64-bit doublewords in a vector

  8718. is 2 + 2 * X1.  If an aggregate has a single SVE vector and 16

  8719. additional bytes, its total size is 32 + 16 * X1 bytes.

  8720. 

  8721.  The header file 'poly-int-types.h' provides typedefs for the most

  8722. common forms of 'poly_int', all having 'NUM_POLY_INT_COEFFS'

  8723. coefficients:

  8724. 

  8725. 'poly_uint16'

  8726.      a 'poly_int' with 'unsigned short' coefficients.

  8727. 

  8728. 'poly_int64'

  8729.      a 'poly_int' with 'HOST_WIDE_INT' coefficients.

  8730. 

  8731. 'poly_uint64'

  8732.      a 'poly_int' with 'unsigned HOST_WIDE_INT' coefficients.

  8733. 

  8734. 'poly_offset_int'

  8735.      a 'poly_int' with 'offset_int' coefficients.

  8736. 

  8737. 'poly_wide_int'

  8738.      a 'poly_int' with 'wide_int' coefficients.

  8739. 

  8740. 'poly_widest_int'

  8741.      a 'poly_int' with 'widest_int' coefficients.

  8742. 

  8743.  Since the main purpose of 'poly_int' is to represent sizes and offsets,

  8744. the last two typedefs are only rarely used.

  8745. 

  8746. 

  8747. File: gccint.info,  Node: Consequences of using poly_int,  Next: Comparisons involving poly_int,  Prev: Overview of poly_int,  Up: poly_int

  8748. 

  8749. 10.2 Consequences of using 'poly_int'

  8750. =====================================

  8751. 

  8752. The two main consequences of using polynomial sizes and offsets are

  8753. that:

  8754. 

  8755.    * there is no total ordering between the values at compile time, and

  8756. 

  8757.    * some operations might yield results that cannot be expressed as a

  8758.      'poly_int'.

  8759. 

  8760.  For example, if X is a runtime invariant, we cannot tell at compile

  8761. time whether:

  8762. 

  8763.      3 + 4X <= 1 + 5X

  8764. 

  8765.  since the condition is false when X <= 1 and true when X >= 2.

  8766. 

  8767.  Similarly, 'poly_int' cannot represent the result of:

  8768. 

  8769.      (3 + 4X) * (1 + 5X)

  8770. 

  8771.  since it cannot (and in practice does not need to) store powers greater

  8772. than one.  It also cannot represent the result of:

  8773. 

  8774.      (3 + 4X) / (1 + 5X)

  8775. 

  8776.  The following sections describe how we deal with these restrictions.

  8777. 

  8778.  As described earlier, a 'poly_int<1, T>' has no indeterminates and so

  8779. degenerates to a compile-time constant of type T.  It would be possible

  8780. in that case to do all normal arithmetic on the T, and to compare the T

  8781. using the normal C++ operators.  We deliberately prevent

  8782. target-independent code from doing this, since the compiler needs to

  8783. support other 'poly_int<N, T>' as well, regardless of the current

  8784. target's 'NUM_POLY_INT_COEFFS'.

  8785. 

  8786.  However, it would be very artificial to force target-specific code to

  8787. follow these restrictions if the target has no runtime indeterminates.

  8788. There is therefore an implicit conversion from 'poly_int<1, T>' to T

  8789. when compiling target-specific translation units.

  8790. 

  8791. 

  8792. File: gccint.info,  Node: Comparisons involving poly_int,  Next: Arithmetic on poly_ints,  Prev: Consequences of using poly_int,  Up: poly_int

  8793. 

  8794. 10.3 Comparisons involving 'poly_int'

  8795. =====================================

  8796. 

  8797. In general we need to compare sizes and offsets in two situations: those

  8798. in which the values need to be ordered, and those in which the values

  8799. can be unordered.  More loosely, the distinction is often between values

  8800. that have a definite link (usually because they refer to the same

  8801. underlying register or memory location) and values that have no definite

  8802. link.  An example of the former is the relationship between the inner

  8803. and outer sizes of a subreg, where we must know at compile time whether

  8804. the subreg is paradoxical, partial, or complete.  An example of the

  8805. latter is alias analysis: we might want to check whether two arbitrary

  8806. memory references overlap.

  8807. 

  8808.  Referring back to the examples in the previous section, it makes sense

  8809. to ask whether a memory reference of size '3 + 4X' overlaps one of size

  8810. '1 + 5X', but it does not make sense to have a subreg in which the outer

  8811. mode has '3 + 4X' bytes and the inner mode has '1 + 5X' bytes (or vice

  8812. versa).  Such subregs are always invalid and should trigger an internal

  8813. compiler error if formed.

  8814. 

  8815.  The underlying operators are the same in both cases, but the

  8816. distinction affects how they are used.

  8817. 

  8818. * Menu:

  8819. 

  8820. * Comparison functions for poly_int::

  8821. * Properties of the poly_int comparisons::

  8822. * Comparing potentially-unordered poly_ints::

  8823. * Comparing ordered poly_ints::

  8824. * Checking for a poly_int marker value::

  8825. * Range checks on poly_ints::

  8826. * Sorting poly_ints::

  8827. 

  8828. 

  8829. File: gccint.info,  Node: Comparison functions for poly_int,  Next: Properties of the poly_int comparisons,  Up: Comparisons involving poly_int

  8830. 

  8831. 10.3.1 Comparison functions for 'poly_int'

  8832. ------------------------------------------

  8833. 

  8834. 'poly_int' provides the following routines for checking whether a

  8835. particular condition "may be" (might be) true:

  8836. 

  8837.      maybe_lt maybe_le maybe_eq maybe_ge maybe_gt

  8838.                        maybe_ne

  8839. 

  8840.  The functions have their natural meaning:

  8841. 

  8842. 'maybe_lt(A, B)'

  8843.      Return true if A might be less than B.

  8844. 

  8845. 'maybe_le(A, B)'

  8846.      Return true if A might be less than or equal to B.

  8847. 

  8848. 'maybe_eq(A, B)'

  8849.      Return true if A might be equal to B.

  8850. 

  8851. 'maybe_ne(A, B)'

  8852.      Return true if A might not be equal to B.

  8853. 

  8854. 'maybe_ge(A, B)'

  8855.      Return true if A might be greater than or equal to B.

  8856. 

  8857. 'maybe_gt(A, B)'

  8858.      Return true if A might be greater than B.

  8859. 

  8860.  For readability, 'poly_int' also provides "known" inverses of these

  8861. functions:

  8862. 

  8863.      known_lt (A, B) == !maybe_ge (A, B)

  8864.      known_le (A, B) == !maybe_gt (A, B)

  8865.      known_eq (A, B) == !maybe_ne (A, B)

  8866.      known_ge (A, B) == !maybe_lt (A, B)

  8867.      known_gt (A, B) == !maybe_le (A, B)

  8868.      known_ne (A, B) == !maybe_eq (A, B)

  8869. 

  8870. 

  8871. File: gccint.info,  Node: Properties of the poly_int comparisons,  Next: Comparing potentially-unordered poly_ints,  Prev: Comparison functions for poly_int,  Up: Comparisons involving poly_int

  8872. 

  8873. 10.3.2 Properties of the 'poly_int' comparisons

  8874. -----------------------------------------------

  8875. 

  8876. All "maybe" relations except 'maybe_ne' are transitive, so for example:

  8877. 

  8878.      maybe_lt (A, B) && maybe_lt (B, C) implies maybe_lt (A, C)

  8879. 

  8880.  for all A, B and C.  'maybe_lt', 'maybe_gt' and 'maybe_ne' are

  8881. irreflexive, so for example:

  8882. 

  8883.      !maybe_lt (A, A)

  8884. 

  8885.  is true for all A.  'maybe_le', 'maybe_eq' and 'maybe_ge' are

  8886. reflexive, so for example:

  8887. 

  8888.      maybe_le (A, A)

  8889. 

  8890.  is true for all A.  'maybe_eq' and 'maybe_ne' are symmetric, so:

  8891. 

  8892.      maybe_eq (A, B) == maybe_eq (B, A)

  8893.      maybe_ne (A, B) == maybe_ne (B, A)

  8894. 

  8895.  for all A and B.  In addition:

  8896. 

  8897.      maybe_le (A, B) == maybe_lt (A, B) || maybe_eq (A, B)

  8898.      maybe_ge (A, B) == maybe_gt (A, B) || maybe_eq (A, B)

  8899.      maybe_lt (A, B) == maybe_gt (B, A)

  8900.      maybe_le (A, B) == maybe_ge (B, A)

  8901. 

  8902.  However:

  8903. 

  8904.      maybe_le (A, B) && maybe_le (B, A) does not imply !maybe_ne (A, B) [== known_eq (A, B)]

  8905.      maybe_ge (A, B) && maybe_ge (B, A) does not imply !maybe_ne (A, B) [== known_eq (A, B)]

  8906. 

  8907.  One example is again 'A == 3 + 4X' and 'B == 1 + 5X', where 'maybe_le

  8908. (A, B)', 'maybe_ge (A, B)' and 'maybe_ne (A, B)' all hold.  'maybe_le'

  8909. and 'maybe_ge' are therefore not antisymetric and do not form a partial

  8910. order.

  8911. 

  8912.  From the above, it follows that:

  8913. 

  8914.    * All "known" relations except 'known_ne' are transitive.

  8915. 

  8916.    * 'known_lt', 'known_ne' and 'known_gt' are irreflexive.

  8917. 

  8918.    * 'known_le', 'known_eq' and 'known_ge' are reflexive.

  8919. 

  8920.  Also:

  8921. 

  8922.      known_lt (A, B) == known_gt (B, A)

  8923.      known_le (A, B) == known_ge (B, A)

  8924.      known_lt (A, B) implies !known_lt (B, A)  [asymmetry]

  8925.      known_gt (A, B) implies !known_gt (B, A)

  8926.      known_le (A, B) && known_le (B, A) == known_eq (A, B) [== !maybe_ne (A, B)]

  8927.      known_ge (A, B) && known_ge (B, A) == known_eq (A, B) [== !maybe_ne (A, B)]

  8928. 

  8929.  'known_le' and 'known_ge' are therefore antisymmetric and are partial

  8930. orders.  However:

  8931. 

  8932.      known_le (A, B) does not imply known_lt (A, B) || known_eq (A, B)

  8933.      known_ge (A, B) does not imply known_gt (A, B) || known_eq (A, B)

  8934. 

  8935.  For example, 'known_le (4, 4 + 4X)' holds because the runtime

  8936. indeterminate X is a nonnegative integer, but neither 'known_lt (4, 4 +

  8937. 4X)' nor 'known_eq (4, 4 + 4X)' hold.

  8938. 

  8939. 

  8940. File: gccint.info,  Node: Comparing potentially-unordered poly_ints,  Next: Comparing ordered poly_ints,  Prev: Properties of the poly_int comparisons,  Up: Comparisons involving poly_int

  8941. 

  8942. 10.3.3 Comparing potentially-unordered 'poly_int's

  8943. --------------------------------------------------

  8944. 

  8945. In cases where there is no definite link between two 'poly_int's, we can

  8946. usually make a conservatively-correct assumption.  For example, the

  8947. conservative assumption for alias analysis is that two references

  8948. _might_ alias.

  8949. 

  8950.  One way of checking whether [BEGIN1, END1) might overlap [BEGIN2, END2)

  8951. using the 'poly_int' comparisons is:

  8952. 

  8953.      maybe_gt (END1, BEGIN2) && maybe_gt (END2, BEGIN1)

  8954. 

  8955.  and another (equivalent) way is:

  8956. 

  8957.      !(known_le (END1, BEGIN2) || known_le (END2, BEGIN1))

  8958. 

  8959.  However, in this particular example, it is better to use the range

  8960. helper functions instead.  *Note Range checks on poly_ints::.

  8961. 

  8962. 

  8963. File: gccint.info,  Node: Comparing ordered poly_ints,  Next: Checking for a poly_int marker value,  Prev: Comparing potentially-unordered poly_ints,  Up: Comparisons involving poly_int

  8964. 

  8965. 10.3.4 Comparing ordered 'poly_int's

  8966. ------------------------------------

  8967. 

  8968. In cases where there is a definite link between two 'poly_int's, such as

  8969. the outer and inner sizes of subregs, we usually require the sizes to be

  8970. ordered by the 'known_le' partial order.  'poly_int' provides the

  8971. following utility functions for ordered values:

  8972. 

  8973. 'ordered_p (A, B)'

  8974.      Return true if A and B are ordered by the 'known_le' partial order.

  8975. 

  8976. 'ordered_min (A, B)'

  8977.      Assert that A and B are ordered by 'known_le' and return the

  8978.      minimum of the two.  When using this function, please add a comment

  8979.      explaining why the values are known to be ordered.

  8980. 

  8981. 'ordered_max (A, B)'

  8982.      Assert that A and B are ordered by 'known_le' and return the

  8983.      maximum of the two.  When using this function, please add a comment

  8984.      explaining why the values are known to be ordered.

  8985. 

  8986.  For example, if a subreg has an outer mode of size OUTER and an inner

  8987. mode of size INNER:

  8988. 

  8989.    * the subreg is complete if known_eq (INNER, OUTER)

  8990. 

  8991.    * otherwise, the subreg is paradoxical if known_le (INNER, OUTER)

  8992. 

  8993.    * otherwise, the subreg is partial if known_le (OUTER, INNER)

  8994. 

  8995.    * otherwise, the subreg is ill-formed

  8996. 

  8997.  Thus the subreg is only valid if 'ordered_p (OUTER, INNER)' is true.

  8998. If this condition is already known to be true then:

  8999. 

  9000.    * the subreg is complete if known_eq (INNER, OUTER)

  9001. 

  9002.    * the subreg is paradoxical if maybe_lt (INNER, OUTER)

  9003. 

  9004.    * the subreg is partial if maybe_lt (OUTER, INNER)

  9005. 

  9006.  with the three conditions being mutually exclusive.

  9007. 

  9008.  Code that checks whether a subreg is valid would therefore generally

  9009. check whether 'ordered_p' holds (in addition to whatever other checks

  9010. are required for subreg validity).  Code that is dealing with existing

  9011. subregs can assert that 'ordered_p' holds and use either of the

  9012. classifications above.

  9013. 

  9014. 

  9015. File: gccint.info,  Node: Checking for a poly_int marker value,  Next: Range checks on poly_ints,  Prev: Comparing ordered poly_ints,  Up: Comparisons involving poly_int

  9016. 

  9017. 10.3.5 Checking for a 'poly_int' marker value

  9018. ---------------------------------------------

  9019. 

  9020. It is sometimes useful to have a special "marker value" that is not

  9021. meant to be taken literally.  For example, some code uses a size of -1

  9022. to represent an unknown size, rather than having to carry around a

  9023. separate boolean to say whether the size is known.

  9024. 

  9025.  The best way of checking whether something is a marker value is

  9026. 'known_eq'.  Conversely the best way of checking whether something is

  9027. _not_ a marker value is 'maybe_ne'.

  9028. 

  9029.  Thus in the size example just mentioned, 'known_eq (size, -1)' would

  9030. check for an unknown size and 'maybe_ne (size, -1)' would check for a

  9031. known size.

  9032. 

  9033. 

  9034. File: gccint.info,  Node: Range checks on poly_ints,  Next: Sorting poly_ints,  Prev: Checking for a poly_int marker value,  Up: Comparisons involving poly_int

  9035. 

  9036. 10.3.6 Range checks on 'poly_int's

  9037. ----------------------------------

  9038. 

  9039. As well as the core comparisons (*note Comparison functions for

  9040. poly_int::), 'poly_int' provides utilities for various kinds of range

  9041. check.  In each case the range is represented by a start position and a

  9042. size rather than a start position and an end position; this is because

  9043. the former is used much more often than the latter in GCC.  Also, the

  9044. sizes can be -1 (or all ones for unsigned sizes) to indicate a range

  9045. with a known start position but an unknown size.  All other sizes must

  9046. be nonnegative.  A range of size 0 does not contain anything or overlap

  9047. anything.

  9048. 

  9049. 'known_size_p (SIZE)'

  9050.      Return true if SIZE represents a known range size, false if it is

  9051.      -1 or all ones (for signed and unsigned types respectively).

  9052. 

  9053. 'ranges_maybe_overlap_p (POS1, SIZE1, POS2, SIZE2)'

  9054.      Return true if the range described by POS1 and SIZE1 _might_

  9055.      overlap the range described by POS2 and SIZE2 (in other words,

  9056.      return true if we cannot prove that the ranges are disjoint).

  9057. 

  9058. 'ranges_known_overlap_p (POS1, SIZE1, POS2, SIZE2)'

  9059.      Return true if the range described by POS1 and SIZE1 is known to

  9060.      overlap the range described by POS2 and SIZE2.

  9061. 

  9062. 'known_subrange_p (POS1, SIZE1, POS2, SIZE2)'

  9063.      Return true if the range described by POS1 and SIZE1 is known to be

  9064.      contained in the range described by POS2 and SIZE2.

  9065. 

  9066. 'maybe_in_range_p (VALUE, POS, SIZE)'

  9067.      Return true if VALUE _might_ be in the range described by POS and

  9068.      SIZE (in other words, return true if we cannot prove that VALUE is

  9069.      outside that range).

  9070. 

  9071. 'known_in_range_p (VALUE, POS, SIZE)'

  9072.      Return true if VALUE is known to be in the range described by POS

  9073.      and SIZE.

  9074. 

  9075. 'endpoint_representable_p (POS, SIZE)'

  9076.      Return true if the range described by POS and SIZE is open-ended or

  9077.      if the endpoint (POS + SIZE) is representable in the same type as

  9078.      POS and SIZE.  The function returns false if adding SIZE to POS

  9079.      makes conceptual sense but could overflow.

  9080. 

  9081.  There is also a 'poly_int' version of the 'IN_RANGE_P' macro:

  9082. 

  9083. 'coeffs_in_range_p (X, LOWER, UPPER)'

  9084.      Return true if every coefficient of X is in the inclusive range

  9085.      [LOWER, UPPER].  This function can be useful when testing whether

  9086.      an operation would cause the values of coefficients to overflow.

  9087. 

  9088.      Note that the function does not indicate whether X itself is in the

  9089.      given range.  X can be either a constant or a 'poly_int'.

  9090. 

  9091. 

  9092. File: gccint.info,  Node: Sorting poly_ints,  Prev: Range checks on poly_ints,  Up: Comparisons involving poly_int

  9093. 

  9094. 10.3.7 Sorting 'poly_int's

  9095. --------------------------

  9096. 

  9097. 'poly_int' provides the following routine for sorting:

  9098. 

  9099. 'compare_sizes_for_sort (A, B)'

  9100.      Compare A and B in reverse lexicographical order (that is, compare

  9101.      the highest-indexed coefficients first).  This can be useful when

  9102.      sorting data structures, since it has the effect of separating

  9103.      constant and non-constant values.  If all values are nonnegative,

  9104.      the constant values come first.

  9105. 

  9106.      Note that the values do not necessarily end up in numerical order.

  9107.      For example, '1 + 1X' would come after '100' in the sort order, but

  9108.      may well be less than '100' at run time.

  9109. 

  9110. 

  9111. File: gccint.info,  Node: Arithmetic on poly_ints,  Next: Alignment of poly_ints,  Prev: Comparisons involving poly_int,  Up: poly_int

  9112. 

  9113. 10.4 Arithmetic on 'poly_int's

  9114. ==============================

  9115. 

  9116. Addition, subtraction, negation and bit inversion all work normally for

  9117. 'poly_int's.  Multiplication by a constant multiplier and left shifting

  9118. by a constant shift amount also work normally.  General multiplication

  9119. of two 'poly_int's is not supported and is not useful in practice.

  9120. 

  9121.  Other operations are only conditionally supported: the operation might

  9122. succeed or might fail, depending on the inputs.

  9123. 

  9124.  This section describes both types of operation.

  9125. 

  9126. * Menu:

  9127. 

  9128. * Using poly_int with C++ arithmetic operators::

  9129. * wi arithmetic on poly_ints::

  9130. * Division of poly_ints::

  9131. * Other poly_int arithmetic::

  9132. 

  9133. 

  9134. File: gccint.info,  Node: Using poly_int with C++ arithmetic operators,  Next: wi arithmetic on poly_ints,  Up: Arithmetic on poly_ints

  9135. 

  9136. 10.4.1 Using 'poly_int' with C++ arithmetic operators

  9137. -----------------------------------------------------

  9138. 

  9139. The following C++ expressions are supported, where P1 and P2 are

  9140. 'poly_int's and where C1 and C2 are scalars:

  9141. 

  9142.      -P1

  9143.      ~P1

  9144. 

  9145.      P1 + P2

  9146.      P1 + C2

  9147.      C1 + P2

  9148. 

  9149.      P1 - P2

  9150.      P1 - C2

  9151.      C1 - P2

  9152. 

  9153.      C1 * P2

  9154.      P1 * C2

  9155. 

  9156.      P1 << C2

  9157. 

  9158.      P1 += P2

  9159.      P1 += C2

  9160. 

  9161.      P1 -= P2

  9162.      P1 -= C2

  9163. 

  9164.      P1 *= C2

  9165.      P1 <<= C2

  9166. 

  9167.  These arithmetic operations handle integer ranks in a similar way to

  9168. C++.  The main difference is that every coefficient narrower than

  9169. 'HOST_WIDE_INT' promotes to 'HOST_WIDE_INT', whereas in C++ everything

  9170. narrower than 'int' promotes to 'int'.  For example:

  9171. 

  9172.      poly_uint16     + int          -> poly_int64

  9173.      unsigned int    + poly_uint16  -> poly_int64

  9174.      poly_int64      + int          -> poly_int64

  9175.      poly_int32      + poly_uint64  -> poly_uint64

  9176.      uint64          + poly_int64   -> poly_uint64

  9177.      poly_offset_int + int32        -> poly_offset_int

  9178.      offset_int      + poly_uint16  -> poly_offset_int

  9179. 

  9180.  In the first two examples, both coefficients are narrower than

  9181. 'HOST_WIDE_INT', so the result has coefficients of type 'HOST_WIDE_INT'.

  9182. In the other examples, the coefficient with the highest rank "wins".

  9183. 

  9184.  If one of the operands is 'wide_int' or 'poly_wide_int', the rules are

  9185. the same as for 'wide_int' arithmetic.

  9186. 

  9187. 

  9188. File: gccint.info,  Node: wi arithmetic on poly_ints,  Next: Division of poly_ints,  Prev: Using poly_int with C++ arithmetic operators,  Up: Arithmetic on poly_ints

  9189. 

  9190. 10.4.2 'wi' arithmetic on 'poly_int's

  9191. -------------------------------------

  9192. 

  9193. As well as the C++ operators, 'poly_int' supports the following 'wi'

  9194. routines:

  9195. 

  9196.      wi::neg (P1, &OVERFLOW)

  9197. 

  9198.      wi::add (P1, P2)

  9199.      wi::add (P1, C2)

  9200.      wi::add (C1, P1)

  9201.      wi::add (P1, P2, SIGN, &OVERFLOW)

  9202. 

  9203.      wi::sub (P1, P2)

  9204.      wi::sub (P1, C2)

  9205.      wi::sub (C1, P1)

  9206.      wi::sub (P1, P2, SIGN, &OVERFLOW)

  9207. 

  9208.      wi::mul (P1, C2)

  9209.      wi::mul (C1, P1)

  9210.      wi::mul (P1, C2, SIGN, &OVERFLOW)

  9211. 

  9212.      wi::lshift (P1, C2)

  9213. 

  9214.  These routines just check whether overflow occurs on any individual

  9215. coefficient; it is not possible to know at compile time whether the

  9216. final runtime value would overflow.

  9217. 

  9218. 

  9219. File: gccint.info,  Node: Division of poly_ints,  Next: Other poly_int arithmetic,  Prev: wi arithmetic on poly_ints,  Up: Arithmetic on poly_ints

  9220. 

  9221. 10.4.3 Division of 'poly_int's

  9222. ------------------------------

  9223. 

  9224. Division of 'poly_int's is possible for certain inputs.  The functions

  9225. for division return true if the operation is possible and in most cases

  9226. return the results by pointer.  The routines are:

  9227. 

  9228. 'multiple_p (A, B)'

  9229. 'multiple_p (A, B, &QUOTIENT)'

  9230.      Return true if A is an exact multiple of B, storing the result in

  9231.      QUOTIENT if so.  There are overloads for various combinations of

  9232.      polynomial and constant A, B and QUOTIENT.

  9233. 

  9234. 'constant_multiple_p (A, B)'

  9235. 'constant_multiple_p (A, B, &QUOTIENT)'

  9236.      Like 'multiple_p', but also test whether the multiple is a

  9237.      compile-time constant.

  9238. 

  9239. 'can_div_trunc_p (A, B, &QUOTIENT)'

  9240. 'can_div_trunc_p (A, B, &QUOTIENT, &REMAINDER)'

  9241.      Return true if we can calculate 'trunc (A / B)' at compile time,

  9242.      storing the result in QUOTIENT and REMAINDER if so.

  9243. 

  9244. 'can_div_away_from_zero_p (A, B, &QUOTIENT)'

  9245.      Return true if we can calculate 'A / B' at compile time, rounding

  9246.      away from zero.  Store the result in QUOTIENT if so.

  9247. 

  9248.      Note that this is true if and only if 'can_div_trunc_p' is true.

  9249.      The only difference is in the rounding of the result.

  9250. 

  9251.  There is also an asserting form of division:

  9252. 

  9253. 'exact_div (A, B)'

  9254.      Assert that A is a multiple of B and return 'A / B'.  The result is

  9255.      a 'poly_int' if A is a 'poly_int'.

  9256. 

  9257. 

  9258. File: gccint.info,  Node: Other poly_int arithmetic,  Prev: Division of poly_ints,  Up: Arithmetic on poly_ints

  9259. 

  9260. 10.4.4 Other 'poly_int' arithmetic

  9261. ----------------------------------

  9262. 

  9263. There are tentative routines for other operations besides division:

  9264. 

  9265. 'can_ior_p (A, B, &RESULT)'

  9266.      Return true if we can calculate 'A | B' at compile time, storing

  9267.      the result in RESULT if so.

  9268. 

  9269.  Also, ANDs with a value '(1 << Y) - 1' or its inverse can be treated as

  9270. alignment operations.  *Note Alignment of poly_ints::.

  9271. 

  9272.  In addition, the following miscellaneous routines are available:

  9273. 

  9274. 'coeff_gcd (A)'

  9275.      Return the greatest common divisor of all nonzero coefficients in

  9276.      A, or zero if A is known to be zero.

  9277. 

  9278. 'common_multiple (A, B)'

  9279.      Return a value that is a multiple of both A and B, where one value

  9280.      is a 'poly_int' and the other is a scalar.  The result will be the

  9281.      least common multiple for some indeterminate values but not

  9282.      necessarily for all.

  9283. 

  9284. 'force_common_multiple (A, B)'

  9285.      Return a value that is a multiple of both 'poly_int' A and

  9286.      'poly_int' B, asserting that such a value exists.  The result will

  9287.      be the least common multiple for some indeterminate values but not

  9288.      necessarily for all.

  9289. 

  9290.      When using this routine, please add a comment explaining why the

  9291.      assertion is known to hold.

  9292. 

  9293.  Please add any other operations that you find to be useful.

  9294. 

  9295. 

  9296. File: gccint.info,  Node: Alignment of poly_ints,  Next: Computing bounds on poly_ints,  Prev: Arithmetic on poly_ints,  Up: poly_int

  9297. 

  9298. 10.5 Alignment of 'poly_int's

  9299. =============================

  9300. 

  9301. 'poly_int' provides various routines for aligning values and for

  9302. querying misalignments.  In each case the alignment must be a power of

  9303. 2.

  9304. 

  9305. 'can_align_p (VALUE, ALIGN)'

  9306.      Return true if we can align VALUE up or down to the nearest

  9307.      multiple of ALIGN at compile time.  The answer is the same for both

  9308.      directions.

  9309. 

  9310. 'can_align_down (VALUE, ALIGN, &ALIGNED)'

  9311.      Return true if 'can_align_p'; if so, set ALIGNED to the greatest

  9312.      aligned value that is less than or equal to VALUE.

  9313. 

  9314. 'can_align_up (VALUE, ALIGN, &ALIGNED)'

  9315.      Return true if 'can_align_p'; if so, set ALIGNED to the lowest

  9316.      aligned value that is greater than or equal to VALUE.

  9317. 

  9318. 'known_equal_after_align_down (A, B, ALIGN)'

  9319.      Return true if we can align A and B down to the nearest ALIGN

  9320.      boundary at compile time and if the two results are equal.

  9321. 

  9322. 'known_equal_after_align_up (A, B, ALIGN)'

  9323.      Return true if we can align A and B up to the nearest ALIGN

  9324.      boundary at compile time and if the two results are equal.

  9325. 

  9326. 'aligned_lower_bound (VALUE, ALIGN)'

  9327.      Return a result that is no greater than VALUE and that is aligned

  9328.      to ALIGN.  The result will the closest aligned value for some

  9329.      indeterminate values but not necessarily for all.

  9330. 

  9331.      For example, suppose we are allocating an object of SIZE bytes in a

  9332.      downward-growing stack whose current limit is given by LIMIT.  If

  9333.      the object requires ALIGN bytes of alignment, the new stack limit

  9334.      is given by:

  9335. 

  9336.           aligned_lower_bound (LIMIT - SIZE, ALIGN)

  9337. 

  9338. 'aligned_upper_bound (VALUE, ALIGN)'

  9339.      Likewise return a result that is no less than VALUE and that is

  9340.      aligned to ALIGN.  This is the routine that would be used for

  9341.      upward-growing stacks in the scenario just described.

  9342. 

  9343. 'known_misalignment (VALUE, ALIGN, &MISALIGN)'

  9344.      Return true if we can calculate the misalignment of VALUE with

  9345.      respect to ALIGN at compile time, storing the result in MISALIGN if

  9346.      so.

  9347. 

  9348. 'known_alignment (VALUE)'

  9349.      Return the minimum alignment that VALUE is known to have (in other

  9350.      words, the largest alignment that can be guaranteed whatever the

  9351.      values of the indeterminates turn out to be).  Return 0 if VALUE is

  9352.      known to be 0.

  9353. 

  9354. 'force_align_down (VALUE, ALIGN)'

  9355.      Assert that VALUE can be aligned down to ALIGN at compile time and

  9356.      return the result.  When using this routine, please add a comment

  9357.      explaining why the assertion is known to hold.

  9358. 

  9359. 'force_align_up (VALUE, ALIGN)'

  9360.      Likewise, but aligning up.

  9361. 

  9362. 'force_align_down_and_div (VALUE, ALIGN)'

  9363.      Divide the result of 'force_align_down' by ALIGN.  Again, please

  9364.      add a comment explaining why the assertion in 'force_align_down' is

  9365.      known to hold.

  9366. 

  9367. 'force_align_up_and_div (VALUE, ALIGN)'

  9368.      Likewise for 'force_align_up'.

  9369. 

  9370. 'force_get_misalignment (VALUE, ALIGN)'

  9371.      Assert that we can calculate the misalignment of VALUE with respect

  9372.      to ALIGN at compile time and return the misalignment.  When using

  9373.      this function, please add a comment explaining why the assertion is

  9374.      known to hold.

  9375. 

  9376. 

  9377. File: gccint.info,  Node: Computing bounds on poly_ints,  Next: Converting poly_ints,  Prev: Alignment of poly_ints,  Up: poly_int

  9378. 

  9379. 10.6 Computing bounds on 'poly_int's

  9380. ====================================

  9381. 

  9382. 'poly_int' also provides routines for calculating lower and upper

  9383. bounds:

  9384. 

  9385. 'constant_lower_bound (A)'

  9386.      Assert that A is nonnegative and return the smallest value it can

  9387.      have.

  9388. 

  9389. 'constant_lower_bound_with_limit (A, B)'

  9390.      Return the least value A can have, given that the context in which

  9391.      A appears guarantees that the answer is no less than B.  In other

  9392.      words, the caller is asserting that A is greater than or equal to B

  9393.      even if 'known_ge (A, B)' doesn't hold.

  9394. 

  9395. 'constant_upper_bound_with_limit (A, B)'

  9396.      Return the greatest value A can have, given that the context in

  9397.      which A appears guarantees that the answer is no greater than B.

  9398.      In other words, the caller is asserting that A is less than or

  9399.      equal to B even if 'known_le (A, B)' doesn't hold.

  9400. 

  9401. 'lower_bound (A, B)'

  9402.      Return a value that is always less than or equal to both A and B.

  9403.      It will be the greatest such value for some indeterminate values

  9404.      but necessarily for all.

  9405. 

  9406. 'upper_bound (A, B)'

  9407.      Return a value that is always greater than or equal to both A and

  9408.      B.  It will be the least such value for some indeterminate values

  9409.      but necessarily for all.

  9410. 

  9411. 

  9412. File: gccint.info,  Node: Converting poly_ints,  Next: Miscellaneous poly_int routines,  Prev: Computing bounds on poly_ints,  Up: poly_int

  9413. 

  9414. 10.7 Converting 'poly_int's

  9415. ===========================

  9416. 

  9417. A 'poly_int<N, T>' can be constructed from up to N individual T

  9418. coefficients, with the remaining coefficients being implicitly zero.  In

  9419. particular, this means that every 'poly_int<N, T>' can be constructed

  9420. from a single scalar T, or something compatible with T.

  9421. 

  9422.  Also, a 'poly_int<N, T>' can be constructed from a 'poly_int<N, U>' if

  9423. T can be constructed from U.

  9424. 

  9425.  The following functions provide other forms of conversion, or test

  9426. whether such a conversion would succeed.

  9427. 

  9428. 'VALUE.is_constant ()'

  9429.      Return true if 'poly_int' VALUE is a compile-time constant.

  9430. 

  9431. 'VALUE.is_constant (&C1)'

  9432.      Return true if 'poly_int' VALUE is a compile-time constant, storing

  9433.      it in C1 if so.  C1 must be able to hold all constant values of

  9434.      VALUE without loss of precision.

  9435. 

  9436. 'VALUE.to_constant ()'

  9437.      Assert that VALUE is a compile-time constant and return its value.

  9438.      When using this function, please add a comment explaining why the

  9439.      condition is known to hold (for example, because an earlier phase

  9440.      of analysis rejected non-constants).

  9441. 

  9442. 'VALUE.to_shwi (&P2)'

  9443.      Return true if 'poly_int<N, T>' VALUE can be represented without

  9444.      loss of precision as a 'poly_int<N, 'HOST_WIDE_INT'>', storing it

  9445.      in that form in P2 if so.

  9446. 

  9447. 'VALUE.to_uhwi (&P2)'

  9448.      Return true if 'poly_int<N, T>' VALUE can be represented without

  9449.      loss of precision as a 'poly_int<N, 'unsigned HOST_WIDE_INT'>',

  9450.      storing it in that form in P2 if so.

  9451. 

  9452. 'VALUE.force_shwi ()'

  9453.      Forcibly convert each coefficient of 'poly_int<N, T>' VALUE to

  9454.      'HOST_WIDE_INT', truncating any that are out of range.  Return the

  9455.      result as a 'poly_int<N, 'HOST_WIDE_INT'>'.

  9456. 

  9457. 'VALUE.force_uhwi ()'

  9458.      Forcibly convert each coefficient of 'poly_int<N, T>' VALUE to

  9459.      'unsigned HOST_WIDE_INT', truncating any that are out of range.

  9460.      Return the result as a 'poly_int<N, 'unsigned HOST_WIDE_INT'>'.

  9461. 

  9462. 'wi::shwi (VALUE, PRECISION)'

  9463.      Return a 'poly_int' with the same value as VALUE, but with the

  9464.      coefficients converted from 'HOST_WIDE_INT' to 'wide_int'.

  9465.      PRECISION specifies the precision of the 'wide_int' cofficients; if

  9466.      this is wider than a 'HOST_WIDE_INT', the coefficients of VALUE

  9467.      will be sign-extended to fit.

  9468. 

  9469. 'wi::uhwi (VALUE, PRECISION)'

  9470.      Like 'wi::shwi', except that VALUE has coefficients of type

  9471.      'unsigned HOST_WIDE_INT'.  If PRECISION is wider than a

  9472.      'HOST_WIDE_INT', the coefficients of VALUE will be zero-extended to

  9473.      fit.

  9474. 

  9475. 'wi::sext (VALUE, PRECISION)'

  9476.      Return a 'poly_int' of the same type as VALUE, sign-extending every

  9477.      coefficient from the low PRECISION bits.  This in effect applies

  9478.      'wi::sext' to each coefficient individually.

  9479. 

  9480. 'wi::zext (VALUE, PRECISION)'

  9481.      Like 'wi::sext', but for zero extension.

  9482. 

  9483. 'poly_wide_int::from (VALUE, PRECISION, SIGN)'

  9484.      Convert VALUE to a 'poly_wide_int' in which each coefficient has

  9485.      PRECISION bits.  Extend the coefficients according to SIGN if the

  9486.      coefficients have fewer bits.

  9487. 

  9488. 'poly_offset_int::from (VALUE, SIGN)'

  9489.      Convert VALUE to a 'poly_offset_int', extending its coefficients

  9490.      according to SIGN if they have fewer bits than 'offset_int'.

  9491. 

  9492. 'poly_widest_int::from (VALUE, SIGN)'

  9493.      Convert VALUE to a 'poly_widest_int', extending its coefficients

  9494.      according to SIGN if they have fewer bits than 'widest_int'.

  9495. 

  9496. 

  9497. File: gccint.info,  Node: Miscellaneous poly_int routines,  Next: Guidelines for using poly_int,  Prev: Converting poly_ints,  Up: poly_int

  9498. 

  9499. 10.8 Miscellaneous 'poly_int' routines

  9500. ======================================

  9501. 

  9502. 'print_dec (VALUE, FILE, SIGN)'

  9503. 'print_dec (VALUE, FILE)'

  9504.      Print VALUE to FILE as a decimal value, interpreting the

  9505.      coefficients according to SIGN.  The final argument is optional if

  9506.      VALUE has an inherent sign; for example, 'poly_int64' values print

  9507.      as signed by default and 'poly_uint64' values print as unsigned by

  9508.      default.

  9509. 

  9510.      This is a simply a 'poly_int' version of a wide-int routine.

  9511. 

  9512. 

  9513. File: gccint.info,  Node: Guidelines for using poly_int,  Prev: Miscellaneous poly_int routines,  Up: poly_int

  9514. 

  9515. 10.9 Guidelines for using 'poly_int'

  9516. ====================================

  9517. 

  9518. One of the main design goals of 'poly_int' was to make it easy to write

  9519. target-independent code that handles variable-sized registers even when

  9520. the current target has fixed-sized registers.  There are two aspects to

  9521. this:

  9522. 

  9523.    * The set of 'poly_int' operations should be complete enough that the

  9524.      question in most cases becomes "Can we do this operation on these

  9525.      particular 'poly_int' values?  If not, bail out" rather than "Are

  9526.      these 'poly_int' values constant?  If so, do the operation,

  9527.      otherwise bail out".

  9528. 

  9529.    * If target-independent code compiles and runs correctly on a target

  9530.      with one value of 'NUM_POLY_INT_COEFFS', and if the code does not

  9531.      use asserting functions like 'to_constant', it is reasonable to

  9532.      assume that the code also works on targets with other values of

  9533.      'NUM_POLY_INT_COEFFS'.  There is no need to check this during

  9534.      everyday development.

  9535. 

  9536.  So the general principle is: if target-independent code is dealing with

  9537. a 'poly_int' value, it is better to operate on it as a 'poly_int' if at

  9538. all possible, choosing conservatively-correct behavior if a particular

  9539. operation fails.  For example, the following code handles an index 'pos'

  9540. into a sequence of vectors that each have 'nunits' elements:

  9541. 

  9542.      /* Calculate which vector contains the result, and which lane of

  9543.         that vector we need.  */

  9544.      if (!can_div_trunc_p (pos, nunits, &vec_entry, &vec_index))

  9545.        {

  9546.          if (dump_enabled_p ())

  9547.            dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,

  9548.                             "Cannot determine which vector holds the"

  9549.                             " final result.\n");

  9550.          return false;

  9551.        }

  9552. 

  9553.  However, there are some contexts in which operating on a 'poly_int' is

  9554. not possible or does not make sense.  One example is when handling

  9555. static initializers, since no current target supports the concept of a

  9556. variable-length static initializer.  In these situations, a reasonable

  9557. fallback is:

  9558. 

  9559.      if (POLY_VALUE.is_constant (&CONST_VALUE))

  9560.        {

  9561.          ...

  9562.          /* Operate on CONST_VALUE.  */

  9563.          ...

  9564.        }

  9565.      else

  9566.        {

  9567.          ...

  9568.          /* Conservatively correct fallback.  */

  9569.          ...

  9570.        }

  9571. 

  9572.  'poly_int' also provides some asserting functions like 'to_constant'.

  9573. Please only use these functions if there is a good theoretical reason to

  9574. believe that the assertion cannot fire.  For example, if some work is

  9575. divided into an analysis phase and an implementation phase, the analysis

  9576. phase might reject inputs that are not 'is_constant', in which case the

  9577. implementation phase can reasonably use 'to_constant' on the remaining

  9578. inputs.  The assertions should not be used to discover whether a

  9579. condition ever occurs "in the field"; in other words, they should not be

  9580. used to restrict code to constants at first, with the intention of only

  9581. implementing a 'poly_int' version if a user hits the assertion.

  9582. 

  9583.  If a particular asserting function like 'to_constant' is needed more

  9584. than once for the same reason, it is probably worth adding a helper

  9585. function or macro for that situation, so that the justification only

  9586. needs to be given once.  For example:

  9587. 

  9588.      /* Return the size of an element in a vector of size SIZE, given that

  9589.         the vector has NELTS elements.  The return value is in the same units

  9590.         as SIZE (either bits or bytes).

  9591. 

  9592.         to_constant () is safe in this situation because vector elements are

  9593.         always constant-sized scalars.  */

  9594.      #define vector_element_size(SIZE, NELTS) \

  9595.        (exact_div (SIZE, NELTS).to_constant ())

  9596. 

  9597.  Target-specific code in 'config/CPU' only needs to handle non-constant

  9598. 'poly_int's if 'NUM_POLY_INT_COEFFS' is greater than one.  For other

  9599. targets, 'poly_int' degenerates to a compile-time constant and is often

  9600. interchangable with a normal scalar integer.  There are two main

  9601. exceptions:

  9602. 

  9603.    * Sometimes an explicit cast to an integer type might be needed, such

  9604.      as to resolve ambiguities in a '?:' expression, or when passing

  9605.      values through '...' to things like print functions.

  9606. 

  9607.    * Target macros are included in target-independent code and so do not

  9608.      have access to the implicit conversion to a scalar integer.  If

  9609.      this becomes a problem for a particular target macro, the possible

  9610.      solutions, in order of preference, are:

  9611. 

  9612.         * Convert the target macro to a target hook (for all targets).

  9613. 

  9614.         * Put the target's implementation of the target macro in its

  9615.           'CPU.c' file and call it from the target macro in the 'CPU.h'

  9616.           file.

  9617. 

  9618.         * Add 'to_constant ()' calls where necessary.  The previous

  9619.           option is preferable because it will help with any future

  9620.           conversion of the macro to a hook.

  9621. 

  9622. 

  9623. File: gccint.info,  Node: GENERIC,  Next: GIMPLE,  Prev: poly_int,  Up: Top

  9624. 

  9625. 11 GENERIC

  9626. **********

  9627. 

  9628. The purpose of GENERIC is simply to provide a language-independent way

  9629. of representing an entire function in trees.  To this end, it was

  9630. necessary to add a few new tree codes to the back end, but almost

  9631. everything was already there.  If you can express it with the codes in

  9632. 'gcc/tree.def', it's GENERIC.

  9633. 

  9634.  Early on, there was a great deal of debate about how to think about

  9635. statements in a tree IL.  In GENERIC, a statement is defined as any

  9636. expression whose value, if any, is ignored.  A statement will always

  9637. have 'TREE_SIDE_EFFECTS' set (or it will be discarded), but a

  9638. non-statement expression may also have side effects.  A 'CALL_EXPR', for

  9639. instance.

  9640. 

  9641.  It would be possible for some local optimizations to work on the

  9642. GENERIC form of a function; indeed, the adapted tree inliner works fine

  9643. on GENERIC, but the current compiler performs inlining after lowering to

  9644. GIMPLE (a restricted form described in the next section).  Indeed,

  9645. currently the frontends perform this lowering before handing off to

  9646. 'tree_rest_of_compilation', but this seems inelegant.

  9647. 

  9648. * Menu:

  9649. 

  9650. * Deficiencies::                Topics net yet covered in this document.

  9651. * Tree overview::               All about 'tree's.

  9652. * Types::                       Fundamental and aggregate types.

  9653. * Declarations::                Type declarations and variables.

  9654. * Attributes::                  Declaration and type attributes.

  9655. * Expressions: Expression trees.            Operating on data.

  9656. * Statements::                  Control flow and related trees.

  9657. * Functions::           	Function bodies, linkage, and other aspects.

  9658. * Language-dependent trees::    Topics and trees specific to language front ends.

  9659. * C and C++ Trees::     	Trees specific to C and C++.

  9660. 

  9661. 

  9662. File: gccint.info,  Node: Deficiencies,  Next: Tree overview,  Up: GENERIC

  9663. 

  9664. 11.1 Deficiencies

  9665. =================

  9666. 

  9667. There are many places in which this document is incomplet and incorrekt.

  9668. It is, as of yet, only _preliminary_ documentation.

  9669. 

  9670. 

  9671. File: gccint.info,  Node: Tree overview,  Next: Types,  Prev: Deficiencies,  Up: GENERIC

  9672. 

  9673. 11.2 Overview

  9674. =============

  9675. 

  9676. The central data structure used by the internal representation is the

  9677. 'tree'.  These nodes, while all of the C type 'tree', are of many

  9678. varieties.  A 'tree' is a pointer type, but the object to which it

  9679. points may be of a variety of types.  From this point forward, we will

  9680. refer to trees in ordinary type, rather than in 'this font', except when

  9681. talking about the actual C type 'tree'.

  9682. 

  9683.  You can tell what kind of node a particular tree is by using the

  9684. 'TREE_CODE' macro.  Many, many macros take trees as input and return

  9685. trees as output.  However, most macros require a certain kind of tree

  9686. node as input.  In other words, there is a type-system for trees, but it

  9687. is not reflected in the C type-system.

  9688. 

  9689.  For safety, it is useful to configure GCC with '--enable-checking'.

  9690. Although this results in a significant performance penalty (since all

  9691. tree types are checked at run-time), and is therefore inappropriate in a

  9692. release version, it is extremely helpful during the development process.

  9693. 

  9694.  Many macros behave as predicates.  Many, although not all, of these

  9695. predicates end in '_P'.  Do not rely on the result type of these macros

  9696. being of any particular type.  You may, however, rely on the fact that

  9697. the type can be compared to '0', so that statements like

  9698.      if (TEST_P (t) && !TEST_P (y))

  9699.        x = 1;

  9700. and

  9701.      int i = (TEST_P (t) != 0);

  9702. are legal.  Macros that return 'int' values now may be changed to return

  9703. 'tree' values, or other pointers in the future.  Even those that

  9704. continue to return 'int' may return multiple nonzero codes where

  9705. previously they returned only zero and one.  Therefore, you should not

  9706. write code like

  9707.      if (TEST_P (t) == 1)

  9708. as this code is not guaranteed to work correctly in the future.

  9709. 

  9710.  You should not take the address of values returned by the macros or

  9711. functions described here.  In particular, no guarantee is given that the

  9712. values are lvalues.

  9713. 

  9714.  In general, the names of macros are all in uppercase, while the names

  9715. of functions are entirely in lowercase.  There are rare exceptions to

  9716. this rule.  You should assume that any macro or function whose name is

  9717. made up entirely of uppercase letters may evaluate its arguments more

  9718. than once.  You may assume that a macro or function whose name is made

  9719. up entirely of lowercase letters will evaluate its arguments only once.

  9720. 

  9721.  The 'error_mark_node' is a special tree.  Its tree code is

  9722. 'ERROR_MARK', but since there is only ever one node with that code, the

  9723. usual practice is to compare the tree against 'error_mark_node'.  (This

  9724. test is just a test for pointer equality.)  If an error has occurred

  9725. during front-end processing the flag 'errorcount' will be set.  If the

  9726. front end has encountered code it cannot handle, it will issue a message

  9727. to the user and set 'sorrycount'.  When these flags are set, any macro

  9728. or function which normally returns a tree of a particular kind may

  9729. instead return the 'error_mark_node'.  Thus, if you intend to do any

  9730. processing of erroneous code, you must be prepared to deal with the

  9731. 'error_mark_node'.

  9732. 

  9733.  Occasionally, a particular tree slot (like an operand to an expression,

  9734. or a particular field in a declaration) will be referred to as "reserved

  9735. for the back end".  These slots are used to store RTL when the tree is

  9736. converted to RTL for use by the GCC back end.  However, if that process

  9737. is not taking place (e.g., if the front end is being hooked up to an

  9738. intelligent editor), then those slots may be used by the back end

  9739. presently in use.

  9740. 

  9741.  If you encounter situations that do not match this documentation, such

  9742. as tree nodes of types not mentioned here, or macros documented to

  9743. return entities of a particular kind that instead return entities of

  9744. some different kind, you have found a bug, either in the front end or in

  9745. the documentation.  Please report these bugs as you would any other bug.

  9746. 

  9747. * Menu:

  9748. 

  9749. * Macros and Functions::Macros and functions that can be used with all trees.

  9750. * Identifiers::         The names of things.

  9751. * Containers::          Lists and vectors.

  9752. 

  9753. 

  9754. File: gccint.info,  Node: Macros and Functions,  Next: Identifiers,  Up: Tree overview

  9755. 

  9756. 11.2.1 Trees

  9757. ------------

  9758. 

  9759. All GENERIC trees have two fields in common.  First, 'TREE_CHAIN' is a

  9760. pointer that can be used as a singly-linked list to other trees.  The

  9761. other is 'TREE_TYPE'.  Many trees store the type of an expression or

  9762. declaration in this field.

  9763. 

  9764.  These are some other functions for handling trees:

  9765. 

  9766. 'tree_size'

  9767.      Return the number of bytes a tree takes.

  9768. 

  9769. 'build0'

  9770. 'build1'

  9771. 'build2'

  9772. 'build3'

  9773. 'build4'

  9774. 'build5'

  9775. 'build6'

  9776. 

  9777.      These functions build a tree and supply values to put in each

  9778.      parameter.  The basic signature is 'code, type, [operands]'.

  9779.      'code' is the 'TREE_CODE', and 'type' is a tree representing the

  9780.      'TREE_TYPE'.  These are followed by the operands, each of which is

  9781.      also a tree.

  9782. 

  9783. 

  9784. File: gccint.info,  Node: Identifiers,  Next: Containers,  Prev: Macros and Functions,  Up: Tree overview

  9785. 

  9786. 11.2.2 Identifiers

  9787. ------------------

  9788. 

  9789. An 'IDENTIFIER_NODE' represents a slightly more general concept than the

  9790. standard C or C++ concept of identifier.  In particular, an

  9791. 'IDENTIFIER_NODE' may contain a '$', or other extraordinary characters.

  9792. 

  9793.  There are never two distinct 'IDENTIFIER_NODE's representing the same

  9794. identifier.  Therefore, you may use pointer equality to compare

  9795. 'IDENTIFIER_NODE's, rather than using a routine like 'strcmp'.  Use

  9796. 'get_identifier' to obtain the unique 'IDENTIFIER_NODE' for a supplied

  9797. string.

  9798. 

  9799.  You can use the following macros to access identifiers:

  9800. 'IDENTIFIER_POINTER'

  9801.      The string represented by the identifier, represented as a 'char*'.

  9802.      This string is always 'NUL'-terminated, and contains no embedded

  9803.      'NUL' characters.

  9804. 

  9805. 'IDENTIFIER_LENGTH'

  9806.      The length of the string returned by 'IDENTIFIER_POINTER', not

  9807.      including the trailing 'NUL'.  This value of 'IDENTIFIER_LENGTH

  9808.      (x)' is always the same as 'strlen (IDENTIFIER_POINTER (x))'.

  9809. 

  9810. 'IDENTIFIER_OPNAME_P'

  9811.      This predicate holds if the identifier represents the name of an

  9812.      overloaded operator.  In this case, you should not depend on the

  9813.      contents of either the 'IDENTIFIER_POINTER' or the

  9814.      'IDENTIFIER_LENGTH'.

  9815. 

  9816. 'IDENTIFIER_TYPENAME_P'

  9817.      This predicate holds if the identifier represents the name of a

  9818.      user-defined conversion operator.  In this case, the 'TREE_TYPE' of

  9819.      the 'IDENTIFIER_NODE' holds the type to which the conversion

  9820.      operator converts.

  9821. 

  9822. 

  9823. File: gccint.info,  Node: Containers,  Prev: Identifiers,  Up: Tree overview

  9824. 

  9825. 11.2.3 Containers

  9826. -----------------

  9827. 

  9828. Two common container data structures can be represented directly with

  9829. tree nodes.  A 'TREE_LIST' is a singly linked list containing two trees

  9830. per node.  These are the 'TREE_PURPOSE' and 'TREE_VALUE' of each node.

  9831. (Often, the 'TREE_PURPOSE' contains some kind of tag, or additional

  9832. information, while the 'TREE_VALUE' contains the majority of the

  9833. payload.  In other cases, the 'TREE_PURPOSE' is simply 'NULL_TREE',

  9834. while in still others both the 'TREE_PURPOSE' and 'TREE_VALUE' are of

  9835. equal stature.)  Given one 'TREE_LIST' node, the next node is found by

  9836. following the 'TREE_CHAIN'.  If the 'TREE_CHAIN' is 'NULL_TREE', then

  9837. you have reached the end of the list.

  9838. 

  9839.  A 'TREE_VEC' is a simple vector.  The 'TREE_VEC_LENGTH' is an integer

  9840. (not a tree) giving the number of nodes in the vector.  The nodes

  9841. themselves are accessed using the 'TREE_VEC_ELT' macro, which takes two

  9842. arguments.  The first is the 'TREE_VEC' in question; the second is an

  9843. integer indicating which element in the vector is desired.  The elements

  9844. are indexed from zero.

  9845. 

  9846. 

  9847. File: gccint.info,  Node: Types,  Next: Declarations,  Prev: Tree overview,  Up: GENERIC

  9848. 

  9849. 11.3 Types

  9850. ==========

  9851. 

  9852. All types have corresponding tree nodes.  However, you should not assume

  9853. that there is exactly one tree node corresponding to each type.  There

  9854. are often multiple nodes corresponding to the same type.

  9855. 

  9856.  For the most part, different kinds of types have different tree codes.

  9857. (For example, pointer types use a 'POINTER_TYPE' code while arrays use

  9858. an 'ARRAY_TYPE' code.)  However, pointers to member functions use the

  9859. 'RECORD_TYPE' code.  Therefore, when writing a 'switch' statement that

  9860. depends on the code associated with a particular type, you should take

  9861. care to handle pointers to member functions under the 'RECORD_TYPE' case

  9862. label.

  9863. 

  9864.  The following functions and macros deal with cv-qualification of types:

  9865. 'TYPE_MAIN_VARIANT'

  9866.      This macro returns the unqualified version of a type.  It may be

  9867.      applied to an unqualified type, but it is not always the identity

  9868.      function in that case.

  9869. 

  9870.  A few other macros and functions are usable with all types:

  9871. 'TYPE_SIZE'

  9872.      The number of bits required to represent the type, represented as

  9873.      an 'INTEGER_CST'.  For an incomplete type, 'TYPE_SIZE' will be

  9874.      'NULL_TREE'.

  9875. 

  9876. 'TYPE_ALIGN'

  9877.      The alignment of the type, in bits, represented as an 'int'.

  9878. 

  9879. 'TYPE_NAME'

  9880.      This macro returns a declaration (in the form of a 'TYPE_DECL') for

  9881.      the type.  (Note this macro does _not_ return an 'IDENTIFIER_NODE',

  9882.      as you might expect, given its name!)  You can look at the

  9883.      'DECL_NAME' of the 'TYPE_DECL' to obtain the actual name of the

  9884.      type.  The 'TYPE_NAME' will be 'NULL_TREE' for a type that is not a

  9885.      built-in type, the result of a typedef, or a named class type.

  9886. 

  9887. 'TYPE_CANONICAL'

  9888.      This macro returns the "canonical" type for the given type node.

  9889.      Canonical types are used to improve performance in the C++ and

  9890.      Objective-C++ front ends by allowing efficient comparison between

  9891.      two type nodes in 'same_type_p': if the 'TYPE_CANONICAL' values of

  9892.      the types are equal, the types are equivalent; otherwise, the types

  9893.      are not equivalent.  The notion of equivalence for canonical types

  9894.      is the same as the notion of type equivalence in the language

  9895.      itself.  For instance,

  9896. 

  9897.      When 'TYPE_CANONICAL' is 'NULL_TREE', there is no canonical type

  9898.      for the given type node.  In this case, comparison between this

  9899.      type and any other type requires the compiler to perform a deep,

  9900.      "structural" comparison to see if the two type nodes have the same

  9901.      form and properties.

  9902. 

  9903.      The canonical type for a node is always the most fundamental type

  9904.      in the equivalence class of types.  For instance, 'int' is its own

  9905.      canonical type.  A typedef 'I' of 'int' will have 'int' as its

  9906.      canonical type.  Similarly, 'I*' and a typedef 'IP' (defined to

  9907.      'I*') will has 'int*' as their canonical type.  When building a new

  9908.      type node, be sure to set 'TYPE_CANONICAL' to the appropriate

  9909.      canonical type.  If the new type is a compound type (built from

  9910.      other types), and any of those other types require structural

  9911.      equality, use 'SET_TYPE_STRUCTURAL_EQUALITY' to ensure that the new

  9912.      type also requires structural equality.  Finally, if for some

  9913.      reason you cannot guarantee that 'TYPE_CANONICAL' will point to the

  9914.      canonical type, use 'SET_TYPE_STRUCTURAL_EQUALITY' to make sure

  9915.      that the new type-and any type constructed based on it-requires

  9916.      structural equality.  If you suspect that the canonical type system

  9917.      is miscomparing types, pass '--param verify-canonical-types=1' to

  9918.      the compiler or configure with '--enable-checking' to force the

  9919.      compiler to verify its canonical-type comparisons against the

  9920.      structural comparisons; the compiler will then print any warnings

  9921.      if the canonical types miscompare.

  9922. 

  9923. 'TYPE_STRUCTURAL_EQUALITY_P'

  9924.      This predicate holds when the node requires structural equality

  9925.      checks, e.g., when 'TYPE_CANONICAL' is 'NULL_TREE'.

  9926. 

  9927. 'SET_TYPE_STRUCTURAL_EQUALITY'

  9928.      This macro states that the type node it is given requires

  9929.      structural equality checks, e.g., it sets 'TYPE_CANONICAL' to

  9930.      'NULL_TREE'.

  9931. 

  9932. 'same_type_p'

  9933.      This predicate takes two types as input, and holds if they are the

  9934.      same type.  For example, if one type is a 'typedef' for the other,

  9935.      or both are 'typedef's for the same type.  This predicate also

  9936.      holds if the two trees given as input are simply copies of one

  9937.      another; i.e., there is no difference between them at the source

  9938.      level, but, for whatever reason, a duplicate has been made in the

  9939.      representation.  You should never use '==' (pointer equality) to

  9940.      compare types; always use 'same_type_p' instead.

  9941. 

  9942.  Detailed below are the various kinds of types, and the macros that can

  9943. be used to access them.  Although other kinds of types are used

  9944. elsewhere in G++, the types described here are the only ones that you

  9945. will encounter while examining the intermediate representation.

  9946. 

  9947. 'VOID_TYPE'

  9948.      Used to represent the 'void' type.

  9949. 

  9950. 'INTEGER_TYPE'

  9951.      Used to represent the various integral types, including 'char',

  9952.      'short', 'int', 'long', and 'long long'.  This code is not used for

  9953.      enumeration types, nor for the 'bool' type.  The 'TYPE_PRECISION'

  9954.      is the number of bits used in the representation, represented as an

  9955.      'unsigned int'.  (Note that in the general case this is not the

  9956.      same value as 'TYPE_SIZE'; suppose that there were a 24-bit integer

  9957.      type, but that alignment requirements for the ABI required 32-bit

  9958.      alignment.  Then, 'TYPE_SIZE' would be an 'INTEGER_CST' for 32,

  9959.      while 'TYPE_PRECISION' would be 24.)  The integer type is unsigned

  9960.      if 'TYPE_UNSIGNED' holds; otherwise, it is signed.

  9961. 

  9962.      The 'TYPE_MIN_VALUE' is an 'INTEGER_CST' for the smallest integer

  9963.      that may be represented by this type.  Similarly, the

  9964.      'TYPE_MAX_VALUE' is an 'INTEGER_CST' for the largest integer that

  9965.      may be represented by this type.

  9966. 

  9967. 'REAL_TYPE'

  9968.      Used to represent the 'float', 'double', and 'long double' types.

  9969.      The number of bits in the floating-point representation is given by

  9970.      'TYPE_PRECISION', as in the 'INTEGER_TYPE' case.

  9971. 

  9972. 'FIXED_POINT_TYPE'

  9973.      Used to represent the 'short _Fract', '_Fract', 'long _Fract',

  9974.      'long long _Fract', 'short _Accum', '_Accum', 'long _Accum', and

  9975.      'long long _Accum' types.  The number of bits in the fixed-point

  9976.      representation is given by 'TYPE_PRECISION', as in the

  9977.      'INTEGER_TYPE' case.  There may be padding bits, fractional bits

  9978.      and integral bits.  The number of fractional bits is given by

  9979.      'TYPE_FBIT', and the number of integral bits is given by

  9980.      'TYPE_IBIT'.  The fixed-point type is unsigned if 'TYPE_UNSIGNED'

  9981.      holds; otherwise, it is signed.  The fixed-point type is saturating

  9982.      if 'TYPE_SATURATING' holds; otherwise, it is not saturating.

  9983. 

  9984. 'COMPLEX_TYPE'

  9985.      Used to represent GCC built-in '__complex__' data types.  The

  9986.      'TREE_TYPE' is the type of the real and imaginary parts.

  9987. 

  9988. 'ENUMERAL_TYPE'

  9989.      Used to represent an enumeration type.  The 'TYPE_PRECISION' gives

  9990.      (as an 'int'), the number of bits used to represent the type.  If

  9991.      there are no negative enumeration constants, 'TYPE_UNSIGNED' will

  9992.      hold.  The minimum and maximum enumeration constants may be

  9993.      obtained with 'TYPE_MIN_VALUE' and 'TYPE_MAX_VALUE', respectively;

  9994.      each of these macros returns an 'INTEGER_CST'.

  9995. 

  9996.      The actual enumeration constants themselves may be obtained by

  9997.      looking at the 'TYPE_VALUES'.  This macro will return a

  9998.      'TREE_LIST', containing the constants.  The 'TREE_PURPOSE' of each

  9999.      node will be an 'IDENTIFIER_NODE' giving the name of the constant;

  10000.      the 'TREE_VALUE' will be an 'INTEGER_CST' giving the value assigned

  10001.      to that constant.  These constants will appear in the order in

  10002.      which they were declared.  The 'TREE_TYPE' of each of these

  10003.      constants will be the type of enumeration type itself.

  10004. 

  10005. 'BOOLEAN_TYPE'

  10006.      Used to represent the 'bool' type.

  10007. 

  10008. 'POINTER_TYPE'

  10009.      Used to represent pointer types, and pointer to data member types.

  10010.      The 'TREE_TYPE' gives the type to which this type points.

  10011. 

  10012. 'REFERENCE_TYPE'

  10013.      Used to represent reference types.  The 'TREE_TYPE' gives the type

  10014.      to which this type refers.

  10015. 

  10016. 'FUNCTION_TYPE'

  10017.      Used to represent the type of non-member functions and of static

  10018.      member functions.  The 'TREE_TYPE' gives the return type of the

  10019.      function.  The 'TYPE_ARG_TYPES' are a 'TREE_LIST' of the argument

  10020.      types.  The 'TREE_VALUE' of each node in this list is the type of

  10021.      the corresponding argument; the 'TREE_PURPOSE' is an expression for

  10022.      the default argument value, if any.  If the last node in the list

  10023.      is 'void_list_node' (a 'TREE_LIST' node whose 'TREE_VALUE' is the

  10024.      'void_type_node'), then functions of this type do not take variable

  10025.      arguments.  Otherwise, they do take a variable number of arguments.

  10026. 

  10027.      Note that in C (but not in C++) a function declared like 'void f()'

  10028.      is an unprototyped function taking a variable number of arguments;

  10029.      the 'TYPE_ARG_TYPES' of such a function will be 'NULL'.

  10030. 

  10031. 'METHOD_TYPE'

  10032.      Used to represent the type of a non-static member function.  Like a

  10033.      'FUNCTION_TYPE', the return type is given by the 'TREE_TYPE'.  The

  10034.      type of '*this', i.e., the class of which functions of this type

  10035.      are a member, is given by the 'TYPE_METHOD_BASETYPE'.  The

  10036.      'TYPE_ARG_TYPES' is the parameter list, as for a 'FUNCTION_TYPE',

  10037.      and includes the 'this' argument.

  10038. 

  10039. 'ARRAY_TYPE'

  10040.      Used to represent array types.  The 'TREE_TYPE' gives the type of

  10041.      the elements in the array.  If the array-bound is present in the

  10042.      type, the 'TYPE_DOMAIN' is an 'INTEGER_TYPE' whose 'TYPE_MIN_VALUE'

  10043.      and 'TYPE_MAX_VALUE' will be the lower and upper bounds of the

  10044.      array, respectively.  The 'TYPE_MIN_VALUE' will always be an

  10045.      'INTEGER_CST' for zero, while the 'TYPE_MAX_VALUE' will be one less

  10046.      than the number of elements in the array, i.e., the highest value

  10047.      which may be used to index an element in the array.

  10048. 

  10049. 'RECORD_TYPE'

  10050.      Used to represent 'struct' and 'class' types, as well as pointers

  10051.      to member functions and similar constructs in other languages.

  10052.      'TYPE_FIELDS' contains the items contained in this type, each of

  10053.      which can be a 'FIELD_DECL', 'VAR_DECL', 'CONST_DECL', or

  10054.      'TYPE_DECL'.  You may not make any assumptions about the ordering

  10055.      of the fields in the type or whether one or more of them overlap.

  10056. 

  10057. 'UNION_TYPE'

  10058.      Used to represent 'union' types.  Similar to 'RECORD_TYPE' except

  10059.      that all 'FIELD_DECL' nodes in 'TYPE_FIELD' start at bit position

  10060.      zero.

  10061. 

  10062. 'QUAL_UNION_TYPE'

  10063.      Used to represent part of a variant record in Ada.  Similar to

  10064.      'UNION_TYPE' except that each 'FIELD_DECL' has a 'DECL_QUALIFIER'

  10065.      field, which contains a boolean expression that indicates whether

  10066.      the field is present in the object.  The type will only have one

  10067.      field, so each field's 'DECL_QUALIFIER' is only evaluated if none

  10068.      of the expressions in the previous fields in 'TYPE_FIELDS' are

  10069.      nonzero.  Normally these expressions will reference a field in the

  10070.      outer object using a 'PLACEHOLDER_EXPR'.

  10071. 

  10072. 'LANG_TYPE'

  10073.      This node is used to represent a language-specific type.  The front

  10074.      end must handle it.

  10075. 

  10076. 'OFFSET_TYPE'

  10077.      This node is used to represent a pointer-to-data member.  For a

  10078.      data member 'X::m' the 'TYPE_OFFSET_BASETYPE' is 'X' and the

  10079.      'TREE_TYPE' is the type of 'm'.

  10080. 

  10081.  There are variables whose values represent some of the basic types.

  10082. These include:

  10083. 'void_type_node'

  10084.      A node for 'void'.

  10085. 

  10086. 'integer_type_node'

  10087.      A node for 'int'.

  10088. 

  10089. 'unsigned_type_node.'

  10090.      A node for 'unsigned int'.

  10091. 

  10092. 'char_type_node.'

  10093.      A node for 'char'.

  10094. It may sometimes be useful to compare one of these variables with a type

  10095. in hand, using 'same_type_p'.

  10096. 

  10097. 

  10098. File: gccint.info,  Node: Declarations,  Next: Attributes,  Prev: Types,  Up: GENERIC

  10099. 

  10100. 11.4 Declarations

  10101. =================

  10102. 

  10103. This section covers the various kinds of declarations that appear in the

  10104. internal representation, except for declarations of functions

  10105. (represented by 'FUNCTION_DECL' nodes), which are described in *note

  10106. Functions::.

  10107. 

  10108. * Menu:

  10109. 

  10110. * Working with declarations::  Macros and functions that work on

  10111. declarations.

  10112. * Internal structure:: How declaration nodes are represented.

  10113. 

  10114. 

  10115. File: gccint.info,  Node: Working with declarations,  Next: Internal structure,  Up: Declarations

  10116. 

  10117. 11.4.1 Working with declarations

  10118. --------------------------------

  10119. 

  10120. Some macros can be used with any kind of declaration.  These include:

  10121. 'DECL_NAME'

  10122.      This macro returns an 'IDENTIFIER_NODE' giving the name of the

  10123.      entity.

  10124. 

  10125. 'TREE_TYPE'

  10126.      This macro returns the type of the entity declared.

  10127. 

  10128. 'EXPR_FILENAME'

  10129.      This macro returns the name of the file in which the entity was

  10130.      declared, as a 'char*'.  For an entity declared implicitly by the

  10131.      compiler (like '__builtin_memcpy'), this will be the string

  10132.      '"<internal>"'.

  10133. 

  10134. 'EXPR_LINENO'

  10135.      This macro returns the line number at which the entity was

  10136.      declared, as an 'int'.

  10137. 

  10138. 'DECL_ARTIFICIAL'

  10139.      This predicate holds if the declaration was implicitly generated by

  10140.      the compiler.  For example, this predicate will hold of an

  10141.      implicitly declared member function, or of the 'TYPE_DECL'

  10142.      implicitly generated for a class type.  Recall that in C++ code

  10143.      like:

  10144.           struct S {};

  10145.      is roughly equivalent to C code like:

  10146.           struct S {};

  10147.           typedef struct S S;

  10148.      The implicitly generated 'typedef' declaration is represented by a

  10149.      'TYPE_DECL' for which 'DECL_ARTIFICIAL' holds.

  10150. 

  10151.  The various kinds of declarations include:

  10152. 'LABEL_DECL'

  10153.      These nodes are used to represent labels in function bodies.  For

  10154.      more information, see *note Functions::.  These nodes only appear

  10155.      in block scopes.

  10156. 

  10157. 'CONST_DECL'

  10158.      These nodes are used to represent enumeration constants.  The value

  10159.      of the constant is given by 'DECL_INITIAL' which will be an

  10160.      'INTEGER_CST' with the same type as the 'TREE_TYPE' of the

  10161.      'CONST_DECL', i.e., an 'ENUMERAL_TYPE'.

  10162. 

  10163. 'RESULT_DECL'

  10164.      These nodes represent the value returned by a function.  When a

  10165.      value is assigned to a 'RESULT_DECL', that indicates that the value

  10166.      should be returned, via bitwise copy, by the function.  You can use

  10167.      'DECL_SIZE' and 'DECL_ALIGN' on a 'RESULT_DECL', just as with a

  10168.      'VAR_DECL'.

  10169. 

  10170. 'TYPE_DECL'

  10171.      These nodes represent 'typedef' declarations.  The 'TREE_TYPE' is

  10172.      the type declared to have the name given by 'DECL_NAME'.  In some

  10173.      cases, there is no associated name.

  10174. 

  10175. 'VAR_DECL'

  10176.      These nodes represent variables with namespace or block scope, as

  10177.      well as static data members.  The 'DECL_SIZE' and 'DECL_ALIGN' are

  10178.      analogous to 'TYPE_SIZE' and 'TYPE_ALIGN'.  For a declaration, you

  10179.      should always use the 'DECL_SIZE' and 'DECL_ALIGN' rather than the

  10180.      'TYPE_SIZE' and 'TYPE_ALIGN' given by the 'TREE_TYPE', since

  10181.      special attributes may have been applied to the variable to give it

  10182.      a particular size and alignment.  You may use the predicates

  10183.      'DECL_THIS_STATIC' or 'DECL_THIS_EXTERN' to test whether the

  10184.      storage class specifiers 'static' or 'extern' were used to declare

  10185.      a variable.

  10186. 

  10187.      If this variable is initialized (but does not require a

  10188.      constructor), the 'DECL_INITIAL' will be an expression for the

  10189.      initializer.  The initializer should be evaluated, and a bitwise

  10190.      copy into the variable performed.  If the 'DECL_INITIAL' is the

  10191.      'error_mark_node', there is an initializer, but it is given by an

  10192.      explicit statement later in the code; no bitwise copy is required.

  10193. 

  10194.      GCC provides an extension that allows either automatic variables,

  10195.      or global variables, to be placed in particular registers.  This

  10196.      extension is being used for a particular 'VAR_DECL' if

  10197.      'DECL_REGISTER' holds for the 'VAR_DECL', and if

  10198.      'DECL_ASSEMBLER_NAME' is not equal to 'DECL_NAME'.  In that case,

  10199.      'DECL_ASSEMBLER_NAME' is the name of the register into which the

  10200.      variable will be placed.

  10201. 

  10202. 'PARM_DECL'

  10203.      Used to represent a parameter to a function.  Treat these nodes

  10204.      similarly to 'VAR_DECL' nodes.  These nodes only appear in the

  10205.      'DECL_ARGUMENTS' for a 'FUNCTION_DECL'.

  10206. 

  10207.      The 'DECL_ARG_TYPE' for a 'PARM_DECL' is the type that will

  10208.      actually be used when a value is passed to this function.  It may

  10209.      be a wider type than the 'TREE_TYPE' of the parameter; for example,

  10210.      the ordinary type might be 'short' while the 'DECL_ARG_TYPE' is

  10211.      'int'.

  10212. 

  10213. 'DEBUG_EXPR_DECL'

  10214.      Used to represent an anonymous debug-information temporary created

  10215.      to hold an expression as it is optimized away, so that its value

  10216.      can be referenced in debug bind statements.

  10217. 

  10218. 'FIELD_DECL'

  10219.      These nodes represent non-static data members.  The 'DECL_SIZE' and

  10220.      'DECL_ALIGN' behave as for 'VAR_DECL' nodes.  The position of the

  10221.      field within the parent record is specified by a combination of

  10222.      three attributes.  'DECL_FIELD_OFFSET' is the position, counting in

  10223.      bytes, of the 'DECL_OFFSET_ALIGN'-bit sized word containing the bit

  10224.      of the field closest to the beginning of the structure.

  10225.      'DECL_FIELD_BIT_OFFSET' is the bit offset of the first bit of the

  10226.      field within this word; this may be nonzero even for fields that

  10227.      are not bit-fields, since 'DECL_OFFSET_ALIGN' may be greater than

  10228.      the natural alignment of the field's type.

  10229. 

  10230.      If 'DECL_C_BIT_FIELD' holds, this field is a bit-field.  In a

  10231.      bit-field, 'DECL_BIT_FIELD_TYPE' also contains the type that was

  10232.      originally specified for it, while DECL_TYPE may be a modified type

  10233.      with lesser precision, according to the size of the bit field.

  10234. 

  10235. 'NAMESPACE_DECL'

  10236.      Namespaces provide a name hierarchy for other declarations.  They

  10237.      appear in the 'DECL_CONTEXT' of other '_DECL' nodes.

  10238. 

  10239. 

  10240. File: gccint.info,  Node: Internal structure,  Prev: Working with declarations,  Up: Declarations

  10241. 

  10242. 11.4.2 Internal structure

  10243. -------------------------

  10244. 

  10245. 'DECL' nodes are represented internally as a hierarchy of structures.

  10246. 

  10247. * Menu:

  10248. 

  10249. * Current structure hierarchy::  The current DECL node structure

  10250. hierarchy.

  10251. * Adding new DECL node types:: How to add a new DECL node to a

  10252. frontend.

  10253. 

  10254. 

  10255. File: gccint.info,  Node: Current structure hierarchy,  Next: Adding new DECL node types,  Up: Internal structure

  10256. 

  10257. 11.4.2.1 Current structure hierarchy

  10258. ....................................

  10259. 

  10260. 'struct tree_decl_minimal'

  10261.      This is the minimal structure to inherit from in order for common

  10262.      'DECL' macros to work.  The fields it contains are a unique ID,

  10263.      source location, context, and name.

  10264. 

  10265. 'struct tree_decl_common'

  10266.      This structure inherits from 'struct tree_decl_minimal'.  It

  10267.      contains fields that most 'DECL' nodes need, such as a field to

  10268.      store alignment, machine mode, size, and attributes.

  10269. 

  10270. 'struct tree_field_decl'

  10271.      This structure inherits from 'struct tree_decl_common'.  It is used

  10272.      to represent 'FIELD_DECL'.

  10273. 

  10274. 'struct tree_label_decl'

  10275.      This structure inherits from 'struct tree_decl_common'.  It is used

  10276.      to represent 'LABEL_DECL'.

  10277. 

  10278. 'struct tree_translation_unit_decl'

  10279.      This structure inherits from 'struct tree_decl_common'.  It is used

  10280.      to represent 'TRANSLATION_UNIT_DECL'.

  10281. 

  10282. 'struct tree_decl_with_rtl'

  10283.      This structure inherits from 'struct tree_decl_common'.  It

  10284.      contains a field to store the low-level RTL associated with a

  10285.      'DECL' node.

  10286. 

  10287. 'struct tree_result_decl'

  10288.      This structure inherits from 'struct tree_decl_with_rtl'.  It is

  10289.      used to represent 'RESULT_DECL'.

  10290. 

  10291. 'struct tree_const_decl'

  10292.      This structure inherits from 'struct tree_decl_with_rtl'.  It is

  10293.      used to represent 'CONST_DECL'.

  10294. 

  10295. 'struct tree_parm_decl'

  10296.      This structure inherits from 'struct tree_decl_with_rtl'.  It is

  10297.      used to represent 'PARM_DECL'.

  10298. 

  10299. 'struct tree_decl_with_vis'

  10300.      This structure inherits from 'struct tree_decl_with_rtl'.  It

  10301.      contains fields necessary to store visibility information, as well

  10302.      as a section name and assembler name.

  10303. 

  10304. 'struct tree_var_decl'

  10305.      This structure inherits from 'struct tree_decl_with_vis'.  It is

  10306.      used to represent 'VAR_DECL'.

  10307. 

  10308. 'struct tree_function_decl'

  10309.      This structure inherits from 'struct tree_decl_with_vis'.  It is

  10310.      used to represent 'FUNCTION_DECL'.

  10311. 

  10312. 

  10313. File: gccint.info,  Node: Adding new DECL node types,  Prev: Current structure hierarchy,  Up: Internal structure

  10314. 

  10315. 11.4.2.2 Adding new DECL node types

  10316. ...................................

  10317. 

  10318. Adding a new 'DECL' tree consists of the following steps

  10319. 

  10320. Add a new tree code for the 'DECL' node

  10321.      For language specific 'DECL' nodes, there is a '.def' file in each

  10322.      frontend directory where the tree code should be added.  For 'DECL'

  10323.      nodes that are part of the middle-end, the code should be added to

  10324.      'tree.def'.

  10325. 

  10326. Create a new structure type for the 'DECL' node

  10327.      These structures should inherit from one of the existing structures

  10328.      in the language hierarchy by using that structure as the first

  10329.      member.

  10330. 

  10331.           struct tree_foo_decl

  10332.           {

  10333.              struct tree_decl_with_vis common;

  10334.           }

  10335. 

  10336.      Would create a structure name 'tree_foo_decl' that inherits from

  10337.      'struct tree_decl_with_vis'.

  10338. 

  10339.      For language specific 'DECL' nodes, this new structure type should

  10340.      go in the appropriate '.h' file.  For 'DECL' nodes that are part of

  10341.      the middle-end, the structure type should go in 'tree.h'.

  10342. 

  10343. Add a member to the tree structure enumerator for the node

  10344.      For garbage collection and dynamic checking purposes, each 'DECL'

  10345.      node structure type is required to have a unique enumerator value

  10346.      specified with it.  For language specific 'DECL' nodes, this new

  10347.      enumerator value should go in the appropriate '.def' file.  For

  10348.      'DECL' nodes that are part of the middle-end, the enumerator values

  10349.      are specified in 'treestruct.def'.

  10350. 

  10351. Update 'union tree_node'

  10352.      In order to make your new structure type usable, it must be added

  10353.      to 'union tree_node'.  For language specific 'DECL' nodes, a new

  10354.      entry should be added to the appropriate '.h' file of the form

  10355.             struct tree_foo_decl GTY ((tag ("TS_VAR_DECL"))) foo_decl;

  10356.      For 'DECL' nodes that are part of the middle-end, the additional

  10357.      member goes directly into 'union tree_node' in 'tree.h'.

  10358. 

  10359. Update dynamic checking info

  10360.      In order to be able to check whether accessing a named portion of

  10361.      'union tree_node' is legal, and whether a certain 'DECL' node

  10362.      contains one of the enumerated 'DECL' node structures in the

  10363.      hierarchy, a simple lookup table is used.  This lookup table needs

  10364.      to be kept up to date with the tree structure hierarchy, or else

  10365.      checking and containment macros will fail inappropriately.

  10366. 

  10367.      For language specific 'DECL' nodes, their is an 'init_ts' function

  10368.      in an appropriate '.c' file, which initializes the lookup table.

  10369.      Code setting up the table for new 'DECL' nodes should be added

  10370.      there.  For each 'DECL' tree code and enumerator value representing

  10371.      a member of the inheritance hierarchy, the table should contain 1

  10372.      if that tree code inherits (directly or indirectly) from that

  10373.      member.  Thus, a 'FOO_DECL' node derived from 'struct

  10374.      decl_with_rtl', and enumerator value 'TS_FOO_DECL', would be set up

  10375.      as follows

  10376.           tree_contains_struct[FOO_DECL][TS_FOO_DECL] = 1;

  10377.           tree_contains_struct[FOO_DECL][TS_DECL_WRTL] = 1;

  10378.           tree_contains_struct[FOO_DECL][TS_DECL_COMMON] = 1;

  10379.           tree_contains_struct[FOO_DECL][TS_DECL_MINIMAL] = 1;

  10380. 

  10381.      For 'DECL' nodes that are part of the middle-end, the setup code

  10382.      goes into 'tree.c'.

  10383. 

  10384. Add macros to access any new fields and flags

  10385. 

  10386.      Each added field or flag should have a macro that is used to access

  10387.      it, that performs appropriate checking to ensure only the right

  10388.      type of 'DECL' nodes access the field.

  10389. 

  10390.      These macros generally take the following form

  10391.           #define FOO_DECL_FIELDNAME(NODE) FOO_DECL_CHECK(NODE)->foo_decl.fieldname

  10392.      However, if the structure is simply a base class for further

  10393.      structures, something like the following should be used

  10394.           #define BASE_STRUCT_CHECK(T) CONTAINS_STRUCT_CHECK(T, TS_BASE_STRUCT)

  10395.           #define BASE_STRUCT_FIELDNAME(NODE) \

  10396.              (BASE_STRUCT_CHECK(NODE)->base_struct.fieldname

  10397. 

  10398.      Reading them from the generated 'all-tree.def' file (which in turn

  10399.      includes all the 'tree.def' files), 'gencheck.c' is used during

  10400.      GCC's build to generate the '*_CHECK' macros for all tree codes.

  10401. 

  10402. 

  10403. File: gccint.info,  Node: Attributes,  Next: Expression trees,  Prev: Declarations,  Up: GENERIC

  10404. 

  10405. 11.5 Attributes in trees

  10406. ========================

  10407. 

  10408. Attributes, as specified using the '__attribute__' keyword, are

  10409. represented internally as a 'TREE_LIST'.  The 'TREE_PURPOSE' is the name

  10410. of the attribute, as an 'IDENTIFIER_NODE'.  The 'TREE_VALUE' is a

  10411. 'TREE_LIST' of the arguments of the attribute, if any, or 'NULL_TREE' if

  10412. there are no arguments; the arguments are stored as the 'TREE_VALUE' of

  10413. successive entries in the list, and may be identifiers or expressions.

  10414. The 'TREE_CHAIN' of the attribute is the next attribute in a list of

  10415. attributes applying to the same declaration or type, or 'NULL_TREE' if

  10416. there are no further attributes in the list.

  10417. 

  10418.  Attributes may be attached to declarations and to types; these

  10419. attributes may be accessed with the following macros.  All attributes

  10420. are stored in this way, and many also cause other changes to the

  10421. declaration or type or to other internal compiler data structures.

  10422. 

  10423.  -- Tree Macro: tree DECL_ATTRIBUTES (tree DECL)

  10424.      This macro returns the attributes on the declaration DECL.

  10425. 

  10426.  -- Tree Macro: tree TYPE_ATTRIBUTES (tree TYPE)

  10427.      This macro returns the attributes on the type TYPE.

  10428. 

  10429. 

  10430. File: gccint.info,  Node: Expression trees,  Next: Statements,  Prev: Attributes,  Up: GENERIC

  10431. 

  10432. 11.6 Expressions

  10433. ================

  10434. 

  10435. The internal representation for expressions is for the most part quite

  10436. straightforward.  However, there are a few facts that one must bear in

  10437. mind.  In particular, the expression "tree" is actually a directed

  10438. acyclic graph.  (For example there may be many references to the integer

  10439. constant zero throughout the source program; many of these will be

  10440. represented by the same expression node.)  You should not rely on

  10441. certain kinds of node being shared, nor should you rely on certain kinds

  10442. of nodes being unshared.

  10443. 

  10444.  The following macros can be used with all expression nodes:

  10445. 

  10446. 'TREE_TYPE'

  10447.      Returns the type of the expression.  This value may not be

  10448.      precisely the same type that would be given the expression in the

  10449.      original program.

  10450. 

  10451.  In what follows, some nodes that one might expect to always have type

  10452. 'bool' are documented to have either integral or boolean type.  At some

  10453. point in the future, the C front end may also make use of this same

  10454. intermediate representation, and at this point these nodes will

  10455. certainly have integral type.  The previous sentence is not meant to

  10456. imply that the C++ front end does not or will not give these nodes

  10457. integral type.

  10458. 

  10459.  Below, we list the various kinds of expression nodes.  Except where

  10460. noted otherwise, the operands to an expression are accessed using the

  10461. 'TREE_OPERAND' macro.  For example, to access the first operand to a

  10462. binary plus expression 'expr', use:

  10463. 

  10464.      TREE_OPERAND (expr, 0)

  10465. 

  10466.  As this example indicates, the operands are zero-indexed.

  10467. 

  10468. * Menu:

  10469. 

  10470. * Constants: Constant expressions.

  10471. * Storage References::

  10472. * Unary and Binary Expressions::

  10473. * Vectors::

  10474. 

  10475. 

  10476. File: gccint.info,  Node: Constant expressions,  Next: Storage References,  Up: Expression trees

  10477. 

  10478. 11.6.1 Constant expressions

  10479. ---------------------------

  10480. 

  10481. The table below begins with constants, moves on to unary expressions,

  10482. then proceeds to binary expressions, and concludes with various other

  10483. kinds of expressions:

  10484. 

  10485. 'INTEGER_CST'

  10486.      These nodes represent integer constants.  Note that the type of

  10487.      these constants is obtained with 'TREE_TYPE'; they are not always

  10488.      of type 'int'.  In particular, 'char' constants are represented

  10489.      with 'INTEGER_CST' nodes.  The value of the integer constant 'e' is

  10490.      represented in an array of HOST_WIDE_INT. There are enough elements

  10491.      in the array to represent the value without taking extra elements

  10492.      for redundant 0s or -1.  The number of elements used to represent

  10493.      'e' is available via 'TREE_INT_CST_NUNITS'.  Element 'i' can be

  10494.      extracted by using 'TREE_INT_CST_ELT (e, i)'.  'TREE_INT_CST_LOW'

  10495.      is a shorthand for 'TREE_INT_CST_ELT (e, 0)'.

  10496. 

  10497.      The functions 'tree_fits_shwi_p' and 'tree_fits_uhwi_p' can be used

  10498.      to tell if the value is small enough to fit in a signed

  10499.      HOST_WIDE_INT or an unsigned HOST_WIDE_INT respectively.  The value

  10500.      can then be extracted using 'tree_to_shwi' and 'tree_to_uhwi'.

  10501. 

  10502. 'REAL_CST'

  10503. 

  10504.      FIXME: Talk about how to obtain representations of this constant,

  10505.      do comparisons, and so forth.

  10506. 

  10507. 'FIXED_CST'

  10508. 

  10509.      These nodes represent fixed-point constants.  The type of these

  10510.      constants is obtained with 'TREE_TYPE'.  'TREE_FIXED_CST_PTR'

  10511.      points to a 'struct fixed_value'; 'TREE_FIXED_CST' returns the

  10512.      structure itself.  'struct fixed_value' contains 'data' with the

  10513.      size of two 'HOST_BITS_PER_WIDE_INT' and 'mode' as the associated

  10514.      fixed-point machine mode for 'data'.

  10515. 

  10516. 'COMPLEX_CST'

  10517.      These nodes are used to represent complex number constants, that is

  10518.      a '__complex__' whose parts are constant nodes.  The

  10519.      'TREE_REALPART' and 'TREE_IMAGPART' return the real and the

  10520.      imaginary parts respectively.

  10521. 

  10522. 'VECTOR_CST'

  10523.      These nodes are used to represent vector constants.  Each vector

  10524.      constant V is treated as a specific instance of an arbitrary-length

  10525.      sequence that itself contains 'VECTOR_CST_NPATTERNS (V)'

  10526.      interleaved patterns.  Each pattern has the form:

  10527. 

  10528.           { BASE0, BASE1, BASE1 + STEP, BASE1 + STEP * 2, ... }

  10529. 

  10530.      The first three elements in each pattern are enough to determine

  10531.      the values of the other elements.  However, if all STEPs are zero,

  10532.      only the first two elements are needed.  If in addition each BASE1

  10533.      is equal to the corresponding BASE0, only the first element in each

  10534.      pattern is needed.  The number of encoded elements per pattern is

  10535.      given by 'VECTOR_CST_NELTS_PER_PATTERN (V)'.

  10536. 

  10537.      For example, the constant:

  10538. 

  10539.           { 0, 1, 2, 6, 3, 8, 4, 10, 5, 12, 6, 14, 7, 16, 8, 18 }

  10540. 

  10541.      is interpreted as an interleaving of the sequences:

  10542. 

  10543.           { 0, 2, 3, 4, 5, 6, 7, 8 }

  10544.           { 1, 6, 8, 10, 12, 14, 16, 18 }

  10545. 

  10546.      where the sequences are represented by the following patterns:

  10547. 

  10548.           BASE0 == 0, BASE1 == 2, STEP == 1

  10549.           BASE0 == 1, BASE1 == 6, STEP == 2

  10550. 

  10551.      In this case:

  10552. 

  10553.           VECTOR_CST_NPATTERNS (V) == 2

  10554.           VECTOR_CST_NELTS_PER_PATTERN (V) == 3

  10555. 

  10556.      The vector is therefore encoded using the first 6 elements ('{ 0,

  10557.      1, 2, 6, 3, 8 }'), with the remaining 10 elements being implicit

  10558.      extensions of them.

  10559. 

  10560.      Sometimes this scheme can create two possible encodings of the same

  10561.      vector.  For example { 0, 1 } could be seen as two patterns with

  10562.      one element each or one pattern with two elements (BASE0 and

  10563.      BASE1).  The canonical encoding is always the one with the fewest

  10564.      patterns or (if both encodings have the same number of petterns)

  10565.      the one with the fewest encoded elements.

  10566. 

  10567.      'vector_cst_encoding_nelts (V)' gives the total number of encoded

  10568.      elements in V, which is 6 in the example above.

  10569.      'VECTOR_CST_ENCODED_ELTS (V)' gives a pointer to the elements

  10570.      encoded in V and 'VECTOR_CST_ENCODED_ELT (V, I)' accesses the value

  10571.      of encoded element I.

  10572. 

  10573.      'VECTOR_CST_DUPLICATE_P (V)' is true if V simply contains repeated

  10574.      instances of 'VECTOR_CST_NPATTERNS (V)' values.  This is a

  10575.      shorthand for testing 'VECTOR_CST_NELTS_PER_PATTERN (V) == 1'.

  10576. 

  10577.      'VECTOR_CST_STEPPED_P (V)' is true if at least one pattern in V has

  10578.      a nonzero step.  This is a shorthand for testing

  10579.      'VECTOR_CST_NELTS_PER_PATTERN (V) == 3'.

  10580. 

  10581.      The utility function 'vector_cst_elt' gives the value of an

  10582.      arbitrary index as a 'tree'.  'vector_cst_int_elt' gives the same

  10583.      value as a 'wide_int'.

  10584. 

  10585. 'STRING_CST'

  10586.      These nodes represent string-constants.  The 'TREE_STRING_LENGTH'

  10587.      returns the length of the string, as an 'int'.  The

  10588.      'TREE_STRING_POINTER' is a 'char*' containing the string itself.

  10589.      The string may not be 'NUL'-terminated, and it may contain embedded

  10590.      'NUL' characters.  Therefore, the 'TREE_STRING_LENGTH' includes the

  10591.      trailing 'NUL' if it is present.

  10592. 

  10593.      For wide string constants, the 'TREE_STRING_LENGTH' is the number

  10594.      of bytes in the string, and the 'TREE_STRING_POINTER' points to an

  10595.      array of the bytes of the string, as represented on the target

  10596.      system (that is, as integers in the target endianness).  Wide and

  10597.      non-wide string constants are distinguished only by the 'TREE_TYPE'

  10598.      of the 'STRING_CST'.

  10599. 

  10600.      FIXME: The formats of string constants are not well-defined when

  10601.      the target system bytes are not the same width as host system

  10602.      bytes.

  10603. 

  10604. 'POLY_INT_CST'

  10605.      These nodes represent invariants that depend on some

  10606.      target-specific runtime parameters.  They consist of

  10607.      'NUM_POLY_INT_COEFFS' coefficients, with the first coefficient

  10608.      being the constant term and the others being multipliers that are

  10609.      applied to the runtime parameters.

  10610. 

  10611.      'POLY_INT_CST_ELT (X, I)' references coefficient number I of

  10612.      'POLY_INT_CST' node X.  Each coefficient is an 'INTEGER_CST'.

  10613. 

  10614. 

  10615. File: gccint.info,  Node: Storage References,  Next: Unary and Binary Expressions,  Prev: Constant expressions,  Up: Expression trees

  10616. 

  10617. 11.6.2 References to storage

  10618. ----------------------------

  10619. 

  10620. 'ARRAY_REF'

  10621.      These nodes represent array accesses.  The first operand is the

  10622.      array; the second is the index.  To calculate the address of the

  10623.      memory accessed, you must scale the index by the size of the type

  10624.      of the array elements.  The type of these expressions must be the

  10625.      type of a component of the array.  The third and fourth operands

  10626.      are used after gimplification to represent the lower bound and

  10627.      component size but should not be used directly; call

  10628.      'array_ref_low_bound' and 'array_ref_element_size' instead.

  10629. 

  10630. 'ARRAY_RANGE_REF'

  10631.      These nodes represent access to a range (or "slice") of an array.

  10632.      The operands are the same as that for 'ARRAY_REF' and have the same

  10633.      meanings.  The type of these expressions must be an array whose

  10634.      component type is the same as that of the first operand.  The range

  10635.      of that array type determines the amount of data these expressions

  10636.      access.

  10637. 

  10638. 'TARGET_MEM_REF'

  10639.      These nodes represent memory accesses whose address directly map to

  10640.      an addressing mode of the target architecture.  The first argument

  10641.      is 'TMR_SYMBOL' and must be a 'VAR_DECL' of an object with a fixed

  10642.      address.  The second argument is 'TMR_BASE' and the third one is

  10643.      'TMR_INDEX'.  The fourth argument is 'TMR_STEP' and must be an

  10644.      'INTEGER_CST'.  The fifth argument is 'TMR_OFFSET' and must be an

  10645.      'INTEGER_CST'.  Any of the arguments may be NULL if the appropriate

  10646.      component does not appear in the address.  Address of the

  10647.      'TARGET_MEM_REF' is determined in the following way.

  10648. 

  10649.           &TMR_SYMBOL + TMR_BASE + TMR_INDEX * TMR_STEP + TMR_OFFSET

  10650. 

  10651.      The sixth argument is the reference to the original memory access,

  10652.      which is preserved for the purposes of the RTL alias analysis.  The

  10653.      seventh argument is a tag representing the results of tree level

  10654.      alias analysis.

  10655. 

  10656. 'ADDR_EXPR'

  10657.      These nodes are used to represent the address of an object.  (These

  10658.      expressions will always have pointer or reference type.)  The

  10659.      operand may be another expression, or it may be a declaration.

  10660. 

  10661.      As an extension, GCC allows users to take the address of a label.

  10662.      In this case, the operand of the 'ADDR_EXPR' will be a

  10663.      'LABEL_DECL'.  The type of such an expression is 'void*'.

  10664. 

  10665.      If the object addressed is not an lvalue, a temporary is created,

  10666.      and the address of the temporary is used.

  10667. 

  10668. 'INDIRECT_REF'

  10669.      These nodes are used to represent the object pointed to by a

  10670.      pointer.  The operand is the pointer being dereferenced; it will

  10671.      always have pointer or reference type.

  10672. 

  10673. 'MEM_REF'

  10674.      These nodes are used to represent the object pointed to by a

  10675.      pointer offset by a constant.  The first operand is the pointer

  10676.      being dereferenced; it will always have pointer or reference type.

  10677.      The second operand is a pointer constant.  Its type is specifying

  10678.      the type to be used for type-based alias analysis.

  10679. 

  10680. 'COMPONENT_REF'

  10681.      These nodes represent non-static data member accesses.  The first

  10682.      operand is the object (rather than a pointer to it); the second

  10683.      operand is the 'FIELD_DECL' for the data member.  The third operand

  10684.      represents the byte offset of the field, but should not be used

  10685.      directly; call 'component_ref_field_offset' instead.

  10686. 

  10687. 

  10688. File: gccint.info,  Node: Unary and Binary Expressions,  Next: Vectors,  Prev: Storage References,  Up: Expression trees

  10689. 

  10690. 11.6.3 Unary and Binary Expressions

  10691. -----------------------------------

  10692. 

  10693. 'NEGATE_EXPR'

  10694.      These nodes represent unary negation of the single operand, for

  10695.      both integer and floating-point types.  The type of negation can be

  10696.      determined by looking at the type of the expression.

  10697. 

  10698.      The behavior of this operation on signed arithmetic overflow is

  10699.      controlled by the 'flag_wrapv' and 'flag_trapv' variables.

  10700. 

  10701. 'ABS_EXPR'

  10702.      These nodes represent the absolute value of the single operand, for

  10703.      both integer and floating-point types.  This is typically used to

  10704.      implement the 'abs', 'labs' and 'llabs' builtins for integer types,

  10705.      and the 'fabs', 'fabsf' and 'fabsl' builtins for floating point

  10706.      types.  The type of abs operation can be determined by looking at

  10707.      the type of the expression.

  10708. 

  10709.      This node is not used for complex types.  To represent the modulus

  10710.      or complex abs of a complex value, use the 'BUILT_IN_CABS',

  10711.      'BUILT_IN_CABSF' or 'BUILT_IN_CABSL' builtins, as used to implement

  10712.      the C99 'cabs', 'cabsf' and 'cabsl' built-in functions.

  10713. 

  10714. 'ABSU_EXPR'

  10715.      These nodes represent the absolute value of the single operand in

  10716.      equivalent unsigned type such that 'ABSU_EXPR' of 'TYPE_MIN' is

  10717.      well defined.

  10718. 

  10719. 'BIT_NOT_EXPR'

  10720.      These nodes represent bitwise complement, and will always have

  10721.      integral type.  The only operand is the value to be complemented.

  10722. 

  10723. 'TRUTH_NOT_EXPR'

  10724.      These nodes represent logical negation, and will always have

  10725.      integral (or boolean) type.  The operand is the value being

  10726.      negated.  The type of the operand and that of the result are always

  10727.      of 'BOOLEAN_TYPE' or 'INTEGER_TYPE'.

  10728. 

  10729. 'PREDECREMENT_EXPR'

  10730. 'PREINCREMENT_EXPR'

  10731. 'POSTDECREMENT_EXPR'

  10732. 'POSTINCREMENT_EXPR'

  10733.      These nodes represent increment and decrement expressions.  The

  10734.      value of the single operand is computed, and the operand

  10735.      incremented or decremented.  In the case of 'PREDECREMENT_EXPR' and

  10736.      'PREINCREMENT_EXPR', the value of the expression is the value

  10737.      resulting after the increment or decrement; in the case of

  10738.      'POSTDECREMENT_EXPR' and 'POSTINCREMENT_EXPR' is the value before

  10739.      the increment or decrement occurs.  The type of the operand, like

  10740.      that of the result, will be either integral, boolean, or

  10741.      floating-point.

  10742. 

  10743. 'FIX_TRUNC_EXPR'

  10744.      These nodes represent conversion of a floating-point value to an

  10745.      integer.  The single operand will have a floating-point type, while

  10746.      the complete expression will have an integral (or boolean) type.

  10747.      The operand is rounded towards zero.

  10748. 

  10749. 'FLOAT_EXPR'

  10750.      These nodes represent conversion of an integral (or boolean) value

  10751.      to a floating-point value.  The single operand will have integral

  10752.      type, while the complete expression will have a floating-point

  10753.      type.

  10754. 

  10755.      FIXME: How is the operand supposed to be rounded?  Is this

  10756.      dependent on '-mieee'?

  10757. 

  10758. 'COMPLEX_EXPR'

  10759.      These nodes are used to represent complex numbers constructed from

  10760.      two expressions of the same (integer or real) type.  The first

  10761.      operand is the real part and the second operand is the imaginary

  10762.      part.

  10763. 

  10764. 'CONJ_EXPR'

  10765.      These nodes represent the conjugate of their operand.

  10766. 

  10767. 'REALPART_EXPR'

  10768. 'IMAGPART_EXPR'

  10769.      These nodes represent respectively the real and the imaginary parts

  10770.      of complex numbers (their sole argument).

  10771. 

  10772. 'NON_LVALUE_EXPR'

  10773.      These nodes indicate that their one and only operand is not an

  10774.      lvalue.  A back end can treat these identically to the single

  10775.      operand.

  10776. 

  10777. 'NOP_EXPR'

  10778.      These nodes are used to represent conversions that do not require

  10779.      any code-generation.  For example, conversion of a 'char*' to an

  10780.      'int*' does not require any code be generated; such a conversion is

  10781.      represented by a 'NOP_EXPR'.  The single operand is the expression

  10782.      to be converted.  The conversion from a pointer to a reference is

  10783.      also represented with a 'NOP_EXPR'.

  10784. 

  10785. 'CONVERT_EXPR'

  10786.      These nodes are similar to 'NOP_EXPR's, but are used in those

  10787.      situations where code may need to be generated.  For example, if an

  10788.      'int*' is converted to an 'int' code may need to be generated on

  10789.      some platforms.  These nodes are never used for C++-specific

  10790.      conversions, like conversions between pointers to different classes

  10791.      in an inheritance hierarchy.  Any adjustments that need to be made

  10792.      in such cases are always indicated explicitly.  Similarly, a

  10793.      user-defined conversion is never represented by a 'CONVERT_EXPR';

  10794.      instead, the function calls are made explicit.

  10795. 

  10796. 'FIXED_CONVERT_EXPR'

  10797.      These nodes are used to represent conversions that involve

  10798.      fixed-point values.  For example, from a fixed-point value to

  10799.      another fixed-point value, from an integer to a fixed-point value,

  10800.      from a fixed-point value to an integer, from a floating-point value

  10801.      to a fixed-point value, or from a fixed-point value to a

  10802.      floating-point value.

  10803. 

  10804. 'LSHIFT_EXPR'

  10805. 'RSHIFT_EXPR'

  10806.      These nodes represent left and right shifts, respectively.  The

  10807.      first operand is the value to shift; it will always be of integral

  10808.      type.  The second operand is an expression for the number of bits

  10809.      by which to shift.  Right shift should be treated as arithmetic,

  10810.      i.e., the high-order bits should be zero-filled when the expression

  10811.      has unsigned type and filled with the sign bit when the expression

  10812.      has signed type.  Note that the result is undefined if the second

  10813.      operand is larger than or equal to the first operand's type size.

  10814.      Unlike most nodes, these can have a vector as first operand and a

  10815.      scalar as second operand.

  10816. 

  10817. 'BIT_IOR_EXPR'

  10818. 'BIT_XOR_EXPR'

  10819. 'BIT_AND_EXPR'

  10820.      These nodes represent bitwise inclusive or, bitwise exclusive or,

  10821.      and bitwise and, respectively.  Both operands will always have

  10822.      integral type.

  10823. 

  10824. 'TRUTH_ANDIF_EXPR'

  10825. 'TRUTH_ORIF_EXPR'

  10826.      These nodes represent logical "and" and logical "or", respectively.

  10827.      These operators are not strict; i.e., the second operand is

  10828.      evaluated only if the value of the expression is not determined by

  10829.      evaluation of the first operand.  The type of the operands and that

  10830.      of the result are always of 'BOOLEAN_TYPE' or 'INTEGER_TYPE'.

  10831. 

  10832. 'TRUTH_AND_EXPR'

  10833. 'TRUTH_OR_EXPR'

  10834. 'TRUTH_XOR_EXPR'

  10835.      These nodes represent logical and, logical or, and logical

  10836.      exclusive or.  They are strict; both arguments are always

  10837.      evaluated.  There are no corresponding operators in C or C++, but

  10838.      the front end will sometimes generate these expressions anyhow, if

  10839.      it can tell that strictness does not matter.  The type of the

  10840.      operands and that of the result are always of 'BOOLEAN_TYPE' or

  10841.      'INTEGER_TYPE'.

  10842. 

  10843. 'POINTER_PLUS_EXPR'

  10844.      This node represents pointer arithmetic.  The first operand is

  10845.      always a pointer/reference type.  The second operand is always an

  10846.      unsigned integer type compatible with sizetype.  This and

  10847.      POINTER_DIFF_EXPR are the only binary arithmetic operators that can

  10848.      operate on pointer types.

  10849. 

  10850. 'POINTER_DIFF_EXPR'

  10851.      This node represents pointer subtraction.  The two operands always

  10852.      have pointer/reference type.  It returns a signed integer of the

  10853.      same precision as the pointers.  The behavior is undefined if the

  10854.      difference of the two pointers, seen as infinite precision

  10855.      non-negative integers, does not fit in the result type.  The result

  10856.      does not depend on the pointer type, it is not divided by the size

  10857.      of the pointed-to type.

  10858. 

  10859. 'PLUS_EXPR'

  10860. 'MINUS_EXPR'

  10861. 'MULT_EXPR'

  10862.      These nodes represent various binary arithmetic operations.

  10863.      Respectively, these operations are addition, subtraction (of the

  10864.      second operand from the first) and multiplication.  Their operands

  10865.      may have either integral or floating type, but there will never be

  10866.      case in which one operand is of floating type and the other is of

  10867.      integral type.

  10868. 

  10869.      The behavior of these operations on signed arithmetic overflow is

  10870.      controlled by the 'flag_wrapv' and 'flag_trapv' variables.

  10871. 

  10872. 'MULT_HIGHPART_EXPR'

  10873.      This node represents the "high-part" of a widening multiplication.

  10874.      For an integral type with B bits of precision, the result is the

  10875.      most significant B bits of the full 2B product.

  10876. 

  10877. 'RDIV_EXPR'

  10878.      This node represents a floating point division operation.

  10879. 

  10880. 'TRUNC_DIV_EXPR'

  10881. 'FLOOR_DIV_EXPR'

  10882. 'CEIL_DIV_EXPR'

  10883. 'ROUND_DIV_EXPR'

  10884.      These nodes represent integer division operations that return an

  10885.      integer result.  'TRUNC_DIV_EXPR' rounds towards zero,

  10886.      'FLOOR_DIV_EXPR' rounds towards negative infinity, 'CEIL_DIV_EXPR'

  10887.      rounds towards positive infinity and 'ROUND_DIV_EXPR' rounds to the

  10888.      closest integer.  Integer division in C and C++ is truncating, i.e.

  10889.      'TRUNC_DIV_EXPR'.

  10890. 

  10891.      The behavior of these operations on signed arithmetic overflow,

  10892.      when dividing the minimum signed integer by minus one, is

  10893.      controlled by the 'flag_wrapv' and 'flag_trapv' variables.

  10894. 

  10895. 'TRUNC_MOD_EXPR'

  10896. 'FLOOR_MOD_EXPR'

  10897. 'CEIL_MOD_EXPR'

  10898. 'ROUND_MOD_EXPR'

  10899.      These nodes represent the integer remainder or modulus operation.

  10900.      The integer modulus of two operands 'a' and 'b' is defined as 'a -

  10901.      (a/b)*b' where the division calculated using the corresponding

  10902.      division operator.  Hence for 'TRUNC_MOD_EXPR' this definition

  10903.      assumes division using truncation towards zero, i.e.

  10904.      'TRUNC_DIV_EXPR'.  Integer remainder in C and C++ uses truncating

  10905.      division, i.e. 'TRUNC_MOD_EXPR'.

  10906. 

  10907. 'EXACT_DIV_EXPR'

  10908.      The 'EXACT_DIV_EXPR' code is used to represent integer divisions

  10909.      where the numerator is known to be an exact multiple of the

  10910.      denominator.  This allows the backend to choose between the faster

  10911.      of 'TRUNC_DIV_EXPR', 'CEIL_DIV_EXPR' and 'FLOOR_DIV_EXPR' for the

  10912.      current target.

  10913. 

  10914. 'LT_EXPR'

  10915. 'LE_EXPR'

  10916. 'GT_EXPR'

  10917. 'GE_EXPR'

  10918. 'LTGT_EXPR'

  10919. 'EQ_EXPR'

  10920. 'NE_EXPR'

  10921.      These nodes represent the less than, less than or equal to, greater

  10922.      than, greater than or equal to, less or greater than, equal, and

  10923.      not equal comparison operators.  The first and second operands will

  10924.      either be both of integral type, both of floating type or both of

  10925.      vector type, except for LTGT_EXPR where they will only be both of

  10926.      floating type.  The result type of these expressions will always be

  10927.      of integral, boolean or signed integral vector type.  These

  10928.      operations return the result type's zero value for false, the

  10929.      result type's one value for true, and a vector whose elements are

  10930.      zero (false) or minus one (true) for vectors.

  10931. 

  10932.      For floating point comparisons, if we honor IEEE NaNs and either

  10933.      operand is NaN, then 'NE_EXPR' always returns true and the

  10934.      remaining operators always return false.  On some targets,

  10935.      comparisons against an IEEE NaN, other than equality and

  10936.      inequality, may generate a floating-point exception.

  10937. 

  10938. 'ORDERED_EXPR'

  10939. 'UNORDERED_EXPR'

  10940.      These nodes represent non-trapping ordered and unordered comparison

  10941.      operators.  These operations take two floating point operands and

  10942.      determine whether they are ordered or unordered relative to each

  10943.      other.  If either operand is an IEEE NaN, their comparison is

  10944.      defined to be unordered, otherwise the comparison is defined to be

  10945.      ordered.  The result type of these expressions will always be of

  10946.      integral or boolean type.  These operations return the result

  10947.      type's zero value for false, and the result type's one value for

  10948.      true.

  10949. 

  10950. 'UNLT_EXPR'

  10951. 'UNLE_EXPR'

  10952. 'UNGT_EXPR'

  10953. 'UNGE_EXPR'

  10954. 'UNEQ_EXPR'

  10955.      These nodes represent the unordered comparison operators.  These

  10956.      operations take two floating point operands and determine whether

  10957.      the operands are unordered or are less than, less than or equal to,

  10958.      greater than, greater than or equal to, or equal respectively.  For

  10959.      example, 'UNLT_EXPR' returns true if either operand is an IEEE NaN

  10960.      or the first operand is less than the second.  All these operations

  10961.      are guaranteed not to generate a floating point exception.  The

  10962.      result type of these expressions will always be of integral or

  10963.      boolean type.  These operations return the result type's zero value

  10964.      for false, and the result type's one value for true.

  10965. 

  10966. 'MODIFY_EXPR'

  10967.      These nodes represent assignment.  The left-hand side is the first

  10968.      operand; the right-hand side is the second operand.  The left-hand

  10969.      side will be a 'VAR_DECL', 'INDIRECT_REF', 'COMPONENT_REF', or

  10970.      other lvalue.

  10971. 

  10972.      These nodes are used to represent not only assignment with '=' but

  10973.      also compound assignments (like '+='), by reduction to '='

  10974.      assignment.  In other words, the representation for 'i += 3' looks

  10975.      just like that for 'i = i + 3'.

  10976. 

  10977. 'INIT_EXPR'

  10978.      These nodes are just like 'MODIFY_EXPR', but are used only when a

  10979.      variable is initialized, rather than assigned to subsequently.

  10980.      This means that we can assume that the target of the initialization

  10981.      is not used in computing its own value; any reference to the lhs in

  10982.      computing the rhs is undefined.

  10983. 

  10984. 'COMPOUND_EXPR'

  10985.      These nodes represent comma-expressions.  The first operand is an

  10986.      expression whose value is computed and thrown away prior to the

  10987.      evaluation of the second operand.  The value of the entire

  10988.      expression is the value of the second operand.

  10989. 

  10990. 'COND_EXPR'

  10991.      These nodes represent '?:' expressions.  The first operand is of

  10992.      boolean or integral type.  If it evaluates to a nonzero value, the

  10993.      second operand should be evaluated, and returned as the value of

  10994.      the expression.  Otherwise, the third operand is evaluated, and

  10995.      returned as the value of the expression.

  10996. 

  10997.      The second operand must have the same type as the entire

  10998.      expression, unless it unconditionally throws an exception or calls

  10999.      a noreturn function, in which case it should have void type.  The

  11000.      same constraints apply to the third operand.  This allows array

  11001.      bounds checks to be represented conveniently as '(i >= 0 && i < 10)

  11002.      ? i : abort()'.

  11003. 

  11004.      As a GNU extension, the C language front-ends allow the second

  11005.      operand of the '?:' operator may be omitted in the source.  For

  11006.      example, 'x ? : 3' is equivalent to 'x ? x : 3', assuming that 'x'

  11007.      is an expression without side effects.  In the tree representation,

  11008.      however, the second operand is always present, possibly protected

  11009.      by 'SAVE_EXPR' if the first argument does cause side effects.

  11010. 

  11011. 'CALL_EXPR'

  11012.      These nodes are used to represent calls to functions, including

  11013.      non-static member functions.  'CALL_EXPR's are implemented as

  11014.      expression nodes with a variable number of operands.  Rather than

  11015.      using 'TREE_OPERAND' to extract them, it is preferable to use the

  11016.      specialized accessor macros and functions that operate specifically

  11017.      on 'CALL_EXPR' nodes.

  11018. 

  11019.      'CALL_EXPR_FN' returns a pointer to the function to call; it is

  11020.      always an expression whose type is a 'POINTER_TYPE'.

  11021. 

  11022.      The number of arguments to the call is returned by

  11023.      'call_expr_nargs', while the arguments themselves can be accessed

  11024.      with the 'CALL_EXPR_ARG' macro.  The arguments are zero-indexed and

  11025.      numbered left-to-right.  You can iterate over the arguments using

  11026.      'FOR_EACH_CALL_EXPR_ARG', as in:

  11027. 

  11028.           tree call, arg;

  11029.           call_expr_arg_iterator iter;

  11030.           FOR_EACH_CALL_EXPR_ARG (arg, iter, call)

  11031.             /* arg is bound to successive arguments of call.  */

  11032.             ...;

  11033. 

  11034.      For non-static member functions, there will be an operand

  11035.      corresponding to the 'this' pointer.  There will always be

  11036.      expressions corresponding to all of the arguments, even if the

  11037.      function is declared with default arguments and some arguments are

  11038.      not explicitly provided at the call sites.

  11039. 

  11040.      'CALL_EXPR's also have a 'CALL_EXPR_STATIC_CHAIN' operand that is

  11041.      used to implement nested functions.  This operand is otherwise

  11042.      null.

  11043. 

  11044. 'CLEANUP_POINT_EXPR'

  11045.      These nodes represent full-expressions.  The single operand is an

  11046.      expression to evaluate.  Any destructor calls engendered by the

  11047.      creation of temporaries during the evaluation of that expression

  11048.      should be performed immediately after the expression is evaluated.

  11049. 

  11050. 'CONSTRUCTOR'

  11051.      These nodes represent the brace-enclosed initializers for a

  11052.      structure or an array.  They contain a sequence of component values

  11053.      made out of a vector of constructor_elt, which is a ('INDEX',

  11054.      'VALUE') pair.

  11055. 

  11056.      If the 'TREE_TYPE' of the 'CONSTRUCTOR' is a 'RECORD_TYPE',

  11057.      'UNION_TYPE' or 'QUAL_UNION_TYPE' then the 'INDEX' of each node in

  11058.      the sequence will be a 'FIELD_DECL' and the 'VALUE' will be the

  11059.      expression used to initialize that field.

  11060. 

  11061.      If the 'TREE_TYPE' of the 'CONSTRUCTOR' is an 'ARRAY_TYPE', then

  11062.      the 'INDEX' of each node in the sequence will be an 'INTEGER_CST'

  11063.      or a 'RANGE_EXPR' of two 'INTEGER_CST's.  A single 'INTEGER_CST'

  11064.      indicates which element of the array is being assigned to.  A

  11065.      'RANGE_EXPR' indicates an inclusive range of elements to

  11066.      initialize.  In both cases the 'VALUE' is the corresponding

  11067.      initializer.  It is re-evaluated for each element of a

  11068.      'RANGE_EXPR'.  If the 'INDEX' is 'NULL_TREE', then the initializer

  11069.      is for the next available array element.

  11070. 

  11071.      In the front end, you should not depend on the fields appearing in

  11072.      any particular order.  However, in the middle end, fields must

  11073.      appear in declaration order.  You should not assume that all fields

  11074.      will be represented.  Unrepresented fields will be cleared

  11075.      (zeroed), unless the CONSTRUCTOR_NO_CLEARING flag is set, in which

  11076.      case their value becomes undefined.

  11077. 

  11078. 'COMPOUND_LITERAL_EXPR'

  11079.      These nodes represent ISO C99 compound literals.  The

  11080.      'COMPOUND_LITERAL_EXPR_DECL_EXPR' is a 'DECL_EXPR' containing an

  11081.      anonymous 'VAR_DECL' for the unnamed object represented by the

  11082.      compound literal; the 'DECL_INITIAL' of that 'VAR_DECL' is a

  11083.      'CONSTRUCTOR' representing the brace-enclosed list of initializers

  11084.      in the compound literal.  That anonymous 'VAR_DECL' can also be

  11085.      accessed directly by the 'COMPOUND_LITERAL_EXPR_DECL' macro.

  11086. 

  11087. 'SAVE_EXPR'

  11088. 

  11089.      A 'SAVE_EXPR' represents an expression (possibly involving side

  11090.      effects) that is used more than once.  The side effects should

  11091.      occur only the first time the expression is evaluated.  Subsequent

  11092.      uses should just reuse the computed value.  The first operand to

  11093.      the 'SAVE_EXPR' is the expression to evaluate.  The side effects

  11094.      should be executed where the 'SAVE_EXPR' is first encountered in a

  11095.      depth-first preorder traversal of the expression tree.

  11096. 

  11097. 'TARGET_EXPR'

  11098.      A 'TARGET_EXPR' represents a temporary object.  The first operand

  11099.      is a 'VAR_DECL' for the temporary variable.  The second operand is

  11100.      the initializer for the temporary.  The initializer is evaluated

  11101.      and, if non-void, copied (bitwise) into the temporary.  If the

  11102.      initializer is void, that means that it will perform the

  11103.      initialization itself.

  11104. 

  11105.      Often, a 'TARGET_EXPR' occurs on the right-hand side of an

  11106.      assignment, or as the second operand to a comma-expression which is

  11107.      itself the right-hand side of an assignment, etc.  In this case, we

  11108.      say that the 'TARGET_EXPR' is "normal"; otherwise, we say it is

  11109.      "orphaned".  For a normal 'TARGET_EXPR' the temporary variable

  11110.      should be treated as an alias for the left-hand side of the

  11111.      assignment, rather than as a new temporary variable.

  11112. 

  11113.      The third operand to the 'TARGET_EXPR', if present, is a

  11114.      cleanup-expression (i.e., destructor call) for the temporary.  If

  11115.      this expression is orphaned, then this expression must be executed

  11116.      when the statement containing this expression is complete.  These

  11117.      cleanups must always be executed in the order opposite to that in

  11118.      which they were encountered.  Note that if a temporary is created

  11119.      on one branch of a conditional operator (i.e., in the second or

  11120.      third operand to a 'COND_EXPR'), the cleanup must be run only if

  11121.      that branch is actually executed.

  11122. 

  11123. 'VA_ARG_EXPR'

  11124.      This node is used to implement support for the C/C++ variable

  11125.      argument-list mechanism.  It represents expressions like 'va_arg

  11126.      (ap, type)'.  Its 'TREE_TYPE' yields the tree representation for

  11127.      'type' and its sole argument yields the representation for 'ap'.

  11128. 

  11129. 'ANNOTATE_EXPR'

  11130.      This node is used to attach markers to an expression.  The first

  11131.      operand is the annotated expression, the second is an 'INTEGER_CST'

  11132.      with a value from 'enum annot_expr_kind', the third is an

  11133.      'INTEGER_CST'.

  11134. 

  11135. 

  11136. File: gccint.info,  Node: Vectors,  Prev: Unary and Binary Expressions,  Up: Expression trees

  11137. 

  11138. 11.6.4 Vectors

  11139. --------------

  11140. 

  11141. 'VEC_DUPLICATE_EXPR'

  11142.      This node has a single operand and represents a vector in which

  11143.      every element is equal to that operand.

  11144. 

  11145. 'VEC_SERIES_EXPR'

  11146.      This node represents a vector formed from a scalar base and step,

  11147.      given as the first and second operands respectively.  Element I of

  11148.      the result is equal to 'BASE + I*STEP'.

  11149. 

  11150.      This node is restricted to integral types, in order to avoid

  11151.      specifying the rounding behavior for floating-point types.

  11152. 

  11153. 'VEC_LSHIFT_EXPR'

  11154. 'VEC_RSHIFT_EXPR'

  11155.      These nodes represent whole vector left and right shifts,

  11156.      respectively.  The first operand is the vector to shift; it will

  11157.      always be of vector type.  The second operand is an expression for

  11158.      the number of bits by which to shift.  Note that the result is

  11159.      undefined if the second operand is larger than or equal to the

  11160.      first operand's type size.

  11161. 

  11162. 'VEC_WIDEN_MULT_HI_EXPR'

  11163. 'VEC_WIDEN_MULT_LO_EXPR'

  11164.      These nodes represent widening vector multiplication of the high

  11165.      and low parts of the two input vectors, respectively.  Their

  11166.      operands are vectors that contain the same number of elements ('N')

  11167.      of the same integral type.  The result is a vector that contains

  11168.      half as many elements, of an integral type whose size is twice as

  11169.      wide.  In the case of 'VEC_WIDEN_MULT_HI_EXPR' the high 'N/2'

  11170.      elements of the two vector are multiplied to produce the vector of

  11171.      'N/2' products.  In the case of 'VEC_WIDEN_MULT_LO_EXPR' the low

  11172.      'N/2' elements of the two vector are multiplied to produce the

  11173.      vector of 'N/2' products.

  11174. 

  11175. 'VEC_UNPACK_HI_EXPR'

  11176. 'VEC_UNPACK_LO_EXPR'

  11177.      These nodes represent unpacking of the high and low parts of the

  11178.      input vector, respectively.  The single operand is a vector that

  11179.      contains 'N' elements of the same integral or floating point type.

  11180.      The result is a vector that contains half as many elements, of an

  11181.      integral or floating point type whose size is twice as wide.  In

  11182.      the case of 'VEC_UNPACK_HI_EXPR' the high 'N/2' elements of the

  11183.      vector are extracted and widened (promoted).  In the case of

  11184.      'VEC_UNPACK_LO_EXPR' the low 'N/2' elements of the vector are

  11185.      extracted and widened (promoted).

  11186. 

  11187. 'VEC_UNPACK_FLOAT_HI_EXPR'

  11188. 'VEC_UNPACK_FLOAT_LO_EXPR'

  11189.      These nodes represent unpacking of the high and low parts of the

  11190.      input vector, where the values are converted from fixed point to

  11191.      floating point.  The single operand is a vector that contains 'N'

  11192.      elements of the same integral type.  The result is a vector that

  11193.      contains half as many elements of a floating point type whose size

  11194.      is twice as wide.  In the case of 'VEC_UNPACK_FLOAT_HI_EXPR' the

  11195.      high 'N/2' elements of the vector are extracted, converted and

  11196.      widened.  In the case of 'VEC_UNPACK_FLOAT_LO_EXPR' the low 'N/2'

  11197.      elements of the vector are extracted, converted and widened.

  11198. 

  11199. 'VEC_UNPACK_FIX_TRUNC_HI_EXPR'

  11200. 'VEC_UNPACK_FIX_TRUNC_LO_EXPR'

  11201.      These nodes represent unpacking of the high and low parts of the

  11202.      input vector, where the values are truncated from floating point to

  11203.      fixed point.  The single operand is a vector that contains 'N'

  11204.      elements of the same floating point type.  The result is a vector

  11205.      that contains half as many elements of an integral type whose size

  11206.      is twice as wide.  In the case of 'VEC_UNPACK_FIX_TRUNC_HI_EXPR'

  11207.      the high 'N/2' elements of the vector are extracted and converted

  11208.      with truncation.  In the case of 'VEC_UNPACK_FIX_TRUNC_LO_EXPR' the

  11209.      low 'N/2' elements of the vector are extracted and converted with

  11210.      truncation.

  11211. 

  11212. 'VEC_PACK_TRUNC_EXPR'

  11213.      This node represents packing of truncated elements of the two input

  11214.      vectors into the output vector.  Input operands are vectors that

  11215.      contain the same number of elements of the same integral or

  11216.      floating point type.  The result is a vector that contains twice as

  11217.      many elements of an integral or floating point type whose size is

  11218.      half as wide.  The elements of the two vectors are demoted and

  11219.      merged (concatenated) to form the output vector.

  11220. 

  11221. 'VEC_PACK_SAT_EXPR'

  11222.      This node represents packing of elements of the two input vectors

  11223.      into the output vector using saturation.  Input operands are

  11224.      vectors that contain the same number of elements of the same

  11225.      integral type.  The result is a vector that contains twice as many

  11226.      elements of an integral type whose size is half as wide.  The

  11227.      elements of the two vectors are demoted and merged (concatenated)

  11228.      to form the output vector.

  11229. 

  11230. 'VEC_PACK_FIX_TRUNC_EXPR'

  11231.      This node represents packing of elements of the two input vectors

  11232.      into the output vector, where the values are converted from

  11233.      floating point to fixed point.  Input operands are vectors that

  11234.      contain the same number of elements of a floating point type.  The

  11235.      result is a vector that contains twice as many elements of an

  11236.      integral type whose size is half as wide.  The elements of the two

  11237.      vectors are merged (concatenated) to form the output vector.

  11238. 

  11239. 'VEC_PACK_FLOAT_EXPR'

  11240.      This node represents packing of elements of the two input vectors

  11241.      into the output vector, where the values are converted from fixed

  11242.      point to floating point.  Input operands are vectors that contain

  11243.      the same number of elements of an integral type.  The result is a

  11244.      vector that contains twice as many elements of floating point type

  11245.      whose size is half as wide.  The elements of the two vectors are

  11246.      merged (concatenated) to form the output vector.

  11247. 

  11248. 'VEC_COND_EXPR'

  11249.      These nodes represent '?:' expressions.  The three operands must be

  11250.      vectors of the same size and number of elements.  The second and

  11251.      third operands must have the same type as the entire expression.

  11252.      The first operand is of signed integral vector type.  If an element

  11253.      of the first operand evaluates to a zero value, the corresponding

  11254.      element of the result is taken from the third operand.  If it

  11255.      evaluates to a minus one value, it is taken from the second

  11256.      operand.  It should never evaluate to any other value currently,

  11257.      but optimizations should not rely on that property.  In contrast

  11258.      with a 'COND_EXPR', all operands are always evaluated.

  11259. 

  11260. 'SAD_EXPR'

  11261.      This node represents the Sum of Absolute Differences operation.

  11262.      The three operands must be vectors of integral types.  The first

  11263.      and second operand must have the same type.  The size of the vector

  11264.      element of the third operand must be at lease twice of the size of

  11265.      the vector element of the first and second one.  The SAD is

  11266.      calculated between the first and second operands, added to the

  11267.      third operand, and returned.

  11268. 

  11269. 

  11270. File: gccint.info,  Node: Statements,  Next: Functions,  Prev: Expression trees,  Up: GENERIC

  11271. 

  11272. 11.7 Statements

  11273. ===============

  11274. 

  11275. Most statements in GIMPLE are assignment statements, represented by

  11276. 'GIMPLE_ASSIGN'.  No other C expressions can appear at statement level;

  11277. a reference to a volatile object is converted into a 'GIMPLE_ASSIGN'.

  11278. 

  11279.  There are also several varieties of complex statements.

  11280. 

  11281. * Menu:

  11282. 

  11283. * Basic Statements::

  11284. * Blocks::

  11285. * Statement Sequences::

  11286. * Empty Statements::

  11287. * Jumps::

  11288. * Cleanups::

  11289. * OpenMP::

  11290. * OpenACC::

  11291. 

  11292. 

  11293. File: gccint.info,  Node: Basic Statements,  Next: Blocks,  Up: Statements

  11294. 

  11295. 11.7.1 Basic Statements

  11296. -----------------------

  11297. 

  11298. 'ASM_EXPR'

  11299. 

  11300.      Used to represent an inline assembly statement.  For an inline

  11301.      assembly statement like:

  11302.           asm ("mov x, y");

  11303.      The 'ASM_STRING' macro will return a 'STRING_CST' node for '"mov x,

  11304.      y"'.  If the original statement made use of the extended-assembly

  11305.      syntax, then 'ASM_OUTPUTS', 'ASM_INPUTS', and 'ASM_CLOBBERS' will

  11306.      be the outputs, inputs, and clobbers for the statement, represented

  11307.      as 'STRING_CST' nodes.  The extended-assembly syntax looks like:

  11308.           asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));

  11309.      The first string is the 'ASM_STRING', containing the instruction

  11310.      template.  The next two strings are the output and inputs,

  11311.      respectively; this statement has no clobbers.  As this example

  11312.      indicates, "plain" assembly statements are merely a special case of

  11313.      extended assembly statements; they have no cv-qualifiers, outputs,

  11314.      inputs, or clobbers.  All of the strings will be 'NUL'-terminated,

  11315.      and will contain no embedded 'NUL'-characters.

  11316. 

  11317.      If the assembly statement is declared 'volatile', or if the

  11318.      statement was not an extended assembly statement, and is therefore

  11319.      implicitly volatile, then the predicate 'ASM_VOLATILE_P' will hold

  11320.      of the 'ASM_EXPR'.

  11321. 

  11322. 'DECL_EXPR'

  11323. 

  11324.      Used to represent a local declaration.  The 'DECL_EXPR_DECL' macro

  11325.      can be used to obtain the entity declared.  This declaration may be

  11326.      a 'LABEL_DECL', indicating that the label declared is a local

  11327.      label.  (As an extension, GCC allows the declaration of labels with

  11328.      scope.)  In C, this declaration may be a 'FUNCTION_DECL',

  11329.      indicating the use of the GCC nested function extension.  For more

  11330.      information, *note Functions::.

  11331. 

  11332. 'LABEL_EXPR'

  11333. 

  11334.      Used to represent a label.  The 'LABEL_DECL' declared by this

  11335.      statement can be obtained with the 'LABEL_EXPR_LABEL' macro.  The

  11336.      'IDENTIFIER_NODE' giving the name of the label can be obtained from

  11337.      the 'LABEL_DECL' with 'DECL_NAME'.

  11338. 

  11339. 'GOTO_EXPR'

  11340. 

  11341.      Used to represent a 'goto' statement.  The 'GOTO_DESTINATION' will

  11342.      usually be a 'LABEL_DECL'.  However, if the "computed goto"

  11343.      extension has been used, the 'GOTO_DESTINATION' will be an

  11344.      arbitrary expression indicating the destination.  This expression

  11345.      will always have pointer type.

  11346. 

  11347. 'RETURN_EXPR'

  11348. 

  11349.      Used to represent a 'return' statement.  Operand 0 represents the

  11350.      value to return.  It should either be the 'RESULT_DECL' for the

  11351.      containing function, or a 'MODIFY_EXPR' or 'INIT_EXPR' setting the

  11352.      function's 'RESULT_DECL'.  It will be 'NULL_TREE' if the statement

  11353.      was just

  11354.           return;

  11355. 

  11356. 'LOOP_EXPR'

  11357.      These nodes represent "infinite" loops.  The 'LOOP_EXPR_BODY'

  11358.      represents the body of the loop.  It should be executed forever,

  11359.      unless an 'EXIT_EXPR' is encountered.

  11360. 

  11361. 'EXIT_EXPR'

  11362.      These nodes represent conditional exits from the nearest enclosing

  11363.      'LOOP_EXPR'.  The single operand is the condition; if it is

  11364.      nonzero, then the loop should be exited.  An 'EXIT_EXPR' will only

  11365.      appear within a 'LOOP_EXPR'.

  11366. 

  11367. 'SWITCH_STMT'

  11368. 

  11369.      Used to represent a 'switch' statement.  The 'SWITCH_STMT_COND' is

  11370.      the expression on which the switch is occurring.  See the

  11371.      documentation for an 'IF_STMT' for more information on the

  11372.      representation used for the condition.  The 'SWITCH_STMT_BODY' is

  11373.      the body of the switch statement.  The 'SWITCH_STMT_TYPE' is the

  11374.      original type of switch expression as given in the source, before

  11375.      any compiler conversions.

  11376. 

  11377. 'CASE_LABEL_EXPR'

  11378. 

  11379.      Use to represent a 'case' label, range of 'case' labels, or a

  11380.      'default' label.  If 'CASE_LOW' is 'NULL_TREE', then this is a

  11381.      'default' label.  Otherwise, if 'CASE_HIGH' is 'NULL_TREE', then

  11382.      this is an ordinary 'case' label.  In this case, 'CASE_LOW' is an

  11383.      expression giving the value of the label.  Both 'CASE_LOW' and

  11384.      'CASE_HIGH' are 'INTEGER_CST' nodes.  These values will have the

  11385.      same type as the condition expression in the switch statement.

  11386. 

  11387.      Otherwise, if both 'CASE_LOW' and 'CASE_HIGH' are defined, the

  11388.      statement is a range of case labels.  Such statements originate

  11389.      with the extension that allows users to write things of the form:

  11390.           case 2 ... 5:

  11391.      The first value will be 'CASE_LOW', while the second will be

  11392.      'CASE_HIGH'.

  11393. 

  11394. 'DEBUG_BEGIN_STMT'

  11395. 

  11396.      Marks the beginning of a source statement, for purposes of debug

  11397.      information generation.

  11398. 

  11399. 

  11400. File: gccint.info,  Node: Blocks,  Next: Statement Sequences,  Prev: Basic Statements,  Up: Statements

  11401. 

  11402. 11.7.2 Blocks

  11403. -------------

  11404. 

  11405. Block scopes and the variables they declare in GENERIC are expressed

  11406. using the 'BIND_EXPR' code, which in previous versions of GCC was

  11407. primarily used for the C statement-expression extension.

  11408. 

  11409.  Variables in a block are collected into 'BIND_EXPR_VARS' in declaration

  11410. order through their 'TREE_CHAIN' field.  Any runtime initialization is

  11411. moved out of 'DECL_INITIAL' and into a statement in the controlled

  11412. block.  When gimplifying from C or C++, this initialization replaces the

  11413. 'DECL_STMT'.  These variables will never require cleanups.  The scope of

  11414. these variables is just the body

  11415. 

  11416.  Variable-length arrays (VLAs) complicate this process, as their size

  11417. often refers to variables initialized earlier in the block and their

  11418. initialization involves an explicit stack allocation.  To handle this,

  11419. we add an indirection and replace them with a pointer to stack space

  11420. allocated by means of 'alloca'.  In most cases, we also arrange for this

  11421. space to be reclaimed when the enclosing 'BIND_EXPR' is exited, the

  11422. exception to this being when there is an explicit call to 'alloca' in

  11423. the source code, in which case the stack is left depressed on exit of

  11424. the 'BIND_EXPR'.

  11425. 

  11426.  A C++ program will usually contain more 'BIND_EXPR's than there are

  11427. syntactic blocks in the source code, since several C++ constructs have

  11428. implicit scopes associated with them.  On the other hand, although the

  11429. C++ front end uses pseudo-scopes to handle cleanups for objects with

  11430. destructors, these don't translate into the GIMPLE form; multiple

  11431. declarations at the same level use the same 'BIND_EXPR'.

  11432. 

  11433. 

  11434. File: gccint.info,  Node: Statement Sequences,  Next: Empty Statements,  Prev: Blocks,  Up: Statements

  11435. 

  11436. 11.7.3 Statement Sequences

  11437. --------------------------

  11438. 

  11439. Multiple statements at the same nesting level are collected into a

  11440. 'STATEMENT_LIST'.  Statement lists are modified and traversed using the

  11441. interface in 'tree-iterator.h'.

  11442. 

  11443. 

  11444. File: gccint.info,  Node: Empty Statements,  Next: Jumps,  Prev: Statement Sequences,  Up: Statements

  11445. 

  11446. 11.7.4 Empty Statements

  11447. -----------------------

  11448. 

  11449. Whenever possible, statements with no effect are discarded.  But if they

  11450. are nested within another construct which cannot be discarded for some

  11451. reason, they are instead replaced with an empty statement, generated by

  11452. 'build_empty_stmt'.  Initially, all empty statements were shared, after

  11453. the pattern of the Java front end, but this caused a lot of trouble in

  11454. practice.

  11455. 

  11456.  An empty statement is represented as '(void)0'.

  11457. 

  11458. 

  11459. File: gccint.info,  Node: Jumps,  Next: Cleanups,  Prev: Empty Statements,  Up: Statements

  11460. 

  11461. 11.7.5 Jumps

  11462. ------------

  11463. 

  11464. Other jumps are expressed by either 'GOTO_EXPR' or 'RETURN_EXPR'.

  11465. 

  11466.  The operand of a 'GOTO_EXPR' must be either a label or a variable

  11467. containing the address to jump to.

  11468. 

  11469.  The operand of a 'RETURN_EXPR' is either 'NULL_TREE', 'RESULT_DECL', or

  11470. a 'MODIFY_EXPR' which sets the return value.  It would be nice to move

  11471. the 'MODIFY_EXPR' into a separate statement, but the special return

  11472. semantics in 'expand_return' make that difficult.  It may still happen

  11473. in the future, perhaps by moving most of that logic into

  11474. 'expand_assignment'.

  11475. 

  11476. 

  11477. File: gccint.info,  Node: Cleanups,  Next: OpenMP,  Prev: Jumps,  Up: Statements

  11478. 

  11479. 11.7.6 Cleanups

  11480. ---------------

  11481. 

  11482. Destructors for local C++ objects and similar dynamic cleanups are

  11483. represented in GIMPLE by a 'TRY_FINALLY_EXPR'.  'TRY_FINALLY_EXPR' has

  11484. two operands, both of which are a sequence of statements to execute.

  11485. The first sequence is executed.  When it completes the second sequence

  11486. is executed.

  11487. 

  11488.  The first sequence may complete in the following ways:

  11489. 

  11490.   1. Execute the last statement in the sequence and fall off the end.

  11491. 

  11492.   2. Execute a goto statement ('GOTO_EXPR') to an ordinary label outside

  11493.      the sequence.

  11494. 

  11495.   3. Execute a return statement ('RETURN_EXPR').

  11496. 

  11497.   4. Throw an exception.  This is currently not explicitly represented

  11498.      in GIMPLE.

  11499. 

  11500.  The second sequence is not executed if the first sequence completes by

  11501. calling 'setjmp' or 'exit' or any other function that does not return.

  11502. The second sequence is also not executed if the first sequence completes

  11503. via a non-local goto or a computed goto (in general the compiler does

  11504. not know whether such a goto statement exits the first sequence or not,

  11505. so we assume that it doesn't).

  11506. 

  11507.  After the second sequence is executed, if it completes normally by

  11508. falling off the end, execution continues wherever the first sequence

  11509. would have continued, by falling off the end, or doing a goto, etc.

  11510. 

  11511.  If the second sequence is an 'EH_ELSE_EXPR' selector, then the sequence

  11512. in its first operand is used when the first sequence completes normally,

  11513. and that in its second operand is used for exceptional cleanups, i.e.,

  11514. when an exception propagates out of the first sequence.

  11515. 

  11516.  'TRY_FINALLY_EXPR' complicates the flow graph, since the cleanup needs

  11517. to appear on every edge out of the controlled block; this reduces the

  11518. freedom to move code across these edges.  Therefore, the EH lowering

  11519. pass which runs before most of the optimization passes eliminates these

  11520. expressions by explicitly adding the cleanup to each edge.  Rethrowing

  11521. the exception is represented using 'RESX_EXPR'.

  11522. 

  11523. 

  11524. File: gccint.info,  Node: OpenMP,  Next: OpenACC,  Prev: Cleanups,  Up: Statements

  11525. 

  11526. 11.7.7 OpenMP

  11527. -------------

  11528. 

  11529. All the statements starting with 'OMP_' represent directives and clauses

  11530. used by the OpenMP API <https://www.openmp.org>.

  11531. 

  11532. 'OMP_PARALLEL'

  11533. 

  11534.      Represents '#pragma omp parallel [clause1 ... clauseN]'.  It has

  11535.      four operands:

  11536. 

  11537.      Operand 'OMP_PARALLEL_BODY' is valid while in GENERIC and High

  11538.      GIMPLE forms.  It contains the body of code to be executed by all

  11539.      the threads.  During GIMPLE lowering, this operand becomes 'NULL'

  11540.      and the body is emitted linearly after 'OMP_PARALLEL'.

  11541. 

  11542.      Operand 'OMP_PARALLEL_CLAUSES' is the list of clauses associated

  11543.      with the directive.

  11544. 

  11545.      Operand 'OMP_PARALLEL_FN' is created by 'pass_lower_omp', it

  11546.      contains the 'FUNCTION_DECL' for the function that will contain the

  11547.      body of the parallel region.

  11548. 

  11549.      Operand 'OMP_PARALLEL_DATA_ARG' is also created by

  11550.      'pass_lower_omp'.  If there are shared variables to be communicated

  11551.      to the children threads, this operand will contain the 'VAR_DECL'

  11552.      that contains all the shared values and variables.

  11553. 

  11554. 'OMP_FOR'

  11555. 

  11556.      Represents '#pragma omp for [clause1 ... clauseN]'.  It has six

  11557.      operands:

  11558. 

  11559.      Operand 'OMP_FOR_BODY' contains the loop body.

  11560. 

  11561.      Operand 'OMP_FOR_CLAUSES' is the list of clauses associated with

  11562.      the directive.

  11563. 

  11564.      Operand 'OMP_FOR_INIT' is the loop initialization code of the form

  11565.      'VAR = N1'.

  11566. 

  11567.      Operand 'OMP_FOR_COND' is the loop conditional expression of the

  11568.      form 'VAR {<,>,<=,>=} N2'.

  11569. 

  11570.      Operand 'OMP_FOR_INCR' is the loop index increment of the form 'VAR

  11571.      {+=,-=} INCR'.

  11572. 

  11573.      Operand 'OMP_FOR_PRE_BODY' contains side effect code from operands

  11574.      'OMP_FOR_INIT', 'OMP_FOR_COND' and 'OMP_FOR_INC'.  These side

  11575.      effects are part of the 'OMP_FOR' block but must be evaluated

  11576.      before the start of loop body.

  11577. 

  11578.      The loop index variable 'VAR' must be a signed integer variable,

  11579.      which is implicitly private to each thread.  Bounds 'N1' and 'N2'

  11580.      and the increment expression 'INCR' are required to be loop

  11581.      invariant integer expressions that are evaluated without any

  11582.      synchronization.  The evaluation order, frequency of evaluation and

  11583.      side effects are unspecified by the standard.

  11584. 

  11585. 'OMP_SECTIONS'

  11586. 

  11587.      Represents '#pragma omp sections [clause1 ... clauseN]'.

  11588. 

  11589.      Operand 'OMP_SECTIONS_BODY' contains the sections body, which in

  11590.      turn contains a set of 'OMP_SECTION' nodes for each of the

  11591.      concurrent sections delimited by '#pragma omp section'.

  11592. 

  11593.      Operand 'OMP_SECTIONS_CLAUSES' is the list of clauses associated

  11594.      with the directive.

  11595. 

  11596. 'OMP_SECTION'

  11597. 

  11598.      Section delimiter for 'OMP_SECTIONS'.

  11599. 

  11600. 'OMP_SINGLE'

  11601. 

  11602.      Represents '#pragma omp single'.

  11603. 

  11604.      Operand 'OMP_SINGLE_BODY' contains the body of code to be executed

  11605.      by a single thread.

  11606. 

  11607.      Operand 'OMP_SINGLE_CLAUSES' is the list of clauses associated with

  11608.      the directive.

  11609. 

  11610. 'OMP_MASTER'

  11611. 

  11612.      Represents '#pragma omp master'.

  11613. 

  11614.      Operand 'OMP_MASTER_BODY' contains the body of code to be executed

  11615.      by the master thread.

  11616. 

  11617. 'OMP_ORDERED'

  11618. 

  11619.      Represents '#pragma omp ordered'.

  11620. 

  11621.      Operand 'OMP_ORDERED_BODY' contains the body of code to be executed

  11622.      in the sequential order dictated by the loop index variable.

  11623. 

  11624. 'OMP_CRITICAL'

  11625. 

  11626.      Represents '#pragma omp critical [name]'.

  11627. 

  11628.      Operand 'OMP_CRITICAL_BODY' is the critical section.

  11629. 

  11630.      Operand 'OMP_CRITICAL_NAME' is an optional identifier to label the

  11631.      critical section.

  11632. 

  11633. 'OMP_RETURN'

  11634. 

  11635.      This does not represent any OpenMP directive, it is an artificial

  11636.      marker to indicate the end of the body of an OpenMP.  It is used by

  11637.      the flow graph ('tree-cfg.c') and OpenMP region building code

  11638.      ('omp-low.c').

  11639. 

  11640. 'OMP_CONTINUE'

  11641. 

  11642.      Similarly, this instruction does not represent an OpenMP directive,

  11643.      it is used by 'OMP_FOR' (and similar codes) as well as

  11644.      'OMP_SECTIONS' to mark the place where the code needs to loop to

  11645.      the next iteration, or the next section, respectively.

  11646. 

  11647.      In some cases, 'OMP_CONTINUE' is placed right before 'OMP_RETURN'.

  11648.      But if there are cleanups that need to occur right after the

  11649.      looping body, it will be emitted between 'OMP_CONTINUE' and

  11650.      'OMP_RETURN'.

  11651. 

  11652. 'OMP_ATOMIC'

  11653. 

  11654.      Represents '#pragma omp atomic'.

  11655. 

  11656.      Operand 0 is the address at which the atomic operation is to be

  11657.      performed.

  11658. 

  11659.      Operand 1 is the expression to evaluate.  The gimplifier tries

  11660.      three alternative code generation strategies.  Whenever possible,

  11661.      an atomic update built-in is used.  If that fails, a

  11662.      compare-and-swap loop is attempted.  If that also fails, a regular

  11663.      critical section around the expression is used.

  11664. 

  11665. 'OMP_CLAUSE'

  11666. 

  11667.      Represents clauses associated with one of the 'OMP_' directives.

  11668.      Clauses are represented by separate subcodes defined in 'tree.h'.

  11669.      Clauses codes can be one of: 'OMP_CLAUSE_PRIVATE',

  11670.      'OMP_CLAUSE_SHARED', 'OMP_CLAUSE_FIRSTPRIVATE',

  11671.      'OMP_CLAUSE_LASTPRIVATE', 'OMP_CLAUSE_COPYIN',

  11672.      'OMP_CLAUSE_COPYPRIVATE', 'OMP_CLAUSE_IF',

  11673.      'OMP_CLAUSE_NUM_THREADS', 'OMP_CLAUSE_SCHEDULE',

  11674.      'OMP_CLAUSE_NOWAIT', 'OMP_CLAUSE_ORDERED', 'OMP_CLAUSE_DEFAULT',

  11675.      'OMP_CLAUSE_REDUCTION', 'OMP_CLAUSE_COLLAPSE', 'OMP_CLAUSE_UNTIED',

  11676.      'OMP_CLAUSE_FINAL', and 'OMP_CLAUSE_MERGEABLE'.  Each code

  11677.      represents the corresponding OpenMP clause.

  11678. 

  11679.      Clauses associated with the same directive are chained together via

  11680.      'OMP_CLAUSE_CHAIN'.  Those clauses that accept a list of variables

  11681.      are restricted to exactly one, accessed with 'OMP_CLAUSE_VAR'.

  11682.      Therefore, multiple variables under the same clause 'C' need to be

  11683.      represented as multiple 'C' clauses chained together.  This

  11684.      facilitates adding new clauses during compilation.

  11685. 

  11686. 

  11687. File: gccint.info,  Node: OpenACC,  Prev: OpenMP,  Up: Statements

  11688. 

  11689. 11.7.8 OpenACC

  11690. --------------

  11691. 

  11692. All the statements starting with 'OACC_' represent directives and

  11693. clauses used by the OpenACC API <https://www.openacc.org>.

  11694. 

  11695. 'OACC_CACHE'

  11696. 

  11697.      Represents '#pragma acc cache (var ...)'.

  11698. 

  11699. 'OACC_DATA'

  11700. 

  11701.      Represents '#pragma acc data [clause1 ... clauseN]'.

  11702. 

  11703. 'OACC_DECLARE'

  11704. 

  11705.      Represents '#pragma acc declare [clause1 ... clauseN]'.

  11706. 

  11707. 'OACC_ENTER_DATA'

  11708. 

  11709.      Represents '#pragma acc enter data [clause1 ... clauseN]'.

  11710. 

  11711. 'OACC_EXIT_DATA'

  11712. 

  11713.      Represents '#pragma acc exit data [clause1 ... clauseN]'.

  11714. 

  11715. 'OACC_HOST_DATA'

  11716. 

  11717.      Represents '#pragma acc host_data [clause1 ... clauseN]'.

  11718. 

  11719. 'OACC_KERNELS'

  11720. 

  11721.      Represents '#pragma acc kernels [clause1 ... clauseN]'.

  11722. 

  11723. 'OACC_LOOP'

  11724. 

  11725.      Represents '#pragma acc loop [clause1 ... clauseN]'.

  11726. 

  11727.      See the description of the 'OMP_FOR' code.

  11728. 

  11729. 'OACC_PARALLEL'

  11730. 

  11731.      Represents '#pragma acc parallel [clause1 ... clauseN]'.

  11732. 

  11733. 'OACC_SERIAL'

  11734. 

  11735.      Represents '#pragma acc serial [clause1 ... clauseN]'.

  11736. 

  11737. 'OACC_UPDATE'

  11738. 

  11739.      Represents '#pragma acc update [clause1 ... clauseN]'.

  11740. 

  11741. 

  11742. File: gccint.info,  Node: Functions,  Next: Language-dependent trees,  Prev: Statements,  Up: GENERIC

  11743. 

  11744. 11.8 Functions

  11745. ==============

  11746. 

  11747. A function is represented by a 'FUNCTION_DECL' node.  It stores the

  11748. basic pieces of the function such as body, parameters, and return type

  11749. as well as information on the surrounding context, visibility, and

  11750. linkage.

  11751. 

  11752. * Menu:

  11753. 

  11754. * Function Basics::     Function names, body, and parameters.

  11755. * Function Properties:: Context, linkage, etc.

  11756. 

  11757. 

  11758. File: gccint.info,  Node: Function Basics,  Next: Function Properties,  Up: Functions

  11759. 

  11760. 11.8.1 Function Basics

  11761. ----------------------

  11762. 

  11763. A function has four core parts: the name, the parameters, the result,

  11764. and the body.  The following macros and functions access these parts of

  11765. a 'FUNCTION_DECL' as well as other basic features:

  11766. 'DECL_NAME'

  11767.      This macro returns the unqualified name of the function, as an

  11768.      'IDENTIFIER_NODE'.  For an instantiation of a function template,

  11769.      the 'DECL_NAME' is the unqualified name of the template, not

  11770.      something like 'f<int>'.  The value of 'DECL_NAME' is undefined

  11771.      when used on a constructor, destructor, overloaded operator, or

  11772.      type-conversion operator, or any function that is implicitly

  11773.      generated by the compiler.  See below for macros that can be used

  11774.      to distinguish these cases.

  11775. 

  11776. 'DECL_ASSEMBLER_NAME'

  11777.      This macro returns the mangled name of the function, also an

  11778.      'IDENTIFIER_NODE'.  This name does not contain leading underscores

  11779.      on systems that prefix all identifiers with underscores.  The

  11780.      mangled name is computed in the same way on all platforms; if

  11781.      special processing is required to deal with the object file format

  11782.      used on a particular platform, it is the responsibility of the back

  11783.      end to perform those modifications.  (Of course, the back end

  11784.      should not modify 'DECL_ASSEMBLER_NAME' itself.)

  11785. 

  11786.      Using 'DECL_ASSEMBLER_NAME' will cause additional memory to be

  11787.      allocated (for the mangled name of the entity) so it should be used

  11788.      only when emitting assembly code.  It should not be used within the

  11789.      optimizers to determine whether or not two declarations are the

  11790.      same, even though some of the existing optimizers do use it in that

  11791.      way.  These uses will be removed over time.

  11792. 

  11793. 'DECL_ARGUMENTS'

  11794.      This macro returns the 'PARM_DECL' for the first argument to the

  11795.      function.  Subsequent 'PARM_DECL' nodes can be obtained by

  11796.      following the 'TREE_CHAIN' links.

  11797. 

  11798. 'DECL_RESULT'

  11799.      This macro returns the 'RESULT_DECL' for the function.

  11800. 

  11801. 'DECL_SAVED_TREE'

  11802.      This macro returns the complete body of the function.

  11803. 

  11804. 'TREE_TYPE'

  11805.      This macro returns the 'FUNCTION_TYPE' or 'METHOD_TYPE' for the

  11806.      function.

  11807. 

  11808. 'DECL_INITIAL'

  11809.      A function that has a definition in the current translation unit

  11810.      will have a non-'NULL' 'DECL_INITIAL'.  However, back ends should

  11811.      not make use of the particular value given by 'DECL_INITIAL'.

  11812. 

  11813.      It should contain a tree of 'BLOCK' nodes that mirrors the scopes

  11814.      that variables are bound in the function.  Each block contains a

  11815.      list of decls declared in a basic block, a pointer to a chain of

  11816.      blocks at the next lower scope level, then a pointer to the next

  11817.      block at the same level and a backpointer to the parent 'BLOCK' or

  11818.      'FUNCTION_DECL'.  So given a function as follows:

  11819. 

  11820.           void foo()

  11821.           {

  11822.             int a;

  11823.             {

  11824.               int b;

  11825.             }

  11826.             int c;

  11827.           }

  11828. 

  11829.      you would get the following:

  11830. 

  11831.           tree foo = FUNCTION_DECL;

  11832.           tree decl_a = VAR_DECL;

  11833.           tree decl_b = VAR_DECL;

  11834.           tree decl_c = VAR_DECL;

  11835.           tree block_a = BLOCK;

  11836.           tree block_b = BLOCK;

  11837.           tree block_c = BLOCK;

  11838.           BLOCK_VARS(block_a) = decl_a;

  11839.           BLOCK_SUBBLOCKS(block_a) = block_b;

  11840.           BLOCK_CHAIN(block_a) = block_c;

  11841.           BLOCK_SUPERCONTEXT(block_a) = foo;

  11842.           BLOCK_VARS(block_b) = decl_b;

  11843.           BLOCK_SUPERCONTEXT(block_b) = block_a;

  11844.           BLOCK_VARS(block_c) = decl_c;

  11845.           BLOCK_SUPERCONTEXT(block_c) = foo;

  11846.           DECL_INITIAL(foo) = block_a;

  11847. 

  11848. 

  11849. File: gccint.info,  Node: Function Properties,  Prev: Function Basics,  Up: Functions

  11850. 

  11851. 11.8.2 Function Properties

  11852. --------------------------

  11853. 

  11854. To determine the scope of a function, you can use the 'DECL_CONTEXT'

  11855. macro.  This macro will return the class (either a 'RECORD_TYPE' or a

  11856. 'UNION_TYPE') or namespace (a 'NAMESPACE_DECL') of which the function is

  11857. a member.  For a virtual function, this macro returns the class in which

  11858. the function was actually defined, not the base class in which the

  11859. virtual declaration occurred.

  11860. 

  11861.  In C, the 'DECL_CONTEXT' for a function maybe another function.  This

  11862. representation indicates that the GNU nested function extension is in

  11863. use.  For details on the semantics of nested functions, see the GCC

  11864. Manual.  The nested function can refer to local variables in its

  11865. containing function.  Such references are not explicitly marked in the

  11866. tree structure; back ends must look at the 'DECL_CONTEXT' for the

  11867. referenced 'VAR_DECL'.  If the 'DECL_CONTEXT' for the referenced

  11868. 'VAR_DECL' is not the same as the function currently being processed,

  11869. and neither 'DECL_EXTERNAL' nor 'TREE_STATIC' hold, then the reference

  11870. is to a local variable in a containing function, and the back end must

  11871. take appropriate action.

  11872. 

  11873. 'DECL_EXTERNAL'

  11874.      This predicate holds if the function is undefined.

  11875. 

  11876. 'TREE_PUBLIC'

  11877.      This predicate holds if the function has external linkage.

  11878. 

  11879. 'TREE_STATIC'

  11880.      This predicate holds if the function has been defined.

  11881. 

  11882. 'TREE_THIS_VOLATILE'

  11883.      This predicate holds if the function does not return normally.

  11884. 

  11885. 'TREE_READONLY'

  11886.      This predicate holds if the function can only read its arguments.

  11887. 

  11888. 'DECL_PURE_P'

  11889.      This predicate holds if the function can only read its arguments,

  11890.      but may also read global memory.

  11891. 

  11892. 'DECL_VIRTUAL_P'

  11893.      This predicate holds if the function is virtual.

  11894. 

  11895. 'DECL_ARTIFICIAL'

  11896.      This macro holds if the function was implicitly generated by the

  11897.      compiler, rather than explicitly declared.  In addition to

  11898.      implicitly generated class member functions, this macro holds for

  11899.      the special functions created to implement static initialization

  11900.      and destruction, to compute run-time type information, and so

  11901.      forth.

  11902. 

  11903. 'DECL_FUNCTION_SPECIFIC_TARGET'

  11904.      This macro returns a tree node that holds the target options that

  11905.      are to be used to compile this particular function or 'NULL_TREE'

  11906.      if the function is to be compiled with the target options specified

  11907.      on the command line.

  11908. 

  11909. 'DECL_FUNCTION_SPECIFIC_OPTIMIZATION'

  11910.      This macro returns a tree node that holds the optimization options

  11911.      that are to be used to compile this particular function or

  11912.      'NULL_TREE' if the function is to be compiled with the optimization

  11913.      options specified on the command line.

  11914. 

  11915. 

  11916. File: gccint.info,  Node: Language-dependent trees,  Next: C and C++ Trees,  Prev: Functions,  Up: GENERIC

  11917. 

  11918. 11.9 Language-dependent trees

  11919. =============================

  11920. 

  11921. Front ends may wish to keep some state associated with various GENERIC

  11922. trees while parsing.  To support this, trees provide a set of flags that

  11923. may be used by the front end.  They are accessed using

  11924. 'TREE_LANG_FLAG_n' where 'n' is currently 0 through 6.

  11925. 

  11926.  If necessary, a front end can use some language-dependent tree codes in

  11927. its GENERIC representation, so long as it provides a hook for converting

  11928. them to GIMPLE and doesn't expect them to work with any (hypothetical)

  11929. optimizers that run before the conversion to GIMPLE.  The intermediate

  11930. representation used while parsing C and C++ looks very little like

  11931. GENERIC, but the C and C++ gimplifier hooks are perfectly happy to take

  11932. it as input and spit out GIMPLE.

  11933. 

  11934. 

  11935. File: gccint.info,  Node: C and C++ Trees,  Prev: Language-dependent trees,  Up: GENERIC

  11936. 

  11937. 11.10 C and C++ Trees

  11938. =====================

  11939. 

  11940. This section documents the internal representation used by GCC to

  11941. represent C and C++ source programs.  When presented with a C or C++

  11942. source program, GCC parses the program, performs semantic analysis

  11943. (including the generation of error messages), and then produces the

  11944. internal representation described here.  This representation contains a

  11945. complete representation for the entire translation unit provided as

  11946. input to the front end.  This representation is then typically processed

  11947. by a code-generator in order to produce machine code, but could also be

  11948. used in the creation of source browsers, intelligent editors, automatic

  11949. documentation generators, interpreters, and any other programs needing

  11950. the ability to process C or C++ code.

  11951. 

  11952.  This section explains the internal representation.  In particular, it

  11953. documents the internal representation for C and C++ source constructs,

  11954. and the macros, functions, and variables that can be used to access

  11955. these constructs.  The C++ representation is largely a superset of the

  11956. representation used in the C front end.  There is only one construct

  11957. used in C that does not appear in the C++ front end and that is the GNU

  11958. "nested function" extension.  Many of the macros documented here do not

  11959. apply in C because the corresponding language constructs do not appear

  11960. in C.

  11961. 

  11962.  The C and C++ front ends generate a mix of GENERIC trees and ones

  11963. specific to C and C++.  These language-specific trees are higher-level

  11964. constructs than the ones in GENERIC to make the parser's job easier.

  11965. This section describes those trees that aren't part of GENERIC as well

  11966. as aspects of GENERIC trees that are treated in a language-specific

  11967. manner.

  11968. 

  11969.  If you are developing a "back end", be it is a code-generator or some

  11970. other tool, that uses this representation, you may occasionally find

  11971. that you need to ask questions not easily answered by the functions and

  11972. macros available here.  If that situation occurs, it is quite likely

  11973. that GCC already supports the functionality you desire, but that the

  11974. interface is simply not documented here.  In that case, you should ask

  11975. the GCC maintainers (via mail to <gcc@gcc.gnu.org>) about documenting

  11976. the functionality you require.  Similarly, if you find yourself writing

  11977. functions that do not deal directly with your back end, but instead

  11978. might be useful to other people using the GCC front end, you should

  11979. submit your patches for inclusion in GCC.

  11980. 

  11981. * Menu:

  11982. 

  11983. * Types for C++::               Fundamental and aggregate types.

  11984. * Namespaces::                  Namespaces.

  11985. * Classes::                     Classes.

  11986. * Functions for C++::           Overloading and accessors for C++.

  11987. * Statements for C++::          Statements specific to C and C++.

  11988. * C++ Expressions::    From 'typeid' to 'throw'.

  11989. 

  11990. 

  11991. File: gccint.info,  Node: Types for C++,  Next: Namespaces,  Up: C and C++ Trees

  11992. 

  11993. 11.10.1 Types for C++

  11994. ---------------------

  11995. 

  11996. In C++, an array type is not qualified; rather the type of the array

  11997. elements is qualified.  This situation is reflected in the intermediate

  11998. representation.  The macros described here will always examine the

  11999. qualification of the underlying element type when applied to an array

  12000. type.  (If the element type is itself an array, then the recursion

  12001. continues until a non-array type is found, and the qualification of this

  12002. type is examined.)  So, for example, 'CP_TYPE_CONST_P' will hold of the

  12003. type 'const int ()[7]', denoting an array of seven 'int's.

  12004. 

  12005.  The following functions and macros deal with cv-qualification of types:

  12006. 'cp_type_quals'

  12007.      This function returns the set of type qualifiers applied to this

  12008.      type.  This value is 'TYPE_UNQUALIFIED' if no qualifiers have been

  12009.      applied.  The 'TYPE_QUAL_CONST' bit is set if the type is

  12010.      'const'-qualified.  The 'TYPE_QUAL_VOLATILE' bit is set if the type

  12011.      is 'volatile'-qualified.  The 'TYPE_QUAL_RESTRICT' bit is set if

  12012.      the type is 'restrict'-qualified.

  12013. 

  12014. 'CP_TYPE_CONST_P'

  12015.      This macro holds if the type is 'const'-qualified.

  12016. 

  12017. 'CP_TYPE_VOLATILE_P'

  12018.      This macro holds if the type is 'volatile'-qualified.

  12019. 

  12020. 'CP_TYPE_RESTRICT_P'

  12021.      This macro holds if the type is 'restrict'-qualified.

  12022. 

  12023. 'CP_TYPE_CONST_NON_VOLATILE_P'

  12024.      This predicate holds for a type that is 'const'-qualified, but

  12025.      _not_ 'volatile'-qualified; other cv-qualifiers are ignored as

  12026.      well: only the 'const'-ness is tested.

  12027. 

  12028.  A few other macros and functions are usable with all types:

  12029. 'TYPE_SIZE'

  12030.      The number of bits required to represent the type, represented as

  12031.      an 'INTEGER_CST'.  For an incomplete type, 'TYPE_SIZE' will be

  12032.      'NULL_TREE'.

  12033. 

  12034. 'TYPE_ALIGN'

  12035.      The alignment of the type, in bits, represented as an 'int'.

  12036. 

  12037. 'TYPE_NAME'

  12038.      This macro returns a declaration (in the form of a 'TYPE_DECL') for

  12039.      the type.  (Note this macro does _not_ return an 'IDENTIFIER_NODE',

  12040.      as you might expect, given its name!)  You can look at the

  12041.      'DECL_NAME' of the 'TYPE_DECL' to obtain the actual name of the

  12042.      type.  The 'TYPE_NAME' will be 'NULL_TREE' for a type that is not a

  12043.      built-in type, the result of a typedef, or a named class type.

  12044. 

  12045. 'CP_INTEGRAL_TYPE'

  12046.      This predicate holds if the type is an integral type.  Notice that

  12047.      in C++, enumerations are _not_ integral types.

  12048. 

  12049. 'ARITHMETIC_TYPE_P'

  12050.      This predicate holds if the type is an integral type (in the C++

  12051.      sense) or a floating point type.

  12052. 

  12053. 'CLASS_TYPE_P'

  12054.      This predicate holds for a class-type.

  12055. 

  12056. 'TYPE_BUILT_IN'

  12057.      This predicate holds for a built-in type.

  12058. 

  12059. 'TYPE_PTRDATAMEM_P'

  12060.      This predicate holds if the type is a pointer to data member.

  12061. 

  12062. 'TYPE_PTR_P'

  12063.      This predicate holds if the type is a pointer type, and the pointee

  12064.      is not a data member.

  12065. 

  12066. 'TYPE_PTRFN_P'

  12067.      This predicate holds for a pointer to function type.

  12068. 

  12069. 'TYPE_PTROB_P'

  12070.      This predicate holds for a pointer to object type.  Note however

  12071.      that it does not hold for the generic pointer to object type 'void

  12072.      *'.  You may use 'TYPE_PTROBV_P' to test for a pointer to object

  12073.      type as well as 'void *'.

  12074. 

  12075.  The table below describes types specific to C and C++ as well as

  12076. language-dependent info about GENERIC types.

  12077. 

  12078. 'POINTER_TYPE'

  12079.      Used to represent pointer types, and pointer to data member types.

  12080.      If 'TREE_TYPE' is a pointer to data member type, then

  12081.      'TYPE_PTRDATAMEM_P' will hold.  For a pointer to data member type

  12082.      of the form 'T X::*', 'TYPE_PTRMEM_CLASS_TYPE' will be the type

  12083.      'X', while 'TYPE_PTRMEM_POINTED_TO_TYPE' will be the type 'T'.

  12084. 

  12085. 'RECORD_TYPE'

  12086.      Used to represent 'struct' and 'class' types in C and C++.  If

  12087.      'TYPE_PTRMEMFUNC_P' holds, then this type is a pointer-to-member

  12088.      type.  In that case, the 'TYPE_PTRMEMFUNC_FN_TYPE' is a

  12089.      'POINTER_TYPE' pointing to a 'METHOD_TYPE'.  The 'METHOD_TYPE' is

  12090.      the type of a function pointed to by the pointer-to-member

  12091.      function.  If 'TYPE_PTRMEMFUNC_P' does not hold, this type is a

  12092.      class type.  For more information, *note Classes::.

  12093. 

  12094. 'UNKNOWN_TYPE'

  12095.      This node is used to represent a type the knowledge of which is

  12096.      insufficient for a sound processing.

  12097. 

  12098. 'TYPENAME_TYPE'

  12099.      Used to represent a construct of the form 'typename T::A'.  The

  12100.      'TYPE_CONTEXT' is 'T'; the 'TYPE_NAME' is an 'IDENTIFIER_NODE' for

  12101.      'A'.  If the type is specified via a template-id, then

  12102.      'TYPENAME_TYPE_FULLNAME' yields a 'TEMPLATE_ID_EXPR'.  The

  12103.      'TREE_TYPE' is non-'NULL' if the node is implicitly generated in

  12104.      support for the implicit typename extension; in which case the

  12105.      'TREE_TYPE' is a type node for the base-class.

  12106. 

  12107. 'TYPEOF_TYPE'

  12108.      Used to represent the '__typeof__' extension.  The 'TYPE_FIELDS' is

  12109.      the expression the type of which is being represented.

  12110. 

  12111. 

  12112. File: gccint.info,  Node: Namespaces,  Next: Classes,  Prev: Types for C++,  Up: C and C++ Trees

  12113. 

  12114. 11.10.2 Namespaces

  12115. ------------------

  12116. 

  12117. The root of the entire intermediate representation is the variable

  12118. 'global_namespace'.  This is the namespace specified with '::' in C++

  12119. source code.  All other namespaces, types, variables, functions, and so

  12120. forth can be found starting with this namespace.

  12121. 

  12122.  However, except for the fact that it is distinguished as the root of

  12123. the representation, the global namespace is no different from any other

  12124. namespace.  Thus, in what follows, we describe namespaces generally,

  12125. rather than the global namespace in particular.

  12126. 

  12127.  A namespace is represented by a 'NAMESPACE_DECL' node.

  12128. 

  12129.  The following macros and functions can be used on a 'NAMESPACE_DECL':

  12130. 

  12131. 'DECL_NAME'

  12132.      This macro is used to obtain the 'IDENTIFIER_NODE' corresponding to

  12133.      the unqualified name of the name of the namespace (*note

  12134.      Identifiers::).  The name of the global namespace is '::', even

  12135.      though in C++ the global namespace is unnamed.  However, you should

  12136.      use comparison with 'global_namespace', rather than 'DECL_NAME' to

  12137.      determine whether or not a namespace is the global one.  An unnamed

  12138.      namespace will have a 'DECL_NAME' equal to

  12139.      'anonymous_namespace_name'.  Within a single translation unit, all

  12140.      unnamed namespaces will have the same name.

  12141. 

  12142. 'DECL_CONTEXT'

  12143.      This macro returns the enclosing namespace.  The 'DECL_CONTEXT' for

  12144.      the 'global_namespace' is 'NULL_TREE'.

  12145. 

  12146. 'DECL_NAMESPACE_ALIAS'

  12147.      If this declaration is for a namespace alias, then

  12148.      'DECL_NAMESPACE_ALIAS' is the namespace for which this one is an

  12149.      alias.

  12150. 

  12151.      Do not attempt to use 'cp_namespace_decls' for a namespace which is

  12152.      an alias.  Instead, follow 'DECL_NAMESPACE_ALIAS' links until you

  12153.      reach an ordinary, non-alias, namespace, and call

  12154.      'cp_namespace_decls' there.

  12155. 

  12156. 'DECL_NAMESPACE_STD_P'

  12157.      This predicate holds if the namespace is the special '::std'

  12158.      namespace.

  12159. 

  12160. 'cp_namespace_decls'

  12161.      This function will return the declarations contained in the

  12162.      namespace, including types, overloaded functions, other namespaces,

  12163.      and so forth.  If there are no declarations, this function will

  12164.      return 'NULL_TREE'.  The declarations are connected through their

  12165.      'TREE_CHAIN' fields.

  12166. 

  12167.      Although most entries on this list will be declarations,

  12168.      'TREE_LIST' nodes may also appear.  In this case, the 'TREE_VALUE'

  12169.      will be an 'OVERLOAD'.  The value of the 'TREE_PURPOSE' is

  12170.      unspecified; back ends should ignore this value.  As with the other

  12171.      kinds of declarations returned by 'cp_namespace_decls', the

  12172.      'TREE_CHAIN' will point to the next declaration in this list.

  12173. 

  12174.      For more information on the kinds of declarations that can occur on

  12175.      this list, *Note Declarations::.  Some declarations will not appear

  12176.      on this list.  In particular, no 'FIELD_DECL', 'LABEL_DECL', or

  12177.      'PARM_DECL' nodes will appear here.

  12178. 

  12179.      This function cannot be used with namespaces that have

  12180.      'DECL_NAMESPACE_ALIAS' set.

  12181. 

  12182. 

  12183. File: gccint.info,  Node: Classes,  Next: Functions for C++,  Prev: Namespaces,  Up: C and C++ Trees

  12184. 

  12185. 11.10.3 Classes

  12186. ---------------

  12187. 

  12188. Besides namespaces, the other high-level scoping construct in C++ is the

  12189. class.  (Throughout this manual the term "class" is used to mean the

  12190. types referred to in the ANSI/ISO C++ Standard as classes; these include

  12191. types defined with the 'class', 'struct', and 'union' keywords.)

  12192. 

  12193.  A class type is represented by either a 'RECORD_TYPE' or a

  12194. 'UNION_TYPE'.  A class declared with the 'union' tag is represented by a

  12195. 'UNION_TYPE', while classes declared with either the 'struct' or the

  12196. 'class' tag are represented by 'RECORD_TYPE's.  You can use the

  12197. 'CLASSTYPE_DECLARED_CLASS' macro to discern whether or not a particular

  12198. type is a 'class' as opposed to a 'struct'.  This macro will be true

  12199. only for classes declared with the 'class' tag.

  12200. 

  12201.  Almost all members are available on the 'TYPE_FIELDS' list.  Given one

  12202. member, the next can be found by following the 'TREE_CHAIN'.  You should

  12203. not depend in any way on the order in which fields appear on this list.

  12204. All nodes on this list will be 'DECL' nodes.  A 'FIELD_DECL' is used to

  12205. represent a non-static data member, a 'VAR_DECL' is used to represent a

  12206. static data member, and a 'TYPE_DECL' is used to represent a type.  Note

  12207. that the 'CONST_DECL' for an enumeration constant will appear on this

  12208. list, if the enumeration type was declared in the class.  (Of course,

  12209. the 'TYPE_DECL' for the enumeration type will appear here as well.)

  12210. There are no entries for base classes on this list.  In particular,

  12211. there is no 'FIELD_DECL' for the "base-class portion" of an object.  If

  12212. a function member is overloaded, each of the overloaded functions

  12213. appears; no 'OVERLOAD' nodes appear on the 'TYPE_FIELDS' list.

  12214. Implicitly declared functions (including default constructors, copy

  12215. constructors, assignment operators, and destructors) will appear on this

  12216. list as well.

  12217. 

  12218.  The 'TYPE_VFIELD' is a compiler-generated field used to point to

  12219. virtual function tables.  It may or may not appear on the 'TYPE_FIELDS'

  12220. list.  However, back ends should handle the 'TYPE_VFIELD' just like all

  12221. the entries on the 'TYPE_FIELDS' list.

  12222. 

  12223.  Every class has an associated "binfo", which can be obtained with

  12224. 'TYPE_BINFO'.  Binfos are used to represent base-classes.  The binfo

  12225. given by 'TYPE_BINFO' is the degenerate case, whereby every class is

  12226. considered to be its own base-class.  The base binfos for a particular

  12227. binfo are held in a vector, whose length is obtained with

  12228. 'BINFO_N_BASE_BINFOS'.  The base binfos themselves are obtained with

  12229. 'BINFO_BASE_BINFO' and 'BINFO_BASE_ITERATE'.  To add a new binfo, use

  12230. 'BINFO_BASE_APPEND'.  The vector of base binfos can be obtained with

  12231. 'BINFO_BASE_BINFOS', but normally you do not need to use that.  The

  12232. class type associated with a binfo is given by 'BINFO_TYPE'.  It is not

  12233. always the case that 'BINFO_TYPE (TYPE_BINFO (x))', because of typedefs

  12234. and qualified types.  Neither is it the case that 'TYPE_BINFO

  12235. (BINFO_TYPE (y))' is the same binfo as 'y'.  The reason is that if 'y'

  12236. is a binfo representing a base-class 'B' of a derived class 'D', then

  12237. 'BINFO_TYPE (y)' will be 'B', and 'TYPE_BINFO (BINFO_TYPE (y))' will be

  12238. 'B' as its own base-class, rather than as a base-class of 'D'.

  12239. 

  12240.  The access to a base type can be found with 'BINFO_BASE_ACCESS'.  This

  12241. will produce 'access_public_node', 'access_private_node' or

  12242. 'access_protected_node'.  If bases are always public,

  12243. 'BINFO_BASE_ACCESSES' may be 'NULL'.

  12244. 

  12245.  'BINFO_VIRTUAL_P' is used to specify whether the binfo is inherited

  12246. virtually or not.  The other flags, 'BINFO_FLAG_0' to 'BINFO_FLAG_6',

  12247. can be used for language specific use.

  12248. 

  12249.  The following macros can be used on a tree node representing a

  12250. class-type.

  12251. 

  12252. 'LOCAL_CLASS_P'

  12253.      This predicate holds if the class is local class _i.e._ declared

  12254.      inside a function body.

  12255. 

  12256. 'TYPE_POLYMORPHIC_P'

  12257.      This predicate holds if the class has at least one virtual function

  12258.      (declared or inherited).

  12259. 

  12260. 'TYPE_HAS_DEFAULT_CONSTRUCTOR'

  12261.      This predicate holds whenever its argument represents a class-type

  12262.      with default constructor.

  12263. 

  12264. 'CLASSTYPE_HAS_MUTABLE'

  12265. 'TYPE_HAS_MUTABLE_P'

  12266.      These predicates hold for a class-type having a mutable data

  12267.      member.

  12268. 

  12269. 'CLASSTYPE_NON_POD_P'

  12270.      This predicate holds only for class-types that are not PODs.

  12271. 

  12272. 'TYPE_HAS_NEW_OPERATOR'

  12273.      This predicate holds for a class-type that defines 'operator new'.

  12274. 

  12275. 'TYPE_HAS_ARRAY_NEW_OPERATOR'

  12276.      This predicate holds for a class-type for which 'operator new[]' is

  12277.      defined.

  12278. 

  12279. 'TYPE_OVERLOADS_CALL_EXPR'

  12280.      This predicate holds for class-type for which the function call

  12281.      'operator()' is overloaded.

  12282. 

  12283. 'TYPE_OVERLOADS_ARRAY_REF'

  12284.      This predicate holds for a class-type that overloads 'operator[]'

  12285. 

  12286. 'TYPE_OVERLOADS_ARROW'

  12287.      This predicate holds for a class-type for which 'operator->' is

  12288.      overloaded.

  12289. 

  12290. 

  12291. File: gccint.info,  Node: Functions for C++,  Next: Statements for C++,  Prev: Classes,  Up: C and C++ Trees

  12292. 

  12293. 11.10.4 Functions for C++

  12294. -------------------------

  12295. 

  12296. A function is represented by a 'FUNCTION_DECL' node.  A set of

  12297. overloaded functions is sometimes represented by an 'OVERLOAD' node.

  12298. 

  12299.  An 'OVERLOAD' node is not a declaration, so none of the 'DECL_' macros

  12300. should be used on an 'OVERLOAD'.  An 'OVERLOAD' node is similar to a

  12301. 'TREE_LIST'.  Use 'OVL_CURRENT' to get the function associated with an

  12302. 'OVERLOAD' node; use 'OVL_NEXT' to get the next 'OVERLOAD' node in the

  12303. list of overloaded functions.  The macros 'OVL_CURRENT' and 'OVL_NEXT'

  12304. are actually polymorphic; you can use them to work with 'FUNCTION_DECL'

  12305. nodes as well as with overloads.  In the case of a 'FUNCTION_DECL',

  12306. 'OVL_CURRENT' will always return the function itself, and 'OVL_NEXT'

  12307. will always be 'NULL_TREE'.

  12308. 

  12309.  To determine the scope of a function, you can use the 'DECL_CONTEXT'

  12310. macro.  This macro will return the class (either a 'RECORD_TYPE' or a

  12311. 'UNION_TYPE') or namespace (a 'NAMESPACE_DECL') of which the function is

  12312. a member.  For a virtual function, this macro returns the class in which

  12313. the function was actually defined, not the base class in which the

  12314. virtual declaration occurred.

  12315. 

  12316.  If a friend function is defined in a class scope, the

  12317. 'DECL_FRIEND_CONTEXT' macro can be used to determine the class in which

  12318. it was defined.  For example, in

  12319.      class C { friend void f() {} };

  12320. the 'DECL_CONTEXT' for 'f' will be the 'global_namespace', but the

  12321. 'DECL_FRIEND_CONTEXT' will be the 'RECORD_TYPE' for 'C'.

  12322. 

  12323.  The following macros and functions can be used on a 'FUNCTION_DECL':

  12324. 'DECL_MAIN_P'

  12325.      This predicate holds for a function that is the program entry point

  12326.      '::code'.

  12327. 

  12328. 'DECL_LOCAL_FUNCTION_P'

  12329.      This predicate holds if the function was declared at block scope,

  12330.      even though it has a global scope.

  12331. 

  12332. 'DECL_ANTICIPATED'

  12333.      This predicate holds if the function is a built-in function but its

  12334.      prototype is not yet explicitly declared.

  12335. 

  12336. 'DECL_EXTERN_C_FUNCTION_P'

  12337.      This predicate holds if the function is declared as an ''extern

  12338.      "C"'' function.

  12339. 

  12340. 'DECL_LINKONCE_P'

  12341.      This macro holds if multiple copies of this function may be emitted

  12342.      in various translation units.  It is the responsibility of the

  12343.      linker to merge the various copies.  Template instantiations are

  12344.      the most common example of functions for which 'DECL_LINKONCE_P'

  12345.      holds; G++ instantiates needed templates in all translation units

  12346.      which require them, and then relies on the linker to remove

  12347.      duplicate instantiations.

  12348. 

  12349.      FIXME: This macro is not yet implemented.

  12350. 

  12351. 'DECL_FUNCTION_MEMBER_P'

  12352.      This macro holds if the function is a member of a class, rather

  12353.      than a member of a namespace.

  12354. 

  12355. 'DECL_STATIC_FUNCTION_P'

  12356.      This predicate holds if the function a static member function.

  12357. 

  12358. 'DECL_NONSTATIC_MEMBER_FUNCTION_P'

  12359.      This macro holds for a non-static member function.

  12360. 

  12361. 'DECL_CONST_MEMFUNC_P'

  12362.      This predicate holds for a 'const'-member function.

  12363. 

  12364. 'DECL_VOLATILE_MEMFUNC_P'

  12365.      This predicate holds for a 'volatile'-member function.

  12366. 

  12367. 'DECL_CONSTRUCTOR_P'

  12368.      This macro holds if the function is a constructor.

  12369. 

  12370. 'DECL_NONCONVERTING_P'

  12371.      This predicate holds if the constructor is a non-converting

  12372.      constructor.

  12373. 

  12374. 'DECL_COMPLETE_CONSTRUCTOR_P'

  12375.      This predicate holds for a function which is a constructor for an

  12376.      object of a complete type.

  12377. 

  12378. 'DECL_BASE_CONSTRUCTOR_P'

  12379.      This predicate holds for a function which is a constructor for a

  12380.      base class sub-object.

  12381. 

  12382. 'DECL_COPY_CONSTRUCTOR_P'

  12383.      This predicate holds for a function which is a copy-constructor.

  12384. 

  12385. 'DECL_DESTRUCTOR_P'

  12386.      This macro holds if the function is a destructor.

  12387. 

  12388. 'DECL_COMPLETE_DESTRUCTOR_P'

  12389.      This predicate holds if the function is the destructor for an

  12390.      object a complete type.

  12391. 

  12392. 'DECL_OVERLOADED_OPERATOR_P'

  12393.      This macro holds if the function is an overloaded operator.

  12394. 

  12395. 'DECL_CONV_FN_P'

  12396.      This macro holds if the function is a type-conversion operator.

  12397. 

  12398. 'DECL_GLOBAL_CTOR_P'

  12399.      This predicate holds if the function is a file-scope initialization

  12400.      function.

  12401. 

  12402. 'DECL_GLOBAL_DTOR_P'

  12403.      This predicate holds if the function is a file-scope finalization

  12404.      function.

  12405. 

  12406. 'DECL_THUNK_P'

  12407.      This predicate holds if the function is a thunk.

  12408. 

  12409.      These functions represent stub code that adjusts the 'this' pointer

  12410.      and then jumps to another function.  When the jumped-to function

  12411.      returns, control is transferred directly to the caller, without

  12412.      returning to the thunk.  The first parameter to the thunk is always

  12413.      the 'this' pointer; the thunk should add 'THUNK_DELTA' to this

  12414.      value.  (The 'THUNK_DELTA' is an 'int', not an 'INTEGER_CST'.)

  12415. 

  12416.      Then, if 'THUNK_VCALL_OFFSET' (an 'INTEGER_CST') is nonzero the

  12417.      adjusted 'this' pointer must be adjusted again.  The complete

  12418.      calculation is given by the following pseudo-code:

  12419. 

  12420.           this += THUNK_DELTA

  12421.           if (THUNK_VCALL_OFFSET)

  12422.             this += (*((ptrdiff_t **) this))[THUNK_VCALL_OFFSET]

  12423. 

  12424.      Finally, the thunk should jump to the location given by

  12425.      'DECL_INITIAL'; this will always be an expression for the address

  12426.      of a function.

  12427. 

  12428. 'DECL_NON_THUNK_FUNCTION_P'

  12429.      This predicate holds if the function is _not_ a thunk function.

  12430. 

  12431. 'GLOBAL_INIT_PRIORITY'

  12432.      If either 'DECL_GLOBAL_CTOR_P' or 'DECL_GLOBAL_DTOR_P' holds, then

  12433.      this gives the initialization priority for the function.  The

  12434.      linker will arrange that all functions for which

  12435.      'DECL_GLOBAL_CTOR_P' holds are run in increasing order of priority

  12436.      before 'main' is called.  When the program exits, all functions for

  12437.      which 'DECL_GLOBAL_DTOR_P' holds are run in the reverse order.

  12438. 

  12439. 'TYPE_RAISES_EXCEPTIONS'

  12440.      This macro returns the list of exceptions that a (member-)function

  12441.      can raise.  The returned list, if non 'NULL', is comprised of nodes

  12442.      whose 'TREE_VALUE' represents a type.

  12443. 

  12444. 'TYPE_NOTHROW_P'

  12445.      This predicate holds when the exception-specification of its

  12446.      arguments is of the form ''()''.

  12447. 

  12448. 'DECL_ARRAY_DELETE_OPERATOR_P'

  12449.      This predicate holds if the function an overloaded 'operator

  12450.      delete[]'.

  12451. 

  12452. 

  12453. File: gccint.info,  Node: Statements for C++,  Next: C++ Expressions,  Prev: Functions for C++,  Up: C and C++ Trees

  12454. 

  12455. 11.10.5 Statements for C++

  12456. --------------------------

  12457. 

  12458. A function that has a definition in the current translation unit will

  12459. have a non-'NULL' 'DECL_INITIAL'.  However, back ends should not make

  12460. use of the particular value given by 'DECL_INITIAL'.

  12461. 

  12462.  The 'DECL_SAVED_TREE' macro will give the complete body of the

  12463. function.

  12464. 

  12465. 11.10.5.1 Statements

  12466. ....................

  12467. 

  12468. There are tree nodes corresponding to all of the source-level statement

  12469. constructs, used within the C and C++ frontends.  These are enumerated

  12470. here, together with a list of the various macros that can be used to

  12471. obtain information about them.  There are a few macros that can be used

  12472. with all statements:

  12473. 

  12474. 'STMT_IS_FULL_EXPR_P'

  12475.      In C++, statements normally constitute "full expressions";

  12476.      temporaries created during a statement are destroyed when the

  12477.      statement is complete.  However, G++ sometimes represents

  12478.      expressions by statements; these statements will not have

  12479.      'STMT_IS_FULL_EXPR_P' set.  Temporaries created during such

  12480.      statements should be destroyed when the innermost enclosing

  12481.      statement with 'STMT_IS_FULL_EXPR_P' set is exited.

  12482. 

  12483.  Here is the list of the various statement nodes, and the macros used to

  12484. access them.  This documentation describes the use of these nodes in

  12485. non-template functions (including instantiations of template functions).

  12486. In template functions, the same nodes are used, but sometimes in

  12487. slightly different ways.

  12488. 

  12489.  Many of the statements have substatements.  For example, a 'while' loop

  12490. will have a body, which is itself a statement.  If the substatement is

  12491. 'NULL_TREE', it is considered equivalent to a statement consisting of a

  12492. single ';', i.e., an expression statement in which the expression has

  12493. been omitted.  A substatement may in fact be a list of statements,

  12494. connected via their 'TREE_CHAIN's.  So, you should always process the

  12495. statement tree by looping over substatements, like this:

  12496.      void process_stmt (stmt)

  12497.           tree stmt;

  12498.      {

  12499.        while (stmt)

  12500.          {

  12501.            switch (TREE_CODE (stmt))

  12502.              {

  12503.              case IF_STMT:

  12504.                process_stmt (THEN_CLAUSE (stmt));

  12505.                /* More processing here.  */

  12506.                break;

  12507. 

  12508.              ...

  12509.              }

  12510. 

  12511.            stmt = TREE_CHAIN (stmt);

  12512.          }

  12513.      }

  12514.  In other words, while the 'then' clause of an 'if' statement in C++ can

  12515. be only one statement (although that one statement may be a compound

  12516. statement), the intermediate representation will sometimes use several

  12517. statements chained together.

  12518. 

  12519. 'BREAK_STMT'

  12520. 

  12521.      Used to represent a 'break' statement.  There are no additional

  12522.      fields.

  12523. 

  12524. 'CLEANUP_STMT'

  12525. 

  12526.      Used to represent an action that should take place upon exit from

  12527.      the enclosing scope.  Typically, these actions are calls to

  12528.      destructors for local objects, but back ends cannot rely on this

  12529.      fact.  If these nodes are in fact representing such destructors,

  12530.      'CLEANUP_DECL' will be the 'VAR_DECL' destroyed.  Otherwise,

  12531.      'CLEANUP_DECL' will be 'NULL_TREE'.  In any case, the

  12532.      'CLEANUP_EXPR' is the expression to execute.  The cleanups executed

  12533.      on exit from a scope should be run in the reverse order of the

  12534.      order in which the associated 'CLEANUP_STMT's were encountered.

  12535. 

  12536. 'CONTINUE_STMT'

  12537. 

  12538.      Used to represent a 'continue' statement.  There are no additional

  12539.      fields.

  12540. 

  12541. 'CTOR_STMT'

  12542. 

  12543.      Used to mark the beginning (if 'CTOR_BEGIN_P' holds) or end (if

  12544.      'CTOR_END_P' holds of the main body of a constructor.  See also

  12545.      'SUBOBJECT' for more information on how to use these nodes.

  12546. 

  12547. 'DO_STMT'

  12548. 

  12549.      Used to represent a 'do' loop.  The body of the loop is given by

  12550.      'DO_BODY' while the termination condition for the loop is given by

  12551.      'DO_COND'.  The condition for a 'do'-statement is always an

  12552.      expression.

  12553. 

  12554. 'EMPTY_CLASS_EXPR'

  12555. 

  12556.      Used to represent a temporary object of a class with no data whose

  12557.      address is never taken.  (All such objects are interchangeable.)

  12558.      The 'TREE_TYPE' represents the type of the object.

  12559. 

  12560. 'EXPR_STMT'

  12561. 

  12562.      Used to represent an expression statement.  Use 'EXPR_STMT_EXPR' to

  12563.      obtain the expression.

  12564. 

  12565. 'FOR_STMT'

  12566. 

  12567.      Used to represent a 'for' statement.  The 'FOR_INIT_STMT' is the

  12568.      initialization statement for the loop.  The 'FOR_COND' is the

  12569.      termination condition.  The 'FOR_EXPR' is the expression executed

  12570.      right before the 'FOR_COND' on each loop iteration; often, this

  12571.      expression increments a counter.  The body of the loop is given by

  12572.      'FOR_BODY'.  Note that 'FOR_INIT_STMT' and 'FOR_BODY' return

  12573.      statements, while 'FOR_COND' and 'FOR_EXPR' return expressions.

  12574. 

  12575. 'HANDLER'

  12576. 

  12577.      Used to represent a C++ 'catch' block.  The 'HANDLER_TYPE' is the

  12578.      type of exception that will be caught by this handler; it is equal

  12579.      (by pointer equality) to 'NULL' if this handler is for all types.

  12580.      'HANDLER_PARMS' is the 'DECL_STMT' for the catch parameter, and

  12581.      'HANDLER_BODY' is the code for the block itself.

  12582. 

  12583. 'IF_STMT'

  12584. 

  12585.      Used to represent an 'if' statement.  The 'IF_COND' is the

  12586.      expression.

  12587. 

  12588.      If the condition is a 'TREE_LIST', then the 'TREE_PURPOSE' is a

  12589.      statement (usually a 'DECL_STMT').  Each time the condition is

  12590.      evaluated, the statement should be executed.  Then, the

  12591.      'TREE_VALUE' should be used as the conditional expression itself.

  12592.      This representation is used to handle C++ code like this:

  12593. 

  12594.      C++ distinguishes between this and 'COND_EXPR' for handling

  12595.      templates.

  12596. 

  12597.           if (int i = 7) ...

  12598. 

  12599.      where there is a new local variable (or variables) declared within

  12600.      the condition.

  12601. 

  12602.      The 'THEN_CLAUSE' represents the statement given by the 'then'

  12603.      condition, while the 'ELSE_CLAUSE' represents the statement given

  12604.      by the 'else' condition.

  12605. 

  12606. 'SUBOBJECT'

  12607. 

  12608.      In a constructor, these nodes are used to mark the point at which a

  12609.      subobject of 'this' is fully constructed.  If, after this point, an

  12610.      exception is thrown before a 'CTOR_STMT' with 'CTOR_END_P' set is

  12611.      encountered, the 'SUBOBJECT_CLEANUP' must be executed.  The

  12612.      cleanups must be executed in the reverse order in which they

  12613.      appear.

  12614. 

  12615. 'SWITCH_STMT'

  12616. 

  12617.      Used to represent a 'switch' statement.  The 'SWITCH_STMT_COND' is

  12618.      the expression on which the switch is occurring.  See the

  12619.      documentation for an 'IF_STMT' for more information on the

  12620.      representation used for the condition.  The 'SWITCH_STMT_BODY' is

  12621.      the body of the switch statement.  The 'SWITCH_STMT_TYPE' is the

  12622.      original type of switch expression as given in the source, before

  12623.      any compiler conversions.

  12624. 

  12625. 'TRY_BLOCK'

  12626.      Used to represent a 'try' block.  The body of the try block is

  12627.      given by 'TRY_STMTS'.  Each of the catch blocks is a 'HANDLER'

  12628.      node.  The first handler is given by 'TRY_HANDLERS'.  Subsequent

  12629.      handlers are obtained by following the 'TREE_CHAIN' link from one

  12630.      handler to the next.  The body of the handler is given by

  12631.      'HANDLER_BODY'.

  12632. 

  12633.      If 'CLEANUP_P' holds of the 'TRY_BLOCK', then the 'TRY_HANDLERS'

  12634.      will not be a 'HANDLER' node.  Instead, it will be an expression

  12635.      that should be executed if an exception is thrown in the try block.

  12636.      It must rethrow the exception after executing that code.  And, if

  12637.      an exception is thrown while the expression is executing,

  12638.      'terminate' must be called.

  12639. 

  12640. 'USING_STMT'

  12641.      Used to represent a 'using' directive.  The namespace is given by

  12642.      'USING_STMT_NAMESPACE', which will be a NAMESPACE_DECL.  This node

  12643.      is needed inside template functions, to implement using directives

  12644.      during instantiation.

  12645. 

  12646. 'WHILE_STMT'

  12647. 

  12648.      Used to represent a 'while' loop.  The 'WHILE_COND' is the

  12649.      termination condition for the loop.  See the documentation for an

  12650.      'IF_STMT' for more information on the representation used for the

  12651.      condition.

  12652. 

  12653.      The 'WHILE_BODY' is the body of the loop.

  12654. 

  12655. 

  12656. File: gccint.info,  Node: C++ Expressions,  Prev: Statements for C++,  Up: C and C++ Trees

  12657. 

  12658. 11.10.6 C++ Expressions

  12659. -----------------------

  12660. 

  12661. This section describes expressions specific to the C and C++ front ends.

  12662. 

  12663. 'TYPEID_EXPR'

  12664. 

  12665.      Used to represent a 'typeid' expression.

  12666. 

  12667. 'NEW_EXPR'

  12668. 'VEC_NEW_EXPR'

  12669. 

  12670.      Used to represent a call to 'new' and 'new[]' respectively.

  12671. 

  12672. 'DELETE_EXPR'

  12673. 'VEC_DELETE_EXPR'

  12674. 

  12675.      Used to represent a call to 'delete' and 'delete[]' respectively.

  12676. 

  12677. 'MEMBER_REF'

  12678. 

  12679.      Represents a reference to a member of a class.

  12680. 

  12681. 'THROW_EXPR'

  12682. 

  12683.      Represents an instance of 'throw' in the program.  Operand 0, which

  12684.      is the expression to throw, may be 'NULL_TREE'.

  12685. 

  12686. 'AGGR_INIT_EXPR'

  12687.      An 'AGGR_INIT_EXPR' represents the initialization as the return

  12688.      value of a function call, or as the result of a constructor.  An

  12689.      'AGGR_INIT_EXPR' will only appear as a full-expression, or as the

  12690.      second operand of a 'TARGET_EXPR'.  'AGGR_INIT_EXPR's have a

  12691.      representation similar to that of 'CALL_EXPR's.  You can use the

  12692.      'AGGR_INIT_EXPR_FN' and 'AGGR_INIT_EXPR_ARG' macros to access the

  12693.      function to call and the arguments to pass.

  12694. 

  12695.      If 'AGGR_INIT_VIA_CTOR_P' holds of the 'AGGR_INIT_EXPR', then the

  12696.      initialization is via a constructor call.  The address of the

  12697.      'AGGR_INIT_EXPR_SLOT' operand, which is always a 'VAR_DECL', is

  12698.      taken, and this value replaces the first argument in the argument

  12699.      list.

  12700. 

  12701.      In either case, the expression is void.

  12702. 

  12703. 

  12704. File: gccint.info,  Node: GIMPLE,  Next: Tree SSA,  Prev: GENERIC,  Up: Top

  12705. 

  12706. 12 GIMPLE

  12707. *********

  12708. 

  12709. GIMPLE is a three-address representation derived from GENERIC by

  12710. breaking down GENERIC expressions into tuples of no more than 3 operands

  12711. (with some exceptions like function calls).  GIMPLE was heavily

  12712. influenced by the SIMPLE IL used by the McCAT compiler project at McGill

  12713. University, though we have made some different choices.  For one thing,

  12714. SIMPLE doesn't support 'goto'.

  12715. 

  12716.  Temporaries are introduced to hold intermediate values needed to

  12717. compute complex expressions.  Additionally, all the control structures

  12718. used in GENERIC are lowered into conditional jumps, lexical scopes are

  12719. removed and exception regions are converted into an on the side

  12720. exception region tree.

  12721. 

  12722.  The compiler pass which converts GENERIC into GIMPLE is referred to as

  12723. the 'gimplifier'.  The gimplifier works recursively, generating GIMPLE

  12724. tuples out of the original GENERIC expressions.

  12725. 

  12726.  One of the early implementation strategies used for the GIMPLE

  12727. representation was to use the same internal data structures used by

  12728. front ends to represent parse trees.  This simplified implementation

  12729. because we could leverage existing functionality and interfaces.

  12730. However, GIMPLE is a much more restrictive representation than abstract

  12731. syntax trees (AST), therefore it does not require the full structural

  12732. complexity provided by the main tree data structure.

  12733. 

  12734.  The GENERIC representation of a function is stored in the

  12735. 'DECL_SAVED_TREE' field of the associated 'FUNCTION_DECL' tree node.  It

  12736. is converted to GIMPLE by a call to 'gimplify_function_tree'.

  12737. 

  12738.  If a front end wants to include language-specific tree codes in the

  12739. tree representation which it provides to the back end, it must provide a

  12740. definition of 'LANG_HOOKS_GIMPLIFY_EXPR' which knows how to convert the

  12741. front end trees to GIMPLE.  Usually such a hook will involve much of the

  12742. same code for expanding front end trees to RTL.  This function can

  12743. return fully lowered GIMPLE, or it can return GENERIC trees and let the

  12744. main gimplifier lower them the rest of the way; this is often simpler.

  12745. GIMPLE that is not fully lowered is known as "High GIMPLE" and consists

  12746. of the IL before the pass 'pass_lower_cf'.  High GIMPLE contains some

  12747. container statements like lexical scopes (represented by 'GIMPLE_BIND')

  12748. and nested expressions (e.g., 'GIMPLE_TRY'), while "Low GIMPLE" exposes

  12749. all of the implicit jumps for control and exception expressions directly

  12750. in the IL and EH region trees.

  12751. 

  12752.  The C and C++ front ends currently convert directly from front end

  12753. trees to GIMPLE, and hand that off to the back end rather than first

  12754. converting to GENERIC.  Their gimplifier hooks know about all the

  12755. '_STMT' nodes and how to convert them to GENERIC forms.  There was some

  12756. work done on a genericization pass which would run first, but the

  12757. existence of 'STMT_EXPR' meant that in order to convert all of the C

  12758. statements into GENERIC equivalents would involve walking the entire

  12759. tree anyway, so it was simpler to lower all the way.  This might change

  12760. in the future if someone writes an optimization pass which would work

  12761. better with higher-level trees, but currently the optimizers all expect

  12762. GIMPLE.

  12763. 

  12764.  You can request to dump a C-like representation of the GIMPLE form with

  12765. the flag '-fdump-tree-gimple'.

  12766. 

  12767. * Menu:

  12768. 

  12769. * Tuple representation::

  12770. * Class hierarchy of GIMPLE statements::

  12771. * GIMPLE instruction set::

  12772. * GIMPLE Exception Handling::

  12773. * Temporaries::

  12774. * Operands::

  12775. * Manipulating GIMPLE statements::

  12776. * Tuple specific accessors::

  12777. * GIMPLE sequences::

  12778. * Sequence iterators::

  12779. * Adding a new GIMPLE statement code::

  12780. * Statement and operand traversals::

  12781. 

  12782. 

  12783. File: gccint.info,  Node: Tuple representation,  Next: Class hierarchy of GIMPLE statements,  Up: GIMPLE

  12784. 

  12785. 12.1 Tuple representation

  12786. =========================

  12787. 

  12788. GIMPLE instructions are tuples of variable size divided in two groups: a

  12789. header describing the instruction and its locations, and a variable

  12790. length body with all the operands.  Tuples are organized into a

  12791. hierarchy with 3 main classes of tuples.

  12792. 

  12793. 12.1.1 'gimple' (gsbase)

  12794. ------------------------

  12795. 

  12796. This is the root of the hierarchy, it holds basic information needed by

  12797. most GIMPLE statements.  There are some fields that may not be relevant

  12798. to every GIMPLE statement, but those were moved into the base structure

  12799. to take advantage of holes left by other fields (thus making the

  12800. structure more compact).  The structure takes 4 words (32 bytes) on 64

  12801. bit hosts:

  12802. 

  12803. Field                   Size (bits)

  12804. 'code'                  8

  12805. 'subcode'               16

  12806. 'no_warning'            1

  12807. 'visited'               1

  12808. 'nontemporal_move'      1

  12809. 'plf'                   2

  12810. 'modified'              1

  12811. 'has_volatile_ops'      1

  12812. 'references_memory_p'   1

  12813. 'uid'                   32

  12814. 'location'              32

  12815. 'num_ops'               32

  12816. 'bb'                    64

  12817. 'block'                 63

  12818. Total size              32 bytes

  12819. 

  12820.    * 'code' Main identifier for a GIMPLE instruction.

  12821. 

  12822.    * 'subcode' Used to distinguish different variants of the same basic

  12823.      instruction or provide flags applicable to a given code.  The

  12824.      'subcode' flags field has different uses depending on the code of

  12825.      the instruction, but mostly it distinguishes instructions of the

  12826.      same family.  The most prominent use of this field is in

  12827.      assignments, where subcode indicates the operation done on the RHS

  12828.      of the assignment.  For example, a = b + c is encoded as

  12829.      'GIMPLE_ASSIGN <PLUS_EXPR, a, b, c>'.

  12830. 

  12831.    * 'no_warning' Bitflag to indicate whether a warning has already been

  12832.      issued on this statement.

  12833. 

  12834.    * 'visited' General purpose "visited" marker.  Set and cleared by

  12835.      each pass when needed.

  12836. 

  12837.    * 'nontemporal_move' Bitflag used in assignments that represent

  12838.      non-temporal moves.  Although this bitflag is only used in

  12839.      assignments, it was moved into the base to take advantage of the

  12840.      bit holes left by the previous fields.

  12841. 

  12842.    * 'plf' Pass Local Flags.  This 2-bit mask can be used as general

  12843.      purpose markers by any pass.  Passes are responsible for clearing

  12844.      and setting these two flags accordingly.

  12845. 

  12846.    * 'modified' Bitflag to indicate whether the statement has been

  12847.      modified.  Used mainly by the operand scanner to determine when to

  12848.      re-scan a statement for operands.

  12849. 

  12850.    * 'has_volatile_ops' Bitflag to indicate whether this statement

  12851.      contains operands that have been marked volatile.

  12852. 

  12853.    * 'references_memory_p' Bitflag to indicate whether this statement

  12854.      contains memory references (i.e., its operands are either global

  12855.      variables, or pointer dereferences or anything that must reside in

  12856.      memory).

  12857. 

  12858.    * 'uid' This is an unsigned integer used by passes that want to

  12859.      assign IDs to every statement.  These IDs must be assigned and used

  12860.      by each pass.

  12861. 

  12862.    * 'location' This is a 'location_t' identifier to specify source code

  12863.      location for this statement.  It is inherited from the front end.

  12864. 

  12865.    * 'num_ops' Number of operands that this statement has.  This

  12866.      specifies the size of the operand vector embedded in the tuple.

  12867.      Only used in some tuples, but it is declared in the base tuple to

  12868.      take advantage of the 32-bit hole left by the previous fields.

  12869. 

  12870.    * 'bb' Basic block holding the instruction.

  12871. 

  12872.    * 'block' Lexical block holding this statement.  Also used for debug

  12873.      information generation.

  12874. 

  12875. 12.1.2 'gimple_statement_with_ops'

  12876. ----------------------------------

  12877. 

  12878. This tuple is actually split in two: 'gimple_statement_with_ops_base'

  12879. and 'gimple_statement_with_ops'.  This is needed to accommodate the way

  12880. the operand vector is allocated.  The operand vector is defined to be an

  12881. array of 1 element.  So, to allocate a dynamic number of operands, the

  12882. memory allocator ('gimple_alloc') simply allocates enough memory to hold

  12883. the structure itself plus 'N - 1' operands which run "off the end" of

  12884. the structure.  For example, to allocate space for a tuple with 3

  12885. operands, 'gimple_alloc' reserves 'sizeof (struct

  12886. gimple_statement_with_ops) + 2 * sizeof (tree)' bytes.

  12887. 

  12888.  On the other hand, several fields in this tuple need to be shared with

  12889. the 'gimple_statement_with_memory_ops' tuple.  So, these common fields

  12890. are placed in 'gimple_statement_with_ops_base' which is then inherited

  12891. from the other two tuples.

  12892. 

  12893. 'gsbase'    256

  12894. 'def_ops'   64

  12895. 'use_ops'   64

  12896. 'op'        'num_ops' * 64

  12897. Total       48 + 8 * 'num_ops' bytes

  12898. size

  12899. 

  12900.    * 'gsbase' Inherited from 'struct gimple'.

  12901. 

  12902.    * 'def_ops' Array of pointers into the operand array indicating all

  12903.      the slots that contain a variable written-to by the statement.

  12904.      This array is also used for immediate use chaining.  Note that it

  12905.      would be possible to not rely on this array, but the changes

  12906.      required to implement this are pretty invasive.

  12907. 

  12908.    * 'use_ops' Similar to 'def_ops' but for variables read by the

  12909.      statement.

  12910. 

  12911.    * 'op' Array of trees with 'num_ops' slots.

  12912. 

  12913. 12.1.3 'gimple_statement_with_memory_ops'

  12914. -----------------------------------------

  12915. 

  12916. This tuple is essentially identical to 'gimple_statement_with_ops',

  12917. except that it contains 4 additional fields to hold vectors related

  12918. memory stores and loads.  Similar to the previous case, the structure is

  12919. split in two to accommodate for the operand vector

  12920. ('gimple_statement_with_memory_ops_base' and

  12921. 'gimple_statement_with_memory_ops').

  12922. 

  12923. Field        Size (bits)

  12924. 'gsbase'     256

  12925. 'def_ops'    64

  12926. 'use_ops'    64

  12927. 'vdef_ops'   64

  12928. 'vuse_ops'   64

  12929. 'stores'     64

  12930. 'loads'      64

  12931. 'op'         'num_ops' * 64

  12932. Total size   80 + 8 * 'num_ops' bytes

  12933. 

  12934.    * 'vdef_ops' Similar to 'def_ops' but for 'VDEF' operators.  There is

  12935.      one entry per memory symbol written by this statement.  This is

  12936.      used to maintain the memory SSA use-def and def-def chains.

  12937. 

  12938.    * 'vuse_ops' Similar to 'use_ops' but for 'VUSE' operators.  There is

  12939.      one entry per memory symbol loaded by this statement.  This is used

  12940.      to maintain the memory SSA use-def chains.

  12941. 

  12942.    * 'stores' Bitset with all the UIDs for the symbols written-to by the

  12943.      statement.  This is different than 'vdef_ops' in that all the

  12944.      affected symbols are mentioned in this set.  If memory partitioning

  12945.      is enabled, the 'vdef_ops' vector will refer to memory partitions.

  12946.      Furthermore, no SSA information is stored in this set.

  12947. 

  12948.    * 'loads' Similar to 'stores', but for memory loads.  (Note that

  12949.      there is some amount of redundancy here, it should be possible to

  12950.      reduce memory utilization further by removing these sets).

  12951. 

  12952.  All the other tuples are defined in terms of these three basic ones.

  12953. Each tuple will add some fields.

  12954. 

  12955. 

  12956. File: gccint.info,  Node: Class hierarchy of GIMPLE statements,  Next: GIMPLE instruction set,  Prev: Tuple representation,  Up: GIMPLE

  12957. 

  12958. 12.2 Class hierarchy of GIMPLE statements

  12959. =========================================

  12960. 

  12961. The following diagram shows the C++ inheritance hierarchy of statement

  12962. kinds, along with their relationships to 'GSS_' values (layouts) and

  12963. 'GIMPLE_' values (codes):

  12964. 

  12965.         gimple

  12966.           |    layout: GSS_BASE

  12967.           |    used for 4 codes: GIMPLE_ERROR_MARK

  12968.           |                      GIMPLE_NOP

  12969.           |                      GIMPLE_OMP_SECTIONS_SWITCH

  12970.           |                      GIMPLE_PREDICT

  12971.           |

  12972.           + gimple_statement_with_ops_base

  12973.           |   |    (no GSS layout)

  12974.           |   |

  12975.           |   + gimple_statement_with_ops

  12976.           |   |   |    layout: GSS_WITH_OPS

  12977.           |   |   |

  12978.           |   |   + gcond

  12979.           |   |   |     code: GIMPLE_COND

  12980.           |   |   |

  12981.           |   |   + gdebug

  12982.           |   |   |     code: GIMPLE_DEBUG

  12983.           |   |   |

  12984.           |   |   + ggoto

  12985.           |   |   |     code: GIMPLE_GOTO

  12986.           |   |   |

  12987.           |   |   + glabel

  12988.           |   |   |     code: GIMPLE_LABEL

  12989.           |   |   |

  12990.           |   |   + gswitch

  12991.           |   |         code: GIMPLE_SWITCH

  12992.           |   |

  12993.           |   + gimple_statement_with_memory_ops_base

  12994.           |       |    layout: GSS_WITH_MEM_OPS_BASE

  12995.           |       |

  12996.           |       + gimple_statement_with_memory_ops

  12997.           |       |   |    layout: GSS_WITH_MEM_OPS

  12998.           |       |   |

  12999.           |       |   + gassign

  13000.           |       |   |    code GIMPLE_ASSIGN

  13001.           |       |   |

  13002.           |       |   + greturn

  13003.           |       |        code GIMPLE_RETURN

  13004.           |       |

  13005.           |       + gcall

  13006.           |       |        layout: GSS_CALL, code: GIMPLE_CALL

  13007.           |       |

  13008.           |       + gasm

  13009.           |       |        layout: GSS_ASM, code: GIMPLE_ASM

  13010.           |       |

  13011.           |       + gtransaction

  13012.           |                layout: GSS_TRANSACTION, code: GIMPLE_TRANSACTION

  13013.           |

  13014.           + gimple_statement_omp

  13015.           |   |    layout: GSS_OMP.  Used for code GIMPLE_OMP_SECTION

  13016.           |   |

  13017.           |   + gomp_critical

  13018.           |   |        layout: GSS_OMP_CRITICAL, code: GIMPLE_OMP_CRITICAL

  13019.           |   |

  13020.           |   + gomp_for

  13021.           |   |        layout: GSS_OMP_FOR, code: GIMPLE_OMP_FOR

  13022.           |   |

  13023.           |   + gomp_parallel_layout

  13024.           |   |   |    layout: GSS_OMP_PARALLEL_LAYOUT

  13025.           |   |   |

  13026.           |   |   + gimple_statement_omp_taskreg

  13027.           |   |   |   |

  13028.           |   |   |   + gomp_parallel

  13029.           |   |   |   |        code: GIMPLE_OMP_PARALLEL

  13030.           |   |   |   |

  13031.           |   |   |   + gomp_task

  13032.           |   |   |            code: GIMPLE_OMP_TASK

  13033.           |   |   |

  13034.           |   |   + gimple_statement_omp_target

  13035.           |   |            code: GIMPLE_OMP_TARGET

  13036.           |   |

  13037.           |   + gomp_sections

  13038.           |   |        layout: GSS_OMP_SECTIONS, code: GIMPLE_OMP_SECTIONS

  13039.           |   |

  13040.           |   + gimple_statement_omp_single_layout

  13041.           |       |    layout: GSS_OMP_SINGLE_LAYOUT

  13042.           |       |

  13043.           |       + gomp_single

  13044.           |       |        code: GIMPLE_OMP_SINGLE

  13045.           |       |

  13046.           |       + gomp_teams

  13047.           |                code: GIMPLE_OMP_TEAMS

  13048.           |

  13049.           + gbind

  13050.           |        layout: GSS_BIND, code: GIMPLE_BIND

  13051.           |

  13052.           + gcatch

  13053.           |        layout: GSS_CATCH, code: GIMPLE_CATCH

  13054.           |

  13055.           + geh_filter

  13056.           |        layout: GSS_EH_FILTER, code: GIMPLE_EH_FILTER

  13057.           |

  13058.           + geh_else

  13059.           |        layout: GSS_EH_ELSE, code: GIMPLE_EH_ELSE

  13060.           |

  13061.           + geh_mnt

  13062.           |        layout: GSS_EH_MNT, code: GIMPLE_EH_MUST_NOT_THROW

  13063.           |

  13064.           + gphi

  13065.           |        layout: GSS_PHI, code: GIMPLE_PHI

  13066.           |

  13067.           + gimple_statement_eh_ctrl

  13068.           |   |    layout: GSS_EH_CTRL

  13069.           |   |

  13070.           |   + gresx

  13071.           |   |        code: GIMPLE_RESX

  13072.           |   |

  13073.           |   + geh_dispatch

  13074.           |            code: GIMPLE_EH_DISPATCH

  13075.           |

  13076.           + gtry

  13077.           |        layout: GSS_TRY, code: GIMPLE_TRY

  13078.           |

  13079.           + gimple_statement_wce

  13080.           |        layout: GSS_WCE, code: GIMPLE_WITH_CLEANUP_EXPR

  13081.           |

  13082.           + gomp_continue

  13083.           |        layout: GSS_OMP_CONTINUE, code: GIMPLE_OMP_CONTINUE

  13084.           |

  13085.           + gomp_atomic_load

  13086.           |        layout: GSS_OMP_ATOMIC_LOAD, code: GIMPLE_OMP_ATOMIC_LOAD

  13087.           |

  13088.           + gimple_statement_omp_atomic_store_layout

  13089.               |    layout: GSS_OMP_ATOMIC_STORE_LAYOUT,

  13090.               |    code: GIMPLE_OMP_ATOMIC_STORE

  13091.               |

  13092.               + gomp_atomic_store

  13093.               |        code: GIMPLE_OMP_ATOMIC_STORE

  13094.               |

  13095.               + gomp_return

  13096.                        code: GIMPLE_OMP_RETURN

  13097. 

  13098. 

  13099. File: gccint.info,  Node: GIMPLE instruction set,  Next: GIMPLE Exception Handling,  Prev: Class hierarchy of GIMPLE statements,  Up: GIMPLE

  13100. 

  13101. 12.3 GIMPLE instruction set

  13102. ===========================

  13103. 

  13104. The following table briefly describes the GIMPLE instruction set.

  13105. 

  13106. Instruction                    High GIMPLE   Low GIMPLE

  13107. 'GIMPLE_ASM'                   x             x

  13108. 'GIMPLE_ASSIGN'                x             x

  13109. 'GIMPLE_BIND'                  x

  13110. 'GIMPLE_CALL'                  x             x

  13111. 'GIMPLE_CATCH'                 x

  13112. 'GIMPLE_COND'                  x             x

  13113. 'GIMPLE_DEBUG'                 x             x

  13114. 'GIMPLE_EH_FILTER'             x

  13115. 'GIMPLE_GOTO'                  x             x

  13116. 'GIMPLE_LABEL'                 x             x

  13117. 'GIMPLE_NOP'                   x             x

  13118. 'GIMPLE_OMP_ATOMIC_LOAD'       x             x

  13119. 'GIMPLE_OMP_ATOMIC_STORE'      x             x

  13120. 'GIMPLE_OMP_CONTINUE'          x             x

  13121. 'GIMPLE_OMP_CRITICAL'          x             x

  13122. 'GIMPLE_OMP_FOR'               x             x

  13123. 'GIMPLE_OMP_MASTER'            x             x

  13124. 'GIMPLE_OMP_ORDERED'           x             x

  13125. 'GIMPLE_OMP_PARALLEL'          x             x

  13126. 'GIMPLE_OMP_RETURN'            x             x

  13127. 'GIMPLE_OMP_SECTION'           x             x

  13128. 'GIMPLE_OMP_SECTIONS'          x             x

  13129. 'GIMPLE_OMP_SECTIONS_SWITCH'   x             x

  13130. 'GIMPLE_OMP_SINGLE'            x             x

  13131. 'GIMPLE_PHI'                                 x

  13132. 'GIMPLE_RESX'                                x

  13133. 'GIMPLE_RETURN'                x             x

  13134. 'GIMPLE_SWITCH'                x             x

  13135. 'GIMPLE_TRY'                   x

  13136. 

  13137. 

  13138. File: gccint.info,  Node: GIMPLE Exception Handling,  Next: Temporaries,  Prev: GIMPLE instruction set,  Up: GIMPLE

  13139. 

  13140. 12.4 Exception Handling

  13141. =======================

  13142. 

  13143. Other exception handling constructs are represented using

  13144. 'GIMPLE_TRY_CATCH'.  'GIMPLE_TRY_CATCH' has two operands.  The first

  13145. operand is a sequence of statements to execute.  If executing these

  13146. statements does not throw an exception, then the second operand is

  13147. ignored.  Otherwise, if an exception is thrown, then the second operand

  13148. of the 'GIMPLE_TRY_CATCH' is checked.  The second operand may have the

  13149. following forms:

  13150. 

  13151.   1. A sequence of statements to execute.  When an exception occurs,

  13152.      these statements are executed, and then the exception is rethrown.

  13153. 

  13154.   2. A sequence of 'GIMPLE_CATCH' statements.  Each 'GIMPLE_CATCH' has a

  13155.      list of applicable exception types and handler code.  If the thrown

  13156.      exception matches one of the caught types, the associated handler

  13157.      code is executed.  If the handler code falls off the bottom,

  13158.      execution continues after the original 'GIMPLE_TRY_CATCH'.

  13159. 

  13160.   3. A 'GIMPLE_EH_FILTER' statement.  This has a list of permitted

  13161.      exception types, and code to handle a match failure.  If the thrown

  13162.      exception does not match one of the allowed types, the associated

  13163.      match failure code is executed.  If the thrown exception does

  13164.      match, it continues unwinding the stack looking for the next

  13165.      handler.

  13166. 

  13167.  Currently throwing an exception is not directly represented in GIMPLE,

  13168. since it is implemented by calling a function.  At some point in the

  13169. future we will want to add some way to express that the call will throw

  13170. an exception of a known type.

  13171. 

  13172.  Just before running the optimizers, the compiler lowers the high-level

  13173. EH constructs above into a set of 'goto's, magic labels, and EH regions.

  13174. Continuing to unwind at the end of a cleanup is represented with a

  13175. 'GIMPLE_RESX'.

  13176. 

  13177. 

  13178. File: gccint.info,  Node: Temporaries,  Next: Operands,  Prev: GIMPLE Exception Handling,  Up: GIMPLE

  13179. 

  13180. 12.5 Temporaries

  13181. ================

  13182. 

  13183. When gimplification encounters a subexpression that is too complex, it

  13184. creates a new temporary variable to hold the value of the subexpression,

  13185. and adds a new statement to initialize it before the current statement.

  13186. These special temporaries are known as 'expression temporaries', and are

  13187. allocated using 'get_formal_tmp_var'.  The compiler tries to always

  13188. evaluate identical expressions into the same temporary, to simplify

  13189. elimination of redundant calculations.

  13190. 

  13191.  We can only use expression temporaries when we know that it will not be

  13192. reevaluated before its value is used, and that it will not be otherwise

  13193. modified(1).  Other temporaries can be allocated using

  13194. 'get_initialized_tmp_var' or 'create_tmp_var'.

  13195. 

  13196.  Currently, an expression like 'a = b + 5' is not reduced any further.

  13197. We tried converting it to something like

  13198.      T1 = b + 5;

  13199.      a = T1;

  13200.  but this bloated the representation for minimal benefit.  However, a

  13201. variable which must live in memory cannot appear in an expression; its

  13202. value is explicitly loaded into a temporary first.  Similarly, storing

  13203. the value of an expression to a memory variable goes through a

  13204. temporary.

  13205. 

  13206.    ---------- Footnotes ----------

  13207. 

  13208.    (1) These restrictions are derived from those in Morgan 4.8.

  13209. 

  13210. 

  13211. File: gccint.info,  Node: Operands,  Next: Manipulating GIMPLE statements,  Prev: Temporaries,  Up: GIMPLE

  13212. 

  13213. 12.6 Operands

  13214. =============

  13215. 

  13216. In general, expressions in GIMPLE consist of an operation and the

  13217. appropriate number of simple operands; these operands must either be a

  13218. GIMPLE rvalue ('is_gimple_val'), i.e. a constant or a register variable.

  13219. More complex operands are factored out into temporaries, so that

  13220.      a = b + c + d

  13221.  becomes

  13222.      T1 = b + c;

  13223.      a = T1 + d;

  13224. 

  13225.  The same rule holds for arguments to a 'GIMPLE_CALL'.

  13226. 

  13227.  The target of an assignment is usually a variable, but can also be a

  13228. 'MEM_REF' or a compound lvalue as described below.

  13229. 

  13230. * Menu:

  13231. 

  13232. * Compound Expressions::

  13233. * Compound Lvalues::

  13234. * Conditional Expressions::

  13235. * Logical Operators::

  13236. 

  13237. 

  13238. File: gccint.info,  Node: Compound Expressions,  Next: Compound Lvalues,  Up: Operands

  13239. 

  13240. 12.6.1 Compound Expressions

  13241. ---------------------------

  13242. 

  13243. The left-hand side of a C comma expression is simply moved into a

  13244. separate statement.

  13245. 

  13246. 

  13247. File: gccint.info,  Node: Compound Lvalues,  Next: Conditional Expressions,  Prev: Compound Expressions,  Up: Operands

  13248. 

  13249. 12.6.2 Compound Lvalues

  13250. -----------------------

  13251. 

  13252. Currently compound lvalues involving array and structure field

  13253. references are not broken down; an expression like 'a.b[2] = 42' is not

  13254. reduced any further (though complex array subscripts are).  This

  13255. restriction is a workaround for limitations in later optimizers; if we

  13256. were to convert this to

  13257. 

  13258.      T1 = &a.b;

  13259.      T1[2] = 42;

  13260. 

  13261.  alias analysis would not remember that the reference to 'T1[2]' came by

  13262. way of 'a.b', so it would think that the assignment could alias another

  13263. member of 'a'; this broke 'struct-alias-1.c'.  Future optimizer

  13264. improvements may make this limitation unnecessary.

  13265. 

  13266. 

  13267. File: gccint.info,  Node: Conditional Expressions,  Next: Logical Operators,  Prev: Compound Lvalues,  Up: Operands

  13268. 

  13269. 12.6.3 Conditional Expressions

  13270. ------------------------------

  13271. 

  13272. A C '?:' expression is converted into an 'if' statement with each branch

  13273. assigning to the same temporary.  So,

  13274. 

  13275.      a = b ? c : d;

  13276.  becomes

  13277.      if (b == 1)

  13278.        T1 = c;

  13279.      else

  13280.        T1 = d;

  13281.      a = T1;

  13282. 

  13283.  The GIMPLE level if-conversion pass re-introduces '?:' expression, if

  13284. appropriate.  It is used to vectorize loops with conditions using vector

  13285. conditional operations.

  13286. 

  13287.  Note that in GIMPLE, 'if' statements are represented using

  13288. 'GIMPLE_COND', as described below.

  13289. 

  13290. 

  13291. File: gccint.info,  Node: Logical Operators,  Prev: Conditional Expressions,  Up: Operands

  13292. 

  13293. 12.6.4 Logical Operators

  13294. ------------------------

  13295. 

  13296. Except when they appear in the condition operand of a 'GIMPLE_COND',

  13297. logical 'and' and 'or' operators are simplified as follows: 'a = b && c'

  13298. becomes

  13299. 

  13300.      T1 = (bool)b;

  13301.      if (T1 == true)

  13302.        T1 = (bool)c;

  13303.      a = T1;

  13304. 

  13305.  Note that 'T1' in this example cannot be an expression temporary,

  13306. because it has two different assignments.

  13307. 

  13308. 12.6.5 Manipulating operands

  13309. ----------------------------

  13310. 

  13311. All gimple operands are of type 'tree'.  But only certain types of trees

  13312. are allowed to be used as operand tuples.  Basic validation is

  13313. controlled by the function 'get_gimple_rhs_class', which given a tree

  13314. code, returns an 'enum' with the following values of type 'enum

  13315. gimple_rhs_class'

  13316. 

  13317.    * 'GIMPLE_INVALID_RHS' The tree cannot be used as a GIMPLE operand.

  13318. 

  13319.    * 'GIMPLE_TERNARY_RHS' The tree is a valid GIMPLE ternary operation.

  13320. 

  13321.    * 'GIMPLE_BINARY_RHS' The tree is a valid GIMPLE binary operation.

  13322. 

  13323.    * 'GIMPLE_UNARY_RHS' The tree is a valid GIMPLE unary operation.

  13324. 

  13325.    * 'GIMPLE_SINGLE_RHS' The tree is a single object, that cannot be

  13326.      split into simpler operands (for instance, 'SSA_NAME', 'VAR_DECL',

  13327.      'COMPONENT_REF', etc).

  13328. 

  13329.      This operand class also acts as an escape hatch for tree nodes that

  13330.      may be flattened out into the operand vector, but would need more

  13331.      than two slots on the RHS. For instance, a 'COND_EXPR' expression

  13332.      of the form '(a op b) ? x : y' could be flattened out on the

  13333.      operand vector using 4 slots, but it would also require additional

  13334.      processing to distinguish 'c = a op b' from 'c = a op b ? x : y'.

  13335.      Something similar occurs with 'ASSERT_EXPR'.  In time, these

  13336.      special case tree expressions should be flattened into the operand

  13337.      vector.

  13338. 

  13339.  For tree nodes in the categories 'GIMPLE_TERNARY_RHS',

  13340. 'GIMPLE_BINARY_RHS' and 'GIMPLE_UNARY_RHS', they cannot be stored inside

  13341. tuples directly.  They first need to be flattened and separated into

  13342. individual components.  For instance, given the GENERIC expression

  13343. 

  13344.      a = b + c

  13345. 

  13346.  its tree representation is:

  13347. 

  13348.      MODIFY_EXPR <VAR_DECL  <a>, PLUS_EXPR <VAR_DECL <b>, VAR_DECL <c>>>

  13349. 

  13350.  In this case, the GIMPLE form for this statement is logically identical

  13351. to its GENERIC form but in GIMPLE, the 'PLUS_EXPR' on the RHS of the

  13352. assignment is not represented as a tree, instead the two operands are

  13353. taken out of the 'PLUS_EXPR' sub-tree and flattened into the GIMPLE

  13354. tuple as follows:

  13355. 

  13356.      GIMPLE_ASSIGN <PLUS_EXPR, VAR_DECL <a>, VAR_DECL <b>, VAR_DECL <c>>

  13357. 

  13358. 12.6.6 Operand vector allocation

  13359. --------------------------------

  13360. 

  13361. The operand vector is stored at the bottom of the three tuple structures

  13362. that accept operands.  This means, that depending on the code of a given

  13363. statement, its operand vector will be at different offsets from the base

  13364. of the structure.  To access tuple operands use the following accessors

  13365. 

  13366.  -- GIMPLE function: unsigned gimple_num_ops (gimple g)

  13367.      Returns the number of operands in statement G.

  13368. 

  13369.  -- GIMPLE function: tree gimple_op (gimple g, unsigned i)

  13370.      Returns operand 'I' from statement 'G'.

  13371. 

  13372.  -- GIMPLE function: tree * gimple_ops (gimple g)

  13373.      Returns a pointer into the operand vector for statement 'G'.  This

  13374.      is computed using an internal table called 'gimple_ops_offset_'[].

  13375.      This table is indexed by the gimple code of 'G'.

  13376. 

  13377.      When the compiler is built, this table is filled-in using the sizes

  13378.      of the structures used by each statement code defined in

  13379.      gimple.def.  Since the operand vector is at the bottom of the

  13380.      structure, for a gimple code 'C' the offset is computed as sizeof

  13381.      (struct-of 'C') - sizeof (tree).

  13382. 

  13383.      This mechanism adds one memory indirection to every access when

  13384.      using 'gimple_op'(), if this becomes a bottleneck, a pass can

  13385.      choose to memoize the result from 'gimple_ops'() and use that to

  13386.      access the operands.

  13387. 

  13388. 12.6.7 Operand validation

  13389. -------------------------

  13390. 

  13391. When adding a new operand to a gimple statement, the operand will be

  13392. validated according to what each tuple accepts in its operand vector.

  13393. These predicates are called by the 'gimple_NAME_set_...()'.  Each tuple

  13394. will use one of the following predicates (Note, this list is not

  13395. exhaustive):

  13396. 

  13397.  -- GIMPLE function: bool is_gimple_val (tree t)

  13398.      Returns true if t is a "GIMPLE value", which are all the

  13399.      non-addressable stack variables (variables for which

  13400.      'is_gimple_reg' returns true) and constants (expressions for which

  13401.      'is_gimple_min_invariant' returns true).

  13402. 

  13403.  -- GIMPLE function: bool is_gimple_addressable (tree t)

  13404.      Returns true if t is a symbol or memory reference whose address can

  13405.      be taken.

  13406. 

  13407.  -- GIMPLE function: bool is_gimple_asm_val (tree t)

  13408.      Similar to 'is_gimple_val' but it also accepts hard registers.

  13409. 

  13410.  -- GIMPLE function: bool is_gimple_call_addr (tree t)

  13411.      Return true if t is a valid expression to use as the function

  13412.      called by a 'GIMPLE_CALL'.

  13413. 

  13414.  -- GIMPLE function: bool is_gimple_mem_ref_addr (tree t)

  13415.      Return true if t is a valid expression to use as first operand of a

  13416.      'MEM_REF' expression.

  13417. 

  13418.  -- GIMPLE function: bool is_gimple_constant (tree t)

  13419.      Return true if t is a valid gimple constant.

  13420. 

  13421.  -- GIMPLE function: bool is_gimple_min_invariant (tree t)

  13422.      Return true if t is a valid minimal invariant.  This is different

  13423.      from constants, in that the specific value of t may not be known at

  13424.      compile time, but it is known that it doesn't change (e.g., the

  13425.      address of a function local variable).

  13426. 

  13427.  -- GIMPLE function: bool is_gimple_ip_invariant (tree t)

  13428.      Return true if t is an interprocedural invariant.  This means that

  13429.      t is a valid invariant in all functions (e.g. it can be an address

  13430.      of a global variable but not of a local one).

  13431. 

  13432.  -- GIMPLE function: bool is_gimple_ip_invariant_address (tree t)

  13433.      Return true if t is an 'ADDR_EXPR' that does not change once the

  13434.      program is running (and which is valid in all functions).

  13435. 

  13436. 12.6.8 Statement validation

  13437. ---------------------------

  13438. 

  13439.  -- GIMPLE function: bool is_gimple_assign (gimple g)

  13440.      Return true if the code of g is 'GIMPLE_ASSIGN'.

  13441. 

  13442.  -- GIMPLE function: bool is_gimple_call (gimple g)

  13443.      Return true if the code of g is 'GIMPLE_CALL'.

  13444. 

  13445.  -- GIMPLE function: bool is_gimple_debug (gimple g)

  13446.      Return true if the code of g is 'GIMPLE_DEBUG'.

  13447. 

  13448.  -- GIMPLE function: bool gimple_assign_cast_p (const_gimple g)

  13449.      Return true if g is a 'GIMPLE_ASSIGN' that performs a type cast

  13450.      operation.

  13451. 

  13452.  -- GIMPLE function: bool gimple_debug_bind_p (gimple g)

  13453.      Return true if g is a 'GIMPLE_DEBUG' that binds the value of an

  13454.      expression to a variable.

  13455. 

  13456.  -- GIMPLE function: bool is_gimple_omp (gimple g)

  13457.      Return true if g is any of the OpenMP codes.

  13458. 

  13459.  -- GIMPLE function: gimple_debug_begin_stmt_p (gimple g)

  13460.      Return true if g is a 'GIMPLE_DEBUG' that marks the beginning of a

  13461.      source statement.

  13462. 

  13463.  -- GIMPLE function: gimple_debug_inline_entry_p (gimple g)

  13464.      Return true if g is a 'GIMPLE_DEBUG' that marks the entry point of

  13465.      an inlined function.

  13466. 

  13467.  -- GIMPLE function: gimple_debug_nonbind_marker_p (gimple g)

  13468.      Return true if g is a 'GIMPLE_DEBUG' that marks a program location,

  13469.      without any variable binding.

  13470. 

  13471. 

  13472. File: gccint.info,  Node: Manipulating GIMPLE statements,  Next: Tuple specific accessors,  Prev: Operands,  Up: GIMPLE

  13473. 

  13474. 12.7 Manipulating GIMPLE statements

  13475. ===================================

  13476. 

  13477. This section documents all the functions available to handle each of the

  13478. GIMPLE instructions.

  13479. 

  13480. 12.7.1 Common accessors

  13481. -----------------------

  13482. 

  13483. The following are common accessors for gimple statements.

  13484. 

  13485.  -- GIMPLE function: enum gimple_code gimple_code (gimple g)

  13486.      Return the code for statement 'G'.

  13487. 

  13488.  -- GIMPLE function: basic_block gimple_bb (gimple g)

  13489.      Return the basic block to which statement 'G' belongs to.

  13490. 

  13491.  -- GIMPLE function: tree gimple_block (gimple g)

  13492.      Return the lexical scope block holding statement 'G'.

  13493. 

  13494.  -- GIMPLE function: tree gimple_expr_type (gimple stmt)

  13495.      Return the type of the main expression computed by 'STMT'.  Return

  13496.      'void_type_node' if 'STMT' computes nothing.  This will only return

  13497.      something meaningful for 'GIMPLE_ASSIGN', 'GIMPLE_COND' and

  13498.      'GIMPLE_CALL'.  For all other tuple codes, it will return

  13499.      'void_type_node'.

  13500. 

  13501.  -- GIMPLE function: enum tree_code gimple_expr_code (gimple stmt)

  13502.      Return the tree code for the expression computed by 'STMT'.  This

  13503.      is only meaningful for 'GIMPLE_CALL', 'GIMPLE_ASSIGN' and

  13504.      'GIMPLE_COND'.  If 'STMT' is 'GIMPLE_CALL', it will return

  13505.      'CALL_EXPR'.  For 'GIMPLE_COND', it returns the code of the

  13506.      comparison predicate.  For 'GIMPLE_ASSIGN' it returns the code of

  13507.      the operation performed by the 'RHS' of the assignment.

  13508. 

  13509.  -- GIMPLE function: void gimple_set_block (gimple g, tree block)

  13510.      Set the lexical scope block of 'G' to 'BLOCK'.

  13511. 

  13512.  -- GIMPLE function: location_t gimple_locus (gimple g)

  13513.      Return locus information for statement 'G'.

  13514. 

  13515.  -- GIMPLE function: void gimple_set_locus (gimple g, location_t locus)

  13516.      Set locus information for statement 'G'.

  13517. 

  13518.  -- GIMPLE function: bool gimple_locus_empty_p (gimple g)

  13519.      Return true if 'G' does not have locus information.

  13520. 

  13521.  -- GIMPLE function: bool gimple_no_warning_p (gimple stmt)

  13522.      Return true if no warnings should be emitted for statement 'STMT'.

  13523. 

  13524.  -- GIMPLE function: void gimple_set_visited (gimple stmt, bool

  13525.           visited_p)

  13526.      Set the visited status on statement 'STMT' to 'VISITED_P'.

  13527. 

  13528.  -- GIMPLE function: bool gimple_visited_p (gimple stmt)

  13529.      Return the visited status on statement 'STMT'.

  13530. 

  13531.  -- GIMPLE function: void gimple_set_plf (gimple stmt, enum plf_mask

  13532.           plf, bool val_p)

  13533.      Set pass local flag 'PLF' on statement 'STMT' to 'VAL_P'.

  13534. 

  13535.  -- GIMPLE function: unsigned int gimple_plf (gimple stmt, enum plf_mask

  13536.           plf)

  13537.      Return the value of pass local flag 'PLF' on statement 'STMT'.

  13538. 

  13539.  -- GIMPLE function: bool gimple_has_ops (gimple g)

  13540.      Return true if statement 'G' has register or memory operands.

  13541. 

  13542.  -- GIMPLE function: bool gimple_has_mem_ops (gimple g)

  13543.      Return true if statement 'G' has memory operands.

  13544. 

  13545.  -- GIMPLE function: unsigned gimple_num_ops (gimple g)

  13546.      Return the number of operands for statement 'G'.

  13547. 

  13548.  -- GIMPLE function: tree * gimple_ops (gimple g)

  13549.      Return the array of operands for statement 'G'.

  13550. 

  13551.  -- GIMPLE function: tree gimple_op (gimple g, unsigned i)

  13552.      Return operand 'I' for statement 'G'.

  13553. 

  13554.  -- GIMPLE function: tree * gimple_op_ptr (gimple g, unsigned i)

  13555.      Return a pointer to operand 'I' for statement 'G'.

  13556. 

  13557.  -- GIMPLE function: void gimple_set_op (gimple g, unsigned i, tree op)

  13558.      Set operand 'I' of statement 'G' to 'OP'.

  13559. 

  13560.  -- GIMPLE function: bitmap gimple_addresses_taken (gimple stmt)

  13561.      Return the set of symbols that have had their address taken by

  13562.      'STMT'.

  13563. 

  13564.  -- GIMPLE function: struct def_optype_d * gimple_def_ops (gimple g)

  13565.      Return the set of 'DEF' operands for statement 'G'.

  13566. 

  13567.  -- GIMPLE function: void gimple_set_def_ops (gimple g, struct

  13568.           def_optype_d *def)

  13569.      Set 'DEF' to be the set of 'DEF' operands for statement 'G'.

  13570. 

  13571.  -- GIMPLE function: struct use_optype_d * gimple_use_ops (gimple g)

  13572.      Return the set of 'USE' operands for statement 'G'.

  13573. 

  13574.  -- GIMPLE function: void gimple_set_use_ops (gimple g, struct

  13575.           use_optype_d *use)

  13576.      Set 'USE' to be the set of 'USE' operands for statement 'G'.

  13577. 

  13578.  -- GIMPLE function: struct voptype_d * gimple_vuse_ops (gimple g)

  13579.      Return the set of 'VUSE' operands for statement 'G'.

  13580. 

  13581.  -- GIMPLE function: void gimple_set_vuse_ops (gimple g, struct

  13582.           voptype_d *ops)

  13583.      Set 'OPS' to be the set of 'VUSE' operands for statement 'G'.

  13584. 

  13585.  -- GIMPLE function: struct voptype_d * gimple_vdef_ops (gimple g)

  13586.      Return the set of 'VDEF' operands for statement 'G'.

  13587. 

  13588.  -- GIMPLE function: void gimple_set_vdef_ops (gimple g, struct

  13589.           voptype_d *ops)

  13590.      Set 'OPS' to be the set of 'VDEF' operands for statement 'G'.

  13591. 

  13592.  -- GIMPLE function: bitmap gimple_loaded_syms (gimple g)

  13593.      Return the set of symbols loaded by statement 'G'.  Each element of

  13594.      the set is the 'DECL_UID' of the corresponding symbol.

  13595. 

  13596.  -- GIMPLE function: bitmap gimple_stored_syms (gimple g)

  13597.      Return the set of symbols stored by statement 'G'.  Each element of

  13598.      the set is the 'DECL_UID' of the corresponding symbol.

  13599. 

  13600.  -- GIMPLE function: bool gimple_modified_p (gimple g)

  13601.      Return true if statement 'G' has operands and the modified field

  13602.      has been set.

  13603. 

  13604.  -- GIMPLE function: bool gimple_has_volatile_ops (gimple stmt)

  13605.      Return true if statement 'STMT' contains volatile operands.

  13606. 

  13607.  -- GIMPLE function: void gimple_set_has_volatile_ops (gimple stmt, bool

  13608.           volatilep)

  13609.      Return true if statement 'STMT' contains volatile operands.

  13610. 

  13611.  -- GIMPLE function: void update_stmt (gimple s)

  13612.      Mark statement 'S' as modified, and update it.

  13613. 

  13614.  -- GIMPLE function: void update_stmt_if_modified (gimple s)

  13615.      Update statement 'S' if it has been marked modified.

  13616. 

  13617.  -- GIMPLE function: gimple gimple_copy (gimple stmt)

  13618.      Return a deep copy of statement 'STMT'.

  13619. 

  13620. 

  13621. File: gccint.info,  Node: Tuple specific accessors,  Next: GIMPLE sequences,  Prev: Manipulating GIMPLE statements,  Up: GIMPLE

  13622. 

  13623. 12.8 Tuple specific accessors

  13624. =============================

  13625. 

  13626. * Menu:

  13627. 

  13628. * GIMPLE_ASM::

  13629. * GIMPLE_ASSIGN::

  13630. * GIMPLE_BIND::

  13631. * GIMPLE_CALL::

  13632. * GIMPLE_CATCH::

  13633. * GIMPLE_COND::

  13634. * GIMPLE_DEBUG::

  13635. * GIMPLE_EH_FILTER::

  13636. * GIMPLE_LABEL::

  13637. * GIMPLE_GOTO::

  13638. * GIMPLE_NOP::

  13639. * GIMPLE_OMP_ATOMIC_LOAD::

  13640. * GIMPLE_OMP_ATOMIC_STORE::

  13641. * GIMPLE_OMP_CONTINUE::

  13642. * GIMPLE_OMP_CRITICAL::

  13643. * GIMPLE_OMP_FOR::

  13644. * GIMPLE_OMP_MASTER::

  13645. * GIMPLE_OMP_ORDERED::

  13646. * GIMPLE_OMP_PARALLEL::

  13647. * GIMPLE_OMP_RETURN::

  13648. * GIMPLE_OMP_SECTION::

  13649. * GIMPLE_OMP_SECTIONS::

  13650. * GIMPLE_OMP_SINGLE::

  13651. * GIMPLE_PHI::

  13652. * GIMPLE_RESX::

  13653. * GIMPLE_RETURN::

  13654. * GIMPLE_SWITCH::

  13655. * GIMPLE_TRY::

  13656. * GIMPLE_WITH_CLEANUP_EXPR::

  13657. 

  13658. 

  13659. File: gccint.info,  Node: GIMPLE_ASM,  Next: GIMPLE_ASSIGN,  Up: Tuple specific accessors

  13660. 

  13661. 12.8.1 'GIMPLE_ASM'

  13662. -------------------

  13663. 

  13664.  -- GIMPLE function: gasm *gimple_build_asm_vec ( const char *string,

  13665.           vec<tree, va_gc> *inputs, vec<tree, va_gc> *outputs, vec<tree,

  13666.           va_gc> *clobbers, vec<tree, va_gc> *labels)

  13667.      Build a 'GIMPLE_ASM' statement.  This statement is used for

  13668.      building in-line assembly constructs.  'STRING' is the assembly

  13669.      code.  'INPUTS', 'OUTPUTS', 'CLOBBERS' and 'LABELS' are the inputs,

  13670.      outputs, clobbered registers and labels.

  13671. 

  13672.  -- GIMPLE function: unsigned gimple_asm_ninputs (const gasm *g)

  13673.      Return the number of input operands for 'GIMPLE_ASM' 'G'.

  13674. 

  13675.  -- GIMPLE function: unsigned gimple_asm_noutputs (const gasm *g)

  13676.      Return the number of output operands for 'GIMPLE_ASM' 'G'.

  13677. 

  13678.  -- GIMPLE function: unsigned gimple_asm_nclobbers (const gasm *g)

  13679.      Return the number of clobber operands for 'GIMPLE_ASM' 'G'.

  13680. 

  13681.  -- GIMPLE function: tree gimple_asm_input_op (const gasm *g, unsigned

  13682.           index)

  13683.      Return input operand 'INDEX' of 'GIMPLE_ASM' 'G'.

  13684. 

  13685.  -- GIMPLE function: void gimple_asm_set_input_op (gasm *g, unsigned

  13686.           index, tree in_op)

  13687.      Set 'IN_OP' to be input operand 'INDEX' in 'GIMPLE_ASM' 'G'.

  13688. 

  13689.  -- GIMPLE function: tree gimple_asm_output_op (const gasm *g, unsigned

  13690.           index)

  13691.      Return output operand 'INDEX' of 'GIMPLE_ASM' 'G'.

  13692. 

  13693.  -- GIMPLE function: void gimple_asm_set_output_op (gasm *g, unsigned

  13694.           index, tree out_op)

  13695.      Set 'OUT_OP' to be output operand 'INDEX' in 'GIMPLE_ASM' 'G'.

  13696. 

  13697.  -- GIMPLE function: tree gimple_asm_clobber_op (const gasm *g, unsigned

  13698.           index)

  13699.      Return clobber operand 'INDEX' of 'GIMPLE_ASM' 'G'.

  13700. 

  13701.  -- GIMPLE function: void gimple_asm_set_clobber_op (gasm *g, unsigned

  13702.           index, tree clobber_op)

  13703.      Set 'CLOBBER_OP' to be clobber operand 'INDEX' in 'GIMPLE_ASM' 'G'.

  13704. 

  13705.  -- GIMPLE function: const char * gimple_asm_string (const gasm *g)

  13706.      Return the string representing the assembly instruction in

  13707.      'GIMPLE_ASM' 'G'.

  13708. 

  13709.  -- GIMPLE function: bool gimple_asm_volatile_p (const gasm *g)

  13710.      Return true if 'G' is an asm statement marked volatile.

  13711. 

  13712.  -- GIMPLE function: void gimple_asm_set_volatile (gasm *g, bool

  13713.           volatile_p)

  13714.      Mark asm statement 'G' as volatile or non-volatile based on

  13715.      'VOLATILE_P'.

  13716. 

  13717. 

  13718. File: gccint.info,  Node: GIMPLE_ASSIGN,  Next: GIMPLE_BIND,  Prev: GIMPLE_ASM,  Up: Tuple specific accessors

  13719. 

  13720. 12.8.2 'GIMPLE_ASSIGN'

  13721. ----------------------

  13722. 

  13723.  -- GIMPLE function: gassign *gimple_build_assign (tree lhs, tree rhs)

  13724.      Build a 'GIMPLE_ASSIGN' statement.  The left-hand side is an lvalue

  13725.      passed in lhs.  The right-hand side can be either a unary or binary

  13726.      tree expression.  The expression tree rhs will be flattened and its

  13727.      operands assigned to the corresponding operand slots in the new

  13728.      statement.  This function is useful when you already have a tree

  13729.      expression that you want to convert into a tuple.  However, try to

  13730.      avoid building expression trees for the sole purpose of calling

  13731.      this function.  If you already have the operands in separate trees,

  13732.      it is better to use 'gimple_build_assign' with 'enum tree_code'

  13733.      argument and separate arguments for each operand.

  13734. 

  13735.  -- GIMPLE function: gassign *gimple_build_assign (tree lhs, enum

  13736.           tree_code subcode, tree op1, tree op2, tree op3)

  13737.      This function is similar to two operand 'gimple_build_assign', but

  13738.      is used to build a 'GIMPLE_ASSIGN' statement when the operands of

  13739.      the right-hand side of the assignment are already split into

  13740.      different operands.

  13741. 

  13742.      The left-hand side is an lvalue passed in lhs.  Subcode is the

  13743.      'tree_code' for the right-hand side of the assignment.  Op1, op2

  13744.      and op3 are the operands.

  13745. 

  13746.  -- GIMPLE function: gassign *gimple_build_assign (tree lhs, enum

  13747.           tree_code subcode, tree op1, tree op2)

  13748.      Like the above 5 operand 'gimple_build_assign', but with the last

  13749.      argument 'NULL' - this overload should not be used for

  13750.      'GIMPLE_TERNARY_RHS' assignments.

  13751. 

  13752.  -- GIMPLE function: gassign *gimple_build_assign (tree lhs, enum

  13753.           tree_code subcode, tree op1)

  13754.      Like the above 4 operand 'gimple_build_assign', but with the last

  13755.      argument 'NULL' - this overload should be used only for

  13756.      'GIMPLE_UNARY_RHS' and 'GIMPLE_SINGLE_RHS' assignments.

  13757. 

  13758.  -- GIMPLE function: gimple gimplify_assign (tree dst, tree src,

  13759.           gimple_seq *seq_p)

  13760.      Build a new 'GIMPLE_ASSIGN' tuple and append it to the end of

  13761.      '*SEQ_P'.

  13762. 

  13763.  'DST'/'SRC' are the destination and source respectively.  You can pass

  13764. ungimplified trees in 'DST' or 'SRC', in which case they will be

  13765. converted to a gimple operand if necessary.

  13766. 

  13767.  This function returns the newly created 'GIMPLE_ASSIGN' tuple.

  13768. 

  13769.  -- GIMPLE function: enum tree_code gimple_assign_rhs_code (gimple g)

  13770.      Return the code of the expression computed on the 'RHS' of

  13771.      assignment statement 'G'.

  13772. 

  13773.  -- GIMPLE function: enum gimple_rhs_class gimple_assign_rhs_class

  13774.           (gimple g)

  13775.      Return the gimple rhs class of the code for the expression computed

  13776.      on the rhs of assignment statement 'G'.  This will never return

  13777.      'GIMPLE_INVALID_RHS'.

  13778. 

  13779.  -- GIMPLE function: tree gimple_assign_lhs (gimple g)

  13780.      Return the 'LHS' of assignment statement 'G'.

  13781. 

  13782.  -- GIMPLE function: tree * gimple_assign_lhs_ptr (gimple g)

  13783.      Return a pointer to the 'LHS' of assignment statement 'G'.

  13784. 

  13785.  -- GIMPLE function: tree gimple_assign_rhs1 (gimple g)

  13786.      Return the first operand on the 'RHS' of assignment statement 'G'.

  13787. 

  13788.  -- GIMPLE function: tree * gimple_assign_rhs1_ptr (gimple g)

  13789.      Return the address of the first operand on the 'RHS' of assignment

  13790.      statement 'G'.

  13791. 

  13792.  -- GIMPLE function: tree gimple_assign_rhs2 (gimple g)

  13793.      Return the second operand on the 'RHS' of assignment statement 'G'.

  13794. 

  13795.  -- GIMPLE function: tree * gimple_assign_rhs2_ptr (gimple g)

  13796.      Return the address of the second operand on the 'RHS' of assignment

  13797.      statement 'G'.

  13798. 

  13799.  -- GIMPLE function: tree gimple_assign_rhs3 (gimple g)

  13800.      Return the third operand on the 'RHS' of assignment statement 'G'.

  13801. 

  13802.  -- GIMPLE function: tree * gimple_assign_rhs3_ptr (gimple g)

  13803.      Return the address of the third operand on the 'RHS' of assignment

  13804.      statement 'G'.

  13805. 

  13806.  -- GIMPLE function: void gimple_assign_set_lhs (gimple g, tree lhs)

  13807.      Set 'LHS' to be the 'LHS' operand of assignment statement 'G'.

  13808. 

  13809.  -- GIMPLE function: void gimple_assign_set_rhs1 (gimple g, tree rhs)

  13810.      Set 'RHS' to be the first operand on the 'RHS' of assignment

  13811.      statement 'G'.

  13812. 

  13813.  -- GIMPLE function: void gimple_assign_set_rhs2 (gimple g, tree rhs)

  13814.      Set 'RHS' to be the second operand on the 'RHS' of assignment

  13815.      statement 'G'.

  13816. 

  13817.  -- GIMPLE function: void gimple_assign_set_rhs3 (gimple g, tree rhs)

  13818.      Set 'RHS' to be the third operand on the 'RHS' of assignment

  13819.      statement 'G'.

  13820. 

  13821.  -- GIMPLE function: bool gimple_assign_cast_p (const_gimple s)

  13822.      Return true if 'S' is a type-cast assignment.

  13823. 

  13824. 

  13825. File: gccint.info,  Node: GIMPLE_BIND,  Next: GIMPLE_CALL,  Prev: GIMPLE_ASSIGN,  Up: Tuple specific accessors

  13826. 

  13827. 12.8.3 'GIMPLE_BIND'

  13828. --------------------

  13829. 

  13830.  -- GIMPLE function: gbind *gimple_build_bind (tree vars, gimple_seq

  13831.           body)

  13832.      Build a 'GIMPLE_BIND' statement with a list of variables in 'VARS'

  13833.      and a body of statements in sequence 'BODY'.

  13834. 

  13835.  -- GIMPLE function: tree gimple_bind_vars (const gbind *g)

  13836.      Return the variables declared in the 'GIMPLE_BIND' statement 'G'.

  13837. 

  13838.  -- GIMPLE function: void gimple_bind_set_vars (gbind *g, tree vars)

  13839.      Set 'VARS' to be the set of variables declared in the 'GIMPLE_BIND'

  13840.      statement 'G'.

  13841. 

  13842.  -- GIMPLE function: void gimple_bind_append_vars (gbind *g, tree vars)

  13843.      Append 'VARS' to the set of variables declared in the 'GIMPLE_BIND'

  13844.      statement 'G'.

  13845. 

  13846.  -- GIMPLE function: gimple_seq gimple_bind_body (gbind *g)

  13847.      Return the GIMPLE sequence contained in the 'GIMPLE_BIND' statement

  13848.      'G'.

  13849. 

  13850.  -- GIMPLE function: void gimple_bind_set_body (gbind *g, gimple_seq

  13851.           seq)

  13852.      Set 'SEQ' to be sequence contained in the 'GIMPLE_BIND' statement

  13853.      'G'.

  13854. 

  13855.  -- GIMPLE function: void gimple_bind_add_stmt (gbind *gs, gimple stmt)

  13856.      Append a statement to the end of a 'GIMPLE_BIND''s body.

  13857. 

  13858.  -- GIMPLE function: void gimple_bind_add_seq (gbind *gs, gimple_seq

  13859.           seq)

  13860.      Append a sequence of statements to the end of a 'GIMPLE_BIND''s

  13861.      body.

  13862. 

  13863.  -- GIMPLE function: tree gimple_bind_block (const gbind *g)

  13864.      Return the 'TREE_BLOCK' node associated with 'GIMPLE_BIND'

  13865.      statement 'G'.  This is analogous to the 'BIND_EXPR_BLOCK' field in

  13866.      trees.

  13867. 

  13868.  -- GIMPLE function: void gimple_bind_set_block (gbind *g, tree block)

  13869.      Set 'BLOCK' to be the 'TREE_BLOCK' node associated with

  13870.      'GIMPLE_BIND' statement 'G'.

  13871. 

  13872. 

  13873. File: gccint.info,  Node: GIMPLE_CALL,  Next: GIMPLE_CATCH,  Prev: GIMPLE_BIND,  Up: Tuple specific accessors

  13874. 

  13875. 12.8.4 'GIMPLE_CALL'

  13876. --------------------

  13877. 

  13878.  -- GIMPLE function: gcall *gimple_build_call (tree fn, unsigned nargs,

  13879.           ...)

  13880.      Build a 'GIMPLE_CALL' statement to function 'FN'.  The argument

  13881.      'FN' must be either a 'FUNCTION_DECL' or a gimple call address as

  13882.      determined by 'is_gimple_call_addr'.  'NARGS' are the number of

  13883.      arguments.  The rest of the arguments follow the argument 'NARGS',

  13884.      and must be trees that are valid as rvalues in gimple (i.e., each

  13885.      operand is validated with 'is_gimple_operand').

  13886. 

  13887.  -- GIMPLE function: gcall *gimple_build_call_from_tree (tree call_expr,

  13888.           tree fnptrtype)

  13889.      Build a 'GIMPLE_CALL' from a 'CALL_EXPR' node.  The arguments and

  13890.      the function are taken from the expression directly.  The type of

  13891.      the 'GIMPLE_CALL' is set from the second parameter passed by a

  13892.      caller.  This routine assumes that 'call_expr' is already in GIMPLE

  13893.      form.  That is, its operands are GIMPLE values and the function

  13894.      call needs no further simplification.  All the call flags in

  13895.      'call_expr' are copied over to the new 'GIMPLE_CALL'.

  13896. 

  13897.  -- GIMPLE function: gcall *gimple_build_call_vec (tree fn, 'vec<tree>'

  13898.           args)

  13899.      Identical to 'gimple_build_call' but the arguments are stored in a

  13900.      'vec<tree>'.

  13901. 

  13902.  -- GIMPLE function: tree gimple_call_lhs (gimple g)

  13903.      Return the 'LHS' of call statement 'G'.

  13904. 

  13905.  -- GIMPLE function: tree * gimple_call_lhs_ptr (gimple g)

  13906.      Return a pointer to the 'LHS' of call statement 'G'.

  13907. 

  13908.  -- GIMPLE function: void gimple_call_set_lhs (gimple g, tree lhs)

  13909.      Set 'LHS' to be the 'LHS' operand of call statement 'G'.

  13910. 

  13911.  -- GIMPLE function: tree gimple_call_fn (gimple g)

  13912.      Return the tree node representing the function called by call

  13913.      statement 'G'.

  13914. 

  13915.  -- GIMPLE function: void gimple_call_set_fn (gcall *g, tree fn)

  13916.      Set 'FN' to be the function called by call statement 'G'.  This has

  13917.      to be a gimple value specifying the address of the called function.

  13918. 

  13919.  -- GIMPLE function: tree gimple_call_fndecl (gimple g)

  13920.      If a given 'GIMPLE_CALL''s callee is a 'FUNCTION_DECL', return it.

  13921.      Otherwise return 'NULL'.  This function is analogous to

  13922.      'get_callee_fndecl' in 'GENERIC'.

  13923. 

  13924.  -- GIMPLE function: tree gimple_call_set_fndecl (gimple g, tree fndecl)

  13925.      Set the called function to 'FNDECL'.

  13926. 

  13927.  -- GIMPLE function: tree gimple_call_return_type (const gcall *g)

  13928.      Return the type returned by call statement 'G'.

  13929. 

  13930.  -- GIMPLE function: tree gimple_call_chain (gimple g)

  13931.      Return the static chain for call statement 'G'.

  13932. 

  13933.  -- GIMPLE function: void gimple_call_set_chain (gcall *g, tree chain)

  13934.      Set 'CHAIN' to be the static chain for call statement 'G'.

  13935. 

  13936.  -- GIMPLE function: unsigned gimple_call_num_args (gimple g)

  13937.      Return the number of arguments used by call statement 'G'.

  13938. 

  13939.  -- GIMPLE function: tree gimple_call_arg (gimple g, unsigned index)

  13940.      Return the argument at position 'INDEX' for call statement 'G'.

  13941.      The first argument is 0.

  13942. 

  13943.  -- GIMPLE function: tree * gimple_call_arg_ptr (gimple g, unsigned

  13944.           index)

  13945.      Return a pointer to the argument at position 'INDEX' for call

  13946.      statement 'G'.

  13947. 

  13948.  -- GIMPLE function: void gimple_call_set_arg (gimple g, unsigned index,

  13949.           tree arg)

  13950.      Set 'ARG' to be the argument at position 'INDEX' for call statement

  13951.      'G'.

  13952. 

  13953.  -- GIMPLE function: void gimple_call_set_tail (gcall *s)

  13954.      Mark call statement 'S' as being a tail call (i.e., a call just

  13955.      before the exit of a function).  These calls are candidate for tail

  13956.      call optimization.

  13957. 

  13958.  -- GIMPLE function: bool gimple_call_tail_p (gcall *s)

  13959.      Return true if 'GIMPLE_CALL' 'S' is marked as a tail call.

  13960. 

  13961.  -- GIMPLE function: bool gimple_call_noreturn_p (gimple s)

  13962.      Return true if 'S' is a noreturn call.

  13963. 

  13964.  -- GIMPLE function: gimple gimple_call_copy_skip_args (gcall *stmt,

  13965.           bitmap args_to_skip)

  13966.      Build a 'GIMPLE_CALL' identical to 'STMT' but skipping the

  13967.      arguments in the positions marked by the set 'ARGS_TO_SKIP'.

  13968. 

  13969. 

  13970. File: gccint.info,  Node: GIMPLE_CATCH,  Next: GIMPLE_COND,  Prev: GIMPLE_CALL,  Up: Tuple specific accessors

  13971. 

  13972. 12.8.5 'GIMPLE_CATCH'

  13973. ---------------------

  13974. 

  13975.  -- GIMPLE function: gcatch *gimple_build_catch (tree types, gimple_seq

  13976.           handler)

  13977.      Build a 'GIMPLE_CATCH' statement.  'TYPES' are the tree types this

  13978.      catch handles.  'HANDLER' is a sequence of statements with the code

  13979.      for the handler.

  13980. 

  13981.  -- GIMPLE function: tree gimple_catch_types (const gcatch *g)

  13982.      Return the types handled by 'GIMPLE_CATCH' statement 'G'.

  13983. 

  13984.  -- GIMPLE function: tree * gimple_catch_types_ptr (gcatch *g)

  13985.      Return a pointer to the types handled by 'GIMPLE_CATCH' statement

  13986.      'G'.

  13987. 

  13988.  -- GIMPLE function: gimple_seq gimple_catch_handler (gcatch *g)

  13989.      Return the GIMPLE sequence representing the body of the handler of

  13990.      'GIMPLE_CATCH' statement 'G'.

  13991. 

  13992.  -- GIMPLE function: void gimple_catch_set_types (gcatch *g, tree t)

  13993.      Set 'T' to be the set of types handled by 'GIMPLE_CATCH' 'G'.

  13994. 

  13995.  -- GIMPLE function: void gimple_catch_set_handler (gcatch *g,

  13996.           gimple_seq handler)

  13997.      Set 'HANDLER' to be the body of 'GIMPLE_CATCH' 'G'.

  13998. 

  13999. 

  14000. File: gccint.info,  Node: GIMPLE_COND,  Next: GIMPLE_DEBUG,  Prev: GIMPLE_CATCH,  Up: Tuple specific accessors

  14001. 

  14002. 12.8.6 'GIMPLE_COND'

  14003. --------------------

  14004. 

  14005.  -- GIMPLE function: gcond *gimple_build_cond ( enum tree_code

  14006.           pred_code, tree lhs, tree rhs, tree t_label, tree f_label)

  14007.      Build a 'GIMPLE_COND' statement.  'A' 'GIMPLE_COND' statement

  14008.      compares 'LHS' and 'RHS' and if the condition in 'PRED_CODE' is

  14009.      true, jump to the label in 't_label', otherwise jump to the label

  14010.      in 'f_label'.  'PRED_CODE' are relational operator tree codes like

  14011.      'EQ_EXPR', 'LT_EXPR', 'LE_EXPR', 'NE_EXPR', etc.

  14012. 

  14013.  -- GIMPLE function: gcond *gimple_build_cond_from_tree (tree cond, tree

  14014.           t_label, tree f_label)

  14015.      Build a 'GIMPLE_COND' statement from the conditional expression

  14016.      tree 'COND'.  'T_LABEL' and 'F_LABEL' are as in

  14017.      'gimple_build_cond'.

  14018. 

  14019.  -- GIMPLE function: enum tree_code gimple_cond_code (gimple g)

  14020.      Return the code of the predicate computed by conditional statement

  14021.      'G'.

  14022. 

  14023.  -- GIMPLE function: void gimple_cond_set_code (gcond *g, enum tree_code

  14024.           code)

  14025.      Set 'CODE' to be the predicate code for the conditional statement

  14026.      'G'.

  14027. 

  14028.  -- GIMPLE function: tree gimple_cond_lhs (gimple g)

  14029.      Return the 'LHS' of the predicate computed by conditional statement

  14030.      'G'.

  14031. 

  14032.  -- GIMPLE function: void gimple_cond_set_lhs (gcond *g, tree lhs)

  14033.      Set 'LHS' to be the 'LHS' operand of the predicate computed by

  14034.      conditional statement 'G'.

  14035. 

  14036.  -- GIMPLE function: tree gimple_cond_rhs (gimple g)

  14037.      Return the 'RHS' operand of the predicate computed by conditional

  14038.      'G'.

  14039. 

  14040.  -- GIMPLE function: void gimple_cond_set_rhs (gcond *g, tree rhs)

  14041.      Set 'RHS' to be the 'RHS' operand of the predicate computed by

  14042.      conditional statement 'G'.

  14043. 

  14044.  -- GIMPLE function: tree gimple_cond_true_label (const gcond *g)

  14045.      Return the label used by conditional statement 'G' when its

  14046.      predicate evaluates to true.

  14047. 

  14048.  -- GIMPLE function: void gimple_cond_set_true_label (gcond *g, tree

  14049.           label)

  14050.      Set 'LABEL' to be the label used by conditional statement 'G' when

  14051.      its predicate evaluates to true.

  14052. 

  14053.  -- GIMPLE function: void gimple_cond_set_false_label (gcond *g, tree

  14054.           label)

  14055.      Set 'LABEL' to be the label used by conditional statement 'G' when

  14056.      its predicate evaluates to false.

  14057. 

  14058.  -- GIMPLE function: tree gimple_cond_false_label (const gcond *g)

  14059.      Return the label used by conditional statement 'G' when its

  14060.      predicate evaluates to false.

  14061. 

  14062.  -- GIMPLE function: void gimple_cond_make_false (gcond *g)

  14063.      Set the conditional 'COND_STMT' to be of the form 'if (1 == 0)'.

  14064. 

  14065.  -- GIMPLE function: void gimple_cond_make_true (gcond *g)

  14066.      Set the conditional 'COND_STMT' to be of the form 'if (1 == 1)'.

  14067. 

  14068. 

  14069. File: gccint.info,  Node: GIMPLE_DEBUG,  Next: GIMPLE_EH_FILTER,  Prev: GIMPLE_COND,  Up: Tuple specific accessors

  14070. 

  14071. 12.8.7 'GIMPLE_DEBUG'

  14072. ---------------------

  14073. 

  14074.  -- GIMPLE function: gdebug *gimple_build_debug_bind (tree var, tree

  14075.           value, gimple stmt)

  14076.      Build a 'GIMPLE_DEBUG' statement with 'GIMPLE_DEBUG_BIND'

  14077.      'subcode'.  The effect of this statement is to tell debug

  14078.      information generation machinery that the value of user variable

  14079.      'var' is given by 'value' at that point, and to remain with that

  14080.      value until 'var' runs out of scope, a dynamically-subsequent debug

  14081.      bind statement overrides the binding, or conflicting values reach a

  14082.      control flow merge point.  Even if components of the 'value'

  14083.      expression change afterwards, the variable is supposed to retain

  14084.      the same value, though not necessarily the same location.

  14085. 

  14086.      It is expected that 'var' be most often a tree for automatic user

  14087.      variables ('VAR_DECL' or 'PARM_DECL') that satisfy the requirements

  14088.      for gimple registers, but it may also be a tree for a scalarized

  14089.      component of a user variable ('ARRAY_REF', 'COMPONENT_REF'), or a

  14090.      debug temporary ('DEBUG_EXPR_DECL').

  14091. 

  14092.      As for 'value', it can be an arbitrary tree expression, but it is

  14093.      recommended that it be in a suitable form for a gimple assignment

  14094.      'RHS'.  It is not expected that user variables that could appear as

  14095.      'var' ever appear in 'value', because in the latter we'd have their

  14096.      'SSA_NAME's instead, but even if they were not in SSA form, user

  14097.      variables appearing in 'value' are to be regarded as part of the

  14098.      executable code space, whereas those in 'var' are to be regarded as

  14099.      part of the source code space.  There is no way to refer to the

  14100.      value bound to a user variable within a 'value' expression.

  14101. 

  14102.      If 'value' is 'GIMPLE_DEBUG_BIND_NOVALUE', debug information

  14103.      generation machinery is informed that the variable 'var' is

  14104.      unbound, i.e., that its value is indeterminate, which sometimes

  14105.      means it is really unavailable, and other times that the compiler

  14106.      could not keep track of it.

  14107. 

  14108.      Block and location information for the newly-created stmt are taken

  14109.      from 'stmt', if given.

  14110. 

  14111.  -- GIMPLE function: tree gimple_debug_bind_get_var (gimple stmt)

  14112.      Return the user variable VAR that is bound at 'stmt'.

  14113. 

  14114.  -- GIMPLE function: tree gimple_debug_bind_get_value (gimple stmt)

  14115.      Return the value expression that is bound to a user variable at

  14116.      'stmt'.

  14117. 

  14118.  -- GIMPLE function: tree * gimple_debug_bind_get_value_ptr (gimple

  14119.           stmt)

  14120.      Return a pointer to the value expression that is bound to a user

  14121.      variable at 'stmt'.

  14122. 

  14123.  -- GIMPLE function: void gimple_debug_bind_set_var (gimple stmt, tree

  14124.           var)

  14125.      Modify the user variable bound at 'stmt' to VAR.

  14126. 

  14127.  -- GIMPLE function: void gimple_debug_bind_set_value (gimple stmt, tree

  14128.           var)

  14129.      Modify the value bound to the user variable bound at 'stmt' to

  14130.      VALUE.

  14131. 

  14132.  -- GIMPLE function: void gimple_debug_bind_reset_value (gimple stmt)

  14133.      Modify the value bound to the user variable bound at 'stmt' so that

  14134.      the variable becomes unbound.

  14135. 

  14136.  -- GIMPLE function: bool gimple_debug_bind_has_value_p (gimple stmt)

  14137.      Return 'TRUE' if 'stmt' binds a user variable to a value, and

  14138.      'FALSE' if it unbinds the variable.

  14139. 

  14140.  -- GIMPLE function: gimple gimple_build_debug_begin_stmt (tree block,

  14141.           location_t location)

  14142.      Build a 'GIMPLE_DEBUG' statement with 'GIMPLE_DEBUG_BEGIN_STMT'

  14143.      'subcode'.  The effect of this statement is to tell debug

  14144.      information generation machinery that the user statement at the

  14145.      given 'location' and 'block' starts at the point at which the

  14146.      statement is inserted.  The intent is that side effects (e.g.

  14147.      variable bindings) of all prior user statements are observable, and

  14148.      that none of the side effects of subsequent user statements are.

  14149. 

  14150.  -- GIMPLE function: gimple gimple_build_debug_inline_entry (tree block,

  14151.           location_t location)

  14152.      Build a 'GIMPLE_DEBUG' statement with 'GIMPLE_DEBUG_INLINE_ENTRY'

  14153.      'subcode'.  The effect of this statement is to tell debug

  14154.      information generation machinery that a function call at 'location'

  14155.      underwent inline substitution, that 'block' is the enclosing

  14156.      lexical block created for the substitution, and that at the point

  14157.      of the program in which the stmt is inserted, all parameters for

  14158.      the inlined function are bound to the respective arguments, and

  14159.      none of the side effects of its stmts are observable.

  14160. 

  14161. 

  14162. File: gccint.info,  Node: GIMPLE_EH_FILTER,  Next: GIMPLE_LABEL,  Prev: GIMPLE_DEBUG,  Up: Tuple specific accessors

  14163. 

  14164. 12.8.8 'GIMPLE_EH_FILTER'

  14165. -------------------------

  14166. 

  14167.  -- GIMPLE function: geh_filter *gimple_build_eh_filter (tree types,

  14168.           gimple_seq failure)

  14169.      Build a 'GIMPLE_EH_FILTER' statement.  'TYPES' are the filter's

  14170.      types.  'FAILURE' is a sequence with the filter's failure action.

  14171. 

  14172.  -- GIMPLE function: tree gimple_eh_filter_types (gimple g)

  14173.      Return the types handled by 'GIMPLE_EH_FILTER' statement 'G'.

  14174. 

  14175.  -- GIMPLE function: tree * gimple_eh_filter_types_ptr (gimple g)

  14176.      Return a pointer to the types handled by 'GIMPLE_EH_FILTER'

  14177.      statement 'G'.

  14178. 

  14179.  -- GIMPLE function: gimple_seq gimple_eh_filter_failure (gimple g)

  14180.      Return the sequence of statement to execute when 'GIMPLE_EH_FILTER'

  14181.      statement fails.

  14182. 

  14183.  -- GIMPLE function: void gimple_eh_filter_set_types (geh_filter *g,

  14184.           tree types)

  14185.      Set 'TYPES' to be the set of types handled by 'GIMPLE_EH_FILTER'

  14186.      'G'.

  14187. 

  14188.  -- GIMPLE function: void gimple_eh_filter_set_failure (geh_filter *g,

  14189.           gimple_seq failure)

  14190.      Set 'FAILURE' to be the sequence of statements to execute on

  14191.      failure for 'GIMPLE_EH_FILTER' 'G'.

  14192. 

  14193.  -- GIMPLE function: tree gimple_eh_must_not_throw_fndecl ( geh_mnt

  14194.           *eh_mnt_stmt)

  14195.      Get the function decl to be called by the MUST_NOT_THROW region.

  14196. 

  14197.  -- GIMPLE function: void gimple_eh_must_not_throw_set_fndecl ( geh_mnt

  14198.           *eh_mnt_stmt, tree decl)

  14199.      Set the function decl to be called by GS to DECL.

  14200. 

  14201. 

  14202. File: gccint.info,  Node: GIMPLE_LABEL,  Next: GIMPLE_GOTO,  Prev: GIMPLE_EH_FILTER,  Up: Tuple specific accessors

  14203. 

  14204. 12.8.9 'GIMPLE_LABEL'

  14205. ---------------------

  14206. 

  14207.  -- GIMPLE function: glabel *gimple_build_label (tree label)

  14208.      Build a 'GIMPLE_LABEL' statement with corresponding to the tree

  14209.      label, 'LABEL'.

  14210. 

  14211.  -- GIMPLE function: tree gimple_label_label (const glabel *g)

  14212.      Return the 'LABEL_DECL' node used by 'GIMPLE_LABEL' statement 'G'.

  14213. 

  14214.  -- GIMPLE function: void gimple_label_set_label (glabel *g, tree label)

  14215.      Set 'LABEL' to be the 'LABEL_DECL' node used by 'GIMPLE_LABEL'

  14216.      statement 'G'.

  14217. 

  14218. 

  14219. File: gccint.info,  Node: GIMPLE_GOTO,  Next: GIMPLE_NOP,  Prev: GIMPLE_LABEL,  Up: Tuple specific accessors

  14220. 

  14221. 12.8.10 'GIMPLE_GOTO'

  14222. ---------------------

  14223. 

  14224.  -- GIMPLE function: ggoto *gimple_build_goto (tree dest)

  14225.      Build a 'GIMPLE_GOTO' statement to label 'DEST'.

  14226. 

  14227.  -- GIMPLE function: tree gimple_goto_dest (gimple g)

  14228.      Return the destination of the unconditional jump 'G'.

  14229. 

  14230.  -- GIMPLE function: void gimple_goto_set_dest (ggoto *g, tree dest)

  14231.      Set 'DEST' to be the destination of the unconditional jump 'G'.

  14232. 

  14233. 

  14234. File: gccint.info,  Node: GIMPLE_NOP,  Next: GIMPLE_OMP_ATOMIC_LOAD,  Prev: GIMPLE_GOTO,  Up: Tuple specific accessors

  14235. 

  14236. 12.8.11 'GIMPLE_NOP'

  14237. --------------------

  14238. 

  14239.  -- GIMPLE function: gimple gimple_build_nop (void)

  14240.      Build a 'GIMPLE_NOP' statement.

  14241. 

  14242.  -- GIMPLE function: bool gimple_nop_p (gimple g)

  14243.      Returns 'TRUE' if statement 'G' is a 'GIMPLE_NOP'.

  14244. 

  14245. 

  14246. File: gccint.info,  Node: GIMPLE_OMP_ATOMIC_LOAD,  Next: GIMPLE_OMP_ATOMIC_STORE,  Prev: GIMPLE_NOP,  Up: Tuple specific accessors

  14247. 

  14248. 12.8.12 'GIMPLE_OMP_ATOMIC_LOAD'

  14249. --------------------------------

  14250. 

  14251.  -- GIMPLE function: gomp_atomic_load *gimple_build_omp_atomic_load (

  14252.           tree lhs, tree rhs)

  14253.      Build a 'GIMPLE_OMP_ATOMIC_LOAD' statement.  'LHS' is the left-hand

  14254.      side of the assignment.  'RHS' is the right-hand side of the

  14255.      assignment.

  14256. 

  14257.  -- GIMPLE function: void gimple_omp_atomic_load_set_lhs (

  14258.           gomp_atomic_load *g, tree lhs)

  14259.      Set the 'LHS' of an atomic load.

  14260. 

  14261.  -- GIMPLE function: tree gimple_omp_atomic_load_lhs ( const

  14262.           gomp_atomic_load *g)

  14263.      Get the 'LHS' of an atomic load.

  14264. 

  14265.  -- GIMPLE function: void gimple_omp_atomic_load_set_rhs (

  14266.           gomp_atomic_load *g, tree rhs)

  14267.      Set the 'RHS' of an atomic set.

  14268. 

  14269.  -- GIMPLE function: tree gimple_omp_atomic_load_rhs ( const

  14270.           gomp_atomic_load *g)

  14271.      Get the 'RHS' of an atomic set.

  14272. 

  14273. 

  14274. File: gccint.info,  Node: GIMPLE_OMP_ATOMIC_STORE,  Next: GIMPLE_OMP_CONTINUE,  Prev: GIMPLE_OMP_ATOMIC_LOAD,  Up: Tuple specific accessors

  14275. 

  14276. 12.8.13 'GIMPLE_OMP_ATOMIC_STORE'

  14277. ---------------------------------

  14278. 

  14279.  -- GIMPLE function: gomp_atomic_store *gimple_build_omp_atomic_store (

  14280.           tree val)

  14281.      Build a 'GIMPLE_OMP_ATOMIC_STORE' statement.  'VAL' is the value to

  14282.      be stored.

  14283. 

  14284.  -- GIMPLE function: void gimple_omp_atomic_store_set_val (

  14285.           gomp_atomic_store *g, tree val)

  14286.      Set the value being stored in an atomic store.

  14287. 

  14288.  -- GIMPLE function: tree gimple_omp_atomic_store_val ( const

  14289.           gomp_atomic_store *g)

  14290.      Return the value being stored in an atomic store.

  14291. 

  14292. 

  14293. File: gccint.info,  Node: GIMPLE_OMP_CONTINUE,  Next: GIMPLE_OMP_CRITICAL,  Prev: GIMPLE_OMP_ATOMIC_STORE,  Up: Tuple specific accessors

  14294. 

  14295. 12.8.14 'GIMPLE_OMP_CONTINUE'

  14296. -----------------------------

  14297. 

  14298.  -- GIMPLE function: gomp_continue *gimple_build_omp_continue ( tree

  14299.           control_def, tree control_use)

  14300.      Build a 'GIMPLE_OMP_CONTINUE' statement.  'CONTROL_DEF' is the

  14301.      definition of the control variable.  'CONTROL_USE' is the use of

  14302.      the control variable.

  14303. 

  14304.  -- GIMPLE function: tree gimple_omp_continue_control_def ( const

  14305.           gomp_continue *s)

  14306.      Return the definition of the control variable on a

  14307.      'GIMPLE_OMP_CONTINUE' in 'S'.

  14308. 

  14309.  -- GIMPLE function: tree gimple_omp_continue_control_def_ptr (

  14310.           gomp_continue *s)

  14311.      Same as above, but return the pointer.

  14312. 

  14313.  -- GIMPLE function: tree gimple_omp_continue_set_control_def (

  14314.           gomp_continue *s)

  14315.      Set the control variable definition for a 'GIMPLE_OMP_CONTINUE'

  14316.      statement in 'S'.

  14317. 

  14318.  -- GIMPLE function: tree gimple_omp_continue_control_use ( const

  14319.           gomp_continue *s)

  14320.      Return the use of the control variable on a 'GIMPLE_OMP_CONTINUE'

  14321.      in 'S'.

  14322. 

  14323.  -- GIMPLE function: tree gimple_omp_continue_control_use_ptr (

  14324.           gomp_continue *s)

  14325.      Same as above, but return the pointer.

  14326. 

  14327.  -- GIMPLE function: tree gimple_omp_continue_set_control_use (

  14328.           gomp_continue *s)

  14329.      Set the control variable use for a 'GIMPLE_OMP_CONTINUE' statement

  14330.      in 'S'.

  14331. 

  14332. 

  14333. File: gccint.info,  Node: GIMPLE_OMP_CRITICAL,  Next: GIMPLE_OMP_FOR,  Prev: GIMPLE_OMP_CONTINUE,  Up: Tuple specific accessors

  14334. 

  14335. 12.8.15 'GIMPLE_OMP_CRITICAL'

  14336. -----------------------------

  14337. 

  14338.  -- GIMPLE function: gomp_critical *gimple_build_omp_critical (

  14339.           gimple_seq body, tree name)

  14340.      Build a 'GIMPLE_OMP_CRITICAL' statement.  'BODY' is the sequence of

  14341.      statements for which only one thread can execute.  'NAME' is an

  14342.      optional identifier for this critical block.

  14343. 

  14344.  -- GIMPLE function: tree gimple_omp_critical_name ( const gomp_critical

  14345.           *g)

  14346.      Return the name associated with 'OMP_CRITICAL' statement 'G'.

  14347. 

  14348.  -- GIMPLE function: tree * gimple_omp_critical_name_ptr ( gomp_critical

  14349.           *g)

  14350.      Return a pointer to the name associated with 'OMP' critical

  14351.      statement 'G'.

  14352. 

  14353.  -- GIMPLE function: void gimple_omp_critical_set_name ( gomp_critical

  14354.           *g, tree name)

  14355.      Set 'NAME' to be the name associated with 'OMP' critical statement

  14356.      'G'.

  14357. 

  14358. 

  14359. File: gccint.info,  Node: GIMPLE_OMP_FOR,  Next: GIMPLE_OMP_MASTER,  Prev: GIMPLE_OMP_CRITICAL,  Up: Tuple specific accessors

  14360. 

  14361. 12.8.16 'GIMPLE_OMP_FOR'

  14362. ------------------------

  14363. 

  14364.  -- GIMPLE function: gomp_for *gimple_build_omp_for (gimple_seq body,

  14365.           tree clauses, tree index, tree initial, tree final, tree incr,

  14366.           gimple_seq pre_body, enum tree_code omp_for_cond)

  14367.      Build a 'GIMPLE_OMP_FOR' statement.  'BODY' is sequence of

  14368.      statements inside the for loop.  'CLAUSES', are any of the loop

  14369.      construct's clauses.  'PRE_BODY' is the sequence of statements that

  14370.      are loop invariant.  'INDEX' is the index variable.  'INITIAL' is

  14371.      the initial value of 'INDEX'.  'FINAL' is final value of 'INDEX'.

  14372.      OMP_FOR_COND is the predicate used to compare 'INDEX' and 'FINAL'.

  14373.      'INCR' is the increment expression.

  14374. 

  14375.  -- GIMPLE function: tree gimple_omp_for_clauses (gimple g)

  14376.      Return the clauses associated with 'OMP_FOR' 'G'.

  14377. 

  14378.  -- GIMPLE function: tree * gimple_omp_for_clauses_ptr (gimple g)

  14379.      Return a pointer to the 'OMP_FOR' 'G'.

  14380. 

  14381.  -- GIMPLE function: void gimple_omp_for_set_clauses (gimple g, tree

  14382.           clauses)

  14383.      Set 'CLAUSES' to be the list of clauses associated with 'OMP_FOR'

  14384.      'G'.

  14385. 

  14386.  -- GIMPLE function: tree gimple_omp_for_index (gimple g)

  14387.      Return the index variable for 'OMP_FOR' 'G'.

  14388. 

  14389.  -- GIMPLE function: tree * gimple_omp_for_index_ptr (gimple g)

  14390.      Return a pointer to the index variable for 'OMP_FOR' 'G'.

  14391. 

  14392.  -- GIMPLE function: void gimple_omp_for_set_index (gimple g, tree

  14393.           index)

  14394.      Set 'INDEX' to be the index variable for 'OMP_FOR' 'G'.

  14395. 

  14396.  -- GIMPLE function: tree gimple_omp_for_initial (gimple g)

  14397.      Return the initial value for 'OMP_FOR' 'G'.

  14398. 

  14399.  -- GIMPLE function: tree * gimple_omp_for_initial_ptr (gimple g)

  14400.      Return a pointer to the initial value for 'OMP_FOR' 'G'.

  14401. 

  14402.  -- GIMPLE function: void gimple_omp_for_set_initial (gimple g, tree

  14403.           initial)

  14404.      Set 'INITIAL' to be the initial value for 'OMP_FOR' 'G'.

  14405. 

  14406.  -- GIMPLE function: tree gimple_omp_for_final (gimple g)

  14407.      Return the final value for 'OMP_FOR' 'G'.

  14408. 

  14409.  -- GIMPLE function: tree * gimple_omp_for_final_ptr (gimple g)

  14410.      turn a pointer to the final value for 'OMP_FOR' 'G'.

  14411. 

  14412.  -- GIMPLE function: void gimple_omp_for_set_final (gimple g, tree

  14413.           final)

  14414.      Set 'FINAL' to be the final value for 'OMP_FOR' 'G'.

  14415. 

  14416.  -- GIMPLE function: tree gimple_omp_for_incr (gimple g)

  14417.      Return the increment value for 'OMP_FOR' 'G'.

  14418. 

  14419.  -- GIMPLE function: tree * gimple_omp_for_incr_ptr (gimple g)

  14420.      Return a pointer to the increment value for 'OMP_FOR' 'G'.

  14421. 

  14422.  -- GIMPLE function: void gimple_omp_for_set_incr (gimple g, tree incr)

  14423.      Set 'INCR' to be the increment value for 'OMP_FOR' 'G'.

  14424. 

  14425.  -- GIMPLE function: gimple_seq gimple_omp_for_pre_body (gimple g)

  14426.      Return the sequence of statements to execute before the 'OMP_FOR'

  14427.      statement 'G' starts.

  14428. 

  14429.  -- GIMPLE function: void gimple_omp_for_set_pre_body (gimple g,

  14430.           gimple_seq pre_body)

  14431.      Set 'PRE_BODY' to be the sequence of statements to execute before

  14432.      the 'OMP_FOR' statement 'G' starts.

  14433. 

  14434.  -- GIMPLE function: void gimple_omp_for_set_cond (gimple g, enum

  14435.           tree_code cond)

  14436.      Set 'COND' to be the condition code for 'OMP_FOR' 'G'.

  14437. 

  14438.  -- GIMPLE function: enum tree_code gimple_omp_for_cond (gimple g)

  14439.      Return the condition code associated with 'OMP_FOR' 'G'.

  14440. 

  14441. 

  14442. File: gccint.info,  Node: GIMPLE_OMP_MASTER,  Next: GIMPLE_OMP_ORDERED,  Prev: GIMPLE_OMP_FOR,  Up: Tuple specific accessors

  14443. 

  14444. 12.8.17 'GIMPLE_OMP_MASTER'

  14445. ---------------------------

  14446. 

  14447.  -- GIMPLE function: gimple gimple_build_omp_master (gimple_seq body)

  14448.      Build a 'GIMPLE_OMP_MASTER' statement.  'BODY' is the sequence of

  14449.      statements to be executed by just the master.

  14450. 

  14451. 

  14452. File: gccint.info,  Node: GIMPLE_OMP_ORDERED,  Next: GIMPLE_OMP_PARALLEL,  Prev: GIMPLE_OMP_MASTER,  Up: Tuple specific accessors

  14453. 

  14454. 12.8.18 'GIMPLE_OMP_ORDERED'

  14455. ----------------------------

  14456. 

  14457.  -- GIMPLE function: gimple gimple_build_omp_ordered (gimple_seq body)

  14458.      Build a 'GIMPLE_OMP_ORDERED' statement.

  14459. 

  14460.  'BODY' is the sequence of statements inside a loop that will executed

  14461. in sequence.

  14462. 

  14463. 

  14464. File: gccint.info,  Node: GIMPLE_OMP_PARALLEL,  Next: GIMPLE_OMP_RETURN,  Prev: GIMPLE_OMP_ORDERED,  Up: Tuple specific accessors

  14465. 

  14466. 12.8.19 'GIMPLE_OMP_PARALLEL'

  14467. -----------------------------

  14468. 

  14469.  -- GIMPLE function: gomp_parallel *gimple_build_omp_parallel

  14470.           (gimple_seq body, tree clauses, tree child_fn, tree data_arg)

  14471.      Build a 'GIMPLE_OMP_PARALLEL' statement.

  14472. 

  14473.  'BODY' is sequence of statements which are executed in parallel.

  14474. 'CLAUSES', are the 'OMP' parallel construct's clauses.  'CHILD_FN' is

  14475. the function created for the parallel threads to execute.  'DATA_ARG'

  14476. are the shared data argument(s).

  14477. 

  14478.  -- GIMPLE function: bool gimple_omp_parallel_combined_p (gimple g)

  14479.      Return true if 'OMP' parallel statement 'G' has the

  14480.      'GF_OMP_PARALLEL_COMBINED' flag set.

  14481. 

  14482.  -- GIMPLE function: void gimple_omp_parallel_set_combined_p (gimple g)

  14483.      Set the 'GF_OMP_PARALLEL_COMBINED' field in 'OMP' parallel

  14484.      statement 'G'.

  14485. 

  14486.  -- GIMPLE function: gimple_seq gimple_omp_body (gimple g)

  14487.      Return the body for the 'OMP' statement 'G'.

  14488. 

  14489.  -- GIMPLE function: void gimple_omp_set_body (gimple g, gimple_seq

  14490.           body)

  14491.      Set 'BODY' to be the body for the 'OMP' statement 'G'.

  14492. 

  14493.  -- GIMPLE function: tree gimple_omp_parallel_clauses (gimple g)

  14494.      Return the clauses associated with 'OMP_PARALLEL' 'G'.

  14495. 

  14496.  -- GIMPLE function: tree * gimple_omp_parallel_clauses_ptr (

  14497.           gomp_parallel *g)

  14498.      Return a pointer to the clauses associated with 'OMP_PARALLEL' 'G'.

  14499. 

  14500.  -- GIMPLE function: void gimple_omp_parallel_set_clauses (

  14501.           gomp_parallel *g, tree clauses)

  14502.      Set 'CLAUSES' to be the list of clauses associated with

  14503.      'OMP_PARALLEL' 'G'.

  14504. 

  14505.  -- GIMPLE function: tree gimple_omp_parallel_child_fn ( const

  14506.           gomp_parallel *g)

  14507.      Return the child function used to hold the body of 'OMP_PARALLEL'

  14508.      'G'.

  14509. 

  14510.  -- GIMPLE function: tree * gimple_omp_parallel_child_fn_ptr (

  14511.           gomp_parallel *g)

  14512.      Return a pointer to the child function used to hold the body of

  14513.      'OMP_PARALLEL' 'G'.

  14514. 

  14515.  -- GIMPLE function: void gimple_omp_parallel_set_child_fn (

  14516.           gomp_parallel *g, tree child_fn)

  14517.      Set 'CHILD_FN' to be the child function for 'OMP_PARALLEL' 'G'.

  14518. 

  14519.  -- GIMPLE function: tree gimple_omp_parallel_data_arg ( const

  14520.           gomp_parallel *g)

  14521.      Return the artificial argument used to send variables and values

  14522.      from the parent to the children threads in 'OMP_PARALLEL' 'G'.

  14523. 

  14524.  -- GIMPLE function: tree * gimple_omp_parallel_data_arg_ptr (

  14525.           gomp_parallel *g)

  14526.      Return a pointer to the data argument for 'OMP_PARALLEL' 'G'.

  14527. 

  14528.  -- GIMPLE function: void gimple_omp_parallel_set_data_arg (

  14529.           gomp_parallel *g, tree data_arg)

  14530.      Set 'DATA_ARG' to be the data argument for 'OMP_PARALLEL' 'G'.

  14531. 

  14532. 

  14533. File: gccint.info,  Node: GIMPLE_OMP_RETURN,  Next: GIMPLE_OMP_SECTION,  Prev: GIMPLE_OMP_PARALLEL,  Up: Tuple specific accessors

  14534. 

  14535. 12.8.20 'GIMPLE_OMP_RETURN'

  14536. ---------------------------

  14537. 

  14538.  -- GIMPLE function: gimple gimple_build_omp_return (bool wait_p)

  14539.      Build a 'GIMPLE_OMP_RETURN' statement.  'WAIT_P' is true if this is

  14540.      a non-waiting return.

  14541. 

  14542.  -- GIMPLE function: void gimple_omp_return_set_nowait (gimple s)

  14543.      Set the nowait flag on 'GIMPLE_OMP_RETURN' statement 'S'.

  14544. 

  14545.  -- GIMPLE function: bool gimple_omp_return_nowait_p (gimple g)

  14546.      Return true if 'OMP' return statement 'G' has the

  14547.      'GF_OMP_RETURN_NOWAIT' flag set.

  14548. 

  14549. 

  14550. File: gccint.info,  Node: GIMPLE_OMP_SECTION,  Next: GIMPLE_OMP_SECTIONS,  Prev: GIMPLE_OMP_RETURN,  Up: Tuple specific accessors

  14551. 

  14552. 12.8.21 'GIMPLE_OMP_SECTION'

  14553. ----------------------------

  14554. 

  14555.  -- GIMPLE function: gimple gimple_build_omp_section (gimple_seq body)

  14556.      Build a 'GIMPLE_OMP_SECTION' statement for a sections statement.

  14557. 

  14558.  'BODY' is the sequence of statements in the section.

  14559. 

  14560.  -- GIMPLE function: bool gimple_omp_section_last_p (gimple g)

  14561.      Return true if 'OMP' section statement 'G' has the

  14562.      'GF_OMP_SECTION_LAST' flag set.

  14563. 

  14564.  -- GIMPLE function: void gimple_omp_section_set_last (gimple g)

  14565.      Set the 'GF_OMP_SECTION_LAST' flag on 'G'.

  14566. 

  14567. 

  14568. File: gccint.info,  Node: GIMPLE_OMP_SECTIONS,  Next: GIMPLE_OMP_SINGLE,  Prev: GIMPLE_OMP_SECTION,  Up: Tuple specific accessors

  14569. 

  14570. 12.8.22 'GIMPLE_OMP_SECTIONS'

  14571. -----------------------------

  14572. 

  14573.  -- GIMPLE function: gomp_sections *gimple_build_omp_sections (

  14574.           gimple_seq body, tree clauses)

  14575.      Build a 'GIMPLE_OMP_SECTIONS' statement.  'BODY' is a sequence of

  14576.      section statements.  'CLAUSES' are any of the 'OMP' sections

  14577.      construct's clauses: private, firstprivate, lastprivate, reduction,

  14578.      and nowait.

  14579. 

  14580.  -- GIMPLE function: gimple gimple_build_omp_sections_switch (void)

  14581.      Build a 'GIMPLE_OMP_SECTIONS_SWITCH' statement.

  14582. 

  14583.  -- GIMPLE function: tree gimple_omp_sections_control (gimple g)

  14584.      Return the control variable associated with the

  14585.      'GIMPLE_OMP_SECTIONS' in 'G'.

  14586. 

  14587.  -- GIMPLE function: tree * gimple_omp_sections_control_ptr (gimple g)

  14588.      Return a pointer to the clauses associated with the

  14589.      'GIMPLE_OMP_SECTIONS' in 'G'.

  14590. 

  14591.  -- GIMPLE function: void gimple_omp_sections_set_control (gimple g,

  14592.           tree control)

  14593.      Set 'CONTROL' to be the set of clauses associated with the

  14594.      'GIMPLE_OMP_SECTIONS' in 'G'.

  14595. 

  14596.  -- GIMPLE function: tree gimple_omp_sections_clauses (gimple g)

  14597.      Return the clauses associated with 'OMP_SECTIONS' 'G'.

  14598. 

  14599.  -- GIMPLE function: tree * gimple_omp_sections_clauses_ptr (gimple g)

  14600.      Return a pointer to the clauses associated with 'OMP_SECTIONS' 'G'.

  14601. 

  14602.  -- GIMPLE function: void gimple_omp_sections_set_clauses (gimple g,

  14603.           tree clauses)

  14604.      Set 'CLAUSES' to be the set of clauses associated with

  14605.      'OMP_SECTIONS' 'G'.

  14606. 

  14607. 

  14608. File: gccint.info,  Node: GIMPLE_OMP_SINGLE,  Next: GIMPLE_PHI,  Prev: GIMPLE_OMP_SECTIONS,  Up: Tuple specific accessors

  14609. 

  14610. 12.8.23 'GIMPLE_OMP_SINGLE'

  14611. ---------------------------

  14612. 

  14613.  -- GIMPLE function: gomp_single *gimple_build_omp_single ( gimple_seq

  14614.           body, tree clauses)

  14615.      Build a 'GIMPLE_OMP_SINGLE' statement.  'BODY' is the sequence of

  14616.      statements that will be executed once.  'CLAUSES' are any of the

  14617.      'OMP' single construct's clauses: private, firstprivate,

  14618.      copyprivate, nowait.

  14619. 

  14620.  -- GIMPLE function: tree gimple_omp_single_clauses (gimple g)

  14621.      Return the clauses associated with 'OMP_SINGLE' 'G'.

  14622. 

  14623.  -- GIMPLE function: tree * gimple_omp_single_clauses_ptr (gimple g)

  14624.      Return a pointer to the clauses associated with 'OMP_SINGLE' 'G'.

  14625. 

  14626.  -- GIMPLE function: void gimple_omp_single_set_clauses ( gomp_single

  14627.           *g, tree clauses)

  14628.      Set 'CLAUSES' to be the clauses associated with 'OMP_SINGLE' 'G'.

  14629. 

  14630. 

  14631. File: gccint.info,  Node: GIMPLE_PHI,  Next: GIMPLE_RESX,  Prev: GIMPLE_OMP_SINGLE,  Up: Tuple specific accessors

  14632. 

  14633. 12.8.24 'GIMPLE_PHI'

  14634. --------------------

  14635. 

  14636.  -- GIMPLE function: unsigned gimple_phi_capacity (gimple g)

  14637.      Return the maximum number of arguments supported by 'GIMPLE_PHI'

  14638.      'G'.

  14639. 

  14640.  -- GIMPLE function: unsigned gimple_phi_num_args (gimple g)

  14641.      Return the number of arguments in 'GIMPLE_PHI' 'G'.  This must

  14642.      always be exactly the number of incoming edges for the basic block

  14643.      holding 'G'.

  14644. 

  14645.  -- GIMPLE function: tree gimple_phi_result (gimple g)

  14646.      Return the 'SSA' name created by 'GIMPLE_PHI' 'G'.

  14647. 

  14648.  -- GIMPLE function: tree * gimple_phi_result_ptr (gimple g)

  14649.      Return a pointer to the 'SSA' name created by 'GIMPLE_PHI' 'G'.

  14650. 

  14651.  -- GIMPLE function: void gimple_phi_set_result (gphi *g, tree result)

  14652.      Set 'RESULT' to be the 'SSA' name created by 'GIMPLE_PHI' 'G'.

  14653. 

  14654.  -- GIMPLE function: struct phi_arg_d * gimple_phi_arg (gimple g, index)

  14655.      Return the 'PHI' argument corresponding to incoming edge 'INDEX'

  14656.      for 'GIMPLE_PHI' 'G'.

  14657. 

  14658.  -- GIMPLE function: void gimple_phi_set_arg (gphi *g, index, struct

  14659.           phi_arg_d * phiarg)

  14660.      Set 'PHIARG' to be the argument corresponding to incoming edge

  14661.      'INDEX' for 'GIMPLE_PHI' 'G'.

  14662. 

  14663. 

  14664. File: gccint.info,  Node: GIMPLE_RESX,  Next: GIMPLE_RETURN,  Prev: GIMPLE_PHI,  Up: Tuple specific accessors

  14665. 

  14666. 12.8.25 'GIMPLE_RESX'

  14667. ---------------------

  14668. 

  14669.  -- GIMPLE function: gresx *gimple_build_resx (int region)

  14670.      Build a 'GIMPLE_RESX' statement which is a statement.  This

  14671.      statement is a placeholder for _Unwind_Resume before we know if a

  14672.      function call or a branch is needed.  'REGION' is the exception

  14673.      region from which control is flowing.

  14674. 

  14675.  -- GIMPLE function: int gimple_resx_region (const gresx *g)

  14676.      Return the region number for 'GIMPLE_RESX' 'G'.

  14677. 

  14678.  -- GIMPLE function: void gimple_resx_set_region (gresx *g, int region)

  14679.      Set 'REGION' to be the region number for 'GIMPLE_RESX' 'G'.

  14680. 

  14681. 

  14682. File: gccint.info,  Node: GIMPLE_RETURN,  Next: GIMPLE_SWITCH,  Prev: GIMPLE_RESX,  Up: Tuple specific accessors

  14683. 

  14684. 12.8.26 'GIMPLE_RETURN'

  14685. -----------------------

  14686. 

  14687.  -- GIMPLE function: greturn *gimple_build_return (tree retval)

  14688.      Build a 'GIMPLE_RETURN' statement whose return value is retval.

  14689. 

  14690.  -- GIMPLE function: tree gimple_return_retval (const greturn *g)

  14691.      Return the return value for 'GIMPLE_RETURN' 'G'.

  14692. 

  14693.  -- GIMPLE function: void gimple_return_set_retval (greturn *g, tree

  14694.           retval)

  14695.      Set 'RETVAL' to be the return value for 'GIMPLE_RETURN' 'G'.

  14696. 

  14697. 

  14698. File: gccint.info,  Node: GIMPLE_SWITCH,  Next: GIMPLE_TRY,  Prev: GIMPLE_RETURN,  Up: Tuple specific accessors

  14699. 

  14700. 12.8.27 'GIMPLE_SWITCH'

  14701. -----------------------

  14702. 

  14703.  -- GIMPLE function: gswitch *gimple_build_switch (tree index, tree

  14704.           default_label, 'vec'<tree> *args)

  14705.      Build a 'GIMPLE_SWITCH' statement.  'INDEX' is the index variable

  14706.      to switch on, and 'DEFAULT_LABEL' represents the default label.

  14707.      'ARGS' is a vector of 'CASE_LABEL_EXPR' trees that contain the

  14708.      non-default case labels.  Each label is a tree of code

  14709.      'CASE_LABEL_EXPR'.

  14710. 

  14711.  -- GIMPLE function: unsigned gimple_switch_num_labels ( const gswitch

  14712.           *g)

  14713.      Return the number of labels associated with the switch statement

  14714.      'G'.

  14715. 

  14716.  -- GIMPLE function: void gimple_switch_set_num_labels (gswitch *g,

  14717.           unsigned nlabels)

  14718.      Set 'NLABELS' to be the number of labels for the switch statement

  14719.      'G'.

  14720. 

  14721.  -- GIMPLE function: tree gimple_switch_index (const gswitch *g)

  14722.      Return the index variable used by the switch statement 'G'.

  14723. 

  14724.  -- GIMPLE function: void gimple_switch_set_index (gswitch *g, tree

  14725.           index)

  14726.      Set 'INDEX' to be the index variable for switch statement 'G'.

  14727. 

  14728.  -- GIMPLE function: tree gimple_switch_label (const gswitch *g,

  14729.           unsigned index)

  14730.      Return the label numbered 'INDEX'.  The default label is 0,

  14731.      followed by any labels in a switch statement.

  14732. 

  14733.  -- GIMPLE function: void gimple_switch_set_label (gswitch *g, unsigned

  14734.           index, tree label)

  14735.      Set the label number 'INDEX' to 'LABEL'.  0 is always the default

  14736.      label.

  14737. 

  14738.  -- GIMPLE function: tree gimple_switch_default_label ( const gswitch

  14739.           *g)

  14740.      Return the default label for a switch statement.

  14741. 

  14742.  -- GIMPLE function: void gimple_switch_set_default_label (gswitch *g,

  14743.           tree label)

  14744.      Set the default label for a switch statement.

  14745. 

  14746. 

  14747. File: gccint.info,  Node: GIMPLE_TRY,  Next: GIMPLE_WITH_CLEANUP_EXPR,  Prev: GIMPLE_SWITCH,  Up: Tuple specific accessors

  14748. 

  14749. 12.8.28 'GIMPLE_TRY'

  14750. --------------------

  14751. 

  14752.  -- GIMPLE function: gtry *gimple_build_try (gimple_seq eval, gimple_seq

  14753.           cleanup, unsigned int kind)

  14754.      Build a 'GIMPLE_TRY' statement.  'EVAL' is a sequence with the

  14755.      expression to evaluate.  'CLEANUP' is a sequence of statements to

  14756.      run at clean-up time.  'KIND' is the enumeration value

  14757.      'GIMPLE_TRY_CATCH' if this statement denotes a try/catch construct

  14758.      or 'GIMPLE_TRY_FINALLY' if this statement denotes a try/finally

  14759.      construct.

  14760. 

  14761.  -- GIMPLE function: enum gimple_try_flags gimple_try_kind (gimple g)

  14762.      Return the kind of try block represented by 'GIMPLE_TRY' 'G'.  This

  14763.      is either 'GIMPLE_TRY_CATCH' or 'GIMPLE_TRY_FINALLY'.

  14764. 

  14765.  -- GIMPLE function: bool gimple_try_catch_is_cleanup (gimple g)

  14766.      Return the 'GIMPLE_TRY_CATCH_IS_CLEANUP' flag.

  14767. 

  14768.  -- GIMPLE function: gimple_seq gimple_try_eval (gimple g)

  14769.      Return the sequence of statements used as the body for 'GIMPLE_TRY'

  14770.      'G'.

  14771. 

  14772.  -- GIMPLE function: gimple_seq gimple_try_cleanup (gimple g)

  14773.      Return the sequence of statements used as the cleanup body for

  14774.      'GIMPLE_TRY' 'G'.

  14775. 

  14776.  -- GIMPLE function: void gimple_try_set_catch_is_cleanup (gimple g,

  14777.           bool catch_is_cleanup)

  14778.      Set the 'GIMPLE_TRY_CATCH_IS_CLEANUP' flag.

  14779. 

  14780.  -- GIMPLE function: void gimple_try_set_eval (gtry *g, gimple_seq eval)

  14781.      Set 'EVAL' to be the sequence of statements to use as the body for

  14782.      'GIMPLE_TRY' 'G'.

  14783. 

  14784.  -- GIMPLE function: void gimple_try_set_cleanup (gtry *g, gimple_seq

  14785.           cleanup)

  14786.      Set 'CLEANUP' to be the sequence of statements to use as the

  14787.      cleanup body for 'GIMPLE_TRY' 'G'.

  14788. 

  14789. 

  14790. File: gccint.info,  Node: GIMPLE_WITH_CLEANUP_EXPR,  Prev: GIMPLE_TRY,  Up: Tuple specific accessors

  14791. 

  14792. 12.8.29 'GIMPLE_WITH_CLEANUP_EXPR'

  14793. ----------------------------------

  14794. 

  14795.  -- GIMPLE function: gimple gimple_build_wce (gimple_seq cleanup)

  14796.      Build a 'GIMPLE_WITH_CLEANUP_EXPR' statement.  'CLEANUP' is the

  14797.      clean-up expression.

  14798. 

  14799.  -- GIMPLE function: gimple_seq gimple_wce_cleanup (gimple g)

  14800.      Return the cleanup sequence for cleanup statement 'G'.

  14801. 

  14802.  -- GIMPLE function: void gimple_wce_set_cleanup (gimple g, gimple_seq

  14803.           cleanup)

  14804.      Set 'CLEANUP' to be the cleanup sequence for 'G'.

  14805. 

  14806.  -- GIMPLE function: bool gimple_wce_cleanup_eh_only (gimple g)

  14807.      Return the 'CLEANUP_EH_ONLY' flag for a 'WCE' tuple.

  14808. 

  14809.  -- GIMPLE function: void gimple_wce_set_cleanup_eh_only (gimple g, bool

  14810.           eh_only_p)

  14811.      Set the 'CLEANUP_EH_ONLY' flag for a 'WCE' tuple.

  14812. 

  14813. 

  14814. File: gccint.info,  Node: GIMPLE sequences,  Next: Sequence iterators,  Prev: Tuple specific accessors,  Up: GIMPLE

  14815. 

  14816. 12.9 GIMPLE sequences

  14817. =====================

  14818. 

  14819. GIMPLE sequences are the tuple equivalent of 'STATEMENT_LIST''s used in

  14820. 'GENERIC'.  They are used to chain statements together, and when used in

  14821. conjunction with sequence iterators, provide a framework for iterating

  14822. through statements.

  14823. 

  14824.  GIMPLE sequences are of type struct 'gimple_sequence', but are more

  14825. commonly passed by reference to functions dealing with sequences.  The

  14826. type for a sequence pointer is 'gimple_seq' which is the same as struct

  14827. 'gimple_sequence' *.  When declaring a local sequence, you can define a

  14828. local variable of type struct 'gimple_sequence'.  When declaring a

  14829. sequence allocated on the garbage collected heap, use the function

  14830. 'gimple_seq_alloc' documented below.

  14831. 

  14832.  There are convenience functions for iterating through sequences in the

  14833. section entitled Sequence Iterators.

  14834. 

  14835.  Below is a list of functions to manipulate and query sequences.

  14836. 

  14837.  -- GIMPLE function: void gimple_seq_add_stmt (gimple_seq *seq, gimple

  14838.           g)

  14839.      Link a gimple statement to the end of the sequence *'SEQ' if 'G' is

  14840.      not 'NULL'.  If *'SEQ' is 'NULL', allocate a sequence before

  14841.      linking.

  14842. 

  14843.  -- GIMPLE function: void gimple_seq_add_seq (gimple_seq *dest,

  14844.           gimple_seq src)

  14845.      Append sequence 'SRC' to the end of sequence *'DEST' if 'SRC' is

  14846.      not 'NULL'.  If *'DEST' is 'NULL', allocate a new sequence before

  14847.      appending.

  14848. 

  14849.  -- GIMPLE function: gimple_seq gimple_seq_deep_copy (gimple_seq src)

  14850.      Perform a deep copy of sequence 'SRC' and return the result.

  14851. 

  14852.  -- GIMPLE function: gimple_seq gimple_seq_reverse (gimple_seq seq)

  14853.      Reverse the order of the statements in the sequence 'SEQ'.  Return

  14854.      'SEQ'.

  14855. 

  14856.  -- GIMPLE function: gimple gimple_seq_first (gimple_seq s)

  14857.      Return the first statement in sequence 'S'.

  14858. 

  14859.  -- GIMPLE function: gimple gimple_seq_last (gimple_seq s)

  14860.      Return the last statement in sequence 'S'.

  14861. 

  14862.  -- GIMPLE function: void gimple_seq_set_last (gimple_seq s, gimple

  14863.           last)

  14864.      Set the last statement in sequence 'S' to the statement in 'LAST'.

  14865. 

  14866.  -- GIMPLE function: void gimple_seq_set_first (gimple_seq s, gimple

  14867.           first)

  14868.      Set the first statement in sequence 'S' to the statement in

  14869.      'FIRST'.

  14870. 

  14871.  -- GIMPLE function: void gimple_seq_init (gimple_seq s)

  14872.      Initialize sequence 'S' to an empty sequence.

  14873. 

  14874.  -- GIMPLE function: gimple_seq gimple_seq_alloc (void)

  14875.      Allocate a new sequence in the garbage collected store and return

  14876.      it.

  14877. 

  14878.  -- GIMPLE function: void gimple_seq_copy (gimple_seq dest, gimple_seq

  14879.           src)

  14880.      Copy the sequence 'SRC' into the sequence 'DEST'.

  14881. 

  14882.  -- GIMPLE function: bool gimple_seq_empty_p (gimple_seq s)

  14883.      Return true if the sequence 'S' is empty.

  14884. 

  14885.  -- GIMPLE function: gimple_seq bb_seq (basic_block bb)

  14886.      Returns the sequence of statements in 'BB'.

  14887. 

  14888.  -- GIMPLE function: void set_bb_seq (basic_block bb, gimple_seq seq)

  14889.      Sets the sequence of statements in 'BB' to 'SEQ'.

  14890. 

  14891.  -- GIMPLE function: bool gimple_seq_singleton_p (gimple_seq seq)

  14892.      Determine whether 'SEQ' contains exactly one statement.

  14893. 

  14894. 

  14895. File: gccint.info,  Node: Sequence iterators,  Next: Adding a new GIMPLE statement code,  Prev: GIMPLE sequences,  Up: GIMPLE

  14896. 

  14897. 12.10 Sequence iterators

  14898. ========================

  14899. 

  14900. Sequence iterators are convenience constructs for iterating through

  14901. statements in a sequence.  Given a sequence 'SEQ', here is a typical use

  14902. of gimple sequence iterators:

  14903. 

  14904.      gimple_stmt_iterator gsi;

  14905. 

  14906.      for (gsi = gsi_start (seq); !gsi_end_p (gsi); gsi_next (&gsi))

  14907.        {

  14908.          gimple g = gsi_stmt (gsi);

  14909.          /* Do something with gimple statement G.  */

  14910.        }

  14911. 

  14912.  Backward iterations are possible:

  14913. 

  14914.              for (gsi = gsi_last (seq); !gsi_end_p (gsi); gsi_prev (&gsi))

  14915. 

  14916.  Forward and backward iterations on basic blocks are possible with

  14917. 'gsi_start_bb' and 'gsi_last_bb'.

  14918. 

  14919.  In the documentation below we sometimes refer to enum

  14920. 'gsi_iterator_update'.  The valid options for this enumeration are:

  14921. 

  14922.    * 'GSI_NEW_STMT' Only valid when a single statement is added.  Move

  14923.      the iterator to it.

  14924. 

  14925.    * 'GSI_SAME_STMT' Leave the iterator at the same statement.

  14926. 

  14927.    * 'GSI_CONTINUE_LINKING' Move iterator to whatever position is

  14928.      suitable for linking other statements in the same direction.

  14929. 

  14930.  Below is a list of the functions used to manipulate and use statement

  14931. iterators.

  14932. 

  14933.  -- GIMPLE function: gimple_stmt_iterator gsi_start (gimple_seq seq)

  14934.      Return a new iterator pointing to the sequence 'SEQ''s first

  14935.      statement.  If 'SEQ' is empty, the iterator's basic block is

  14936.      'NULL'.  Use 'gsi_start_bb' instead when the iterator needs to

  14937.      always have the correct basic block set.

  14938. 

  14939.  -- GIMPLE function: gimple_stmt_iterator gsi_start_bb (basic_block bb)

  14940.      Return a new iterator pointing to the first statement in basic

  14941.      block 'BB'.

  14942. 

  14943.  -- GIMPLE function: gimple_stmt_iterator gsi_last (gimple_seq seq)

  14944.      Return a new iterator initially pointing to the last statement of

  14945.      sequence 'SEQ'.  If 'SEQ' is empty, the iterator's basic block is

  14946.      'NULL'.  Use 'gsi_last_bb' instead when the iterator needs to

  14947.      always have the correct basic block set.

  14948. 

  14949.  -- GIMPLE function: gimple_stmt_iterator gsi_last_bb (basic_block bb)

  14950.      Return a new iterator pointing to the last statement in basic block

  14951.      'BB'.

  14952. 

  14953.  -- GIMPLE function: bool gsi_end_p (gimple_stmt_iterator i)

  14954.      Return 'TRUE' if at the end of 'I'.

  14955. 

  14956.  -- GIMPLE function: bool gsi_one_before_end_p (gimple_stmt_iterator i)

  14957.      Return 'TRUE' if we're one statement before the end of 'I'.

  14958. 

  14959.  -- GIMPLE function: void gsi_next (gimple_stmt_iterator *i)

  14960.      Advance the iterator to the next gimple statement.

  14961. 

  14962.  -- GIMPLE function: void gsi_prev (gimple_stmt_iterator *i)

  14963.      Advance the iterator to the previous gimple statement.

  14964. 

  14965.  -- GIMPLE function: gimple gsi_stmt (gimple_stmt_iterator i)

  14966.      Return the current stmt.

  14967. 

  14968.  -- GIMPLE function: gimple_stmt_iterator gsi_after_labels (basic_block

  14969.           bb)

  14970.      Return a block statement iterator that points to the first

  14971.      non-label statement in block 'BB'.

  14972. 

  14973.  -- GIMPLE function: gimple * gsi_stmt_ptr (gimple_stmt_iterator *i)

  14974.      Return a pointer to the current stmt.

  14975. 

  14976.  -- GIMPLE function: basic_block gsi_bb (gimple_stmt_iterator i)

  14977.      Return the basic block associated with this iterator.

  14978. 

  14979.  -- GIMPLE function: gimple_seq gsi_seq (gimple_stmt_iterator i)

  14980.      Return the sequence associated with this iterator.

  14981. 

  14982.  -- GIMPLE function: void gsi_remove (gimple_stmt_iterator *i, bool

  14983.           remove_eh_info)

  14984.      Remove the current stmt from the sequence.  The iterator is updated

  14985.      to point to the next statement.  When 'REMOVE_EH_INFO' is true we

  14986.      remove the statement pointed to by iterator 'I' from the 'EH'

  14987.      tables.  Otherwise we do not modify the 'EH' tables.  Generally,

  14988.      'REMOVE_EH_INFO' should be true when the statement is going to be

  14989.      removed from the 'IL' and not reinserted elsewhere.

  14990. 

  14991.  -- GIMPLE function: void gsi_link_seq_before (gimple_stmt_iterator *i,

  14992.           gimple_seq seq, enum gsi_iterator_update mode)

  14993.      Links the sequence of statements 'SEQ' before the statement pointed

  14994.      by iterator 'I'.  'MODE' indicates what to do with the iterator

  14995.      after insertion (see 'enum gsi_iterator_update' above).

  14996. 

  14997.  -- GIMPLE function: void gsi_link_before (gimple_stmt_iterator *i,

  14998.           gimple g, enum gsi_iterator_update mode)

  14999.      Links statement 'G' before the statement pointed-to by iterator

  15000.      'I'.  Updates iterator 'I' according to 'MODE'.

  15001. 

  15002.  -- GIMPLE function: void gsi_link_seq_after (gimple_stmt_iterator *i,

  15003.           gimple_seq seq, enum gsi_iterator_update mode)

  15004.      Links sequence 'SEQ' after the statement pointed-to by iterator

  15005.      'I'.  'MODE' is as in 'gsi_insert_after'.

  15006. 

  15007.  -- GIMPLE function: void gsi_link_after (gimple_stmt_iterator *i,

  15008.           gimple g, enum gsi_iterator_update mode)

  15009.      Links statement 'G' after the statement pointed-to by iterator 'I'.

  15010.      'MODE' is as in 'gsi_insert_after'.

  15011. 

  15012.  -- GIMPLE function: gimple_seq gsi_split_seq_after

  15013.           (gimple_stmt_iterator i)

  15014.      Move all statements in the sequence after 'I' to a new sequence.

  15015.      Return this new sequence.

  15016. 

  15017.  -- GIMPLE function: gimple_seq gsi_split_seq_before

  15018.           (gimple_stmt_iterator *i)

  15019.      Move all statements in the sequence before 'I' to a new sequence.

  15020.      Return this new sequence.

  15021. 

  15022.  -- GIMPLE function: void gsi_replace (gimple_stmt_iterator *i, gimple

  15023.           stmt, bool update_eh_info)

  15024.      Replace the statement pointed-to by 'I' to 'STMT'.  If

  15025.      'UPDATE_EH_INFO' is true, the exception handling information of the

  15026.      original statement is moved to the new statement.

  15027. 

  15028.  -- GIMPLE function: void gsi_insert_before (gimple_stmt_iterator *i,

  15029.           gimple stmt, enum gsi_iterator_update mode)

  15030.      Insert statement 'STMT' before the statement pointed-to by iterator

  15031.      'I', update 'STMT''s basic block and scan it for new operands.

  15032.      'MODE' specifies how to update iterator 'I' after insertion (see

  15033.      enum 'gsi_iterator_update').

  15034. 

  15035.  -- GIMPLE function: void gsi_insert_seq_before (gimple_stmt_iterator

  15036.           *i, gimple_seq seq, enum gsi_iterator_update mode)

  15037.      Like 'gsi_insert_before', but for all the statements in 'SEQ'.

  15038. 

  15039.  -- GIMPLE function: void gsi_insert_after (gimple_stmt_iterator *i,

  15040.           gimple stmt, enum gsi_iterator_update mode)

  15041.      Insert statement 'STMT' after the statement pointed-to by iterator

  15042.      'I', update 'STMT''s basic block and scan it for new operands.

  15043.      'MODE' specifies how to update iterator 'I' after insertion (see

  15044.      enum 'gsi_iterator_update').

  15045. 

  15046.  -- GIMPLE function: void gsi_insert_seq_after (gimple_stmt_iterator *i,

  15047.           gimple_seq seq, enum gsi_iterator_update mode)

  15048.      Like 'gsi_insert_after', but for all the statements in 'SEQ'.

  15049. 

  15050.  -- GIMPLE function: gimple_stmt_iterator gsi_for_stmt (gimple stmt)

  15051.      Finds iterator for 'STMT'.

  15052. 

  15053.  -- GIMPLE function: void gsi_move_after (gimple_stmt_iterator *from,

  15054.           gimple_stmt_iterator *to)

  15055.      Move the statement at 'FROM' so it comes right after the statement

  15056.      at 'TO'.

  15057. 

  15058.  -- GIMPLE function: void gsi_move_before (gimple_stmt_iterator *from,

  15059.           gimple_stmt_iterator *to)

  15060.      Move the statement at 'FROM' so it comes right before the statement

  15061.      at 'TO'.

  15062. 

  15063.  -- GIMPLE function: void gsi_move_to_bb_end (gimple_stmt_iterator

  15064.           *from, basic_block bb)

  15065.      Move the statement at 'FROM' to the end of basic block 'BB'.

  15066. 

  15067.  -- GIMPLE function: void gsi_insert_on_edge (edge e, gimple stmt)

  15068.      Add 'STMT' to the pending list of edge 'E'.  No actual insertion is

  15069.      made until a call to 'gsi_commit_edge_inserts'() is made.

  15070. 

  15071.  -- GIMPLE function: void gsi_insert_seq_on_edge (edge e, gimple_seq

  15072.           seq)

  15073.      Add the sequence of statements in 'SEQ' to the pending list of edge

  15074.      'E'.  No actual insertion is made until a call to

  15075.      'gsi_commit_edge_inserts'() is made.

  15076. 

  15077.  -- GIMPLE function: basic_block gsi_insert_on_edge_immediate (edge e,

  15078.           gimple stmt)

  15079.      Similar to 'gsi_insert_on_edge'+'gsi_commit_edge_inserts'.  If a

  15080.      new block has to be created, it is returned.

  15081. 

  15082.  -- GIMPLE function: void gsi_commit_one_edge_insert (edge e,

  15083.           basic_block *new_bb)

  15084.      Commit insertions pending at edge 'E'.  If a new block is created,

  15085.      set 'NEW_BB' to this block, otherwise set it to 'NULL'.

  15086. 

  15087.  -- GIMPLE function: void gsi_commit_edge_inserts (void)

  15088.      This routine will commit all pending edge insertions, creating any

  15089.      new basic blocks which are necessary.

  15090. 

  15091. 

  15092. File: gccint.info,  Node: Adding a new GIMPLE statement code,  Next: Statement and operand traversals,  Prev: Sequence iterators,  Up: GIMPLE

  15093. 

  15094. 12.11 Adding a new GIMPLE statement code

  15095. ========================================

  15096. 

  15097. The first step in adding a new GIMPLE statement code, is modifying the

  15098. file 'gimple.def', which contains all the GIMPLE codes.  Then you must

  15099. add a corresponding gimple subclass located in 'gimple.h'.  This in

  15100. turn, will require you to add a corresponding 'GTY' tag in

  15101. 'gsstruct.def', and code to handle this tag in 'gss_for_code' which is

  15102. located in 'gimple.c'.

  15103. 

  15104.  In order for the garbage collector to know the size of the structure

  15105. you created in 'gimple.h', you need to add a case to handle your new

  15106. GIMPLE statement in 'gimple_size' which is located in 'gimple.c'.

  15107. 

  15108.  You will probably want to create a function to build the new gimple

  15109. statement in 'gimple.c'.  The function should be called

  15110. 'gimple_build_NEW-TUPLE-NAME', and should return the new tuple as a

  15111. pointer to the appropriate gimple subclass.

  15112. 

  15113.  If your new statement requires accessors for any members or operands it

  15114. may have, put simple inline accessors in 'gimple.h' and any non-trivial

  15115. accessors in 'gimple.c' with a corresponding prototype in 'gimple.h'.

  15116. 

  15117.  You should add the new statement subclass to the class hierarchy

  15118. diagram in 'gimple.texi'.

  15119. 

  15120. 

  15121. File: gccint.info,  Node: Statement and operand traversals,  Prev: Adding a new GIMPLE statement code,  Up: GIMPLE

  15122. 

  15123. 12.12 Statement and operand traversals

  15124. ======================================

  15125. 

  15126. There are two functions available for walking statements and sequences:

  15127. 'walk_gimple_stmt' and 'walk_gimple_seq', accordingly, and a third

  15128. function for walking the operands in a statement: 'walk_gimple_op'.

  15129. 

  15130.  -- GIMPLE function: tree walk_gimple_stmt (gimple_stmt_iterator *gsi,

  15131.           walk_stmt_fn callback_stmt, walk_tree_fn callback_op, struct

  15132.           walk_stmt_info *wi)

  15133.      This function is used to walk the current statement in 'GSI',

  15134.      optionally using traversal state stored in 'WI'.  If 'WI' is

  15135.      'NULL', no state is kept during the traversal.

  15136. 

  15137.      The callback 'CALLBACK_STMT' is called.  If 'CALLBACK_STMT' returns

  15138.      true, it means that the callback function has handled all the

  15139.      operands of the statement and it is not necessary to walk its

  15140.      operands.

  15141. 

  15142.      If 'CALLBACK_STMT' is 'NULL' or it returns false, 'CALLBACK_OP' is

  15143.      called on each operand of the statement via 'walk_gimple_op'.  If

  15144.      'walk_gimple_op' returns non-'NULL' for any operand, the remaining

  15145.      operands are not scanned.

  15146. 

  15147.      The return value is that returned by the last call to

  15148.      'walk_gimple_op', or 'NULL_TREE' if no 'CALLBACK_OP' is specified.

  15149. 

  15150.  -- GIMPLE function: tree walk_gimple_op (gimple stmt, walk_tree_fn

  15151.           callback_op, struct walk_stmt_info *wi)

  15152.      Use this function to walk the operands of statement 'STMT'.  Every

  15153.      operand is walked via 'walk_tree' with optional state information

  15154.      in 'WI'.

  15155. 

  15156.      'CALLBACK_OP' is called on each operand of 'STMT' via 'walk_tree'.

  15157.      Additional parameters to 'walk_tree' must be stored in 'WI'.  For

  15158.      each operand 'OP', 'walk_tree' is called as:

  15159. 

  15160.           walk_tree (&OP, CALLBACK_OP, WI, PSET)

  15161. 

  15162.      If 'CALLBACK_OP' returns non-'NULL' for an operand, the remaining

  15163.      operands are not scanned.  The return value is that returned by the

  15164.      last call to 'walk_tree', or 'NULL_TREE' if no 'CALLBACK_OP' is

  15165.      specified.

  15166. 

  15167.  -- GIMPLE function: tree walk_gimple_seq (gimple_seq seq, walk_stmt_fn

  15168.           callback_stmt, walk_tree_fn callback_op, struct walk_stmt_info

  15169.           *wi)

  15170.      This function walks all the statements in the sequence 'SEQ'

  15171.      calling 'walk_gimple_stmt' on each one.  'WI' is as in

  15172.      'walk_gimple_stmt'.  If 'walk_gimple_stmt' returns non-'NULL', the

  15173.      walk is stopped and the value returned.  Otherwise, all the

  15174.      statements are walked and 'NULL_TREE' returned.

  15175. 

  15176. 

  15177. File: gccint.info,  Node: Tree SSA,  Next: RTL,  Prev: GIMPLE,  Up: Top

  15178. 

  15179. 13 Analysis and Optimization of GIMPLE tuples

  15180. *********************************************

  15181. 

  15182. GCC uses three main intermediate languages to represent the program

  15183. during compilation: GENERIC, GIMPLE and RTL.  GENERIC is a

  15184. language-independent representation generated by each front end.  It is

  15185. used to serve as an interface between the parser and optimizer.  GENERIC

  15186. is a common representation that is able to represent programs written in

  15187. all the languages supported by GCC.

  15188. 

  15189.  GIMPLE and RTL are used to optimize the program.  GIMPLE is used for

  15190. target and language independent optimizations (e.g., inlining, constant

  15191. propagation, tail call elimination, redundancy elimination, etc).  Much

  15192. like GENERIC, GIMPLE is a language independent, tree based

  15193. representation.  However, it differs from GENERIC in that the GIMPLE

  15194. grammar is more restrictive: expressions contain no more than 3 operands

  15195. (except function calls), it has no control flow structures and

  15196. expressions with side effects are only allowed on the right hand side of

  15197. assignments.  See the chapter describing GENERIC and GIMPLE for more

  15198. details.

  15199. 

  15200.  This chapter describes the data structures and functions used in the

  15201. GIMPLE optimizers (also known as "tree optimizers" or "middle end").  In

  15202. particular, it focuses on all the macros, data structures, functions and

  15203. programming constructs needed to implement optimization passes for

  15204. GIMPLE.

  15205. 

  15206. * Menu:

  15207. 

  15208. * Annotations::         Attributes for variables.

  15209. * SSA Operands::        SSA names referenced by GIMPLE statements.

  15210. * SSA::                 Static Single Assignment representation.

  15211. * Alias analysis::      Representing aliased loads and stores.

  15212. * Memory model::        Memory model used by the middle-end.

  15213. 

  15214. 

  15215. File: gccint.info,  Node: Annotations,  Next: SSA Operands,  Up: Tree SSA

  15216. 

  15217. 13.1 Annotations

  15218. ================

  15219. 

  15220. The optimizers need to associate attributes with variables during the

  15221. optimization process.  For instance, we need to know whether a variable

  15222. has aliases.  All these attributes are stored in data structures called

  15223. annotations which are then linked to the field 'ann' in 'struct

  15224. tree_common'.

  15225. 

  15226. 

  15227. File: gccint.info,  Node: SSA Operands,  Next: SSA,  Prev: Annotations,  Up: Tree SSA

  15228. 

  15229. 13.2 SSA Operands

  15230. =================

  15231. 

  15232. Almost every GIMPLE statement will contain a reference to a variable or

  15233. memory location.  Since statements come in different shapes and sizes,

  15234. their operands are going to be located at various spots inside the

  15235. statement's tree.  To facilitate access to the statement's operands,

  15236. they are organized into lists associated inside each statement's

  15237. annotation.  Each element in an operand list is a pointer to a

  15238. 'VAR_DECL', 'PARM_DECL' or 'SSA_NAME' tree node.  This provides a very

  15239. convenient way of examining and replacing operands.

  15240. 

  15241.  Data flow analysis and optimization is done on all tree nodes

  15242. representing variables.  Any node for which 'SSA_VAR_P' returns nonzero

  15243. is considered when scanning statement operands.  However, not all

  15244. 'SSA_VAR_P' variables are processed in the same way.  For the purposes

  15245. of optimization, we need to distinguish between references to local

  15246. scalar variables and references to globals, statics, structures, arrays,

  15247. aliased variables, etc.  The reason is simple, the compiler can gather

  15248. complete data flow information for a local scalar.  On the other hand, a

  15249. global variable may be modified by a function call, it may not be

  15250. possible to keep track of all the elements of an array or the fields of

  15251. a structure, etc.

  15252. 

  15253.  The operand scanner gathers two kinds of operands: "real" and

  15254. "virtual".  An operand for which 'is_gimple_reg' returns true is

  15255. considered real, otherwise it is a virtual operand.  We also distinguish

  15256. between uses and definitions.  An operand is used if its value is loaded

  15257. by the statement (e.g., the operand at the RHS of an assignment).  If

  15258. the statement assigns a new value to the operand, the operand is

  15259. considered a definition (e.g., the operand at the LHS of an assignment).

  15260. 

  15261.  Virtual and real operands also have very different data flow

  15262. properties.  Real operands are unambiguous references to the full object

  15263. that they represent.  For instance, given

  15264. 

  15265.      {

  15266.        int a, b;

  15267.        a = b

  15268.      }

  15269. 

  15270.  Since 'a' and 'b' are non-aliased locals, the statement 'a = b' will

  15271. have one real definition and one real use because variable 'a' is

  15272. completely modified with the contents of variable 'b'.  Real definition

  15273. are also known as "killing definitions".  Similarly, the use of 'b'

  15274. reads all its bits.

  15275. 

  15276.  In contrast, virtual operands are used with variables that can have a

  15277. partial or ambiguous reference.  This includes structures, arrays,

  15278. globals, and aliased variables.  In these cases, we have two types of

  15279. definitions.  For globals, structures, and arrays, we can determine from

  15280. a statement whether a variable of these types has a killing definition.

  15281. If the variable does, then the statement is marked as having a "must

  15282. definition" of that variable.  However, if a statement is only defining

  15283. a part of the variable (i.e. a field in a structure), or if we know that

  15284. a statement might define the variable but we cannot say for sure, then

  15285. we mark that statement as having a "may definition".  For instance,

  15286. given

  15287. 

  15288.      {

  15289.        int a, b, *p;

  15290. 

  15291.        if (...)

  15292.          p = &a;

  15293.        else

  15294.          p = &b;

  15295.        *p = 5;

  15296.        return *p;

  15297.      }

  15298. 

  15299.  The assignment '*p = 5' may be a definition of 'a' or 'b'.  If we

  15300. cannot determine statically where 'p' is pointing to at the time of the

  15301. store operation, we create virtual definitions to mark that statement as

  15302. a potential definition site for 'a' and 'b'.  Memory loads are similarly

  15303. marked with virtual use operands.  Virtual operands are shown in tree

  15304. dumps right before the statement that contains them.  To request a tree

  15305. dump with virtual operands, use the '-vops' option to '-fdump-tree':

  15306. 

  15307.      {

  15308.        int a, b, *p;

  15309. 

  15310.        if (...)

  15311.          p = &a;

  15312.        else

  15313.          p = &b;

  15314.        # a = VDEF <a>

  15315.        # b = VDEF <b>

  15316.        *p = 5;

  15317. 

  15318.        # VUSE <a>

  15319.        # VUSE <b>

  15320.        return *p;

  15321.      }

  15322. 

  15323.  Notice that 'VDEF' operands have two copies of the referenced variable.

  15324. This indicates that this is not a killing definition of that variable.

  15325. In this case we refer to it as a "may definition" or "aliased store".

  15326. The presence of the second copy of the variable in the 'VDEF' operand

  15327. will become important when the function is converted into SSA form.

  15328. This will be used to link all the non-killing definitions to prevent

  15329. optimizations from making incorrect assumptions about them.

  15330. 

  15331.  Operands are updated as soon as the statement is finished via a call to

  15332. 'update_stmt'.  If statement elements are changed via 'SET_USE' or

  15333. 'SET_DEF', then no further action is required (i.e., those macros take

  15334. care of updating the statement).  If changes are made by manipulating

  15335. the statement's tree directly, then a call must be made to 'update_stmt'

  15336. when complete.  Calling one of the 'bsi_insert' routines or

  15337. 'bsi_replace' performs an implicit call to 'update_stmt'.

  15338. 

  15339. 13.2.1 Operand Iterators And Access Routines

  15340. --------------------------------------------

  15341. 

  15342. Operands are collected by 'tree-ssa-operands.c'.  They are stored inside

  15343. each statement's annotation and can be accessed through either the

  15344. operand iterators or an access routine.

  15345. 

  15346.  The following access routines are available for examining operands:

  15347. 

  15348.   1. 'SINGLE_SSA_{USE,DEF,TREE}_OPERAND': These accessors will return

  15349.      NULL unless there is exactly one operand matching the specified

  15350.      flags.  If there is exactly one operand, the operand is returned as

  15351.      either a 'tree', 'def_operand_p', or 'use_operand_p'.

  15352. 

  15353.           tree t = SINGLE_SSA_TREE_OPERAND (stmt, flags);

  15354.           use_operand_p u = SINGLE_SSA_USE_OPERAND (stmt, SSA_ALL_VIRTUAL_USES);

  15355.           def_operand_p d = SINGLE_SSA_DEF_OPERAND (stmt, SSA_OP_ALL_DEFS);

  15356. 

  15357.   2. 'ZERO_SSA_OPERANDS': This macro returns true if there are no

  15358.      operands matching the specified flags.

  15359. 

  15360.           if (ZERO_SSA_OPERANDS (stmt, SSA_OP_ALL_VIRTUALS))

  15361.             return;

  15362. 

  15363.   3. 'NUM_SSA_OPERANDS': This macro Returns the number of operands

  15364.      matching 'flags'.  This actually executes a loop to perform the

  15365.      count, so only use this if it is really needed.

  15366. 

  15367.           int count = NUM_SSA_OPERANDS (stmt, flags)

  15368. 

  15369.  If you wish to iterate over some or all operands, use the

  15370. 'FOR_EACH_SSA_{USE,DEF,TREE}_OPERAND' iterator.  For example, to print

  15371. all the operands for a statement:

  15372. 

  15373.      void

  15374.      print_ops (tree stmt)

  15375.      {

  15376.        ssa_op_iter;

  15377.        tree var;

  15378. 

  15379.        FOR_EACH_SSA_TREE_OPERAND (var, stmt, iter, SSA_OP_ALL_OPERANDS)

  15380.          print_generic_expr (stderr, var, TDF_SLIM);

  15381.      }

  15382. 

  15383.  How to choose the appropriate iterator:

  15384. 

  15385.   1. Determine whether you are need to see the operand pointers, or just

  15386.      the trees, and choose the appropriate macro:

  15387. 

  15388.           Need            Macro:

  15389.           ----            -------

  15390.           use_operand_p   FOR_EACH_SSA_USE_OPERAND

  15391.           def_operand_p   FOR_EACH_SSA_DEF_OPERAND

  15392.           tree            FOR_EACH_SSA_TREE_OPERAND

  15393. 

  15394.   2. You need to declare a variable of the type you are interested in,

  15395.      and an ssa_op_iter structure which serves as the loop controlling

  15396.      variable.

  15397. 

  15398.   3. Determine which operands you wish to use, and specify the flags of

  15399.      those you are interested in.  They are documented in

  15400.      'tree-ssa-operands.h':

  15401. 

  15402.           #define SSA_OP_USE              0x01    /* Real USE operands.  */

  15403.           #define SSA_OP_DEF              0x02    /* Real DEF operands.  */

  15404.           #define SSA_OP_VUSE             0x04    /* VUSE operands.  */

  15405.           #define SSA_OP_VDEF             0x08    /* VDEF operands.  */

  15406. 

  15407.           /* These are commonly grouped operand flags.  */

  15408.           #define SSA_OP_VIRTUAL_USES	(SSA_OP_VUSE)

  15409.           #define SSA_OP_VIRTUAL_DEFS	(SSA_OP_VDEF)

  15410.           #define SSA_OP_ALL_VIRTUALS     (SSA_OP_VIRTUAL_USES | SSA_OP_VIRTUAL_DEFS)

  15411.           #define SSA_OP_ALL_USES		(SSA_OP_VIRTUAL_USES | SSA_OP_USE)

  15412.           #define SSA_OP_ALL_DEFS		(SSA_OP_VIRTUAL_DEFS | SSA_OP_DEF)

  15413.           #define SSA_OP_ALL_OPERANDS	(SSA_OP_ALL_USES | SSA_OP_ALL_DEFS)

  15414. 

  15415.  So if you want to look at the use pointers for all the 'USE' and 'VUSE'

  15416. operands, you would do something like:

  15417. 

  15418.        use_operand_p use_p;

  15419.        ssa_op_iter iter;

  15420. 

  15421.        FOR_EACH_SSA_USE_OPERAND (use_p, stmt, iter, (SSA_OP_USE | SSA_OP_VUSE))

  15422.          {

  15423.            process_use_ptr (use_p);

  15424.          }

  15425. 

  15426.  The 'TREE' macro is basically the same as the 'USE' and 'DEF' macros,

  15427. only with the use or def dereferenced via 'USE_FROM_PTR (use_p)' and

  15428. 'DEF_FROM_PTR (def_p)'.  Since we aren't using operand pointers, use and

  15429. defs flags can be mixed.

  15430. 

  15431.        tree var;

  15432.        ssa_op_iter iter;

  15433. 

  15434.        FOR_EACH_SSA_TREE_OPERAND (var, stmt, iter, SSA_OP_VUSE)

  15435.          {

  15436.             print_generic_expr (stderr, var, TDF_SLIM);

  15437.          }

  15438. 

  15439.  'VDEF's are broken into two flags, one for the 'DEF' portion

  15440. ('SSA_OP_VDEF') and one for the USE portion ('SSA_OP_VUSE').

  15441. 

  15442.  There are many examples in the code, in addition to the documentation

  15443. in 'tree-ssa-operands.h' and 'ssa-iterators.h'.

  15444. 

  15445.  There are also a couple of variants on the stmt iterators regarding PHI

  15446. nodes.

  15447. 

  15448.  'FOR_EACH_PHI_ARG' Works exactly like 'FOR_EACH_SSA_USE_OPERAND',

  15449. except it works over 'PHI' arguments instead of statement operands.

  15450. 

  15451.      /* Look at every virtual PHI use.  */

  15452.      FOR_EACH_PHI_ARG (use_p, phi_stmt, iter, SSA_OP_VIRTUAL_USES)

  15453.      {

  15454.         my_code;

  15455.      }

  15456. 

  15457.      /* Look at every real PHI use.  */

  15458.      FOR_EACH_PHI_ARG (use_p, phi_stmt, iter, SSA_OP_USES)

  15459.        my_code;

  15460. 

  15461.      /* Look at every PHI use.  */

  15462.      FOR_EACH_PHI_ARG (use_p, phi_stmt, iter, SSA_OP_ALL_USES)

  15463.        my_code;

  15464. 

  15465.  'FOR_EACH_PHI_OR_STMT_{USE,DEF}' works exactly like

  15466. 'FOR_EACH_SSA_{USE,DEF}_OPERAND', except it will function on either a

  15467. statement or a 'PHI' node.  These should be used when it is appropriate

  15468. but they are not quite as efficient as the individual 'FOR_EACH_PHI' and

  15469. 'FOR_EACH_SSA' routines.

  15470. 

  15471.      FOR_EACH_PHI_OR_STMT_USE (use_operand_p, stmt, iter, flags)

  15472.        {

  15473.           my_code;

  15474.        }

  15475. 

  15476.      FOR_EACH_PHI_OR_STMT_DEF (def_operand_p, phi, iter, flags)

  15477.        {

  15478.           my_code;

  15479.        }

  15480. 

  15481. 13.2.2 Immediate Uses

  15482. ---------------------

  15483. 

  15484. Immediate use information is now always available.  Using the immediate

  15485. use iterators, you may examine every use of any 'SSA_NAME'.  For

  15486. instance, to change each use of 'ssa_var' to 'ssa_var2' and call

  15487. fold_stmt on each stmt after that is done:

  15488. 

  15489.        use_operand_p imm_use_p;

  15490.        imm_use_iterator iterator;

  15491.        tree ssa_var, stmt;

  15492. 

  15493. 

  15494.        FOR_EACH_IMM_USE_STMT (stmt, iterator, ssa_var)

  15495.          {

  15496.            FOR_EACH_IMM_USE_ON_STMT (imm_use_p, iterator)

  15497.              SET_USE (imm_use_p, ssa_var_2);

  15498.            fold_stmt (stmt);

  15499.          }

  15500. 

  15501.  There are 2 iterators which can be used.  'FOR_EACH_IMM_USE_FAST' is

  15502. used when the immediate uses are not changed, i.e., you are looking at

  15503. the uses, but not setting them.

  15504. 

  15505.  If they do get changed, then care must be taken that things are not

  15506. changed under the iterators, so use the 'FOR_EACH_IMM_USE_STMT' and

  15507. 'FOR_EACH_IMM_USE_ON_STMT' iterators.  They attempt to preserve the

  15508. sanity of the use list by moving all the uses for a statement into a

  15509. controlled position, and then iterating over those uses.  Then the

  15510. optimization can manipulate the stmt when all the uses have been

  15511. processed.  This is a little slower than the FAST version since it adds

  15512. a placeholder element and must sort through the list a bit for each

  15513. statement.  This placeholder element must be also be removed if the loop

  15514. is terminated early.  The macro 'BREAK_FROM_IMM_USE_STMT' is provided to

  15515. do this :

  15516. 

  15517.        FOR_EACH_IMM_USE_STMT (stmt, iterator, ssa_var)

  15518.          {

  15519.            if (stmt == last_stmt)

  15520.              BREAK_FROM_IMM_USE_STMT (iterator);

  15521. 

  15522.            FOR_EACH_IMM_USE_ON_STMT (imm_use_p, iterator)

  15523.              SET_USE (imm_use_p, ssa_var_2);

  15524.            fold_stmt (stmt);

  15525.          }

  15526. 

  15527.  There are checks in 'verify_ssa' which verify that the immediate use

  15528. list is up to date, as well as checking that an optimization didn't

  15529. break from the loop without using this macro.  It is safe to simply

  15530. 'break'; from a 'FOR_EACH_IMM_USE_FAST' traverse.

  15531. 

  15532.  Some useful functions and macros:

  15533.   1. 'has_zero_uses (ssa_var)' : Returns true if there are no uses of

  15534.      'ssa_var'.

  15535.   2. 'has_single_use (ssa_var)' : Returns true if there is only a single

  15536.      use of 'ssa_var'.

  15537.   3. 'single_imm_use (ssa_var, use_operand_p *ptr, tree *stmt)' :

  15538.      Returns true if there is only a single use of 'ssa_var', and also

  15539.      returns the use pointer and statement it occurs in, in the second

  15540.      and third parameters.

  15541.   4. 'num_imm_uses (ssa_var)' : Returns the number of immediate uses of

  15542.      'ssa_var'.  It is better not to use this if possible since it

  15543.      simply utilizes a loop to count the uses.

  15544.   5. 'PHI_ARG_INDEX_FROM_USE (use_p)' : Given a use within a 'PHI' node,

  15545.      return the index number for the use.  An assert is triggered if the

  15546.      use isn't located in a 'PHI' node.

  15547.   6. 'USE_STMT (use_p)' : Return the statement a use occurs in.

  15548. 

  15549.  Note that uses are not put into an immediate use list until their

  15550. statement is actually inserted into the instruction stream via a 'bsi_*'

  15551. routine.

  15552. 

  15553.  It is also still possible to utilize lazy updating of statements, but

  15554. this should be used only when absolutely required.  Both alias analysis

  15555. and the dominator optimizations currently do this.

  15556. 

  15557.  When lazy updating is being used, the immediate use information is out

  15558. of date and cannot be used reliably.  Lazy updating is achieved by

  15559. simply marking statements modified via calls to 'gimple_set_modified'

  15560. instead of 'update_stmt'.  When lazy updating is no longer required, all

  15561. the modified statements must have 'update_stmt' called in order to bring

  15562. them up to date.  This must be done before the optimization is finished,

  15563. or 'verify_ssa' will trigger an abort.

  15564. 

  15565.  This is done with a simple loop over the instruction stream:

  15566.        block_stmt_iterator bsi;

  15567.        basic_block bb;

  15568.        FOR_EACH_BB (bb)

  15569.          {

  15570.            for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))

  15571.              update_stmt_if_modified (bsi_stmt (bsi));

  15572.          }

  15573. 

  15574. 

  15575. File: gccint.info,  Node: SSA,  Next: Alias analysis,  Prev: SSA Operands,  Up: Tree SSA

  15576. 

  15577. 13.3 Static Single Assignment

  15578. =============================

  15579. 

  15580. Most of the tree optimizers rely on the data flow information provided

  15581. by the Static Single Assignment (SSA) form.  We implement the SSA form

  15582. as described in 'R. Cytron, J. Ferrante, B. Rosen, M. Wegman, and K.

  15583. Zadeck. Efficiently Computing Static Single Assignment Form and the

  15584. Control Dependence Graph. ACM Transactions on Programming Languages and

  15585. Systems, 13(4):451-490, October 1991'.

  15586. 

  15587.  The SSA form is based on the premise that program variables are

  15588. assigned in exactly one location in the program.  Multiple assignments

  15589. to the same variable create new versions of that variable.  Naturally,

  15590. actual programs are seldom in SSA form initially because variables tend

  15591. to be assigned multiple times.  The compiler modifies the program

  15592. representation so that every time a variable is assigned in the code, a

  15593. new version of the variable is created.  Different versions of the same

  15594. variable are distinguished by subscripting the variable name with its

  15595. version number.  Variables used in the right-hand side of expressions

  15596. are renamed so that their version number matches that of the most recent

  15597. assignment.

  15598. 

  15599.  We represent variable versions using 'SSA_NAME' nodes.  The renaming

  15600. process in 'tree-ssa.c' wraps every real and virtual operand with an

  15601. 'SSA_NAME' node which contains the version number and the statement that

  15602. created the 'SSA_NAME'.  Only definitions and virtual definitions may

  15603. create new 'SSA_NAME' nodes.

  15604. 

  15605.  Sometimes, flow of control makes it impossible to determine the most

  15606. recent version of a variable.  In these cases, the compiler inserts an

  15607. artificial definition for that variable called "PHI function" or "PHI

  15608. node".  This new definition merges all the incoming versions of the

  15609. variable to create a new name for it.  For instance,

  15610. 

  15611.      if (...)

  15612.        a_1 = 5;

  15613.      else if (...)

  15614.        a_2 = 2;

  15615.      else

  15616.        a_3 = 13;

  15617. 

  15618.      # a_4 = PHI <a_1, a_2, a_3>

  15619.      return a_4;

  15620. 

  15621.  Since it is not possible to determine which of the three branches will

  15622. be taken at runtime, we don't know which of 'a_1', 'a_2' or 'a_3' to use

  15623. at the return statement.  So, the SSA renamer creates a new version

  15624. 'a_4' which is assigned the result of "merging" 'a_1', 'a_2' and 'a_3'.

  15625. Hence, PHI nodes mean "one of these operands.  I don't know which".

  15626. 

  15627.  The following functions can be used to examine PHI nodes

  15628. 

  15629.  -- Function: gimple_phi_result (PHI)

  15630.      Returns the 'SSA_NAME' created by PHI node PHI (i.e., PHI's LHS).

  15631. 

  15632.  -- Function: gimple_phi_num_args (PHI)

  15633.      Returns the number of arguments in PHI.  This number is exactly the

  15634.      number of incoming edges to the basic block holding PHI.

  15635. 

  15636.  -- Function: gimple_phi_arg (PHI, I)

  15637.      Returns Ith argument of PHI.

  15638. 

  15639.  -- Function: gimple_phi_arg_edge (PHI, I)

  15640.      Returns the incoming edge for the Ith argument of PHI.

  15641. 

  15642.  -- Function: gimple_phi_arg_def (PHI, I)

  15643.      Returns the 'SSA_NAME' for the Ith argument of PHI.

  15644. 

  15645. 13.3.1 Preserving the SSA form

  15646. ------------------------------

  15647. 

  15648. Some optimization passes make changes to the function that invalidate

  15649. the SSA property.  This can happen when a pass has added new symbols or

  15650. changed the program so that variables that were previously aliased

  15651. aren't anymore.  Whenever something like this happens, the affected

  15652. symbols must be renamed into SSA form again.  Transformations that emit

  15653. new code or replicate existing statements will also need to update the

  15654. SSA form.

  15655. 

  15656.  Since GCC implements two different SSA forms for register and virtual

  15657. variables, keeping the SSA form up to date depends on whether you are

  15658. updating register or virtual names.  In both cases, the general idea

  15659. behind incremental SSA updates is similar: when new SSA names are

  15660. created, they typically are meant to replace other existing names in the

  15661. program.

  15662. 

  15663.  For instance, given the following code:

  15664. 

  15665.           1  L0:

  15666.           2  x_1 = PHI (0, x_5)

  15667.           3  if (x_1 < 10)

  15668.           4    if (x_1 > 7)

  15669.           5      y_2 = 0

  15670.           6    else

  15671.           7      y_3 = x_1 + x_7

  15672.           8    endif

  15673.           9    x_5 = x_1 + 1

  15674.           10   goto L0;

  15675.           11 endif

  15676. 

  15677.  Suppose that we insert new names 'x_10' and 'x_11' (lines '4' and '8').

  15678. 

  15679.           1  L0:

  15680.           2  x_1 = PHI (0, x_5)

  15681.           3  if (x_1 < 10)

  15682.           4    x_10 = ...

  15683.           5    if (x_1 > 7)

  15684.           6      y_2 = 0

  15685.           7    else

  15686.           8      x_11 = ...

  15687.           9      y_3 = x_1 + x_7

  15688.           10   endif

  15689.           11   x_5 = x_1 + 1

  15690.           12   goto L0;

  15691.           13 endif

  15692. 

  15693.  We want to replace all the uses of 'x_1' with the new definitions of

  15694. 'x_10' and 'x_11'.  Note that the only uses that should be replaced are

  15695. those at lines '5', '9' and '11'.  Also, the use of 'x_7' at line '9'

  15696. should _not_ be replaced (this is why we cannot just mark symbol 'x' for

  15697. renaming).

  15698. 

  15699.  Additionally, we may need to insert a PHI node at line '11' because

  15700. that is a merge point for 'x_10' and 'x_11'.  So the use of 'x_1' at

  15701. line '11' will be replaced with the new PHI node.  The insertion of PHI

  15702. nodes is optional.  They are not strictly necessary to preserve the SSA

  15703. form, and depending on what the caller inserted, they may not even be

  15704. useful for the optimizers.

  15705. 

  15706.  Updating the SSA form is a two step process.  First, the pass has to

  15707. identify which names need to be updated and/or which symbols need to be

  15708. renamed into SSA form for the first time.  When new names are introduced

  15709. to replace existing names in the program, the mapping between the old

  15710. and the new names are registered by calling 'register_new_name_mapping'

  15711. (note that if your pass creates new code by duplicating basic blocks,

  15712. the call to 'tree_duplicate_bb' will set up the necessary mappings

  15713. automatically).

  15714. 

  15715.  After the replacement mappings have been registered and new symbols

  15716. marked for renaming, a call to 'update_ssa' makes the registered

  15717. changes.  This can be done with an explicit call or by creating 'TODO'

  15718. flags in the 'tree_opt_pass' structure for your pass.  There are several

  15719. 'TODO' flags that control the behavior of 'update_ssa':

  15720. 

  15721.    * 'TODO_update_ssa'.  Update the SSA form inserting PHI nodes for

  15722.      newly exposed symbols and virtual names marked for updating.  When

  15723.      updating real names, only insert PHI nodes for a real name 'O_j' in

  15724.      blocks reached by all the new and old definitions for 'O_j'.  If

  15725.      the iterated dominance frontier for 'O_j' is not pruned, we may end

  15726.      up inserting PHI nodes in blocks that have one or more edges with

  15727.      no incoming definition for 'O_j'.  This would lead to uninitialized

  15728.      warnings for 'O_j''s symbol.

  15729. 

  15730.    * 'TODO_update_ssa_no_phi'.  Update the SSA form without inserting

  15731.      any new PHI nodes at all.  This is used by passes that have either

  15732.      inserted all the PHI nodes themselves or passes that need only to

  15733.      patch use-def and def-def chains for virtuals (e.g., DCE).

  15734. 

  15735.    * 'TODO_update_ssa_full_phi'.  Insert PHI nodes everywhere they are

  15736.      needed.  No pruning of the IDF is done.  This is used by passes

  15737.      that need the PHI nodes for 'O_j' even if it means that some

  15738.      arguments will come from the default definition of 'O_j''s symbol

  15739.      (e.g., 'pass_linear_transform').

  15740. 

  15741.      WARNING: If you need to use this flag, chances are that your pass

  15742.      may be doing something wrong.  Inserting PHI nodes for an old name

  15743.      where not all edges carry a new replacement may lead to silent

  15744.      codegen errors or spurious uninitialized warnings.

  15745. 

  15746.    * 'TODO_update_ssa_only_virtuals'.  Passes that update the SSA form

  15747.      on their own may want to delegate the updating of virtual names to

  15748.      the generic updater.  Since FUD chains are easier to maintain, this

  15749.      simplifies the work they need to do.  NOTE: If this flag is used,

  15750.      any OLD->NEW mappings for real names are explicitly destroyed and

  15751.      only the symbols marked for renaming are processed.

  15752. 

  15753. 13.3.2 Examining 'SSA_NAME' nodes

  15754. ---------------------------------

  15755. 

  15756. The following macros can be used to examine 'SSA_NAME' nodes

  15757. 

  15758.  -- Macro: SSA_NAME_DEF_STMT (VAR)

  15759.      Returns the statement S that creates the 'SSA_NAME' VAR.  If S is

  15760.      an empty statement (i.e., 'IS_EMPTY_STMT (S)' returns 'true'), it

  15761.      means that the first reference to this variable is a USE or a VUSE.

  15762. 

  15763.  -- Macro: SSA_NAME_VERSION (VAR)

  15764.      Returns the version number of the 'SSA_NAME' object VAR.

  15765. 

  15766. 13.3.3 Walking the dominator tree

  15767. ---------------------------------

  15768. 

  15769.  -- Tree SSA function: void walk_dominator_tree (WALK_DATA, BB)

  15770. 

  15771.      This function walks the dominator tree for the current CFG calling

  15772.      a set of callback functions defined in STRUCT DOM_WALK_DATA in

  15773.      'domwalk.h'.  The call back functions you need to define give you

  15774.      hooks to execute custom code at various points during traversal:

  15775. 

  15776.        1. Once to initialize any local data needed while processing BB

  15777.           and its children.  This local data is pushed into an internal

  15778.           stack which is automatically pushed and popped as the walker

  15779.           traverses the dominator tree.

  15780. 

  15781.        2. Once before traversing all the statements in the BB.

  15782. 

  15783.        3. Once for every statement inside BB.

  15784. 

  15785.        4. Once after traversing all the statements and before recursing

  15786.           into BB's dominator children.

  15787. 

  15788.        5. It then recurses into all the dominator children of BB.

  15789. 

  15790.        6. After recursing into all the dominator children of BB it can,

  15791.           optionally, traverse every statement in BB again (i.e.,

  15792.           repeating steps 2 and 3).

  15793. 

  15794.        7. Once after walking the statements in BB and BB's dominator

  15795.           children.  At this stage, the block local data stack is

  15796.           popped.

  15797. 

  15798. 

  15799. File: gccint.info,  Node: Alias analysis,  Next: Memory model,  Prev: SSA,  Up: Tree SSA

  15800. 

  15801. 13.4 Alias analysis

  15802. ===================

  15803. 

  15804. Alias analysis in GIMPLE SSA form consists of two pieces.  First the

  15805. virtual SSA web ties conflicting memory accesses and provides a SSA

  15806. use-def chain and SSA immediate-use chains for walking possibly

  15807. dependent memory accesses.  Second an alias-oracle can be queried to

  15808. disambiguate explicit and implicit memory references.

  15809. 

  15810.   1. Memory SSA form.

  15811. 

  15812.      All statements that may use memory have exactly one accompanied use

  15813.      of a virtual SSA name that represents the state of memory at the

  15814.      given point in the IL.

  15815. 

  15816.      All statements that may define memory have exactly one accompanied

  15817.      definition of a virtual SSA name using the previous state of memory

  15818.      and defining the new state of memory after the given point in the

  15819.      IL.

  15820. 

  15821.           int i;

  15822.           int foo (void)

  15823.           {

  15824.             # .MEM_3 = VDEF <.MEM_2(D)>

  15825.             i = 1;

  15826.             # VUSE <.MEM_3>

  15827.             return i;

  15828.           }

  15829. 

  15830.      The virtual SSA names in this case are '.MEM_2(D)' and '.MEM_3'.

  15831.      The store to the global variable 'i' defines '.MEM_3' invalidating

  15832.      '.MEM_2(D)'.  The load from 'i' uses that new state '.MEM_3'.

  15833. 

  15834.      The virtual SSA web serves as constraints to SSA optimizers

  15835.      preventing illegitimate code-motion and optimization.  It also

  15836.      provides a way to walk related memory statements.

  15837. 

  15838.   2. Points-to and escape analysis.

  15839. 

  15840.      Points-to analysis builds a set of constraints from the GIMPLE SSA

  15841.      IL representing all pointer operations and facts we do or do not

  15842.      know about pointers.  Solving this set of constraints yields a

  15843.      conservatively correct solution for each pointer variable in the

  15844.      program (though we are only interested in SSA name pointers) as to

  15845.      what it may possibly point to.

  15846. 

  15847.      This points-to solution for a given SSA name pointer is stored in

  15848.      the 'pt_solution' sub-structure of the 'SSA_NAME_PTR_INFO' record.

  15849.      The following accessor functions are available:

  15850. 

  15851.         * 'pt_solution_includes'

  15852.         * 'pt_solutions_intersect'

  15853. 

  15854.      Points-to analysis also computes the solution for two special set

  15855.      of pointers, 'ESCAPED' and 'CALLUSED'.  Those represent all memory

  15856.      that has escaped the scope of analysis or that is used by pure or

  15857.      nested const calls.

  15858. 

  15859.   3. Type-based alias analysis

  15860. 

  15861.      Type-based alias analysis is frontend dependent though generic

  15862.      support is provided by the middle-end in 'alias.c'.  TBAA code is

  15863.      used by both tree optimizers and RTL optimizers.

  15864. 

  15865.      Every language that wishes to perform language-specific alias

  15866.      analysis should define a function that computes, given a 'tree'

  15867.      node, an alias set for the node.  Nodes in different alias sets are

  15868.      not allowed to alias.  For an example, see the C front-end function

  15869.      'c_get_alias_set'.

  15870. 

  15871.   4. Tree alias-oracle

  15872. 

  15873.      The tree alias-oracle provides means to disambiguate two memory

  15874.      references and memory references against statements.  The following

  15875.      queries are available:

  15876. 

  15877.         * 'refs_may_alias_p'

  15878.         * 'ref_maybe_used_by_stmt_p'

  15879.         * 'stmt_may_clobber_ref_p'

  15880. 

  15881.      In addition to those two kind of statement walkers are available

  15882.      walking statements related to a reference ref.

  15883.      'walk_non_aliased_vuses' walks over dominating memory defining

  15884.      statements and calls back if the statement does not clobber ref

  15885.      providing the non-aliased VUSE. The walk stops at the first

  15886.      clobbering statement or if asked to.  'walk_aliased_vdefs' walks

  15887.      over dominating memory defining statements and calls back on each

  15888.      statement clobbering ref providing its aliasing VDEF. The walk

  15889.      stops if asked to.

  15890. 

  15891. 

  15892. File: gccint.info,  Node: Memory model,  Prev: Alias analysis,  Up: Tree SSA

  15893. 

  15894. 13.5 Memory model

  15895. =================

  15896. 

  15897. The memory model used by the middle-end models that of the C/C++

  15898. languages.  The middle-end has the notion of an effective type of a

  15899. memory region which is used for type-based alias analysis.

  15900. 

  15901.  The following is a refinement of ISO C99 6.5/6, clarifying the block

  15902. copy case to follow common sense and extending the concept of a dynamic

  15903. effective type to objects with a declared type as required for C++.

  15904. 

  15905.      The effective type of an object for an access to its stored value is

  15906.      the declared type of the object or the effective type determined by

  15907.      a previous store to it.  If a value is stored into an object through

  15908.      an lvalue having a type that is not a character type, then the

  15909.      type of the lvalue becomes the effective type of the object for that

  15910.      access and for subsequent accesses that do not modify the stored value.

  15911.      If a value is copied into an object using memcpy or memmove,

  15912.      or is copied as an array of character type, then the effective type

  15913.      of the modified object for that access and for subsequent accesses that

  15914.      do not modify the value is undetermined.  For all other accesses to an

  15915.      object, the effective type of the object is simply the type of the

  15916.      lvalue used for the access.

  15917. 

  15918. 

  15919. File: gccint.info,  Node: RTL,  Next: Control Flow,  Prev: Tree SSA,  Up: Top

  15920. 

  15921. 14 RTL Representation

  15922. *********************

  15923. 

  15924. The last part of the compiler work is done on a low-level intermediate

  15925. representation called Register Transfer Language.  In this language, the

  15926. instructions to be output are described, pretty much one by one, in an

  15927. algebraic form that describes what the instruction does.

  15928. 

  15929.  RTL is inspired by Lisp lists.  It has both an internal form, made up

  15930. of structures that point at other structures, and a textual form that is

  15931. used in the machine description and in printed debugging dumps.  The

  15932. textual form uses nested parentheses to indicate the pointers in the

  15933. internal form.

  15934. 

  15935. * Menu:

  15936. 

  15937. * RTL Objects::       Expressions vs vectors vs strings vs integers.

  15938. * RTL Classes::       Categories of RTL expression objects, and their structure.

  15939. * Accessors::         Macros to access expression operands or vector elts.

  15940. * Special Accessors:: Macros to access specific annotations on RTL.

  15941. * Flags::             Other flags in an RTL expression.

  15942. * Machine Modes::     Describing the size and format of a datum.

  15943. * Constants::         Expressions with constant values.

  15944. * Regs and Memory::   Expressions representing register contents or memory.

  15945. * Arithmetic::        Expressions representing arithmetic on other expressions.

  15946. * Comparisons::       Expressions representing comparison of expressions.

  15947. * Bit-Fields::        Expressions representing bit-fields in memory or reg.

  15948. * Vector Operations:: Expressions involving vector datatypes.

  15949. * Conversions::       Extending, truncating, floating or fixing.

  15950. * RTL Declarations::  Declaring volatility, constancy, etc.

  15951. * Side Effects::      Expressions for storing in registers, etc.

  15952. * Incdec::            Embedded side-effects for autoincrement addressing.

  15953. * Assembler::         Representing 'asm' with operands.

  15954. * Debug Information:: Expressions representing debugging information.

  15955. * Insns::             Expression types for entire insns.

  15956. * Calls::             RTL representation of function call insns.

  15957. * Sharing::           Some expressions are unique; others *must* be copied.

  15958. * Reading RTL::       Reading textual RTL from a file.

  15959. 

  15960. 

  15961. File: gccint.info,  Node: RTL Objects,  Next: RTL Classes,  Up: RTL

  15962. 

  15963. 14.1 RTL Object Types

  15964. =====================

  15965. 

  15966. RTL uses five kinds of objects: expressions, integers, wide integers,

  15967. strings and vectors.  Expressions are the most important ones.  An RTL

  15968. expression ("RTX", for short) is a C structure, but it is usually

  15969. referred to with a pointer; a type that is given the typedef name 'rtx'.

  15970. 

  15971.  An integer is simply an 'int'; their written form uses decimal digits.

  15972. A wide integer is an integral object whose type is 'HOST_WIDE_INT';

  15973. their written form uses decimal digits.

  15974. 

  15975.  A string is a sequence of characters.  In core it is represented as a

  15976. 'char *' in usual C fashion, and it is written in C syntax as well.

  15977. However, strings in RTL may never be null.  If you write an empty string

  15978. in a machine description, it is represented in core as a null pointer

  15979. rather than as a pointer to a null character.  In certain contexts,

  15980. these null pointers instead of strings are valid.  Within RTL code,

  15981. strings are most commonly found inside 'symbol_ref' expressions, but

  15982. they appear in other contexts in the RTL expressions that make up

  15983. machine descriptions.

  15984. 

  15985.  In a machine description, strings are normally written with double

  15986. quotes, as you would in C.  However, strings in machine descriptions may

  15987. extend over many lines, which is invalid C, and adjacent string

  15988. constants are not concatenated as they are in C.  Any string constant

  15989. may be surrounded with a single set of parentheses.  Sometimes this

  15990. makes the machine description easier to read.

  15991. 

  15992.  There is also a special syntax for strings, which can be useful when C

  15993. code is embedded in a machine description.  Wherever a string can

  15994. appear, it is also valid to write a C-style brace block.  The entire

  15995. brace block, including the outermost pair of braces, is considered to be

  15996. the string constant.  Double quote characters inside the braces are not

  15997. special.  Therefore, if you write string constants in the C code, you

  15998. need not escape each quote character with a backslash.

  15999. 

  16000.  A vector contains an arbitrary number of pointers to expressions.  The

  16001. number of elements in the vector is explicitly present in the vector.

  16002. The written form of a vector consists of square brackets ('[...]')

  16003. surrounding the elements, in sequence and with whitespace separating

  16004. them.  Vectors of length zero are not created; null pointers are used

  16005. instead.

  16006. 

  16007.  Expressions are classified by "expression codes" (also called RTX

  16008. codes).  The expression code is a name defined in 'rtl.def', which is

  16009. also (in uppercase) a C enumeration constant.  The possible expression

  16010. codes and their meanings are machine-independent.  The code of an RTX

  16011. can be extracted with the macro 'GET_CODE (X)' and altered with

  16012. 'PUT_CODE (X, NEWCODE)'.

  16013. 

  16014.  The expression code determines how many operands the expression

  16015. contains, and what kinds of objects they are.  In RTL, unlike Lisp, you

  16016. cannot tell by looking at an operand what kind of object it is.

  16017. Instead, you must know from its context--from the expression code of the

  16018. containing expression.  For example, in an expression of code 'subreg',

  16019. the first operand is to be regarded as an expression and the second

  16020. operand as a polynomial integer.  In an expression of code 'plus', there

  16021. are two operands, both of which are to be regarded as expressions.  In a

  16022. 'symbol_ref' expression, there is one operand, which is to be regarded

  16023. as a string.

  16024. 

  16025.  Expressions are written as parentheses containing the name of the

  16026. expression type, its flags and machine mode if any, and then the

  16027. operands of the expression (separated by spaces).

  16028. 

  16029.  Expression code names in the 'md' file are written in lowercase, but

  16030. when they appear in C code they are written in uppercase.  In this

  16031. manual, they are shown as follows: 'const_int'.

  16032. 

  16033.  In a few contexts a null pointer is valid where an expression is

  16034. normally wanted.  The written form of this is '(nil)'.

  16035. 

  16036. 

  16037. File: gccint.info,  Node: RTL Classes,  Next: Accessors,  Prev: RTL Objects,  Up: RTL

  16038. 

  16039. 14.2 RTL Classes and Formats

  16040. ============================

  16041. 

  16042. The various expression codes are divided into several "classes", which

  16043. are represented by single characters.  You can determine the class of an

  16044. RTX code with the macro 'GET_RTX_CLASS (CODE)'.  Currently, 'rtl.def'

  16045. defines these classes:

  16046. 

  16047. 'RTX_OBJ'

  16048.      An RTX code that represents an actual object, such as a register

  16049.      ('REG') or a memory location ('MEM', 'SYMBOL_REF').  'LO_SUM') is

  16050.      also included; instead, 'SUBREG' and 'STRICT_LOW_PART' are not in

  16051.      this class, but in class 'RTX_EXTRA'.

  16052. 

  16053. 'RTX_CONST_OBJ'

  16054.      An RTX code that represents a constant object.  'HIGH' is also

  16055.      included in this class.

  16056. 

  16057. 'RTX_COMPARE'

  16058.      An RTX code for a non-symmetric comparison, such as 'GEU' or 'LT'.

  16059. 

  16060. 'RTX_COMM_COMPARE'

  16061.      An RTX code for a symmetric (commutative) comparison, such as 'EQ'

  16062.      or 'ORDERED'.

  16063. 

  16064. 'RTX_UNARY'

  16065.      An RTX code for a unary arithmetic operation, such as 'NEG', 'NOT',

  16066.      or 'ABS'.  This category also includes value extension (sign or

  16067.      zero) and conversions between integer and floating point.

  16068. 

  16069. 'RTX_COMM_ARITH'

  16070.      An RTX code for a commutative binary operation, such as 'PLUS' or

  16071.      'AND'.  'NE' and 'EQ' are comparisons, so they have class

  16072.      'RTX_COMM_COMPARE'.

  16073. 

  16074. 'RTX_BIN_ARITH'

  16075.      An RTX code for a non-commutative binary operation, such as

  16076.      'MINUS', 'DIV', or 'ASHIFTRT'.

  16077. 

  16078. 'RTX_BITFIELD_OPS'

  16079.      An RTX code for a bit-field operation.  Currently only

  16080.      'ZERO_EXTRACT' and 'SIGN_EXTRACT'.  These have three inputs and are

  16081.      lvalues (so they can be used for insertion as well).  *Note

  16082.      Bit-Fields::.

  16083. 

  16084. 'RTX_TERNARY'

  16085.      An RTX code for other three input operations.  Currently only

  16086.      'IF_THEN_ELSE', 'VEC_MERGE', 'SIGN_EXTRACT', 'ZERO_EXTRACT', and

  16087.      'FMA'.

  16088. 

  16089. 'RTX_INSN'

  16090.      An RTX code for an entire instruction: 'INSN', 'JUMP_INSN', and

  16091.      'CALL_INSN'.  *Note Insns::.

  16092. 

  16093. 'RTX_MATCH'

  16094.      An RTX code for something that matches in insns, such as

  16095.      'MATCH_DUP'.  These only occur in machine descriptions.

  16096. 

  16097. 'RTX_AUTOINC'

  16098.      An RTX code for an auto-increment addressing mode, such as

  16099.      'POST_INC'.  'XEXP (X, 0)' gives the auto-modified register.

  16100. 

  16101. 'RTX_EXTRA'

  16102.      All other RTX codes.  This category includes the remaining codes

  16103.      used only in machine descriptions ('DEFINE_*', etc.).  It also

  16104.      includes all the codes describing side effects ('SET', 'USE',

  16105.      'CLOBBER', etc.)  and the non-insns that may appear on an insn

  16106.      chain, such as 'NOTE', 'BARRIER', and 'CODE_LABEL'.  'SUBREG' is

  16107.      also part of this class.

  16108. 

  16109.  For each expression code, 'rtl.def' specifies the number of contained

  16110. objects and their kinds using a sequence of characters called the

  16111. "format" of the expression code.  For example, the format of 'subreg' is

  16112. 'ep'.

  16113. 

  16114.  These are the most commonly used format characters:

  16115. 

  16116. 'e'

  16117.      An expression (actually a pointer to an expression).

  16118. 

  16119. 'i'

  16120.      An integer.

  16121. 

  16122. 'w'

  16123.      A wide integer.

  16124. 

  16125. 's'

  16126.      A string.

  16127. 

  16128. 'E'

  16129.      A vector of expressions.

  16130. 

  16131.  A few other format characters are used occasionally:

  16132. 

  16133. 'u'

  16134.      'u' is equivalent to 'e' except that it is printed differently in

  16135.      debugging dumps.  It is used for pointers to insns.

  16136. 

  16137. 'n'

  16138.      'n' is equivalent to 'i' except that it is printed differently in

  16139.      debugging dumps.  It is used for the line number or code number of

  16140.      a 'note' insn.

  16141. 

  16142. 'S'

  16143.      'S' indicates a string which is optional.  In the RTL objects in

  16144.      core, 'S' is equivalent to 's', but when the object is read, from

  16145.      an 'md' file, the string value of this operand may be omitted.  An

  16146.      omitted string is taken to be the null string.

  16147. 

  16148. 'V'

  16149.      'V' indicates a vector which is optional.  In the RTL objects in

  16150.      core, 'V' is equivalent to 'E', but when the object is read from an

  16151.      'md' file, the vector value of this operand may be omitted.  An

  16152.      omitted vector is effectively the same as a vector of no elements.

  16153. 

  16154. 'B'

  16155.      'B' indicates a pointer to basic block structure.

  16156. 

  16157. 'p'

  16158.      A polynomial integer.  At present this is used only for

  16159.      'SUBREG_BYTE'.

  16160. 

  16161. '0'

  16162.      '0' means a slot whose contents do not fit any normal category.

  16163.      '0' slots are not printed at all in dumps, and are often used in

  16164.      special ways by small parts of the compiler.

  16165. 

  16166.  There are macros to get the number of operands and the format of an

  16167. expression code:

  16168. 

  16169. 'GET_RTX_LENGTH (CODE)'

  16170.      Number of operands of an RTX of code CODE.

  16171. 

  16172. 'GET_RTX_FORMAT (CODE)'

  16173.      The format of an RTX of code CODE, as a C string.

  16174. 

  16175.  Some classes of RTX codes always have the same format.  For example, it

  16176. is safe to assume that all comparison operations have format 'ee'.

  16177. 

  16178. 'RTX_UNARY'

  16179.      All codes of this class have format 'e'.

  16180. 

  16181. 'RTX_BIN_ARITH'

  16182. 'RTX_COMM_ARITH'

  16183. 'RTX_COMM_COMPARE'

  16184. 'RTX_COMPARE'

  16185.      All codes of these classes have format 'ee'.

  16186. 

  16187. 'RTX_BITFIELD_OPS'

  16188. 'RTX_TERNARY'

  16189.      All codes of these classes have format 'eee'.

  16190. 

  16191. 'RTX_INSN'

  16192.      All codes of this class have formats that begin with 'iuueiee'.

  16193.      *Note Insns::.  Note that not all RTL objects linked onto an insn

  16194.      chain are of class 'RTX_INSN'.

  16195. 

  16196. 'RTX_CONST_OBJ'

  16197. 'RTX_OBJ'

  16198. 'RTX_MATCH'

  16199. 'RTX_EXTRA'

  16200.      You can make no assumptions about the format of these codes.

  16201. 

  16202. 

  16203. File: gccint.info,  Node: Accessors,  Next: Special Accessors,  Prev: RTL Classes,  Up: RTL

  16204. 

  16205. 14.3 Access to Operands

  16206. =======================

  16207. 

  16208. Operands of expressions are accessed using the macros 'XEXP', 'XINT',

  16209. 'XWINT' and 'XSTR'.  Each of these macros takes two arguments: an

  16210. expression-pointer (RTX) and an operand number (counting from zero).

  16211. Thus,

  16212. 

  16213.      XEXP (X, 2)

  16214. 

  16215. accesses operand 2 of expression X, as an expression.

  16216. 

  16217.      XINT (X, 2)

  16218. 

  16219. accesses the same operand as an integer.  'XSTR', used in the same

  16220. fashion, would access it as a string.

  16221. 

  16222.  Any operand can be accessed as an integer, as an expression or as a

  16223. string.  You must choose the correct method of access for the kind of

  16224. value actually stored in the operand.  You would do this based on the

  16225. expression code of the containing expression.  That is also how you

  16226. would know how many operands there are.

  16227. 

  16228.  For example, if X is an 'int_list' expression, you know that it has two

  16229. operands which can be correctly accessed as 'XINT (X, 0)' and 'XEXP (X,

  16230. 1)'.  Incorrect accesses like 'XEXP (X, 0)' and 'XINT (X, 1)' would

  16231. compile, but would trigger an internal compiler error when rtl checking

  16232. is enabled.  Nothing stops you from writing 'XEXP (X, 28)' either, but

  16233. this will access memory past the end of the expression with

  16234. unpredictable results.

  16235. 

  16236.  Access to operands which are vectors is more complicated.  You can use

  16237. the macro 'XVEC' to get the vector-pointer itself, or the macros

  16238. 'XVECEXP' and 'XVECLEN' to access the elements and length of a vector.

  16239. 

  16240. 'XVEC (EXP, IDX)'

  16241.      Access the vector-pointer which is operand number IDX in EXP.

  16242. 

  16243. 'XVECLEN (EXP, IDX)'

  16244.      Access the length (number of elements) in the vector which is in

  16245.      operand number IDX in EXP.  This value is an 'int'.

  16246. 

  16247. 'XVECEXP (EXP, IDX, ELTNUM)'

  16248.      Access element number ELTNUM in the vector which is in operand

  16249.      number IDX in EXP.  This value is an RTX.

  16250. 

  16251.      It is up to you to make sure that ELTNUM is not negative and is

  16252.      less than 'XVECLEN (EXP, IDX)'.

  16253. 

  16254.  All the macros defined in this section expand into lvalues and

  16255. therefore can be used to assign the operands, lengths and vector

  16256. elements as well as to access them.

  16257. 

  16258. 

  16259. File: gccint.info,  Node: Special Accessors,  Next: Flags,  Prev: Accessors,  Up: RTL

  16260. 

  16261. 14.4 Access to Special Operands

  16262. ===============================

  16263. 

  16264. Some RTL nodes have special annotations associated with them.

  16265. 

  16266. 'MEM'

  16267.      'MEM_ALIAS_SET (X)'

  16268.           If 0, X is not in any alias set, and may alias anything.

  16269.           Otherwise, X can only alias 'MEM's in a conflicting alias set.

  16270.           This value is set in a language-dependent manner in the

  16271.           front-end, and should not be altered in the back-end.  In some

  16272.           front-ends, these numbers may correspond in some way to types,

  16273.           or other language-level entities, but they need not, and the

  16274.           back-end makes no such assumptions.  These set numbers are

  16275.           tested with 'alias_sets_conflict_p'.

  16276. 

  16277.      'MEM_EXPR (X)'

  16278.           If this register is known to hold the value of some user-level

  16279.           declaration, this is that tree node.  It may also be a

  16280.           'COMPONENT_REF', in which case this is some field reference,

  16281.           and 'TREE_OPERAND (X, 0)' contains the declaration, or another

  16282.           'COMPONENT_REF', or null if there is no compile-time object

  16283.           associated with the reference.

  16284. 

  16285.      'MEM_OFFSET_KNOWN_P (X)'

  16286.           True if the offset of the memory reference from 'MEM_EXPR' is

  16287.           known.  'MEM_OFFSET (X)' provides the offset if so.

  16288. 

  16289.      'MEM_OFFSET (X)'

  16290.           The offset from the start of 'MEM_EXPR'.  The value is only

  16291.           valid if 'MEM_OFFSET_KNOWN_P (X)' is true.

  16292. 

  16293.      'MEM_SIZE_KNOWN_P (X)'

  16294.           True if the size of the memory reference is known.  'MEM_SIZE

  16295.           (X)' provides its size if so.

  16296. 

  16297.      'MEM_SIZE (X)'

  16298.           The size in bytes of the memory reference.  This is mostly

  16299.           relevant for 'BLKmode' references as otherwise the size is

  16300.           implied by the mode.  The value is only valid if

  16301.           'MEM_SIZE_KNOWN_P (X)' is true.

  16302. 

  16303.      'MEM_ALIGN (X)'

  16304.           The known alignment in bits of the memory reference.

  16305. 

  16306.      'MEM_ADDR_SPACE (X)'

  16307.           The address space of the memory reference.  This will commonly

  16308.           be zero for the generic address space.

  16309. 

  16310. 'REG'

  16311.      'ORIGINAL_REGNO (X)'

  16312.           This field holds the number the register "originally" had; for

  16313.           a pseudo register turned into a hard reg this will hold the

  16314.           old pseudo register number.

  16315. 

  16316.      'REG_EXPR (X)'

  16317.           If this register is known to hold the value of some user-level

  16318.           declaration, this is that tree node.

  16319. 

  16320.      'REG_OFFSET (X)'

  16321.           If this register is known to hold the value of some user-level

  16322.           declaration, this is the offset into that logical storage.

  16323. 

  16324. 'SYMBOL_REF'

  16325.      'SYMBOL_REF_DECL (X)'

  16326.           If the 'symbol_ref' X was created for a 'VAR_DECL' or a

  16327.           'FUNCTION_DECL', that tree is recorded here.  If this value is

  16328.           null, then X was created by back end code generation routines,

  16329.           and there is no associated front end symbol table entry.

  16330. 

  16331.           'SYMBOL_REF_DECL' may also point to a tree of class ''c'',

  16332.           that is, some sort of constant.  In this case, the

  16333.           'symbol_ref' is an entry in the per-file constant pool; again,

  16334.           there is no associated front end symbol table entry.

  16335. 

  16336.      'SYMBOL_REF_CONSTANT (X)'

  16337.           If 'CONSTANT_POOL_ADDRESS_P (X)' is true, this is the constant

  16338.           pool entry for X.  It is null otherwise.

  16339. 

  16340.      'SYMBOL_REF_DATA (X)'

  16341.           A field of opaque type used to store 'SYMBOL_REF_DECL' or

  16342.           'SYMBOL_REF_CONSTANT'.

  16343. 

  16344.      'SYMBOL_REF_FLAGS (X)'

  16345.           In a 'symbol_ref', this is used to communicate various

  16346.           predicates about the symbol.  Some of these are common enough

  16347.           to be computed by common code, some are specific to the

  16348.           target.  The common bits are:

  16349. 

  16350.           'SYMBOL_FLAG_FUNCTION'

  16351.                Set if the symbol refers to a function.

  16352. 

  16353.           'SYMBOL_FLAG_LOCAL'

  16354.                Set if the symbol is local to this "module".  See

  16355.                'TARGET_BINDS_LOCAL_P'.

  16356. 

  16357.           'SYMBOL_FLAG_EXTERNAL'

  16358.                Set if this symbol is not defined in this translation

  16359.                unit.  Note that this is not the inverse of

  16360.                'SYMBOL_FLAG_LOCAL'.

  16361. 

  16362.           'SYMBOL_FLAG_SMALL'

  16363.                Set if the symbol is located in the small data section.

  16364.                See 'TARGET_IN_SMALL_DATA_P'.

  16365. 

  16366.           'SYMBOL_REF_TLS_MODEL (X)'

  16367.                This is a multi-bit field accessor that returns the

  16368.                'tls_model' to be used for a thread-local storage symbol.

  16369.                It returns zero for non-thread-local symbols.

  16370. 

  16371.           'SYMBOL_FLAG_HAS_BLOCK_INFO'

  16372.                Set if the symbol has 'SYMBOL_REF_BLOCK' and

  16373.                'SYMBOL_REF_BLOCK_OFFSET' fields.

  16374. 

  16375.           'SYMBOL_FLAG_ANCHOR'

  16376.                Set if the symbol is used as a section anchor.  "Section

  16377.                anchors" are symbols that have a known position within an

  16378.                'object_block' and that can be used to access nearby

  16379.                members of that block.  They are used to implement

  16380.                '-fsection-anchors'.

  16381. 

  16382.                If this flag is set, then 'SYMBOL_FLAG_HAS_BLOCK_INFO'

  16383.                will be too.

  16384. 

  16385.           Bits beginning with 'SYMBOL_FLAG_MACH_DEP' are available for

  16386.           the target's use.

  16387. 

  16388. 'SYMBOL_REF_BLOCK (X)'

  16389.      If 'SYMBOL_REF_HAS_BLOCK_INFO_P (X)', this is the 'object_block'

  16390.      structure to which the symbol belongs, or 'NULL' if it has not been

  16391.      assigned a block.

  16392. 

  16393. 'SYMBOL_REF_BLOCK_OFFSET (X)'

  16394.      If 'SYMBOL_REF_HAS_BLOCK_INFO_P (X)', this is the offset of X from

  16395.      the first object in 'SYMBOL_REF_BLOCK (X)'.  The value is negative

  16396.      if X has not yet been assigned to a block, or it has not been given

  16397.      an offset within that block.

  16398. 

  16399. 

  16400. File: gccint.info,  Node: Flags,  Next: Machine Modes,  Prev: Special Accessors,  Up: RTL

  16401. 

  16402. 14.5 Flags in an RTL Expression

  16403. ===============================

  16404. 

  16405. RTL expressions contain several flags (one-bit bit-fields) that are used

  16406. in certain types of expression.  Most often they are accessed with the

  16407. following macros, which expand into lvalues.

  16408. 

  16409. 'CROSSING_JUMP_P (X)'

  16410.      Nonzero in a 'jump_insn' if it crosses between hot and cold

  16411.      sections, which could potentially be very far apart in the

  16412.      executable.  The presence of this flag indicates to other

  16413.      optimizations that this branching instruction should not be

  16414.      "collapsed" into a simpler branching construct.  It is used when

  16415.      the optimization to partition basic blocks into hot and cold

  16416.      sections is turned on.

  16417. 

  16418. 'CONSTANT_POOL_ADDRESS_P (X)'

  16419.      Nonzero in a 'symbol_ref' if it refers to part of the current

  16420.      function's constant pool.  For most targets these addresses are in

  16421.      a '.rodata' section entirely separate from the function, but for

  16422.      some targets the addresses are close to the beginning of the

  16423.      function.  In either case GCC assumes these addresses can be

  16424.      addressed directly, perhaps with the help of base registers.

  16425.      Stored in the 'unchanging' field and printed as '/u'.

  16426. 

  16427. 'INSN_ANNULLED_BRANCH_P (X)'

  16428.      In a 'jump_insn', 'call_insn', or 'insn' indicates that the branch

  16429.      is an annulling one.  See the discussion under 'sequence' below.

  16430.      Stored in the 'unchanging' field and printed as '/u'.

  16431. 

  16432. 'INSN_DELETED_P (X)'

  16433.      In an 'insn', 'call_insn', 'jump_insn', 'code_label',

  16434.      'jump_table_data', 'barrier', or 'note', nonzero if the insn has

  16435.      been deleted.  Stored in the 'volatil' field and printed as '/v'.

  16436. 

  16437. 'INSN_FROM_TARGET_P (X)'

  16438.      In an 'insn' or 'jump_insn' or 'call_insn' in a delay slot of a

  16439.      branch, indicates that the insn is from the target of the branch.

  16440.      If the branch insn has 'INSN_ANNULLED_BRANCH_P' set, this insn will

  16441.      only be executed if the branch is taken.  For annulled branches

  16442.      with 'INSN_FROM_TARGET_P' clear, the insn will be executed only if

  16443.      the branch is not taken.  When 'INSN_ANNULLED_BRANCH_P' is not set,

  16444.      this insn will always be executed.  Stored in the 'in_struct' field

  16445.      and printed as '/s'.

  16446. 

  16447. 'LABEL_PRESERVE_P (X)'

  16448.      In a 'code_label' or 'note', indicates that the label is referenced

  16449.      by code or data not visible to the RTL of a given function.  Labels

  16450.      referenced by a non-local goto will have this bit set.  Stored in

  16451.      the 'in_struct' field and printed as '/s'.

  16452. 

  16453. 'LABEL_REF_NONLOCAL_P (X)'

  16454.      In 'label_ref' and 'reg_label' expressions, nonzero if this is a

  16455.      reference to a non-local label.  Stored in the 'volatil' field and

  16456.      printed as '/v'.

  16457. 

  16458. 'MEM_KEEP_ALIAS_SET_P (X)'

  16459.      In 'mem' expressions, 1 if we should keep the alias set for this

  16460.      mem unchanged when we access a component.  Set to 1, for example,

  16461.      when we are already in a non-addressable component of an aggregate.

  16462.      Stored in the 'jump' field and printed as '/j'.

  16463. 

  16464. 'MEM_VOLATILE_P (X)'

  16465.      In 'mem', 'asm_operands', and 'asm_input' expressions, nonzero for

  16466.      volatile memory references.  Stored in the 'volatil' field and

  16467.      printed as '/v'.

  16468. 

  16469. 'MEM_NOTRAP_P (X)'

  16470.      In 'mem', nonzero for memory references that will not trap.  Stored

  16471.      in the 'call' field and printed as '/c'.

  16472. 

  16473. 'MEM_POINTER (X)'

  16474.      Nonzero in a 'mem' if the memory reference holds a pointer.  Stored

  16475.      in the 'frame_related' field and printed as '/f'.

  16476. 

  16477. 'MEM_READONLY_P (X)'

  16478.      Nonzero in a 'mem', if the memory is statically allocated and

  16479.      read-only.

  16480. 

  16481.      Read-only in this context means never modified during the lifetime

  16482.      of the program, not necessarily in ROM or in write-disabled pages.

  16483.      A common example of the later is a shared library's global offset

  16484.      table.  This table is initialized by the runtime loader, so the

  16485.      memory is technically writable, but after control is transferred

  16486.      from the runtime loader to the application, this memory will never

  16487.      be subsequently modified.

  16488. 

  16489.      Stored in the 'unchanging' field and printed as '/u'.

  16490. 

  16491. 'PREFETCH_SCHEDULE_BARRIER_P (X)'

  16492.      In a 'prefetch', indicates that the prefetch is a scheduling

  16493.      barrier.  No other INSNs will be moved over it.  Stored in the

  16494.      'volatil' field and printed as '/v'.

  16495. 

  16496. 'REG_FUNCTION_VALUE_P (X)'

  16497.      Nonzero in a 'reg' if it is the place in which this function's

  16498.      value is going to be returned.  (This happens only in a hard

  16499.      register.)  Stored in the 'return_val' field and printed as '/i'.

  16500. 

  16501. 'REG_POINTER (X)'

  16502.      Nonzero in a 'reg' if the register holds a pointer.  Stored in the

  16503.      'frame_related' field and printed as '/f'.

  16504. 

  16505. 'REG_USERVAR_P (X)'

  16506.      In a 'reg', nonzero if it corresponds to a variable present in the

  16507.      user's source code.  Zero for temporaries generated internally by

  16508.      the compiler.  Stored in the 'volatil' field and printed as '/v'.

  16509. 

  16510.      The same hard register may be used also for collecting the values

  16511.      of functions called by this one, but 'REG_FUNCTION_VALUE_P' is zero

  16512.      in this kind of use.

  16513. 

  16514. 'RTL_CONST_CALL_P (X)'

  16515.      In a 'call_insn' indicates that the insn represents a call to a

  16516.      const function.  Stored in the 'unchanging' field and printed as

  16517.      '/u'.

  16518. 

  16519. 'RTL_PURE_CALL_P (X)'

  16520.      In a 'call_insn' indicates that the insn represents a call to a

  16521.      pure function.  Stored in the 'return_val' field and printed as

  16522.      '/i'.

  16523. 

  16524. 'RTL_CONST_OR_PURE_CALL_P (X)'

  16525.      In a 'call_insn', true if 'RTL_CONST_CALL_P' or 'RTL_PURE_CALL_P'

  16526.      is true.

  16527. 

  16528. 'RTL_LOOPING_CONST_OR_PURE_CALL_P (X)'

  16529.      In a 'call_insn' indicates that the insn represents a possibly

  16530.      infinite looping call to a const or pure function.  Stored in the

  16531.      'call' field and printed as '/c'.  Only true if one of

  16532.      'RTL_CONST_CALL_P' or 'RTL_PURE_CALL_P' is true.

  16533. 

  16534. 'RTX_FRAME_RELATED_P (X)'

  16535.      Nonzero in an 'insn', 'call_insn', 'jump_insn', 'barrier', or 'set'

  16536.      which is part of a function prologue and sets the stack pointer,

  16537.      sets the frame pointer, or saves a register.  This flag should also

  16538.      be set on an instruction that sets up a temporary register to use

  16539.      in place of the frame pointer.  Stored in the 'frame_related' field

  16540.      and printed as '/f'.

  16541. 

  16542.      In particular, on RISC targets where there are limits on the sizes

  16543.      of immediate constants, it is sometimes impossible to reach the

  16544.      register save area directly from the stack pointer.  In that case,

  16545.      a temporary register is used that is near enough to the register

  16546.      save area, and the Canonical Frame Address, i.e., DWARF2's logical

  16547.      frame pointer, register must (temporarily) be changed to be this

  16548.      temporary register.  So, the instruction that sets this temporary

  16549.      register must be marked as 'RTX_FRAME_RELATED_P'.

  16550. 

  16551.      If the marked instruction is overly complex (defined in terms of

  16552.      what 'dwarf2out_frame_debug_expr' can handle), you will also have

  16553.      to create a 'REG_FRAME_RELATED_EXPR' note and attach it to the

  16554.      instruction.  This note should contain a simple expression of the

  16555.      computation performed by this instruction, i.e., one that

  16556.      'dwarf2out_frame_debug_expr' can handle.

  16557. 

  16558.      This flag is required for exception handling support on targets

  16559.      with RTL prologues.

  16560. 

  16561. 'SCHED_GROUP_P (X)'

  16562.      During instruction scheduling, in an 'insn', 'call_insn',

  16563.      'jump_insn' or 'jump_table_data', indicates that the previous insn

  16564.      must be scheduled together with this insn.  This is used to ensure

  16565.      that certain groups of instructions will not be split up by the

  16566.      instruction scheduling pass, for example, 'use' insns before a

  16567.      'call_insn' may not be separated from the 'call_insn'.  Stored in

  16568.      the 'in_struct' field and printed as '/s'.

  16569. 

  16570. 'SET_IS_RETURN_P (X)'

  16571.      For a 'set', nonzero if it is for a return.  Stored in the 'jump'

  16572.      field and printed as '/j'.

  16573. 

  16574. 'SIBLING_CALL_P (X)'

  16575.      For a 'call_insn', nonzero if the insn is a sibling call.  Stored

  16576.      in the 'jump' field and printed as '/j'.

  16577. 

  16578. 'STRING_POOL_ADDRESS_P (X)'

  16579.      For a 'symbol_ref' expression, nonzero if it addresses this

  16580.      function's string constant pool.  Stored in the 'frame_related'

  16581.      field and printed as '/f'.

  16582. 

  16583. 'SUBREG_PROMOTED_UNSIGNED_P (X)'

  16584.      Returns a value greater then zero for a 'subreg' that has

  16585.      'SUBREG_PROMOTED_VAR_P' nonzero if the object being referenced is

  16586.      kept zero-extended, zero if it is kept sign-extended, and less then

  16587.      zero if it is extended some other way via the 'ptr_extend'

  16588.      instruction.  Stored in the 'unchanging' field and 'volatil' field,

  16589.      printed as '/u' and '/v'.  This macro may only be used to get the

  16590.      value it may not be used to change the value.  Use

  16591.      'SUBREG_PROMOTED_UNSIGNED_SET' to change the value.

  16592. 

  16593. 'SUBREG_PROMOTED_UNSIGNED_SET (X)'

  16594.      Set the 'unchanging' and 'volatil' fields in a 'subreg' to reflect

  16595.      zero, sign, or other extension.  If 'volatil' is zero, then

  16596.      'unchanging' as nonzero means zero extension and as zero means sign

  16597.      extension.  If 'volatil' is nonzero then some other type of

  16598.      extension was done via the 'ptr_extend' instruction.

  16599. 

  16600. 'SUBREG_PROMOTED_VAR_P (X)'

  16601.      Nonzero in a 'subreg' if it was made when accessing an object that

  16602.      was promoted to a wider mode in accord with the 'PROMOTED_MODE'

  16603.      machine description macro (*note Storage Layout::).  In this case,

  16604.      the mode of the 'subreg' is the declared mode of the object and the

  16605.      mode of 'SUBREG_REG' is the mode of the register that holds the

  16606.      object.  Promoted variables are always either sign- or

  16607.      zero-extended to the wider mode on every assignment.  Stored in the

  16608.      'in_struct' field and printed as '/s'.

  16609. 

  16610. 'SYMBOL_REF_USED (X)'

  16611.      In a 'symbol_ref', indicates that X has been used.  This is

  16612.      normally only used to ensure that X is only declared external once.

  16613.      Stored in the 'used' field.

  16614. 

  16615. 'SYMBOL_REF_WEAK (X)'

  16616.      In a 'symbol_ref', indicates that X has been declared weak.  Stored

  16617.      in the 'return_val' field and printed as '/i'.

  16618. 

  16619. 'SYMBOL_REF_FLAG (X)'

  16620.      In a 'symbol_ref', this is used as a flag for machine-specific

  16621.      purposes.  Stored in the 'volatil' field and printed as '/v'.

  16622. 

  16623.      Most uses of 'SYMBOL_REF_FLAG' are historic and may be subsumed by

  16624.      'SYMBOL_REF_FLAGS'.  Certainly use of 'SYMBOL_REF_FLAGS' is

  16625.      mandatory if the target requires more than one bit of storage.

  16626. 

  16627.  These are the fields to which the above macros refer:

  16628. 

  16629. 'call'

  16630.      In a 'mem', 1 means that the memory reference will not trap.

  16631. 

  16632.      In a 'call', 1 means that this pure or const call may possibly

  16633.      infinite loop.

  16634. 

  16635.      In an RTL dump, this flag is represented as '/c'.

  16636. 

  16637. 'frame_related'

  16638.      In an 'insn' or 'set' expression, 1 means that it is part of a

  16639.      function prologue and sets the stack pointer, sets the frame

  16640.      pointer, saves a register, or sets up a temporary register to use

  16641.      in place of the frame pointer.

  16642. 

  16643.      In 'reg' expressions, 1 means that the register holds a pointer.

  16644. 

  16645.      In 'mem' expressions, 1 means that the memory reference holds a

  16646.      pointer.

  16647. 

  16648.      In 'symbol_ref' expressions, 1 means that the reference addresses

  16649.      this function's string constant pool.

  16650. 

  16651.      In an RTL dump, this flag is represented as '/f'.

  16652. 

  16653. 'in_struct'

  16654.      In 'reg' expressions, it is 1 if the register has its entire life

  16655.      contained within the test expression of some loop.

  16656. 

  16657.      In 'subreg' expressions, 1 means that the 'subreg' is accessing an

  16658.      object that has had its mode promoted from a wider mode.

  16659. 

  16660.      In 'label_ref' expressions, 1 means that the referenced label is

  16661.      outside the innermost loop containing the insn in which the

  16662.      'label_ref' was found.

  16663. 

  16664.      In 'code_label' expressions, it is 1 if the label may never be

  16665.      deleted.  This is used for labels which are the target of non-local

  16666.      gotos.  Such a label that would have been deleted is replaced with

  16667.      a 'note' of type 'NOTE_INSN_DELETED_LABEL'.

  16668. 

  16669.      In an 'insn' during dead-code elimination, 1 means that the insn is

  16670.      dead code.

  16671. 

  16672.      In an 'insn' or 'jump_insn' during reorg for an insn in the delay

  16673.      slot of a branch, 1 means that this insn is from the target of the

  16674.      branch.

  16675. 

  16676.      In an 'insn' during instruction scheduling, 1 means that this insn

  16677.      must be scheduled as part of a group together with the previous

  16678.      insn.

  16679. 

  16680.      In an RTL dump, this flag is represented as '/s'.

  16681. 

  16682. 'return_val'

  16683.      In 'reg' expressions, 1 means the register contains the value to be

  16684.      returned by the current function.  On machines that pass parameters

  16685.      in registers, the same register number may be used for parameters

  16686.      as well, but this flag is not set on such uses.

  16687. 

  16688.      In 'symbol_ref' expressions, 1 means the referenced symbol is weak.

  16689. 

  16690.      In 'call' expressions, 1 means the call is pure.

  16691. 

  16692.      In an RTL dump, this flag is represented as '/i'.

  16693. 

  16694. 'jump'

  16695.      In a 'mem' expression, 1 means we should keep the alias set for

  16696.      this mem unchanged when we access a component.

  16697. 

  16698.      In a 'set', 1 means it is for a return.

  16699. 

  16700.      In a 'call_insn', 1 means it is a sibling call.

  16701. 

  16702.      In a 'jump_insn', 1 means it is a crossing jump.

  16703. 

  16704.      In an RTL dump, this flag is represented as '/j'.

  16705. 

  16706. 'unchanging'

  16707.      In 'reg' and 'mem' expressions, 1 means that the value of the

  16708.      expression never changes.

  16709. 

  16710.      In 'subreg' expressions, it is 1 if the 'subreg' references an

  16711.      unsigned object whose mode has been promoted to a wider mode.

  16712. 

  16713.      In an 'insn' or 'jump_insn' in the delay slot of a branch

  16714.      instruction, 1 means an annulling branch should be used.

  16715. 

  16716.      In a 'symbol_ref' expression, 1 means that this symbol addresses

  16717.      something in the per-function constant pool.

  16718. 

  16719.      In a 'call_insn' 1 means that this instruction is a call to a const

  16720.      function.

  16721. 

  16722.      In an RTL dump, this flag is represented as '/u'.

  16723. 

  16724. 'used'

  16725.      This flag is used directly (without an access macro) at the end of

  16726.      RTL generation for a function, to count the number of times an

  16727.      expression appears in insns.  Expressions that appear more than

  16728.      once are copied, according to the rules for shared structure (*note

  16729.      Sharing::).

  16730. 

  16731.      For a 'reg', it is used directly (without an access macro) by the

  16732.      leaf register renumbering code to ensure that each register is only

  16733.      renumbered once.

  16734. 

  16735.      In a 'symbol_ref', it indicates that an external declaration for

  16736.      the symbol has already been written.

  16737. 

  16738. 'volatil'

  16739.      In a 'mem', 'asm_operands', or 'asm_input' expression, it is 1 if

  16740.      the memory reference is volatile.  Volatile memory references may

  16741.      not be deleted, reordered or combined.

  16742. 

  16743.      In a 'symbol_ref' expression, it is used for machine-specific

  16744.      purposes.

  16745. 

  16746.      In a 'reg' expression, it is 1 if the value is a user-level

  16747.      variable.  0 indicates an internal compiler temporary.

  16748. 

  16749.      In an 'insn', 1 means the insn has been deleted.

  16750. 

  16751.      In 'label_ref' and 'reg_label' expressions, 1 means a reference to

  16752.      a non-local label.

  16753. 

  16754.      In 'prefetch' expressions, 1 means that the containing insn is a

  16755.      scheduling barrier.

  16756. 

  16757.      In an RTL dump, this flag is represented as '/v'.

  16758. 

  16759. 

  16760. File: gccint.info,  Node: Machine Modes,  Next: Constants,  Prev: Flags,  Up: RTL

  16761. 

  16762. 14.6 Machine Modes

  16763. ==================

  16764. 

  16765. A machine mode describes a size of data object and the representation

  16766. used for it.  In the C code, machine modes are represented by an

  16767. enumeration type, 'machine_mode', defined in 'machmode.def'.  Each RTL

  16768. expression has room for a machine mode and so do certain kinds of tree

  16769. expressions (declarations and types, to be precise).

  16770. 

  16771.  In debugging dumps and machine descriptions, the machine mode of an RTL

  16772. expression is written after the expression code with a colon to separate

  16773. them.  The letters 'mode' which appear at the end of each machine mode

  16774. name are omitted.  For example, '(reg:SI 38)' is a 'reg' expression with

  16775. machine mode 'SImode'.  If the mode is 'VOIDmode', it is not written at

  16776. all.

  16777. 

  16778.  Here is a table of machine modes.  The term "byte" below refers to an

  16779. object of 'BITS_PER_UNIT' bits (*note Storage Layout::).

  16780. 

  16781. 'BImode'

  16782.      "Bit" mode represents a single bit, for predicate registers.

  16783. 

  16784. 'QImode'

  16785.      "Quarter-Integer" mode represents a single byte treated as an

  16786.      integer.

  16787. 

  16788. 'HImode'

  16789.      "Half-Integer" mode represents a two-byte integer.

  16790. 

  16791. 'PSImode'

  16792.      "Partial Single Integer" mode represents an integer which occupies

  16793.      four bytes but which doesn't really use all four.  On some

  16794.      machines, this is the right mode to use for pointers.

  16795. 

  16796. 'SImode'

  16797.      "Single Integer" mode represents a four-byte integer.

  16798. 

  16799. 'PDImode'

  16800.      "Partial Double Integer" mode represents an integer which occupies

  16801.      eight bytes but which doesn't really use all eight.  On some

  16802.      machines, this is the right mode to use for certain pointers.

  16803. 

  16804. 'DImode'

  16805.      "Double Integer" mode represents an eight-byte integer.

  16806. 

  16807. 'TImode'

  16808.      "Tetra Integer" (?)  mode represents a sixteen-byte integer.

  16809. 

  16810. 'OImode'

  16811.      "Octa Integer" (?)  mode represents a thirty-two-byte integer.

  16812. 

  16813. 'XImode'

  16814.      "Hexadeca Integer" (?)  mode represents a sixty-four-byte integer.

  16815. 

  16816. 'QFmode'

  16817.      "Quarter-Floating" mode represents a quarter-precision (single

  16818.      byte) floating point number.

  16819. 

  16820. 'HFmode'

  16821.      "Half-Floating" mode represents a half-precision (two byte)

  16822.      floating point number.

  16823. 

  16824. 'TQFmode'

  16825.      "Three-Quarter-Floating" (?)  mode represents a

  16826.      three-quarter-precision (three byte) floating point number.

  16827. 

  16828. 'SFmode'

  16829.      "Single Floating" mode represents a four byte floating point

  16830.      number.  In the common case, of a processor with IEEE arithmetic

  16831.      and 8-bit bytes, this is a single-precision IEEE floating point

  16832.      number; it can also be used for double-precision (on processors

  16833.      with 16-bit bytes) and single-precision VAX and IBM types.

  16834. 

  16835. 'DFmode'

  16836.      "Double Floating" mode represents an eight byte floating point

  16837.      number.  In the common case, of a processor with IEEE arithmetic

  16838.      and 8-bit bytes, this is a double-precision IEEE floating point

  16839.      number.

  16840. 

  16841. 'XFmode'

  16842.      "Extended Floating" mode represents an IEEE extended floating point

  16843.      number.  This mode only has 80 meaningful bits (ten bytes).  Some

  16844.      processors require such numbers to be padded to twelve bytes,

  16845.      others to sixteen; this mode is used for either.

  16846. 

  16847. 'SDmode'

  16848.      "Single Decimal Floating" mode represents a four byte decimal

  16849.      floating point number (as distinct from conventional binary

  16850.      floating point).

  16851. 

  16852. 'DDmode'

  16853.      "Double Decimal Floating" mode represents an eight byte decimal

  16854.      floating point number.

  16855. 

  16856. 'TDmode'

  16857.      "Tetra Decimal Floating" mode represents a sixteen byte decimal

  16858.      floating point number all 128 of whose bits are meaningful.

  16859. 

  16860. 'TFmode'

  16861.      "Tetra Floating" mode represents a sixteen byte floating point

  16862.      number all 128 of whose bits are meaningful.  One common use is the

  16863.      IEEE quad-precision format.

  16864. 

  16865. 'QQmode'

  16866.      "Quarter-Fractional" mode represents a single byte treated as a

  16867.      signed fractional number.  The default format is "s.7".

  16868. 

  16869. 'HQmode'

  16870.      "Half-Fractional" mode represents a two-byte signed fractional

  16871.      number.  The default format is "s.15".

  16872. 

  16873. 'SQmode'

  16874.      "Single Fractional" mode represents a four-byte signed fractional

  16875.      number.  The default format is "s.31".

  16876. 

  16877. 'DQmode'

  16878.      "Double Fractional" mode represents an eight-byte signed fractional

  16879.      number.  The default format is "s.63".

  16880. 

  16881. 'TQmode'

  16882.      "Tetra Fractional" mode represents a sixteen-byte signed fractional

  16883.      number.  The default format is "s.127".

  16884. 

  16885. 'UQQmode'

  16886.      "Unsigned Quarter-Fractional" mode represents a single byte treated

  16887.      as an unsigned fractional number.  The default format is ".8".

  16888. 

  16889. 'UHQmode'

  16890.      "Unsigned Half-Fractional" mode represents a two-byte unsigned

  16891.      fractional number.  The default format is ".16".

  16892. 

  16893. 'USQmode'

  16894.      "Unsigned Single Fractional" mode represents a four-byte unsigned

  16895.      fractional number.  The default format is ".32".

  16896. 

  16897. 'UDQmode'

  16898.      "Unsigned Double Fractional" mode represents an eight-byte unsigned

  16899.      fractional number.  The default format is ".64".

  16900. 

  16901. 'UTQmode'

  16902.      "Unsigned Tetra Fractional" mode represents a sixteen-byte unsigned

  16903.      fractional number.  The default format is ".128".

  16904. 

  16905. 'HAmode'

  16906.      "Half-Accumulator" mode represents a two-byte signed accumulator.

  16907.      The default format is "s8.7".

  16908. 

  16909. 'SAmode'

  16910.      "Single Accumulator" mode represents a four-byte signed

  16911.      accumulator.  The default format is "s16.15".

  16912. 

  16913. 'DAmode'

  16914.      "Double Accumulator" mode represents an eight-byte signed

  16915.      accumulator.  The default format is "s32.31".

  16916. 

  16917. 'TAmode'

  16918.      "Tetra Accumulator" mode represents a sixteen-byte signed

  16919.      accumulator.  The default format is "s64.63".

  16920. 

  16921. 'UHAmode'

  16922.      "Unsigned Half-Accumulator" mode represents a two-byte unsigned

  16923.      accumulator.  The default format is "8.8".

  16924. 

  16925. 'USAmode'

  16926.      "Unsigned Single Accumulator" mode represents a four-byte unsigned

  16927.      accumulator.  The default format is "16.16".

  16928. 

  16929. 'UDAmode'

  16930.      "Unsigned Double Accumulator" mode represents an eight-byte

  16931.      unsigned accumulator.  The default format is "32.32".

  16932. 

  16933. 'UTAmode'

  16934.      "Unsigned Tetra Accumulator" mode represents a sixteen-byte

  16935.      unsigned accumulator.  The default format is "64.64".

  16936. 

  16937. 'CCmode'

  16938.      "Condition Code" mode represents the value of a condition code,

  16939.      which is a machine-specific set of bits used to represent the

  16940.      result of a comparison operation.  Other machine-specific modes may

  16941.      also be used for the condition code.  These modes are not used on

  16942.      machines that use 'cc0' (*note Condition Code::).

  16943. 

  16944. 'BLKmode'

  16945.      "Block" mode represents values that are aggregates to which none of

  16946.      the other modes apply.  In RTL, only memory references can have

  16947.      this mode, and only if they appear in string-move or vector

  16948.      instructions.  On machines which have no such instructions,

  16949.      'BLKmode' will not appear in RTL.

  16950. 

  16951. 'VOIDmode'

  16952.      Void mode means the absence of a mode or an unspecified mode.  For

  16953.      example, RTL expressions of code 'const_int' have mode 'VOIDmode'

  16954.      because they can be taken to have whatever mode the context

  16955.      requires.  In debugging dumps of RTL, 'VOIDmode' is expressed by

  16956.      the absence of any mode.

  16957. 

  16958. 'QCmode, HCmode, SCmode, DCmode, XCmode, TCmode'

  16959.      These modes stand for a complex number represented as a pair of

  16960.      floating point values.  The floating point values are in 'QFmode',

  16961.      'HFmode', 'SFmode', 'DFmode', 'XFmode', and 'TFmode', respectively.

  16962. 

  16963. 'CQImode, CHImode, CSImode, CDImode, CTImode, COImode, CPSImode'

  16964.      These modes stand for a complex number represented as a pair of

  16965.      integer values.  The integer values are in 'QImode', 'HImode',

  16966.      'SImode', 'DImode', 'TImode', 'OImode', and 'PSImode',

  16967.      respectively.

  16968. 

  16969. 'BND32mode BND64mode'

  16970.      These modes stand for bounds for pointer of 32 and 64 bit size

  16971.      respectively.  Mode size is double pointer mode size.

  16972. 

  16973.  The machine description defines 'Pmode' as a C macro which expands into

  16974. the machine mode used for addresses.  Normally this is the mode whose

  16975. size is 'BITS_PER_WORD', 'SImode' on 32-bit machines.

  16976. 

  16977.  The only modes which a machine description must support are 'QImode',

  16978. and the modes corresponding to 'BITS_PER_WORD', 'FLOAT_TYPE_SIZE' and

  16979. 'DOUBLE_TYPE_SIZE'.  The compiler will attempt to use 'DImode' for

  16980. 8-byte structures and unions, but this can be prevented by overriding

  16981. the definition of 'MAX_FIXED_MODE_SIZE'.  Alternatively, you can have

  16982. the compiler use 'TImode' for 16-byte structures and unions.  Likewise,

  16983. you can arrange for the C type 'short int' to avoid using 'HImode'.

  16984. 

  16985.  Very few explicit references to machine modes remain in the compiler

  16986. and these few references will soon be removed.  Instead, the machine

  16987. modes are divided into mode classes.  These are represented by the

  16988. enumeration type 'enum mode_class' defined in 'machmode.h'.  The

  16989. possible mode classes are:

  16990. 

  16991. 'MODE_INT'

  16992.      Integer modes.  By default these are 'BImode', 'QImode', 'HImode',

  16993.      'SImode', 'DImode', 'TImode', and 'OImode'.

  16994. 

  16995. 'MODE_PARTIAL_INT'

  16996.      The "partial integer" modes, 'PQImode', 'PHImode', 'PSImode' and

  16997.      'PDImode'.

  16998. 

  16999. 'MODE_FLOAT'

  17000.      Floating point modes.  By default these are 'QFmode', 'HFmode',

  17001.      'TQFmode', 'SFmode', 'DFmode', 'XFmode' and 'TFmode'.

  17002. 

  17003. 'MODE_DECIMAL_FLOAT'

  17004.      Decimal floating point modes.  By default these are 'SDmode',

  17005.      'DDmode' and 'TDmode'.

  17006. 

  17007. 'MODE_FRACT'

  17008.      Signed fractional modes.  By default these are 'QQmode', 'HQmode',

  17009.      'SQmode', 'DQmode' and 'TQmode'.

  17010. 

  17011. 'MODE_UFRACT'

  17012.      Unsigned fractional modes.  By default these are 'UQQmode',

  17013.      'UHQmode', 'USQmode', 'UDQmode' and 'UTQmode'.

  17014. 

  17015. 'MODE_ACCUM'

  17016.      Signed accumulator modes.  By default these are 'HAmode', 'SAmode',

  17017.      'DAmode' and 'TAmode'.

  17018. 

  17019. 'MODE_UACCUM'

  17020.      Unsigned accumulator modes.  By default these are 'UHAmode',

  17021.      'USAmode', 'UDAmode' and 'UTAmode'.

  17022. 

  17023. 'MODE_COMPLEX_INT'

  17024.      Complex integer modes.  (These are not currently implemented).

  17025. 

  17026. 'MODE_COMPLEX_FLOAT'

  17027.      Complex floating point modes.  By default these are 'QCmode',

  17028.      'HCmode', 'SCmode', 'DCmode', 'XCmode', and 'TCmode'.

  17029. 

  17030. 'MODE_CC'

  17031.      Modes representing condition code values.  These are 'CCmode' plus

  17032.      any 'CC_MODE' modes listed in the 'MACHINE-modes.def'.  *Note Jump

  17033.      Patterns::, also see *note Condition Code::.

  17034. 

  17035. 'MODE_POINTER_BOUNDS'

  17036.      Pointer bounds modes.  Used to represent values of pointer bounds

  17037.      type.  Operations in these modes may be executed as NOPs depending

  17038.      on hardware features and environment setup.

  17039. 

  17040. 'MODE_RANDOM'

  17041.      This is a catchall mode class for modes which don't fit into the

  17042.      above classes.  Currently 'VOIDmode' and 'BLKmode' are in

  17043.      'MODE_RANDOM'.

  17044. 

  17045.  'machmode.h' also defines various wrapper classes that combine a

  17046. 'machine_mode' with a static assertion that a particular condition

  17047. holds.  The classes are:

  17048. 

  17049. 'scalar_int_mode'

  17050.      A mode that has class 'MODE_INT' or 'MODE_PARTIAL_INT'.

  17051. 

  17052. 'scalar_float_mode'

  17053.      A mode that has class 'MODE_FLOAT' or 'MODE_DECIMAL_FLOAT'.

  17054. 

  17055. 'scalar_mode'

  17056.      A mode that holds a single numerical value.  In practice this means

  17057.      that the mode is a 'scalar_int_mode', is a 'scalar_float_mode', or

  17058.      has class 'MODE_FRACT', 'MODE_UFRACT', 'MODE_ACCUM', 'MODE_UACCUM'

  17059.      or 'MODE_POINTER_BOUNDS'.

  17060. 

  17061. 'complex_mode'

  17062.      A mode that has class 'MODE_COMPLEX_INT' or 'MODE_COMPLEX_FLOAT'.

  17063. 

  17064. 'fixed_size_mode'

  17065.      A mode whose size is known at compile time.

  17066. 

  17067.  Named modes use the most constrained of the available wrapper classes,

  17068. if one exists, otherwise they use 'machine_mode'.  For example, 'QImode'

  17069. is a 'scalar_int_mode', 'SFmode' is a 'scalar_float_mode' and 'BLKmode'

  17070. is a plain 'machine_mode'.  It is possible to refer to any mode as a raw

  17071. 'machine_mode' by adding the 'E_' prefix, where 'E' stands for

  17072. "enumeration".  For example, the raw 'machine_mode' names of the modes

  17073. just mentioned are 'E_QImode', 'E_SFmode' and 'E_BLKmode' respectively.

  17074. 

  17075.  The wrapper classes implicitly convert to 'machine_mode' and to any

  17076. wrapper class that represents a more general condition; for example

  17077. 'scalar_int_mode' and 'scalar_float_mode' both convert to 'scalar_mode'

  17078. and all three convert to 'fixed_size_mode'.  The classes act like

  17079. 'machine_mode's that accept only certain named modes.

  17080. 

  17081.  'machmode.h' also defines a template class 'opt_mode<T>' that holds a

  17082. 'T' or nothing, where 'T' can be either 'machine_mode' or one of the

  17083. wrapper classes above.  The main operations on an 'opt_mode<T>' X are as

  17084. follows:

  17085. 

  17086. 'X.exists ()'

  17087.      Return true if X holds a mode rather than nothing.

  17088. 

  17089. 'X.exists (&Y)'

  17090.      Return true if X holds a mode rather than nothing, storing the mode

  17091.      in Y if so.  Y must be assignment-compatible with T.

  17092. 

  17093. 'X.require ()'

  17094.      Assert that X holds a mode rather than nothing and return that

  17095.      mode.

  17096. 

  17097. 'X = Y'

  17098.      Set X to Y, where Y is a T or implicitly converts to a T.

  17099. 

  17100.  The default constructor sets an 'opt_mode<T>' to nothing.  There is

  17101. also a constructor that takes an initial value of type T.

  17102. 

  17103.  It is possible to use the 'is-a.h' accessors on a 'machine_mode' or

  17104. machine mode wrapper X:

  17105. 

  17106. 'is_a <T> (X)'

  17107.      Return true if X meets the conditions for wrapper class T.

  17108. 

  17109. 'is_a <T> (X, &Y)'

  17110.      Return true if X meets the conditions for wrapper class T, storing

  17111.      it in Y if so.  Y must be assignment-compatible with T.

  17112. 

  17113. 'as_a <T> (X)'

  17114.      Assert that X meets the conditions for wrapper class T and return

  17115.      it as a T.

  17116. 

  17117. 'dyn_cast <T> (X)'

  17118.      Return an 'opt_mode<T>' that holds X if X meets the conditions for

  17119.      wrapper class T and that holds nothing otherwise.

  17120. 

  17121.  The purpose of these wrapper classes is to give stronger static type

  17122. checking.  For example, if a function takes a 'scalar_int_mode', a

  17123. caller that has a general 'machine_mode' must either check or assert

  17124. that the code is indeed a scalar integer first, using one of the

  17125. functions above.

  17126. 

  17127.  The wrapper classes are normal C++ classes, with user-defined

  17128. constructors.  Sometimes it is useful to have a POD version of the same

  17129. type, particularly if the type appears in a 'union'.  The template class

  17130. 'pod_mode<T>' provides a POD version of wrapper class T.  It is

  17131. assignment-compatible with T and implicitly converts to both

  17132. 'machine_mode' and T.

  17133. 

  17134.  Here are some C macros that relate to machine modes:

  17135. 

  17136. 'GET_MODE (X)'

  17137.      Returns the machine mode of the RTX X.

  17138. 

  17139. 'PUT_MODE (X, NEWMODE)'

  17140.      Alters the machine mode of the RTX X to be NEWMODE.

  17141. 

  17142. 'NUM_MACHINE_MODES'

  17143.      Stands for the number of machine modes available on the target

  17144.      machine.  This is one greater than the largest numeric value of any

  17145.      machine mode.

  17146. 

  17147. 'GET_MODE_NAME (M)'

  17148.      Returns the name of mode M as a string.

  17149. 

  17150. 'GET_MODE_CLASS (M)'

  17151.      Returns the mode class of mode M.

  17152. 

  17153. 'GET_MODE_WIDER_MODE (M)'

  17154.      Returns the next wider natural mode.  For example, the expression

  17155.      'GET_MODE_WIDER_MODE (QImode)' returns 'HImode'.

  17156. 

  17157. 'GET_MODE_SIZE (M)'

  17158.      Returns the size in bytes of a datum of mode M.

  17159. 

  17160. 'GET_MODE_BITSIZE (M)'

  17161.      Returns the size in bits of a datum of mode M.

  17162. 

  17163. 'GET_MODE_IBIT (M)'

  17164.      Returns the number of integral bits of a datum of fixed-point mode

  17165.      M.

  17166. 

  17167. 'GET_MODE_FBIT (M)'

  17168.      Returns the number of fractional bits of a datum of fixed-point

  17169.      mode M.

  17170. 

  17171. 'GET_MODE_MASK (M)'

  17172.      Returns a bitmask containing 1 for all bits in a word that fit

  17173.      within mode M.  This macro can only be used for modes whose bitsize

  17174.      is less than or equal to 'HOST_BITS_PER_INT'.

  17175. 

  17176. 'GET_MODE_ALIGNMENT (M)'

  17177.      Return the required alignment, in bits, for an object of mode M.

  17178. 

  17179. 'GET_MODE_UNIT_SIZE (M)'

  17180.      Returns the size in bytes of the subunits of a datum of mode M.

  17181.      This is the same as 'GET_MODE_SIZE' except in the case of complex

  17182.      modes.  For them, the unit size is the size of the real or

  17183.      imaginary part.

  17184. 

  17185. 'GET_MODE_NUNITS (M)'

  17186.      Returns the number of units contained in a mode, i.e.,

  17187.      'GET_MODE_SIZE' divided by 'GET_MODE_UNIT_SIZE'.

  17188. 

  17189. 'GET_CLASS_NARROWEST_MODE (C)'

  17190.      Returns the narrowest mode in mode class C.

  17191. 

  17192.  The following 3 variables are defined on every target.  They can be

  17193. used to allocate buffers that are guaranteed to be large enough to hold

  17194. any value that can be represented on the target.  The first two can be

  17195. overridden by defining them in the target's mode.def file, however, the

  17196. value must be a constant that can determined very early in the

  17197. compilation process.  The third symbol cannot be overridden.

  17198. 

  17199. 'BITS_PER_UNIT'

  17200.      The number of bits in an addressable storage unit (byte).  If you

  17201.      do not define this, the default is 8.

  17202. 

  17203. 'MAX_BITSIZE_MODE_ANY_INT'

  17204.      The maximum bitsize of any mode that is used in integer math.  This

  17205.      should be overridden by the target if it uses large integers as

  17206.      containers for larger vectors but otherwise never uses the contents

  17207.      to compute integer values.

  17208. 

  17209. 'MAX_BITSIZE_MODE_ANY_MODE'

  17210.      The bitsize of the largest mode on the target.  The default value

  17211.      is the largest mode size given in the mode definition file, which

  17212.      is always correct for targets whose modes have a fixed size.

  17213.      Targets that might increase the size of a mode beyond this default

  17214.      should define 'MAX_BITSIZE_MODE_ANY_MODE' to the actual upper limit

  17215.      in 'MACHINE-modes.def'.

  17216. 

  17217.  The global variables 'byte_mode' and 'word_mode' contain modes whose

  17218. classes are 'MODE_INT' and whose bitsizes are either 'BITS_PER_UNIT' or

  17219. 'BITS_PER_WORD', respectively.  On 32-bit machines, these are 'QImode'

  17220. and 'SImode', respectively.

  17221. 

  17222. 

  17223. File: gccint.info,  Node: Constants,  Next: Regs and Memory,  Prev: Machine Modes,  Up: RTL

  17224. 

  17225. 14.7 Constant Expression Types

  17226. ==============================

  17227. 

  17228. The simplest RTL expressions are those that represent constant values.

  17229. 

  17230. '(const_int I)'

  17231.      This type of expression represents the integer value I.  I is

  17232.      customarily accessed with the macro 'INTVAL' as in 'INTVAL (EXP)',

  17233.      which is equivalent to 'XWINT (EXP, 0)'.

  17234. 

  17235.      Constants generated for modes with fewer bits than in

  17236.      'HOST_WIDE_INT' must be sign extended to full width (e.g., with

  17237.      'gen_int_mode').  For constants for modes with more bits than in

  17238.      'HOST_WIDE_INT' the implied high order bits of that constant are

  17239.      copies of the top bit.  Note however that values are neither

  17240.      inherently signed nor inherently unsigned; where necessary,

  17241.      signedness is determined by the rtl operation instead.

  17242. 

  17243.      There is only one expression object for the integer value zero; it

  17244.      is the value of the variable 'const0_rtx'.  Likewise, the only

  17245.      expression for integer value one is found in 'const1_rtx', the only

  17246.      expression for integer value two is found in 'const2_rtx', and the

  17247.      only expression for integer value negative one is found in

  17248.      'constm1_rtx'.  Any attempt to create an expression of code

  17249.      'const_int' and value zero, one, two or negative one will return

  17250.      'const0_rtx', 'const1_rtx', 'const2_rtx' or 'constm1_rtx' as

  17251.      appropriate.

  17252. 

  17253.      Similarly, there is only one object for the integer whose value is

  17254.      'STORE_FLAG_VALUE'.  It is found in 'const_true_rtx'.  If

  17255.      'STORE_FLAG_VALUE' is one, 'const_true_rtx' and 'const1_rtx' will

  17256.      point to the same object.  If 'STORE_FLAG_VALUE' is -1,

  17257.      'const_true_rtx' and 'constm1_rtx' will point to the same object.

  17258. 

  17259. '(const_double:M I0 I1 ...)'

  17260.      This represents either a floating-point constant of mode M or (on

  17261.      older ports that do not define 'TARGET_SUPPORTS_WIDE_INT') an

  17262.      integer constant too large to fit into 'HOST_BITS_PER_WIDE_INT'

  17263.      bits but small enough to fit within twice that number of bits.  In

  17264.      the latter case, M will be 'VOIDmode'.  For integral values

  17265.      constants for modes with more bits than twice the number in

  17266.      'HOST_WIDE_INT' the implied high order bits of that constant are

  17267.      copies of the top bit of 'CONST_DOUBLE_HIGH'.  Note however that

  17268.      integral values are neither inherently signed nor inherently

  17269.      unsigned; where necessary, signedness is determined by the rtl

  17270.      operation instead.

  17271. 

  17272.      On more modern ports, 'CONST_DOUBLE' only represents floating point

  17273.      values.  New ports define 'TARGET_SUPPORTS_WIDE_INT' to make this

  17274.      designation.

  17275. 

  17276.      If M is 'VOIDmode', the bits of the value are stored in I0 and I1.

  17277.      I0 is customarily accessed with the macro 'CONST_DOUBLE_LOW' and I1

  17278.      with 'CONST_DOUBLE_HIGH'.

  17279. 

  17280.      If the constant is floating point (regardless of its precision),

  17281.      then the number of integers used to store the value depends on the

  17282.      size of 'REAL_VALUE_TYPE' (*note Floating Point::).  The integers

  17283.      represent a floating point number, but not precisely in the target

  17284.      machine's or host machine's floating point format.  To convert them

  17285.      to the precise bit pattern used by the target machine, use the

  17286.      macro 'REAL_VALUE_TO_TARGET_DOUBLE' and friends (*note Data

  17287.      Output::).

  17288. 

  17289. '(const_wide_int:M NUNITS ELT0 ...)'

  17290.      This contains an array of 'HOST_WIDE_INT's that is large enough to

  17291.      hold any constant that can be represented on the target.  This form

  17292.      of rtl is only used on targets that define

  17293.      'TARGET_SUPPORTS_WIDE_INT' to be nonzero and then 'CONST_DOUBLE's

  17294.      are only used to hold floating-point values.  If the target leaves

  17295.      'TARGET_SUPPORTS_WIDE_INT' defined as 0, 'CONST_WIDE_INT's are not

  17296.      used and 'CONST_DOUBLE's are as they were before.

  17297. 

  17298.      The values are stored in a compressed format.  The higher-order 0s

  17299.      or -1s are not represented if they are just the logical sign

  17300.      extension of the number that is represented.

  17301. 

  17302. 'CONST_WIDE_INT_VEC (CODE)'

  17303.      Returns the entire array of 'HOST_WIDE_INT's that are used to store

  17304.      the value.  This macro should be rarely used.

  17305. 

  17306. 'CONST_WIDE_INT_NUNITS (CODE)'

  17307.      The number of 'HOST_WIDE_INT's used to represent the number.  Note

  17308.      that this generally is smaller than the number of 'HOST_WIDE_INT's

  17309.      implied by the mode size.

  17310. 

  17311. 'CONST_WIDE_INT_ELT (CODE,I)'

  17312.      Returns the 'i'th element of the array.  Element 0 is contains the

  17313.      low order bits of the constant.

  17314. 

  17315. '(const_fixed:M ...)'

  17316.      Represents a fixed-point constant of mode M.  The operand is a data

  17317.      structure of type 'struct fixed_value' and is accessed with the

  17318.      macro 'CONST_FIXED_VALUE'.  The high part of data is accessed with

  17319.      'CONST_FIXED_VALUE_HIGH'; the low part is accessed with

  17320.      'CONST_FIXED_VALUE_LOW'.

  17321. 

  17322. '(const_poly_int:M [C0 C1 ...])'

  17323.      Represents a 'poly_int'-style polynomial integer with coefficients

  17324.      C0, C1, ....  The coefficients are 'wide_int'-based integers rather

  17325.      than rtxes.  'CONST_POLY_INT_COEFFS' gives the values of individual

  17326.      coefficients (which is mostly only useful in low-level routines)

  17327.      and 'const_poly_int_value' gives the full 'poly_int' value.

  17328. 

  17329. '(const_vector:M [X0 X1 ...])'

  17330.      Represents a vector constant.  The values in square brackets are

  17331.      elements of the vector, which are always 'const_int',

  17332.      'const_wide_int', 'const_double' or 'const_fixed' expressions.

  17333. 

  17334.      Each vector constant V is treated as a specific instance of an

  17335.      arbitrary-length sequence that itself contains

  17336.      'CONST_VECTOR_NPATTERNS (V)' interleaved patterns.  Each pattern

  17337.      has the form:

  17338. 

  17339.           { BASE0, BASE1, BASE1 + STEP, BASE1 + STEP * 2, ... }

  17340. 

  17341.      The first three elements in each pattern are enough to determine

  17342.      the values of the other elements.  However, if all STEPs are zero,

  17343.      only the first two elements are needed.  If in addition each BASE1

  17344.      is equal to the corresponding BASE0, only the first element in each

  17345.      pattern is needed.  The number of determining elements per pattern

  17346.      is given by 'CONST_VECTOR_NELTS_PER_PATTERN (V)'.

  17347. 

  17348.      For example, the constant:

  17349. 

  17350.           { 0, 1, 2, 6, 3, 8, 4, 10, 5, 12, 6, 14, 7, 16, 8, 18 }

  17351. 

  17352.      is interpreted as an interleaving of the sequences:

  17353. 

  17354.           { 0, 2, 3, 4, 5, 6, 7, 8 }

  17355.           { 1, 6, 8, 10, 12, 14, 16, 18 }

  17356. 

  17357.      where the sequences are represented by the following patterns:

  17358. 

  17359.           BASE0 == 0, BASE1 == 2, STEP == 1

  17360.           BASE0 == 1, BASE1 == 6, STEP == 2

  17361. 

  17362.      In this case:

  17363. 

  17364.           CONST_VECTOR_NPATTERNS (V) == 2

  17365.           CONST_VECTOR_NELTS_PER_PATTERN (V) == 3

  17366. 

  17367.      Thus the first 6 elements ('{ 0, 1, 2, 6, 3, 8 }') are enough to

  17368.      determine the whole sequence; we refer to them as the "encoded"

  17369.      elements.  They are the only elements present in the square

  17370.      brackets for variable-length 'const_vector's (i.e. for

  17371.      'const_vector's whose mode M has a variable number of elements).

  17372.      However, as a convenience to code that needs to handle both

  17373.      'const_vector's and 'parallel's, all elements are present in the

  17374.      square brackets for fixed-length 'const_vector's; the encoding

  17375.      scheme simply reduces the amount of work involved in processing

  17376.      constants that follow a regular pattern.

  17377. 

  17378.      Sometimes this scheme can create two possible encodings of the same

  17379.      vector.  For example { 0, 1 } could be seen as two patterns with

  17380.      one element each or one pattern with two elements (BASE0 and

  17381.      BASE1).  The canonical encoding is always the one with the fewest

  17382.      patterns or (if both encodings have the same number of petterns)

  17383.      the one with the fewest encoded elements.

  17384. 

  17385.      'const_vector_encoding_nelts (V)' gives the total number of encoded

  17386.      elements in V, which is 6 in the example above.

  17387.      'CONST_VECTOR_ENCODED_ELT (V, I)' accesses the value of encoded

  17388.      element I.

  17389. 

  17390.      'CONST_VECTOR_DUPLICATE_P (V)' is true if V simply contains

  17391.      repeated instances of 'CONST_VECTOR_NPATTERNS (V)' values.  This is

  17392.      a shorthand for testing 'CONST_VECTOR_NELTS_PER_PATTERN (V) == 1'.

  17393. 

  17394.      'CONST_VECTOR_STEPPED_P (V)' is true if at least one pattern in V

  17395.      has a nonzero step.  This is a shorthand for testing

  17396.      'CONST_VECTOR_NELTS_PER_PATTERN (V) == 3'.

  17397. 

  17398.      'CONST_VECTOR_NUNITS (V)' gives the total number of elements in V;

  17399.      it is a shorthand for getting the number of units in 'GET_MODE

  17400.      (V)'.

  17401. 

  17402.      The utility function 'const_vector_elt' gives the value of an

  17403.      arbitrary element as an 'rtx'.  'const_vector_int_elt' gives the

  17404.      same value as a 'wide_int'.

  17405. 

  17406. '(const_string STR)'

  17407.      Represents a constant string with value STR.  Currently this is

  17408.      used only for insn attributes (*note Insn Attributes::) since

  17409.      constant strings in C are placed in memory.

  17410. 

  17411. '(symbol_ref:MODE SYMBOL)'

  17412.      Represents the value of an assembler label for data.  SYMBOL is a

  17413.      string that describes the name of the assembler label.  If it

  17414.      starts with a '*', the label is the rest of SYMBOL not including

  17415.      the '*'.  Otherwise, the label is SYMBOL, usually prefixed with

  17416.      '_'.

  17417. 

  17418.      The 'symbol_ref' contains a mode, which is usually 'Pmode'.

  17419.      Usually that is the only mode for which a symbol is directly valid.

  17420. 

  17421. '(label_ref:MODE LABEL)'

  17422.      Represents the value of an assembler label for code.  It contains

  17423.      one operand, an expression, which must be a 'code_label' or a

  17424.      'note' of type 'NOTE_INSN_DELETED_LABEL' that appears in the

  17425.      instruction sequence to identify the place where the label should

  17426.      go.

  17427. 

  17428.      The reason for using a distinct expression type for code label

  17429.      references is so that jump optimization can distinguish them.

  17430. 

  17431.      The 'label_ref' contains a mode, which is usually 'Pmode'.  Usually

  17432.      that is the only mode for which a label is directly valid.

  17433. 

  17434. '(const:M EXP)'

  17435.      Represents a constant that is the result of an assembly-time

  17436.      arithmetic computation.  The operand, EXP, contains only

  17437.      'const_int', 'symbol_ref', 'label_ref' or 'unspec' expressions,

  17438.      combined with 'plus' and 'minus'.  Any such 'unspec's are

  17439.      target-specific and typically represent some form of relocation

  17440.      operator.  M should be a valid address mode.

  17441. 

  17442. '(high:M EXP)'

  17443.      Represents the high-order bits of EXP.  The number of bits is

  17444.      machine-dependent and is normally the number of bits specified in

  17445.      an instruction that initializes the high order bits of a register.

  17446.      It is used with 'lo_sum' to represent the typical two-instruction

  17447.      sequence used in RISC machines to reference large immediate values

  17448.      and/or link-time constants such as global memory addresses.  In the

  17449.      latter case, M is 'Pmode' and EXP is usually a constant expression

  17450.      involving 'symbol_ref'.

  17451. 

  17452.  The macro 'CONST0_RTX (MODE)' refers to an expression with value 0 in

  17453. mode MODE.  If mode MODE is of mode class 'MODE_INT', it returns

  17454. 'const0_rtx'.  If mode MODE is of mode class 'MODE_FLOAT', it returns a

  17455. 'CONST_DOUBLE' expression in mode MODE.  Otherwise, it returns a

  17456. 'CONST_VECTOR' expression in mode MODE.  Similarly, the macro

  17457. 'CONST1_RTX (MODE)' refers to an expression with value 1 in mode MODE

  17458. and similarly for 'CONST2_RTX'.  The 'CONST1_RTX' and 'CONST2_RTX'

  17459. macros are undefined for vector modes.

  17460. 

  17461. 

  17462. File: gccint.info,  Node: Regs and Memory,  Next: Arithmetic,  Prev: Constants,  Up: RTL

  17463. 

  17464. 14.8 Registers and Memory

  17465. =========================

  17466. 

  17467. Here are the RTL expression types for describing access to machine

  17468. registers and to main memory.

  17469. 

  17470. '(reg:M N)'

  17471.      For small values of the integer N (those that are less than

  17472.      'FIRST_PSEUDO_REGISTER'), this stands for a reference to machine

  17473.      register number N: a "hard register".  For larger values of N, it

  17474.      stands for a temporary value or "pseudo register".  The compiler's

  17475.      strategy is to generate code assuming an unlimited number of such

  17476.      pseudo registers, and later convert them into hard registers or

  17477.      into memory references.

  17478. 

  17479.      M is the machine mode of the reference.  It is necessary because

  17480.      machines can generally refer to each register in more than one

  17481.      mode.  For example, a register may contain a full word but there

  17482.      may be instructions to refer to it as a half word or as a single

  17483.      byte, as well as instructions to refer to it as a floating point

  17484.      number of various precisions.

  17485. 

  17486.      Even for a register that the machine can access in only one mode,

  17487.      the mode must always be specified.

  17488. 

  17489.      The symbol 'FIRST_PSEUDO_REGISTER' is defined by the machine

  17490.      description, since the number of hard registers on the machine is

  17491.      an invariant characteristic of the machine.  Note, however, that

  17492.      not all of the machine registers must be general registers.  All

  17493.      the machine registers that can be used for storage of data are

  17494.      given hard register numbers, even those that can be used only in

  17495.      certain instructions or can hold only certain types of data.

  17496. 

  17497.      A hard register may be accessed in various modes throughout one

  17498.      function, but each pseudo register is given a natural mode and is

  17499.      accessed only in that mode.  When it is necessary to describe an

  17500.      access to a pseudo register using a nonnatural mode, a 'subreg'

  17501.      expression is used.

  17502. 

  17503.      A 'reg' expression with a machine mode that specifies more than one

  17504.      word of data may actually stand for several consecutive registers.

  17505.      If in addition the register number specifies a hardware register,

  17506.      then it actually represents several consecutive hardware registers

  17507.      starting with the specified one.

  17508. 

  17509.      Each pseudo register number used in a function's RTL code is

  17510.      represented by a unique 'reg' expression.

  17511. 

  17512.      Some pseudo register numbers, those within the range of

  17513.      'FIRST_VIRTUAL_REGISTER' to 'LAST_VIRTUAL_REGISTER' only appear

  17514.      during the RTL generation phase and are eliminated before the

  17515.      optimization phases.  These represent locations in the stack frame

  17516.      that cannot be determined until RTL generation for the function has

  17517.      been completed.  The following virtual register numbers are

  17518.      defined:

  17519. 

  17520.      'VIRTUAL_INCOMING_ARGS_REGNUM'

  17521.           This points to the first word of the incoming arguments passed

  17522.           on the stack.  Normally these arguments are placed there by

  17523.           the caller, but the callee may have pushed some arguments that

  17524.           were previously passed in registers.

  17525. 

  17526.           When RTL generation is complete, this virtual register is

  17527.           replaced by the sum of the register given by

  17528.           'ARG_POINTER_REGNUM' and the value of 'FIRST_PARM_OFFSET'.

  17529. 

  17530.      'VIRTUAL_STACK_VARS_REGNUM'

  17531.           If 'FRAME_GROWS_DOWNWARD' is defined to a nonzero value, this

  17532.           points to immediately above the first variable on the stack.

  17533.           Otherwise, it points to the first variable on the stack.

  17534. 

  17535.           'VIRTUAL_STACK_VARS_REGNUM' is replaced with the sum of the

  17536.           register given by 'FRAME_POINTER_REGNUM' and the value

  17537.           'TARGET_STARTING_FRAME_OFFSET'.

  17538. 

  17539.      'VIRTUAL_STACK_DYNAMIC_REGNUM'

  17540.           This points to the location of dynamically allocated memory on

  17541.           the stack immediately after the stack pointer has been

  17542.           adjusted by the amount of memory desired.

  17543. 

  17544.           This virtual register is replaced by the sum of the register

  17545.           given by 'STACK_POINTER_REGNUM' and the value

  17546.           'STACK_DYNAMIC_OFFSET'.

  17547. 

  17548.      'VIRTUAL_OUTGOING_ARGS_REGNUM'

  17549.           This points to the location in the stack at which outgoing

  17550.           arguments should be written when the stack is pre-pushed

  17551.           (arguments pushed using push insns should always use

  17552.           'STACK_POINTER_REGNUM').

  17553. 

  17554.           This virtual register is replaced by the sum of the register

  17555.           given by 'STACK_POINTER_REGNUM' and the value

  17556.           'STACK_POINTER_OFFSET'.

  17557. 

  17558. '(subreg:M1 REG:M2 BYTENUM)'

  17559. 

  17560.      'subreg' expressions are used to refer to a register in a machine

  17561.      mode other than its natural one, or to refer to one register of a

  17562.      multi-part 'reg' that actually refers to several registers.

  17563. 

  17564.      Each pseudo register has a natural mode.  If it is necessary to

  17565.      operate on it in a different mode, the register must be enclosed in

  17566.      a 'subreg'.

  17567. 

  17568.      There are currently three supported types for the first operand of

  17569.      a 'subreg':

  17570.         * pseudo registers This is the most common case.  Most 'subreg's

  17571.           have pseudo 'reg's as their first operand.

  17572. 

  17573.         * mem 'subreg's of 'mem' were common in earlier versions of GCC

  17574.           and are still supported.  During the reload pass these are

  17575.           replaced by plain 'mem's.  On machines that do not do

  17576.           instruction scheduling, use of 'subreg's of 'mem' are still

  17577.           used, but this is no longer recommended.  Such 'subreg's are

  17578.           considered to be 'register_operand's rather than

  17579.           'memory_operand's before and during reload.  Because of this,

  17580.           the scheduling passes cannot properly schedule instructions

  17581.           with 'subreg's of 'mem', so for machines that do scheduling,

  17582.           'subreg's of 'mem' should never be used.  To support this, the

  17583.           combine and recog passes have explicit code to inhibit the

  17584.           creation of 'subreg's of 'mem' when 'INSN_SCHEDULING' is

  17585.           defined.

  17586. 

  17587.           The use of 'subreg's of 'mem' after the reload pass is an area

  17588.           that is not well understood and should be avoided.  There is

  17589.           still some code in the compiler to support this, but this code

  17590.           has possibly rotted.  This use of 'subreg's is discouraged and

  17591.           will most likely not be supported in the future.

  17592. 

  17593.         * hard registers It is seldom necessary to wrap hard registers

  17594.           in 'subreg's; such registers would normally reduce to a single

  17595.           'reg' rtx.  This use of 'subreg's is discouraged and may not

  17596.           be supported in the future.

  17597. 

  17598.      'subreg's of 'subreg's are not supported.  Using

  17599.      'simplify_gen_subreg' is the recommended way to avoid this problem.

  17600. 

  17601.      'subreg's come in two distinct flavors, each having its own usage

  17602.      and rules:

  17603. 

  17604.      Paradoxical subregs

  17605.           When M1 is strictly wider than M2, the 'subreg' expression is

  17606.           called "paradoxical".  The canonical test for this class of

  17607.           'subreg' is:

  17608. 

  17609.                paradoxical_subreg_p (M1, M2)

  17610. 

  17611.           Paradoxical 'subreg's can be used as both lvalues and rvalues.

  17612.           When used as an lvalue, the low-order bits of the source value

  17613.           are stored in REG and the high-order bits are discarded.  When

  17614.           used as an rvalue, the low-order bits of the 'subreg' are

  17615.           taken from REG while the high-order bits may or may not be

  17616.           defined.

  17617. 

  17618.           The high-order bits of rvalues are defined in the following

  17619.           circumstances:

  17620. 

  17621.              * 'subreg's of 'mem' When M2 is smaller than a word, the

  17622.                macro 'LOAD_EXTEND_OP', can control how the high-order

  17623.                bits are defined.

  17624. 

  17625.              * 'subreg' of 'reg's The upper bits are defined when

  17626.                'SUBREG_PROMOTED_VAR_P' is true.

  17627.                'SUBREG_PROMOTED_UNSIGNED_P' describes what the upper

  17628.                bits hold.  Such subregs usually represent local

  17629.                variables, register variables and parameter pseudo

  17630.                variables that have been promoted to a wider mode.

  17631. 

  17632.           BYTENUM is always zero for a paradoxical 'subreg', even on

  17633.           big-endian targets.

  17634. 

  17635.           For example, the paradoxical 'subreg':

  17636. 

  17637.                (set (subreg:SI (reg:HI X) 0) Y)

  17638. 

  17639.           stores the lower 2 bytes of Y in X and discards the upper 2

  17640.           bytes.  A subsequent:

  17641. 

  17642.                (set Z (subreg:SI (reg:HI X) 0))

  17643. 

  17644.           would set the lower two bytes of Z to Y and set the upper two

  17645.           bytes to an unknown value assuming 'SUBREG_PROMOTED_VAR_P' is

  17646.           false.

  17647. 

  17648.      Normal subregs

  17649.           When M1 is at least as narrow as M2 the 'subreg' expression is

  17650.           called "normal".

  17651. 

  17652.           Normal 'subreg's restrict consideration to certain bits of

  17653.           REG.  For this purpose, REG is divided into

  17654.           individually-addressable blocks in which each block has:

  17655. 

  17656.                REGMODE_NATURAL_SIZE (M2)

  17657. 

  17658.           bytes.  Usually the value is 'UNITS_PER_WORD'; that is, most

  17659.           targets usually treat each word of a register as being

  17660.           independently addressable.

  17661. 

  17662.           There are two types of normal 'subreg'.  If M1 is known to be

  17663.           no bigger than a block, the 'subreg' refers to the

  17664.           least-significant part (or "lowpart") of one block of REG.  If

  17665.           M1 is known to be larger than a block, the 'subreg' refers to

  17666.           two or more complete blocks.

  17667. 

  17668.           When used as an lvalue, 'subreg' is a block-based accessor.

  17669.           Storing to a 'subreg' modifies all the blocks of REG that

  17670.           overlap the 'subreg', but it leaves the other blocks of REG

  17671.           alone.

  17672. 

  17673.           When storing to a normal 'subreg' that is smaller than a

  17674.           block, the other bits of the referenced block are usually left

  17675.           in an undefined state.  This laxity makes it easier to

  17676.           generate efficient code for such instructions.  To represent

  17677.           an instruction that preserves all the bits outside of those in

  17678.           the 'subreg', use 'strict_low_part' or 'zero_extract' around

  17679.           the 'subreg'.

  17680. 

  17681.           BYTENUM must identify the offset of the first byte of the

  17682.           'subreg' from the start of REG, assuming that REG is laid out

  17683.           in memory order.  The memory order of bytes is defined by two

  17684.           target macros, 'WORDS_BIG_ENDIAN' and 'BYTES_BIG_ENDIAN':

  17685. 

  17686.              * 'WORDS_BIG_ENDIAN', if set to 1, says that byte number

  17687.                zero is part of the most significant word; otherwise, it

  17688.                is part of the least significant word.

  17689. 

  17690.              * 'BYTES_BIG_ENDIAN', if set to 1, says that byte number

  17691.                zero is the most significant byte within a word;

  17692.                otherwise, it is the least significant byte within a

  17693.                word.

  17694. 

  17695.           On a few targets, 'FLOAT_WORDS_BIG_ENDIAN' disagrees with

  17696.           'WORDS_BIG_ENDIAN'.  However, most parts of the compiler treat

  17697.           floating point values as if they had the same endianness as

  17698.           integer values.  This works because they handle them solely as

  17699.           a collection of integer values, with no particular numerical

  17700.           value.  Only real.c and the runtime libraries care about

  17701.           'FLOAT_WORDS_BIG_ENDIAN'.

  17702. 

  17703.           Thus,

  17704. 

  17705.                (subreg:HI (reg:SI X) 2)

  17706. 

  17707.           on a 'BYTES_BIG_ENDIAN', 'UNITS_PER_WORD == 4' target is the

  17708.           same as

  17709. 

  17710.                (subreg:HI (reg:SI X) 0)

  17711. 

  17712.           on a little-endian, 'UNITS_PER_WORD == 4' target.  Both

  17713.           'subreg's access the lower two bytes of register X.

  17714. 

  17715.           Note that the byte offset is a polynomial integer; it may not

  17716.           be a compile-time constant on targets with variable-sized

  17717.           modes.  However, the restrictions above mean that there are

  17718.           only a certain set of acceptable offsets for a given

  17719.           combination of M1 and M2.  The compiler can always tell which

  17720.           blocks a valid subreg occupies, and whether the subreg is a

  17721.           lowpart of a block.

  17722. 

  17723.      A 'MODE_PARTIAL_INT' mode behaves as if it were as wide as the

  17724.      corresponding 'MODE_INT' mode, except that it has an unknown number

  17725.      of undefined bits.  For example:

  17726. 

  17727.           (subreg:PSI (reg:SI 0) 0)

  17728. 

  17729.      accesses the whole of '(reg:SI 0)', but the exact relationship

  17730.      between the 'PSImode' value and the 'SImode' value is not defined.

  17731.      If we assume 'REGMODE_NATURAL_SIZE (DImode) <= 4', then the

  17732.      following two 'subreg's:

  17733. 

  17734.           (subreg:PSI (reg:DI 0) 0)

  17735.           (subreg:PSI (reg:DI 0) 4)

  17736. 

  17737.      represent independent 4-byte accesses to the two halves of '(reg:DI

  17738.      0)'.  Both 'subreg's have an unknown number of undefined bits.

  17739. 

  17740.      If 'REGMODE_NATURAL_SIZE (PSImode) <= 2' then these two 'subreg's:

  17741. 

  17742.           (subreg:HI (reg:PSI 0) 0)

  17743.           (subreg:HI (reg:PSI 0) 2)

  17744. 

  17745.      represent independent 2-byte accesses that together span the whole

  17746.      of '(reg:PSI 0)'.  Storing to the first 'subreg' does not affect

  17747.      the value of the second, and vice versa.  '(reg:PSI 0)' has an

  17748.      unknown number of undefined bits, so the assignment:

  17749. 

  17750.           (set (subreg:HI (reg:PSI 0) 0) (reg:HI 4))

  17751. 

  17752.      does not guarantee that '(subreg:HI (reg:PSI 0) 0)' has the value

  17753.      '(reg:HI 4)'.

  17754. 

  17755.      The rules above apply to both pseudo REGs and hard REGs.  If the

  17756.      semantics are not correct for particular combinations of M1, M2 and

  17757.      hard REG, the target-specific code must ensure that those

  17758.      combinations are never used.  For example:

  17759. 

  17760.           TARGET_CAN_CHANGE_MODE_CLASS (M2, M1, CLASS)

  17761. 

  17762.      must be false for every class CLASS that includes REG.

  17763. 

  17764.      GCC must be able to determine at compile time whether a subreg is

  17765.      paradoxical, whether it occupies a whole number of blocks, or

  17766.      whether it is a lowpart of a block.  This means that certain

  17767.      combinations of variable-sized mode are not permitted.  For

  17768.      example, if M2 holds N 'SI' values, where N is greater than zero,

  17769.      it is not possible to form a 'DI' 'subreg' of it; such a 'subreg'

  17770.      would be paradoxical when N is 1 but not when N is greater than 1.

  17771. 

  17772.      The first operand of a 'subreg' expression is customarily accessed

  17773.      with the 'SUBREG_REG' macro and the second operand is customarily

  17774.      accessed with the 'SUBREG_BYTE' macro.

  17775. 

  17776.      It has been several years since a platform in which

  17777.      'BYTES_BIG_ENDIAN' not equal to 'WORDS_BIG_ENDIAN' has been tested.

  17778.      Anyone wishing to support such a platform in the future may be

  17779.      confronted with code rot.

  17780. 

  17781. '(scratch:M)'

  17782.      This represents a scratch register that will be required for the

  17783.      execution of a single instruction and not used subsequently.  It is

  17784.      converted into a 'reg' by either the local register allocator or

  17785.      the reload pass.

  17786. 

  17787.      'scratch' is usually present inside a 'clobber' operation (*note

  17788.      Side Effects::).

  17789. 

  17790. '(cc0)'

  17791.      This refers to the machine's condition code register.  It has no

  17792.      operands and may not have a machine mode.  There are two ways to

  17793.      use it:

  17794. 

  17795.         * To stand for a complete set of condition code flags.  This is

  17796.           best on most machines, where each comparison sets the entire

  17797.           series of flags.

  17798. 

  17799.           With this technique, '(cc0)' may be validly used in only two

  17800.           contexts: as the destination of an assignment (in test and

  17801.           compare instructions) and in comparison operators comparing

  17802.           against zero ('const_int' with value zero; that is to say,

  17803.           'const0_rtx').

  17804. 

  17805.         * To stand for a single flag that is the result of a single

  17806.           condition.  This is useful on machines that have only a single

  17807.           flag bit, and in which comparison instructions must specify

  17808.           the condition to test.

  17809. 

  17810.           With this technique, '(cc0)' may be validly used in only two

  17811.           contexts: as the destination of an assignment (in test and

  17812.           compare instructions) where the source is a comparison

  17813.           operator, and as the first operand of 'if_then_else' (in a

  17814.           conditional branch).

  17815. 

  17816.      There is only one expression object of code 'cc0'; it is the value

  17817.      of the variable 'cc0_rtx'.  Any attempt to create an expression of

  17818.      code 'cc0' will return 'cc0_rtx'.

  17819. 

  17820.      Instructions can set the condition code implicitly.  On many

  17821.      machines, nearly all instructions set the condition code based on

  17822.      the value that they compute or store.  It is not necessary to

  17823.      record these actions explicitly in the RTL because the machine

  17824.      description includes a prescription for recognizing the

  17825.      instructions that do so (by means of the macro 'NOTICE_UPDATE_CC').

  17826.      *Note Condition Code::.  Only instructions whose sole purpose is to

  17827.      set the condition code, and instructions that use the condition

  17828.      code, need mention '(cc0)'.

  17829. 

  17830.      On some machines, the condition code register is given a register

  17831.      number and a 'reg' is used instead of '(cc0)'.  This is usually the

  17832.      preferable approach if only a small subset of instructions modify

  17833.      the condition code.  Other machines store condition codes in

  17834.      general registers; in such cases a pseudo register should be used.

  17835. 

  17836.      Some machines, such as the SPARC and RS/6000, have two sets of

  17837.      arithmetic instructions, one that sets and one that does not set

  17838.      the condition code.  This is best handled by normally generating

  17839.      the instruction that does not set the condition code, and making a

  17840.      pattern that both performs the arithmetic and sets the condition

  17841.      code register (which would not be '(cc0)' in this case).  For

  17842.      examples, search for 'addcc' and 'andcc' in 'sparc.md'.

  17843. 

  17844. '(pc)'

  17845.      This represents the machine's program counter.  It has no operands

  17846.      and may not have a machine mode.  '(pc)' may be validly used only

  17847.      in certain specific contexts in jump instructions.

  17848. 

  17849.      There is only one expression object of code 'pc'; it is the value

  17850.      of the variable 'pc_rtx'.  Any attempt to create an expression of

  17851.      code 'pc' will return 'pc_rtx'.

  17852. 

  17853.      All instructions that do not jump alter the program counter

  17854.      implicitly by incrementing it, but there is no need to mention this

  17855.      in the RTL.

  17856. 

  17857. '(mem:M ADDR ALIAS)'

  17858.      This RTX represents a reference to main memory at an address

  17859.      represented by the expression ADDR.  M specifies how large a unit

  17860.      of memory is accessed.  ALIAS specifies an alias set for the

  17861.      reference.  In general two items are in different alias sets if

  17862.      they cannot reference the same memory address.

  17863. 

  17864.      The construct '(mem:BLK (scratch))' is considered to alias all

  17865.      other memories.  Thus it may be used as a memory barrier in

  17866.      epilogue stack deallocation patterns.

  17867. 

  17868. '(concatM RTX RTX)'

  17869.      This RTX represents the concatenation of two other RTXs.  This is

  17870.      used for complex values.  It should only appear in the RTL attached

  17871.      to declarations and during RTL generation.  It should not appear in

  17872.      the ordinary insn chain.

  17873. 

  17874. '(concatnM [RTX ...])'

  17875.      This RTX represents the concatenation of all the RTX to make a

  17876.      single value.  Like 'concat', this should only appear in

  17877.      declarations, and not in the insn chain.

  17878. 

  17879. 

  17880. File: gccint.info,  Node: Arithmetic,  Next: Comparisons,  Prev: Regs and Memory,  Up: RTL

  17881. 

  17882. 14.9 RTL Expressions for Arithmetic

  17883. ===================================

  17884. 

  17885. Unless otherwise specified, all the operands of arithmetic expressions

  17886. must be valid for mode M.  An operand is valid for mode M if it has mode

  17887. M, or if it is a 'const_int' or 'const_double' and M is a mode of class

  17888. 'MODE_INT'.

  17889. 

  17890.  For commutative binary operations, constants should be placed in the

  17891. second operand.

  17892. 

  17893. '(plus:M X Y)'

  17894. '(ss_plus:M X Y)'

  17895. '(us_plus:M X Y)'

  17896. 

  17897.      These three expressions all represent the sum of the values

  17898.      represented by X and Y carried out in machine mode M.  They differ

  17899.      in their behavior on overflow of integer modes.  'plus' wraps round

  17900.      modulo the width of M; 'ss_plus' saturates at the maximum signed

  17901.      value representable in M; 'us_plus' saturates at the maximum

  17902.      unsigned value.

  17903. 

  17904. '(lo_sum:M X Y)'

  17905. 

  17906.      This expression represents the sum of X and the low-order bits of

  17907.      Y.  It is used with 'high' (*note Constants::) to represent the

  17908.      typical two-instruction sequence used in RISC machines to reference

  17909.      large immediate values and/or link-time constants such as global

  17910.      memory addresses.  In the latter case, M is 'Pmode' and Y is

  17911.      usually a constant expression involving 'symbol_ref'.

  17912. 

  17913.      The number of low order bits is machine-dependent but is normally

  17914.      the number of bits in mode M minus the number of bits set by

  17915.      'high'.

  17916. 

  17917. '(minus:M X Y)'

  17918. '(ss_minus:M X Y)'

  17919. '(us_minus:M X Y)'

  17920. 

  17921.      These three expressions represent the result of subtracting Y from

  17922.      X, carried out in mode M.  Behavior on overflow is the same as for

  17923.      the three variants of 'plus' (see above).

  17924. 

  17925. '(compare:M X Y)'

  17926.      Represents the result of subtracting Y from X for purposes of

  17927.      comparison.  The result is computed without overflow, as if with

  17928.      infinite precision.

  17929. 

  17930.      Of course, machines cannot really subtract with infinite precision.

  17931.      However, they can pretend to do so when only the sign of the result

  17932.      will be used, which is the case when the result is stored in the

  17933.      condition code.  And that is the _only_ way this kind of expression

  17934.      may validly be used: as a value to be stored in the condition

  17935.      codes, either '(cc0)' or a register.  *Note Comparisons::.

  17936. 

  17937.      The mode M is not related to the modes of X and Y, but instead is

  17938.      the mode of the condition code value.  If '(cc0)' is used, it is

  17939.      'VOIDmode'.  Otherwise it is some mode in class 'MODE_CC', often

  17940.      'CCmode'.  *Note Condition Code::.  If M is 'VOIDmode' or 'CCmode',

  17941.      the operation returns sufficient information (in an unspecified

  17942.      format) so that any comparison operator can be applied to the

  17943.      result of the 'COMPARE' operation.  For other modes in class

  17944.      'MODE_CC', the operation only returns a subset of this information.

  17945. 

  17946.      Normally, X and Y must have the same mode.  Otherwise, 'compare' is

  17947.      valid only if the mode of X is in class 'MODE_INT' and Y is a

  17948.      'const_int' or 'const_double' with mode 'VOIDmode'.  The mode of X

  17949.      determines what mode the comparison is to be done in; thus it must

  17950.      not be 'VOIDmode'.

  17951. 

  17952.      If one of the operands is a constant, it should be placed in the

  17953.      second operand and the comparison code adjusted as appropriate.

  17954. 

  17955.      A 'compare' specifying two 'VOIDmode' constants is not valid since

  17956.      there is no way to know in what mode the comparison is to be

  17957.      performed; the comparison must either be folded during the

  17958.      compilation or the first operand must be loaded into a register

  17959.      while its mode is still known.

  17960. 

  17961. '(neg:M X)'

  17962. '(ss_neg:M X)'

  17963. '(us_neg:M X)'

  17964.      These two expressions represent the negation (subtraction from

  17965.      zero) of the value represented by X, carried out in mode M.  They

  17966.      differ in the behavior on overflow of integer modes.  In the case

  17967.      of 'neg', the negation of the operand may be a number not

  17968.      representable in mode M, in which case it is truncated to M.

  17969.      'ss_neg' and 'us_neg' ensure that an out-of-bounds result saturates

  17970.      to the maximum or minimum signed or unsigned value.

  17971. 

  17972. '(mult:M X Y)'

  17973. '(ss_mult:M X Y)'

  17974. '(us_mult:M X Y)'

  17975.      Represents the signed product of the values represented by X and Y

  17976.      carried out in machine mode M.  'ss_mult' and 'us_mult' ensure that

  17977.      an out-of-bounds result saturates to the maximum or minimum signed

  17978.      or unsigned value.

  17979. 

  17980.      Some machines support a multiplication that generates a product

  17981.      wider than the operands.  Write the pattern for this as

  17982. 

  17983.           (mult:M (sign_extend:M X) (sign_extend:M Y))

  17984. 

  17985.      where M is wider than the modes of X and Y, which need not be the

  17986.      same.

  17987. 

  17988.      For unsigned widening multiplication, use the same idiom, but with

  17989.      'zero_extend' instead of 'sign_extend'.

  17990. 

  17991. '(fma:M X Y Z)'

  17992.      Represents the 'fma', 'fmaf', and 'fmal' builtin functions, which

  17993.      compute 'X * Y + Z' without doing an intermediate rounding step.

  17994. 

  17995. '(div:M X Y)'

  17996. '(ss_div:M X Y)'

  17997.      Represents the quotient in signed division of X by Y, carried out

  17998.      in machine mode M.  If M is a floating point mode, it represents

  17999.      the exact quotient; otherwise, the integerized quotient.  'ss_div'

  18000.      ensures that an out-of-bounds result saturates to the maximum or

  18001.      minimum signed value.

  18002. 

  18003.      Some machines have division instructions in which the operands and

  18004.      quotient widths are not all the same; you should represent such

  18005.      instructions using 'truncate' and 'sign_extend' as in,

  18006. 

  18007.           (truncate:M1 (div:M2 X (sign_extend:M2 Y)))

  18008. 

  18009. '(udiv:M X Y)'

  18010. '(us_div:M X Y)'

  18011.      Like 'div' but represents unsigned division.  'us_div' ensures that

  18012.      an out-of-bounds result saturates to the maximum or minimum

  18013.      unsigned value.

  18014. 

  18015. '(mod:M X Y)'

  18016. '(umod:M X Y)'

  18017.      Like 'div' and 'udiv' but represent the remainder instead of the

  18018.      quotient.

  18019. 

  18020. '(smin:M X Y)'

  18021. '(smax:M X Y)'

  18022.      Represents the smaller (for 'smin') or larger (for 'smax') of X and

  18023.      Y, interpreted as signed values in mode M.  When used with floating

  18024.      point, if both operands are zeros, or if either operand is 'NaN',

  18025.      then it is unspecified which of the two operands is returned as the

  18026.      result.

  18027. 

  18028. '(umin:M X Y)'

  18029. '(umax:M X Y)'

  18030.      Like 'smin' and 'smax', but the values are interpreted as unsigned

  18031.      integers.

  18032. 

  18033. '(not:M X)'

  18034.      Represents the bitwise complement of the value represented by X,

  18035.      carried out in mode M, which must be a fixed-point machine mode.

  18036. 

  18037. '(and:M X Y)'

  18038.      Represents the bitwise logical-and of the values represented by X

  18039.      and Y, carried out in machine mode M, which must be a fixed-point

  18040.      machine mode.

  18041. 

  18042. '(ior:M X Y)'

  18043.      Represents the bitwise inclusive-or of the values represented by X

  18044.      and Y, carried out in machine mode M, which must be a fixed-point

  18045.      mode.

  18046. 

  18047. '(xor:M X Y)'

  18048.      Represents the bitwise exclusive-or of the values represented by X

  18049.      and Y, carried out in machine mode M, which must be a fixed-point

  18050.      mode.

  18051. 

  18052. '(ashift:M X C)'

  18053. '(ss_ashift:M X C)'

  18054. '(us_ashift:M X C)'

  18055.      These three expressions represent the result of arithmetically

  18056.      shifting X left by C places.  They differ in their behavior on

  18057.      overflow of integer modes.  An 'ashift' operation is a plain shift

  18058.      with no special behavior in case of a change in the sign bit;

  18059.      'ss_ashift' and 'us_ashift' saturates to the minimum or maximum

  18060.      representable value if any of the bits shifted out differs from the

  18061.      final sign bit.

  18062. 

  18063.      X have mode M, a fixed-point machine mode.  C be a fixed-point mode

  18064.      or be a constant with mode 'VOIDmode'; which mode is determined by

  18065.      the mode called for in the machine description entry for the

  18066.      left-shift instruction.  For example, on the VAX, the mode of C is

  18067.      'QImode' regardless of M.

  18068. 

  18069. '(lshiftrt:M X C)'

  18070. '(ashiftrt:M X C)'

  18071.      Like 'ashift' but for right shift.  Unlike the case for left shift,

  18072.      these two operations are distinct.

  18073. 

  18074. '(rotate:M X C)'

  18075. '(rotatert:M X C)'

  18076.      Similar but represent left and right rotate.  If C is a constant,

  18077.      use 'rotate'.

  18078. 

  18079. '(abs:M X)'

  18080. '(ss_abs:M X)'

  18081.      Represents the absolute value of X, computed in mode M.  'ss_abs'

  18082.      ensures that an out-of-bounds result saturates to the maximum

  18083.      signed value.

  18084. 

  18085. '(sqrt:M X)'

  18086.      Represents the square root of X, computed in mode M.  Most often M

  18087.      will be a floating point mode.

  18088. 

  18089. '(ffs:M X)'

  18090.      Represents one plus the index of the least significant 1-bit in X,

  18091.      represented as an integer of mode M.  (The value is zero if X is

  18092.      zero.)  The mode of X must be M or 'VOIDmode'.

  18093. 

  18094. '(clrsb:M X)'

  18095.      Represents the number of redundant leading sign bits in X,

  18096.      represented as an integer of mode M, starting at the most

  18097.      significant bit position.  This is one less than the number of

  18098.      leading sign bits (either 0 or 1), with no special cases.  The mode

  18099.      of X must be M or 'VOIDmode'.

  18100. 

  18101. '(clz:M X)'

  18102.      Represents the number of leading 0-bits in X, represented as an

  18103.      integer of mode M, starting at the most significant bit position.

  18104.      If X is zero, the value is determined by

  18105.      'CLZ_DEFINED_VALUE_AT_ZERO' (*note Misc::).  Note that this is one

  18106.      of the few expressions that is not invariant under widening.  The

  18107.      mode of X must be M or 'VOIDmode'.

  18108. 

  18109. '(ctz:M X)'

  18110.      Represents the number of trailing 0-bits in X, represented as an

  18111.      integer of mode M, starting at the least significant bit position.

  18112.      If X is zero, the value is determined by

  18113.      'CTZ_DEFINED_VALUE_AT_ZERO' (*note Misc::).  Except for this case,

  18114.      'ctz(x)' is equivalent to 'ffs(X) - 1'.  The mode of X must be M or

  18115.      'VOIDmode'.

  18116. 

  18117. '(popcount:M X)'

  18118.      Represents the number of 1-bits in X, represented as an integer of

  18119.      mode M.  The mode of X must be M or 'VOIDmode'.

  18120. 

  18121. '(parity:M X)'

  18122.      Represents the number of 1-bits modulo 2 in X, represented as an

  18123.      integer of mode M.  The mode of X must be M or 'VOIDmode'.

  18124. 

  18125. '(bswap:M X)'

  18126.      Represents the value X with the order of bytes reversed, carried

  18127.      out in mode M, which must be a fixed-point machine mode.  The mode

  18128.      of X must be M or 'VOIDmode'.

  18129. 

  18130. 

  18131. File: gccint.info,  Node: Comparisons,  Next: Bit-Fields,  Prev: Arithmetic,  Up: RTL

  18132. 

  18133. 14.10 Comparison Operations

  18134. ===========================

  18135. 

  18136. Comparison operators test a relation on two operands and are considered

  18137. to represent a machine-dependent nonzero value described by, but not

  18138. necessarily equal to, 'STORE_FLAG_VALUE' (*note Misc::) if the relation

  18139. holds, or zero if it does not, for comparison operators whose results

  18140. have a 'MODE_INT' mode, 'FLOAT_STORE_FLAG_VALUE' (*note Misc::) if the

  18141. relation holds, or zero if it does not, for comparison operators that

  18142. return floating-point values, and a vector of either

  18143. 'VECTOR_STORE_FLAG_VALUE' (*note Misc::) if the relation holds, or of

  18144. zeros if it does not, for comparison operators that return vector

  18145. results.  The mode of the comparison operation is independent of the

  18146. mode of the data being compared.  If the comparison operation is being

  18147. tested (e.g., the first operand of an 'if_then_else'), the mode must be

  18148. 'VOIDmode'.

  18149. 

  18150.  There are two ways that comparison operations may be used.  The

  18151. comparison operators may be used to compare the condition codes '(cc0)'

  18152. against zero, as in '(eq (cc0) (const_int 0))'.  Such a construct

  18153. actually refers to the result of the preceding instruction in which the

  18154. condition codes were set.  The instruction setting the condition code

  18155. must be adjacent to the instruction using the condition code; only

  18156. 'note' insns may separate them.

  18157. 

  18158.  Alternatively, a comparison operation may directly compare two data

  18159. objects.  The mode of the comparison is determined by the operands; they

  18160. must both be valid for a common machine mode.  A comparison with both

  18161. operands constant would be invalid as the machine mode could not be

  18162. deduced from it, but such a comparison should never exist in RTL due to

  18163. constant folding.

  18164. 

  18165.  In the example above, if '(cc0)' were last set to '(compare X Y)', the

  18166. comparison operation is identical to '(eq X Y)'.  Usually only one style

  18167. of comparisons is supported on a particular machine, but the combine

  18168. pass will try to merge the operations to produce the 'eq' shown in case

  18169. it exists in the context of the particular insn involved.

  18170. 

  18171.  Inequality comparisons come in two flavors, signed and unsigned.  Thus,

  18172. there are distinct expression codes 'gt' and 'gtu' for signed and

  18173. unsigned greater-than.  These can produce different results for the same

  18174. pair of integer values: for example, 1 is signed greater-than -1 but not

  18175. unsigned greater-than, because -1 when regarded as unsigned is actually

  18176. '0xffffffff' which is greater than 1.

  18177. 

  18178.  The signed comparisons are also used for floating point values.

  18179. Floating point comparisons are distinguished by the machine modes of the

  18180. operands.

  18181. 

  18182. '(eq:M X Y)'

  18183.      'STORE_FLAG_VALUE' if the values represented by X and Y are equal,

  18184.      otherwise 0.

  18185. 

  18186. '(ne:M X Y)'

  18187.      'STORE_FLAG_VALUE' if the values represented by X and Y are not

  18188.      equal, otherwise 0.

  18189. 

  18190. '(gt:M X Y)'

  18191.      'STORE_FLAG_VALUE' if the X is greater than Y.  If they are

  18192.      fixed-point, the comparison is done in a signed sense.

  18193. 

  18194. '(gtu:M X Y)'

  18195.      Like 'gt' but does unsigned comparison, on fixed-point numbers

  18196.      only.

  18197. 

  18198. '(lt:M X Y)'

  18199. '(ltu:M X Y)'

  18200.      Like 'gt' and 'gtu' but test for "less than".

  18201. 

  18202. '(ge:M X Y)'

  18203. '(geu:M X Y)'

  18204.      Like 'gt' and 'gtu' but test for "greater than or equal".

  18205. 

  18206. '(le:M X Y)'

  18207. '(leu:M X Y)'

  18208.      Like 'gt' and 'gtu' but test for "less than or equal".

  18209. 

  18210. '(if_then_else COND THEN ELSE)'

  18211.      This is not a comparison operation but is listed here because it is

  18212.      always used in conjunction with a comparison operation.  To be

  18213.      precise, COND is a comparison expression.  This expression

  18214.      represents a choice, according to COND, between the value

  18215.      represented by THEN and the one represented by ELSE.

  18216. 

  18217.      On most machines, 'if_then_else' expressions are valid only to

  18218.      express conditional jumps.

  18219. 

  18220. '(cond [TEST1 VALUE1 TEST2 VALUE2 ...] DEFAULT)'

  18221.      Similar to 'if_then_else', but more general.  Each of TEST1, TEST2,

  18222.      ... is performed in turn.  The result of this expression is the

  18223.      VALUE corresponding to the first nonzero test, or DEFAULT if none

  18224.      of the tests are nonzero expressions.

  18225. 

  18226.      This is currently not valid for instruction patterns and is

  18227.      supported only for insn attributes.  *Note Insn Attributes::.

  18228. 

  18229. 

  18230. File: gccint.info,  Node: Bit-Fields,  Next: Vector Operations,  Prev: Comparisons,  Up: RTL

  18231. 

  18232. 14.11 Bit-Fields

  18233. ================

  18234. 

  18235. Special expression codes exist to represent bit-field instructions.

  18236. 

  18237. '(sign_extract:M LOC SIZE POS)'

  18238.      This represents a reference to a sign-extended bit-field contained

  18239.      or starting in LOC (a memory or register reference).  The bit-field

  18240.      is SIZE bits wide and starts at bit POS.  The compilation option

  18241.      'BITS_BIG_ENDIAN' says which end of the memory unit POS counts

  18242.      from.

  18243. 

  18244.      If LOC is in memory, its mode must be a single-byte integer mode.

  18245.      If LOC is in a register, the mode to use is specified by the

  18246.      operand of the 'insv' or 'extv' pattern (*note Standard Names::)

  18247.      and is usually a full-word integer mode, which is the default if

  18248.      none is specified.

  18249. 

  18250.      The mode of POS is machine-specific and is also specified in the

  18251.      'insv' or 'extv' pattern.

  18252. 

  18253.      The mode M is the same as the mode that would be used for LOC if it

  18254.      were a register.

  18255. 

  18256.      A 'sign_extract' cannot appear as an lvalue, or part thereof, in

  18257.      RTL.

  18258. 

  18259. '(zero_extract:M LOC SIZE POS)'

  18260.      Like 'sign_extract' but refers to an unsigned or zero-extended

  18261.      bit-field.  The same sequence of bits are extracted, but they are

  18262.      filled to an entire word with zeros instead of by sign-extension.

  18263. 

  18264.      Unlike 'sign_extract', this type of expressions can be lvalues in

  18265.      RTL; they may appear on the left side of an assignment, indicating

  18266.      insertion of a value into the specified bit-field.

  18267. 

  18268. 

  18269. File: gccint.info,  Node: Vector Operations,  Next: Conversions,  Prev: Bit-Fields,  Up: RTL

  18270. 

  18271. 14.12 Vector Operations

  18272. =======================

  18273. 

  18274. All normal RTL expressions can be used with vector modes; they are

  18275. interpreted as operating on each part of the vector independently.

  18276. Additionally, there are a few new expressions to describe specific

  18277. vector operations.

  18278. 

  18279. '(vec_merge:M VEC1 VEC2 ITEMS)'

  18280.      This describes a merge operation between two vectors.  The result

  18281.      is a vector of mode M; its elements are selected from either VEC1

  18282.      or VEC2.  Which elements are selected is described by ITEMS, which

  18283.      is a bit mask represented by a 'const_int'; a zero bit indicates

  18284.      the corresponding element in the result vector is taken from VEC2

  18285.      while a set bit indicates it is taken from VEC1.

  18286. 

  18287. '(vec_select:M VEC1 SELECTION)'

  18288.      This describes an operation that selects parts of a vector.  VEC1

  18289.      is the source vector, and SELECTION is a 'parallel' that contains a

  18290.      'const_int' (or another expression, if the selection can be made at

  18291.      runtime) for each of the subparts of the result vector, giving the

  18292.      number of the source subpart that should be stored into it.  The

  18293.      result mode M is either the submode for a single element of VEC1

  18294.      (if only one subpart is selected), or another vector mode with that

  18295.      element submode (if multiple subparts are selected).

  18296. 

  18297. '(vec_concat:M X1 X2)'

  18298.      Describes a vector concat operation.  The result is a concatenation

  18299.      of the vectors or scalars X1 and X2; its length is the sum of the

  18300.      lengths of the two inputs.

  18301. 

  18302. '(vec_duplicate:M X)'

  18303.      This operation converts a scalar into a vector or a small vector

  18304.      into a larger one by duplicating the input values.  The output

  18305.      vector mode must have the same submodes as the input vector mode or

  18306.      the scalar modes, and the number of output parts must be an integer

  18307.      multiple of the number of input parts.

  18308. 

  18309. '(vec_series:M BASE STEP)'

  18310.      This operation creates a vector in which element I is equal to

  18311.      'BASE + I*STEP'.  M must be a vector integer mode.

  18312. 

  18313. 

  18314. File: gccint.info,  Node: Conversions,  Next: RTL Declarations,  Prev: Vector Operations,  Up: RTL

  18315. 

  18316. 14.13 Conversions

  18317. =================

  18318. 

  18319. All conversions between machine modes must be represented by explicit

  18320. conversion operations.  For example, an expression which is the sum of a

  18321. byte and a full word cannot be written as '(plus:SI (reg:QI 34) (reg:SI

  18322. 80))' because the 'plus' operation requires two operands of the same

  18323. machine mode.  Therefore, the byte-sized operand is enclosed in a

  18324. conversion operation, as in

  18325. 

  18326.      (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))

  18327. 

  18328.  The conversion operation is not a mere placeholder, because there may

  18329. be more than one way of converting from a given starting mode to the

  18330. desired final mode.  The conversion operation code says how to do it.

  18331. 

  18332.  For all conversion operations, X must not be 'VOIDmode' because the

  18333. mode in which to do the conversion would not be known.  The conversion

  18334. must either be done at compile-time or X must be placed into a register.

  18335. 

  18336. '(sign_extend:M X)'

  18337.      Represents the result of sign-extending the value X to machine mode

  18338.      M.  M must be a fixed-point mode and X a fixed-point value of a

  18339.      mode narrower than M.

  18340. 

  18341. '(zero_extend:M X)'

  18342.      Represents the result of zero-extending the value X to machine mode

  18343.      M.  M must be a fixed-point mode and X a fixed-point value of a

  18344.      mode narrower than M.

  18345. 

  18346. '(float_extend:M X)'

  18347.      Represents the result of extending the value X to machine mode M.

  18348.      M must be a floating point mode and X a floating point value of a

  18349.      mode narrower than M.

  18350. 

  18351. '(truncate:M X)'

  18352.      Represents the result of truncating the value X to machine mode M.

  18353.      M must be a fixed-point mode and X a fixed-point value of a mode

  18354.      wider than M.

  18355. 

  18356. '(ss_truncate:M X)'

  18357.      Represents the result of truncating the value X to machine mode M,

  18358.      using signed saturation in the case of overflow.  Both M and the

  18359.      mode of X must be fixed-point modes.

  18360. 

  18361. '(us_truncate:M X)'

  18362.      Represents the result of truncating the value X to machine mode M,

  18363.      using unsigned saturation in the case of overflow.  Both M and the

  18364.      mode of X must be fixed-point modes.

  18365. 

  18366. '(float_truncate:M X)'

  18367.      Represents the result of truncating the value X to machine mode M.

  18368.      M must be a floating point mode and X a floating point value of a

  18369.      mode wider than M.

  18370. 

  18371. '(float:M X)'

  18372.      Represents the result of converting fixed point value X, regarded

  18373.      as signed, to floating point mode M.

  18374. 

  18375. '(unsigned_float:M X)'

  18376.      Represents the result of converting fixed point value X, regarded

  18377.      as unsigned, to floating point mode M.

  18378. 

  18379. '(fix:M X)'

  18380.      When M is a floating-point mode, represents the result of

  18381.      converting floating point value X (valid for mode M) to an integer,

  18382.      still represented in floating point mode M, by rounding towards

  18383.      zero.

  18384. 

  18385.      When M is a fixed-point mode, represents the result of converting

  18386.      floating point value X to mode M, regarded as signed.  How rounding

  18387.      is done is not specified, so this operation may be used validly in

  18388.      compiling C code only for integer-valued operands.

  18389. 

  18390. '(unsigned_fix:M X)'

  18391.      Represents the result of converting floating point value X to fixed

  18392.      point mode M, regarded as unsigned.  How rounding is done is not

  18393.      specified.

  18394. 

  18395. '(fract_convert:M X)'

  18396.      Represents the result of converting fixed-point value X to

  18397.      fixed-point mode M, signed integer value X to fixed-point mode M,

  18398.      floating-point value X to fixed-point mode M, fixed-point value X

  18399.      to integer mode M regarded as signed, or fixed-point value X to

  18400.      floating-point mode M.  When overflows or underflows happen, the

  18401.      results are undefined.

  18402. 

  18403. '(sat_fract:M X)'

  18404.      Represents the result of converting fixed-point value X to

  18405.      fixed-point mode M, signed integer value X to fixed-point mode M,

  18406.      or floating-point value X to fixed-point mode M.  When overflows or

  18407.      underflows happen, the results are saturated to the maximum or the

  18408.      minimum.

  18409. 

  18410. '(unsigned_fract_convert:M X)'

  18411.      Represents the result of converting fixed-point value X to integer

  18412.      mode M regarded as unsigned, or unsigned integer value X to

  18413.      fixed-point mode M.  When overflows or underflows happen, the

  18414.      results are undefined.

  18415. 

  18416. '(unsigned_sat_fract:M X)'

  18417.      Represents the result of converting unsigned integer value X to

  18418.      fixed-point mode M.  When overflows or underflows happen, the

  18419.      results are saturated to the maximum or the minimum.

  18420. 

  18421. 

  18422. File: gccint.info,  Node: RTL Declarations,  Next: Side Effects,  Prev: Conversions,  Up: RTL

  18423. 

  18424. 14.14 Declarations

  18425. ==================

  18426. 

  18427. Declaration expression codes do not represent arithmetic operations but

  18428. rather state assertions about their operands.

  18429. 

  18430. '(strict_low_part (subreg:M (reg:N R) 0))'

  18431.      This expression code is used in only one context: as the

  18432.      destination operand of a 'set' expression.  In addition, the

  18433.      operand of this expression must be a non-paradoxical 'subreg'

  18434.      expression.

  18435. 

  18436.      The presence of 'strict_low_part' says that the part of the

  18437.      register which is meaningful in mode N, but is not part of mode M,

  18438.      is not to be altered.  Normally, an assignment to such a subreg is

  18439.      allowed to have undefined effects on the rest of the register when

  18440.      M is smaller than 'REGMODE_NATURAL_SIZE (N)'.

  18441. 

  18442. 

  18443. File: gccint.info,  Node: Side Effects,  Next: Incdec,  Prev: RTL Declarations,  Up: RTL

  18444. 

  18445. 14.15 Side Effect Expressions

  18446. =============================

  18447. 

  18448. The expression codes described so far represent values, not actions.

  18449. But machine instructions never produce values; they are meaningful only

  18450. for their side effects on the state of the machine.  Special expression

  18451. codes are used to represent side effects.

  18452. 

  18453.  The body of an instruction is always one of these side effect codes;

  18454. the codes described above, which represent values, appear only as the

  18455. operands of these.

  18456. 

  18457. '(set LVAL X)'

  18458.      Represents the action of storing the value of X into the place

  18459.      represented by LVAL.  LVAL must be an expression representing a

  18460.      place that can be stored in: 'reg' (or 'subreg', 'strict_low_part'

  18461.      or 'zero_extract'), 'mem', 'pc', 'parallel', or 'cc0'.

  18462. 

  18463.      If LVAL is a 'reg', 'subreg' or 'mem', it has a machine mode; then

  18464.      X must be valid for that mode.

  18465. 

  18466.      If LVAL is a 'reg' whose machine mode is less than the full width

  18467.      of the register, then it means that the part of the register

  18468.      specified by the machine mode is given the specified value and the

  18469.      rest of the register receives an undefined value.  Likewise, if

  18470.      LVAL is a 'subreg' whose machine mode is narrower than the mode of

  18471.      the register, the rest of the register can be changed in an

  18472.      undefined way.

  18473. 

  18474.      If LVAL is a 'strict_low_part' of a subreg, then the part of the

  18475.      register specified by the machine mode of the 'subreg' is given the

  18476.      value X and the rest of the register is not changed.

  18477. 

  18478.      If LVAL is a 'zero_extract', then the referenced part of the

  18479.      bit-field (a memory or register reference) specified by the

  18480.      'zero_extract' is given the value X and the rest of the bit-field

  18481.      is not changed.  Note that 'sign_extract' cannot appear in LVAL.

  18482. 

  18483.      If LVAL is '(cc0)', it has no machine mode, and X may be either a

  18484.      'compare' expression or a value that may have any mode.  The latter

  18485.      case represents a "test" instruction.  The expression '(set (cc0)

  18486.      (reg:M N))' is equivalent to '(set (cc0) (compare (reg:M N)

  18487.      (const_int 0)))'.  Use the former expression to save space during

  18488.      the compilation.

  18489. 

  18490.      If LVAL is a 'parallel', it is used to represent the case of a

  18491.      function returning a structure in multiple registers.  Each element

  18492.      of the 'parallel' is an 'expr_list' whose first operand is a 'reg'

  18493.      and whose second operand is a 'const_int' representing the offset

  18494.      (in bytes) into the structure at which the data in that register

  18495.      corresponds.  The first element may be null to indicate that the

  18496.      structure is also passed partly in memory.

  18497. 

  18498.      If LVAL is '(pc)', we have a jump instruction, and the

  18499.      possibilities for X are very limited.  It may be a 'label_ref'

  18500.      expression (unconditional jump).  It may be an 'if_then_else'

  18501.      (conditional jump), in which case either the second or the third

  18502.      operand must be '(pc)' (for the case which does not jump) and the

  18503.      other of the two must be a 'label_ref' (for the case which does

  18504.      jump).  X may also be a 'mem' or '(plus:SI (pc) Y)', where Y may be

  18505.      a 'reg' or a 'mem'; these unusual patterns are used to represent

  18506.      jumps through branch tables.

  18507. 

  18508.      If LVAL is neither '(cc0)' nor '(pc)', the mode of LVAL must not be

  18509.      'VOIDmode' and the mode of X must be valid for the mode of LVAL.

  18510. 

  18511.      LVAL is customarily accessed with the 'SET_DEST' macro and X with

  18512.      the 'SET_SRC' macro.

  18513. 

  18514. '(return)'

  18515.      As the sole expression in a pattern, represents a return from the

  18516.      current function, on machines where this can be done with one

  18517.      instruction, such as VAXen.  On machines where a multi-instruction

  18518.      "epilogue" must be executed in order to return from the function,

  18519.      returning is done by jumping to a label which precedes the

  18520.      epilogue, and the 'return' expression code is never used.

  18521. 

  18522.      Inside an 'if_then_else' expression, represents the value to be

  18523.      placed in 'pc' to return to the caller.

  18524. 

  18525.      Note that an insn pattern of '(return)' is logically equivalent to

  18526.      '(set (pc) (return))', but the latter form is never used.

  18527. 

  18528. '(simple_return)'

  18529.      Like '(return)', but truly represents only a function return, while

  18530.      '(return)' may represent an insn that also performs other functions

  18531.      of the function epilogue.  Like '(return)', this may also occur in

  18532.      conditional jumps.

  18533. 

  18534. '(call FUNCTION NARGS)'

  18535.      Represents a function call.  FUNCTION is a 'mem' expression whose

  18536.      address is the address of the function to be called.  NARGS is an

  18537.      expression which can be used for two purposes: on some machines it

  18538.      represents the number of bytes of stack argument; on others, it

  18539.      represents the number of argument registers.

  18540. 

  18541.      Each machine has a standard machine mode which FUNCTION must have.

  18542.      The machine description defines macro 'FUNCTION_MODE' to expand

  18543.      into the requisite mode name.  The purpose of this mode is to

  18544.      specify what kind of addressing is allowed, on machines where the

  18545.      allowed kinds of addressing depend on the machine mode being

  18546.      addressed.

  18547. 

  18548. '(clobber X)'

  18549.      Represents the storing or possible storing of an unpredictable,

  18550.      undescribed value into X, which must be a 'reg', 'scratch',

  18551.      'parallel' or 'mem' expression.

  18552. 

  18553.      One place this is used is in string instructions that store

  18554.      standard values into particular hard registers.  It may not be

  18555.      worth the trouble to describe the values that are stored, but it is

  18556.      essential to inform the compiler that the registers will be

  18557.      altered, lest it attempt to keep data in them across the string

  18558.      instruction.

  18559. 

  18560.      If X is '(mem:BLK (const_int 0))' or '(mem:BLK (scratch))', it

  18561.      means that all memory locations must be presumed clobbered.  If X

  18562.      is a 'parallel', it has the same meaning as a 'parallel' in a 'set'

  18563.      expression.

  18564. 

  18565.      Note that the machine description classifies certain hard registers

  18566.      as "call-clobbered".  All function call instructions are assumed by

  18567.      default to clobber these registers, so there is no need to use

  18568.      'clobber' expressions to indicate this fact.  Also, each function

  18569.      call is assumed to have the potential to alter any memory location,

  18570.      unless the function is declared 'const'.

  18571. 

  18572.      If the last group of expressions in a 'parallel' are each a

  18573.      'clobber' expression whose arguments are 'reg' or 'match_scratch'

  18574.      (*note RTL Template::) expressions, the combiner phase can add the

  18575.      appropriate 'clobber' expressions to an insn it has constructed

  18576.      when doing so will cause a pattern to be matched.

  18577. 

  18578.      This feature can be used, for example, on a machine that whose

  18579.      multiply and add instructions don't use an MQ register but which

  18580.      has an add-accumulate instruction that does clobber the MQ

  18581.      register.  Similarly, a combined instruction might require a

  18582.      temporary register while the constituent instructions might not.

  18583. 

  18584.      When a 'clobber' expression for a register appears inside a

  18585.      'parallel' with other side effects, the register allocator

  18586.      guarantees that the register is unoccupied both before and after

  18587.      that insn if it is a hard register clobber.  For pseudo-register

  18588.      clobber, the register allocator and the reload pass do not assign

  18589.      the same hard register to the clobber and the input operands if

  18590.      there is an insn alternative containing the '&' constraint (*note

  18591.      Modifiers::) for the clobber and the hard register is in register

  18592.      classes of the clobber in the alternative.  You can clobber either

  18593.      a specific hard register, a pseudo register, or a 'scratch'

  18594.      expression; in the latter two cases, GCC will allocate a hard

  18595.      register that is available there for use as a temporary.

  18596. 

  18597.      For instructions that require a temporary register, you should use

  18598.      'scratch' instead of a pseudo-register because this will allow the

  18599.      combiner phase to add the 'clobber' when required.  You do this by

  18600.      coding ('clobber' ('match_scratch' ...)).  If you do clobber a

  18601.      pseudo register, use one which appears nowhere else--generate a new

  18602.      one each time.  Otherwise, you may confuse CSE.

  18603. 

  18604.      There is one other known use for clobbering a pseudo register in a

  18605.      'parallel': when one of the input operands of the insn is also

  18606.      clobbered by the insn.  In this case, using the same pseudo

  18607.      register in the clobber and elsewhere in the insn produces the

  18608.      expected results.

  18609. 

  18610. '(use X)'

  18611.      Represents the use of the value of X.  It indicates that the value

  18612.      in X at this point in the program is needed, even though it may not

  18613.      be apparent why this is so.  Therefore, the compiler will not

  18614.      attempt to delete previous instructions whose only effect is to

  18615.      store a value in X.  X must be a 'reg' expression.

  18616. 

  18617.      In some situations, it may be tempting to add a 'use' of a register

  18618.      in a 'parallel' to describe a situation where the value of a

  18619.      special register will modify the behavior of the instruction.  A

  18620.      hypothetical example might be a pattern for an addition that can

  18621.      either wrap around or use saturating addition depending on the

  18622.      value of a special control register:

  18623. 

  18624.           (parallel [(set (reg:SI 2) (unspec:SI [(reg:SI 3)

  18625.                                                  (reg:SI 4)] 0))

  18626.                      (use (reg:SI 1))])

  18627. 

  18628. 

  18629.      This will not work, several of the optimizers only look at

  18630.      expressions locally; it is very likely that if you have multiple

  18631.      insns with identical inputs to the 'unspec', they will be optimized

  18632.      away even if register 1 changes in between.

  18633. 

  18634.      This means that 'use' can _only_ be used to describe that the

  18635.      register is live.  You should think twice before adding 'use'

  18636.      statements, more often you will want to use 'unspec' instead.  The

  18637.      'use' RTX is most commonly useful to describe that a fixed register

  18638.      is implicitly used in an insn.  It is also safe to use in patterns

  18639.      where the compiler knows for other reasons that the result of the

  18640.      whole pattern is variable, such as 'cpymemM' or 'call' patterns.

  18641. 

  18642.      During the reload phase, an insn that has a 'use' as pattern can

  18643.      carry a reg_equal note.  These 'use' insns will be deleted before

  18644.      the reload phase exits.

  18645. 

  18646.      During the delayed branch scheduling phase, X may be an insn.  This

  18647.      indicates that X previously was located at this place in the code

  18648.      and its data dependencies need to be taken into account.  These

  18649.      'use' insns will be deleted before the delayed branch scheduling

  18650.      phase exits.

  18651. 

  18652. '(parallel [X0 X1 ...])'

  18653.      Represents several side effects performed in parallel.  The square

  18654.      brackets stand for a vector; the operand of 'parallel' is a vector

  18655.      of expressions.  X0, X1 and so on are individual side effect

  18656.      expressions--expressions of code 'set', 'call', 'return',

  18657.      'simple_return', 'clobber' or 'use'.

  18658. 

  18659.      "In parallel" means that first all the values used in the

  18660.      individual side-effects are computed, and second all the actual

  18661.      side-effects are performed.  For example,

  18662. 

  18663.           (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))

  18664.                      (set (mem:SI (reg:SI 1)) (reg:SI 1))])

  18665. 

  18666.      says unambiguously that the values of hard register 1 and the

  18667.      memory location addressed by it are interchanged.  In both places

  18668.      where '(reg:SI 1)' appears as a memory address it refers to the

  18669.      value in register 1 _before_ the execution of the insn.

  18670. 

  18671.      It follows that it is _incorrect_ to use 'parallel' and expect the

  18672.      result of one 'set' to be available for the next one.  For example,

  18673.      people sometimes attempt to represent a jump-if-zero instruction

  18674.      this way:

  18675. 

  18676.           (parallel [(set (cc0) (reg:SI 34))

  18677.                      (set (pc) (if_then_else

  18678.                                   (eq (cc0) (const_int 0))

  18679.                                   (label_ref ...)

  18680.                                   (pc)))])

  18681. 

  18682.      But this is incorrect, because it says that the jump condition

  18683.      depends on the condition code value _before_ this instruction, not

  18684.      on the new value that is set by this instruction.

  18685. 

  18686.      Peephole optimization, which takes place together with final

  18687.      assembly code output, can produce insns whose patterns consist of a

  18688.      'parallel' whose elements are the operands needed to output the

  18689.      resulting assembler code--often 'reg', 'mem' or constant

  18690.      expressions.  This would not be well-formed RTL at any other stage

  18691.      in compilation, but it is OK then because no further optimization

  18692.      remains to be done.  However, the definition of the macro

  18693.      'NOTICE_UPDATE_CC', if any, must deal with such insns if you define

  18694.      any peephole optimizations.

  18695. 

  18696. '(cond_exec [COND EXPR])'

  18697.      Represents a conditionally executed expression.  The EXPR is

  18698.      executed only if the COND is nonzero.  The COND expression must not

  18699.      have side-effects, but the EXPR may very well have side-effects.

  18700. 

  18701. '(sequence [INSNS ...])'

  18702.      Represents a sequence of insns.  If a 'sequence' appears in the

  18703.      chain of insns, then each of the INSNS that appears in the sequence

  18704.      must be suitable for appearing in the chain of insns, i.e. must

  18705.      satisfy the 'INSN_P' predicate.

  18706. 

  18707.      After delay-slot scheduling is completed, an insn and all the insns

  18708.      that reside in its delay slots are grouped together into a

  18709.      'sequence'.  The insn requiring the delay slot is the first insn in

  18710.      the vector; subsequent insns are to be placed in the delay slot.

  18711. 

  18712.      'INSN_ANNULLED_BRANCH_P' is set on an insn in a delay slot to

  18713.      indicate that a branch insn should be used that will conditionally

  18714.      annul the effect of the insns in the delay slots.  In such a case,

  18715.      'INSN_FROM_TARGET_P' indicates that the insn is from the target of

  18716.      the branch and should be executed only if the branch is taken;

  18717.      otherwise the insn should be executed only if the branch is not

  18718.      taken.  *Note Delay Slots::.

  18719. 

  18720.      Some back ends also use 'sequence' objects for purposes other than

  18721.      delay-slot groups.  This is not supported in the common parts of

  18722.      the compiler, which treat such sequences as delay-slot groups.

  18723. 

  18724.      DWARF2 Call Frame Address (CFA) adjustments are sometimes also

  18725.      expressed using 'sequence' objects as the value of a

  18726.      'RTX_FRAME_RELATED_P' note.  This only happens if the CFA

  18727.      adjustments cannot be easily derived from the pattern of the

  18728.      instruction to which the note is attached.  In such cases, the

  18729.      value of the note is used instead of best-guesing the semantics of

  18730.      the instruction.  The back end can attach notes containing a

  18731.      'sequence' of 'set' patterns that express the effect of the parent

  18732.      instruction.

  18733. 

  18734.  These expression codes appear in place of a side effect, as the body of

  18735. an insn, though strictly speaking they do not always describe side

  18736. effects as such:

  18737. 

  18738. '(asm_input S)'

  18739.      Represents literal assembler code as described by the string S.

  18740. 

  18741. '(unspec [OPERANDS ...] INDEX)'

  18742. '(unspec_volatile [OPERANDS ...] INDEX)'

  18743.      Represents a machine-specific operation on OPERANDS.  INDEX selects

  18744.      between multiple machine-specific operations.  'unspec_volatile' is

  18745.      used for volatile operations and operations that may trap; 'unspec'

  18746.      is used for other operations.

  18747. 

  18748.      These codes may appear inside a 'pattern' of an insn, inside a

  18749.      'parallel', or inside an expression.

  18750. 

  18751. '(addr_vec:M [LR0 LR1 ...])'

  18752.      Represents a table of jump addresses.  The vector elements LR0,

  18753.      etc., are 'label_ref' expressions.  The mode M specifies how much

  18754.      space is given to each address; normally M would be 'Pmode'.

  18755. 

  18756. '(addr_diff_vec:M BASE [LR0 LR1 ...] MIN MAX FLAGS)'

  18757.      Represents a table of jump addresses expressed as offsets from

  18758.      BASE.  The vector elements LR0, etc., are 'label_ref' expressions

  18759.      and so is BASE.  The mode M specifies how much space is given to

  18760.      each address-difference.  MIN and MAX are set up by branch

  18761.      shortening and hold a label with a minimum and a maximum address,

  18762.      respectively.  FLAGS indicates the relative position of BASE, MIN

  18763.      and MAX to the containing insn and of MIN and MAX to BASE.  See

  18764.      rtl.def for details.

  18765. 

  18766. '(prefetch:M ADDR RW LOCALITY)'

  18767.      Represents prefetch of memory at address ADDR.  Operand RW is 1 if

  18768.      the prefetch is for data to be written, 0 otherwise; targets that

  18769.      do not support write prefetches should treat this as a normal

  18770.      prefetch.  Operand LOCALITY specifies the amount of temporal

  18771.      locality; 0 if there is none or 1, 2, or 3 for increasing levels of

  18772.      temporal locality; targets that do not support locality hints

  18773.      should ignore this.

  18774. 

  18775.      This insn is used to minimize cache-miss latency by moving data

  18776.      into a cache before it is accessed.  It should use only

  18777.      non-faulting data prefetch instructions.

  18778. 

  18779. 

  18780. File: gccint.info,  Node: Incdec,  Next: Assembler,  Prev: Side Effects,  Up: RTL

  18781. 

  18782. 14.16 Embedded Side-Effects on Addresses

  18783. ========================================

  18784. 

  18785. Six special side-effect expression codes appear as memory addresses.

  18786. 

  18787. '(pre_dec:M X)'

  18788.      Represents the side effect of decrementing X by a standard amount

  18789.      and represents also the value that X has after being decremented.

  18790.      X must be a 'reg' or 'mem', but most machines allow only a 'reg'.

  18791.      M must be the machine mode for pointers on the machine in use.  The

  18792.      amount X is decremented by is the length in bytes of the machine

  18793.      mode of the containing memory reference of which this expression

  18794.      serves as the address.  Here is an example of its use:

  18795. 

  18796.           (mem:DF (pre_dec:SI (reg:SI 39)))

  18797. 

  18798.      This says to decrement pseudo register 39 by the length of a

  18799.      'DFmode' value and use the result to address a 'DFmode' value.

  18800. 

  18801. '(pre_inc:M X)'

  18802.      Similar, but specifies incrementing X instead of decrementing it.

  18803. 

  18804. '(post_dec:M X)'

  18805.      Represents the same side effect as 'pre_dec' but a different value.

  18806.      The value represented here is the value X has before being

  18807.      decremented.

  18808. 

  18809. '(post_inc:M X)'

  18810.      Similar, but specifies incrementing X instead of decrementing it.

  18811. 

  18812. '(post_modify:M X Y)'

  18813. 

  18814.      Represents the side effect of setting X to Y and represents X

  18815.      before X is modified.  X must be a 'reg' or 'mem', but most

  18816.      machines allow only a 'reg'.  M must be the machine mode for

  18817.      pointers on the machine in use.

  18818. 

  18819.      The expression Y must be one of three forms: '(plus:M X Z)',

  18820.      '(minus:M X Z)', or '(plus:M X I)', where Z is an index register

  18821.      and I is a constant.

  18822. 

  18823.      Here is an example of its use:

  18824. 

  18825.           (mem:SF (post_modify:SI (reg:SI 42) (plus (reg:SI 42)

  18826.                                                     (reg:SI 48))))

  18827. 

  18828.      This says to modify pseudo register 42 by adding the contents of

  18829.      pseudo register 48 to it, after the use of what ever 42 points to.

  18830. 

  18831. '(pre_modify:M X EXPR)'

  18832.      Similar except side effects happen before the use.

  18833. 

  18834.  These embedded side effect expressions must be used with care.

  18835. Instruction patterns may not use them.  Until the 'flow' pass of the

  18836. compiler, they may occur only to represent pushes onto the stack.  The

  18837. 'flow' pass finds cases where registers are incremented or decremented

  18838. in one instruction and used as an address shortly before or after; these

  18839. cases are then transformed to use pre- or post-increment or -decrement.

  18840. 

  18841.  If a register used as the operand of these expressions is used in

  18842. another address in an insn, the original value of the register is used.

  18843. Uses of the register outside of an address are not permitted within the

  18844. same insn as a use in an embedded side effect expression because such

  18845. insns behave differently on different machines and hence must be treated

  18846. as ambiguous and disallowed.

  18847. 

  18848.  An instruction that can be represented with an embedded side effect

  18849. could also be represented using 'parallel' containing an additional

  18850. 'set' to describe how the address register is altered.  This is not done

  18851. because machines that allow these operations at all typically allow them

  18852. wherever a memory address is called for.  Describing them as additional

  18853. parallel stores would require doubling the number of entries in the

  18854. machine description.

  18855. 

  18856. 

  18857. File: gccint.info,  Node: Assembler,  Next: Debug Information,  Prev: Incdec,  Up: RTL

  18858. 

  18859. 14.17 Assembler Instructions as Expressions

  18860. ===========================================

  18861. 

  18862. The RTX code 'asm_operands' represents a value produced by a

  18863. user-specified assembler instruction.  It is used to represent an 'asm'

  18864. statement with arguments.  An 'asm' statement with a single output

  18865. operand, like this:

  18866. 

  18867.      asm ("foo %1,%2,%0" : "=a" (outputvar) : "g" (x + y), "di" (*z));

  18868. 

  18869. is represented using a single 'asm_operands' RTX which represents the

  18870. value that is stored in 'outputvar':

  18871. 

  18872.      (set RTX-FOR-OUTPUTVAR

  18873.           (asm_operands "foo %1,%2,%0" "a" 0

  18874.                         [RTX-FOR-ADDITION-RESULT RTX-FOR-*Z]

  18875.                         [(asm_input:M1 "g")

  18876.                          (asm_input:M2 "di")]))

  18877. 

  18878. Here the operands of the 'asm_operands' RTX are the assembler template

  18879. string, the output-operand's constraint, the index-number of the output

  18880. operand among the output operands specified, a vector of input operand

  18881. RTX's, and a vector of input-operand modes and constraints.  The mode M1

  18882. is the mode of the sum 'x+y'; M2 is that of '*z'.

  18883. 

  18884.  When an 'asm' statement has multiple output values, its insn has

  18885. several such 'set' RTX's inside of a 'parallel'.  Each 'set' contains an

  18886. 'asm_operands'; all of these share the same assembler template and

  18887. vectors, but each contains the constraint for the respective output

  18888. operand.  They are also distinguished by the output-operand index

  18889. number, which is 0, 1, ... for successive output operands.

  18890. 

  18891. 

  18892. File: gccint.info,  Node: Debug Information,  Next: Insns,  Prev: Assembler,  Up: RTL

  18893. 

  18894. 14.18 Variable Location Debug Information in RTL

  18895. ================================================

  18896. 

  18897. Variable tracking relies on 'MEM_EXPR' and 'REG_EXPR' annotations to

  18898. determine what user variables memory and register references refer to.

  18899. 

  18900.  Variable tracking at assignments uses these notes only when they refer

  18901. to variables that live at fixed locations (e.g., addressable variables,

  18902. global non-automatic variables).  For variables whose location may vary,

  18903. it relies on the following types of notes.

  18904. 

  18905. '(var_location:MODE VAR EXP STAT)'

  18906.      Binds variable 'var', a tree, to value EXP, an RTL expression.  It

  18907.      appears only in 'NOTE_INSN_VAR_LOCATION' and 'DEBUG_INSN's, with

  18908.      slightly different meanings.  MODE, if present, represents the mode

  18909.      of EXP, which is useful if it is a modeless expression.  STAT is

  18910.      only meaningful in notes, indicating whether the variable is known

  18911.      to be initialized or uninitialized.

  18912. 

  18913. '(debug_expr:MODE DECL)'

  18914.      Stands for the value bound to the 'DEBUG_EXPR_DECL' DECL, that

  18915.      points back to it, within value expressions in 'VAR_LOCATION'

  18916.      nodes.

  18917. 

  18918. '(debug_implicit_ptr:MODE DECL)'

  18919.      Stands for the location of a DECL that is no longer addressable.

  18920. 

  18921. '(entry_value:MODE DECL)'

  18922.      Stands for the value a DECL had at the entry point of the

  18923.      containing function.

  18924. 

  18925. '(debug_parameter_ref:MODE DECL)'

  18926.      Refers to a parameter that was completely optimized out.

  18927. 

  18928. '(debug_marker:MODE)'

  18929.      Marks a program location.  With 'VOIDmode', it stands for the

  18930.      beginning of a statement, a recommended inspection point logically

  18931.      after all prior side effects, and before any subsequent side

  18932.      effects.  With 'BLKmode', it indicates an inline entry point: the

  18933.      lexical block encoded in the 'INSN_LOCATION' is the enclosing block

  18934.      that encloses the inlined function.

  18935. 

  18936. 

  18937. File: gccint.info,  Node: Insns,  Next: Calls,  Prev: Debug Information,  Up: RTL

  18938. 

  18939. 14.19 Insns

  18940. ===========

  18941. 

  18942. The RTL representation of the code for a function is a doubly-linked

  18943. chain of objects called "insns".  Insns are expressions with special

  18944. codes that are used for no other purpose.  Some insns are actual

  18945. instructions; others represent dispatch tables for 'switch' statements;

  18946. others represent labels to jump to or various sorts of declarative

  18947. information.

  18948. 

  18949.  In addition to its own specific data, each insn must have a unique

  18950. id-number that distinguishes it from all other insns in the current

  18951. function (after delayed branch scheduling, copies of an insn with the

  18952. same id-number may be present in multiple places in a function, but

  18953. these copies will always be identical and will only appear inside a

  18954. 'sequence'), and chain pointers to the preceding and following insns.

  18955. These three fields occupy the same position in every insn, independent

  18956. of the expression code of the insn.  They could be accessed with 'XEXP'

  18957. and 'XINT', but instead three special macros are always used:

  18958. 

  18959. 'INSN_UID (I)'

  18960.      Accesses the unique id of insn I.

  18961. 

  18962. 'PREV_INSN (I)'

  18963.      Accesses the chain pointer to the insn preceding I.  If I is the

  18964.      first insn, this is a null pointer.

  18965. 

  18966. 'NEXT_INSN (I)'

  18967.      Accesses the chain pointer to the insn following I.  If I is the

  18968.      last insn, this is a null pointer.

  18969. 

  18970.  The first insn in the chain is obtained by calling 'get_insns'; the

  18971. last insn is the result of calling 'get_last_insn'.  Within the chain

  18972. delimited by these insns, the 'NEXT_INSN' and 'PREV_INSN' pointers must

  18973. always correspond: if INSN is not the first insn,

  18974. 

  18975.      NEXT_INSN (PREV_INSN (INSN)) == INSN

  18976. 

  18977. is always true and if INSN is not the last insn,

  18978. 

  18979.      PREV_INSN (NEXT_INSN (INSN)) == INSN

  18980. 

  18981. is always true.

  18982. 

  18983.  After delay slot scheduling, some of the insns in the chain might be

  18984. 'sequence' expressions, which contain a vector of insns.  The value of

  18985. 'NEXT_INSN' in all but the last of these insns is the next insn in the

  18986. vector; the value of 'NEXT_INSN' of the last insn in the vector is the

  18987. same as the value of 'NEXT_INSN' for the 'sequence' in which it is

  18988. contained.  Similar rules apply for 'PREV_INSN'.

  18989. 

  18990.  This means that the above invariants are not necessarily true for insns

  18991. inside 'sequence' expressions.  Specifically, if INSN is the first insn

  18992. in a 'sequence', 'NEXT_INSN (PREV_INSN (INSN))' is the insn containing

  18993. the 'sequence' expression, as is the value of 'PREV_INSN (NEXT_INSN

  18994. (INSN))' if INSN is the last insn in the 'sequence' expression.  You can

  18995. use these expressions to find the containing 'sequence' expression.

  18996. 

  18997.  Every insn has one of the following expression codes:

  18998. 

  18999. 'insn'

  19000.      The expression code 'insn' is used for instructions that do not

  19001.      jump and do not do function calls.  'sequence' expressions are

  19002.      always contained in insns with code 'insn' even if one of those

  19003.      insns should jump or do function calls.

  19004. 

  19005.      Insns with code 'insn' have four additional fields beyond the three

  19006.      mandatory ones listed above.  These four are described in a table

  19007.      below.

  19008. 

  19009. 'jump_insn'

  19010.      The expression code 'jump_insn' is used for instructions that may

  19011.      jump (or, more generally, may contain 'label_ref' expressions to

  19012.      which 'pc' can be set in that instruction).  If there is an

  19013.      instruction to return from the current function, it is recorded as

  19014.      a 'jump_insn'.

  19015. 

  19016.      'jump_insn' insns have the same extra fields as 'insn' insns,

  19017.      accessed in the same way and in addition contain a field

  19018.      'JUMP_LABEL' which is defined once jump optimization has completed.

  19019. 

  19020.      For simple conditional and unconditional jumps, this field contains

  19021.      the 'code_label' to which this insn will (possibly conditionally)

  19022.      branch.  In a more complex jump, 'JUMP_LABEL' records one of the

  19023.      labels that the insn refers to; other jump target labels are

  19024.      recorded as 'REG_LABEL_TARGET' notes.  The exception is 'addr_vec'

  19025.      and 'addr_diff_vec', where 'JUMP_LABEL' is 'NULL_RTX' and the only

  19026.      way to find the labels is to scan the entire body of the insn.

  19027. 

  19028.      Return insns count as jumps, but their 'JUMP_LABEL' is 'RETURN' or

  19029.      'SIMPLE_RETURN'.

  19030. 

  19031. 'call_insn'

  19032.      The expression code 'call_insn' is used for instructions that may

  19033.      do function calls.  It is important to distinguish these

  19034.      instructions because they imply that certain registers and memory

  19035.      locations may be altered unpredictably.

  19036. 

  19037.      'call_insn' insns have the same extra fields as 'insn' insns,

  19038.      accessed in the same way and in addition contain a field

  19039.      'CALL_INSN_FUNCTION_USAGE', which contains a list (chain of

  19040.      'expr_list' expressions) containing 'use', 'clobber' and sometimes

  19041.      'set' expressions that denote hard registers and 'mem's used or

  19042.      clobbered by the called function.

  19043. 

  19044.      A 'mem' generally points to a stack slot in which arguments passed

  19045.      to the libcall by reference (*note TARGET_PASS_BY_REFERENCE:

  19046.      Register Arguments.) are stored.  If the argument is caller-copied

  19047.      (*note TARGET_CALLEE_COPIES: Register Arguments.), the stack slot

  19048.      will be mentioned in 'clobber' and 'use' entries; if it's

  19049.      callee-copied, only a 'use' will appear, and the 'mem' may point to

  19050.      addresses that are not stack slots.

  19051. 

  19052.      Registers occurring inside a 'clobber' in this list augment

  19053.      registers specified in 'CALL_USED_REGISTERS' (*note Register

  19054.      Basics::).

  19055. 

  19056.      If the list contains a 'set' involving two registers, it indicates

  19057.      that the function returns one of its arguments.  Such a 'set' may

  19058.      look like a no-op if the same register holds the argument and the

  19059.      return value.

  19060. 

  19061. 'code_label'

  19062.      A 'code_label' insn represents a label that a jump insn can jump

  19063.      to.  It contains two special fields of data in addition to the

  19064.      three standard ones.  'CODE_LABEL_NUMBER' is used to hold the

  19065.      "label number", a number that identifies this label uniquely among

  19066.      all the labels in the compilation (not just in the current

  19067.      function).  Ultimately, the label is represented in the assembler

  19068.      output as an assembler label, usually of the form 'LN' where N is

  19069.      the label number.

  19070. 

  19071.      When a 'code_label' appears in an RTL expression, it normally

  19072.      appears within a 'label_ref' which represents the address of the

  19073.      label, as a number.

  19074. 

  19075.      Besides as a 'code_label', a label can also be represented as a

  19076.      'note' of type 'NOTE_INSN_DELETED_LABEL'.

  19077. 

  19078.      The field 'LABEL_NUSES' is only defined once the jump optimization

  19079.      phase is completed.  It contains the number of times this label is

  19080.      referenced in the current function.

  19081. 

  19082.      The field 'LABEL_KIND' differentiates four different types of

  19083.      labels: 'LABEL_NORMAL', 'LABEL_STATIC_ENTRY', 'LABEL_GLOBAL_ENTRY',

  19084.      and 'LABEL_WEAK_ENTRY'.  The only labels that do not have type

  19085.      'LABEL_NORMAL' are "alternate entry points" to the current

  19086.      function.  These may be static (visible only in the containing

  19087.      translation unit), global (exposed to all translation units), or

  19088.      weak (global, but can be overridden by another symbol with the same

  19089.      name).

  19090. 

  19091.      Much of the compiler treats all four kinds of label identically.

  19092.      Some of it needs to know whether or not a label is an alternate

  19093.      entry point; for this purpose, the macro 'LABEL_ALT_ENTRY_P' is

  19094.      provided.  It is equivalent to testing whether 'LABEL_KIND (label)

  19095.      == LABEL_NORMAL'.  The only place that cares about the distinction

  19096.      between static, global, and weak alternate entry points, besides

  19097.      the front-end code that creates them, is the function

  19098.      'output_alternate_entry_point', in 'final.c'.

  19099. 

  19100.      To set the kind of a label, use the 'SET_LABEL_KIND' macro.

  19101. 

  19102. 'jump_table_data'

  19103.      A 'jump_table_data' insn is a placeholder for the jump-table data

  19104.      of a 'casesi' or 'tablejump' insn.  They are placed after a

  19105.      'tablejump_p' insn.  A 'jump_table_data' insn is not part o a basic

  19106.      blockm but it is associated with the basic block that ends with the

  19107.      'tablejump_p' insn.  The 'PATTERN' of a 'jump_table_data' is always

  19108.      either an 'addr_vec' or an 'addr_diff_vec', and a 'jump_table_data'

  19109.      insn is always preceded by a 'code_label'.  The 'tablejump_p' insn

  19110.      refers to that 'code_label' via its 'JUMP_LABEL'.

  19111. 

  19112. 'barrier'

  19113.      Barriers are placed in the instruction stream when control cannot

  19114.      flow past them.  They are placed after unconditional jump

  19115.      instructions to indicate that the jumps are unconditional and after

  19116.      calls to 'volatile' functions, which do not return (e.g., 'exit').

  19117.      They contain no information beyond the three standard fields.

  19118. 

  19119. 'note'

  19120.      'note' insns are used to represent additional debugging and

  19121.      declarative information.  They contain two nonstandard fields, an

  19122.      integer which is accessed with the macro 'NOTE_LINE_NUMBER' and a

  19123.      string accessed with 'NOTE_SOURCE_FILE'.

  19124. 

  19125.      If 'NOTE_LINE_NUMBER' is positive, the note represents the position

  19126.      of a source line and 'NOTE_SOURCE_FILE' is the source file name

  19127.      that the line came from.  These notes control generation of line

  19128.      number data in the assembler output.

  19129. 

  19130.      Otherwise, 'NOTE_LINE_NUMBER' is not really a line number but a

  19131.      code with one of the following values (and 'NOTE_SOURCE_FILE' must

  19132.      contain a null pointer):

  19133. 

  19134.      'NOTE_INSN_DELETED'

  19135.           Such a note is completely ignorable.  Some passes of the

  19136.           compiler delete insns by altering them into notes of this

  19137.           kind.

  19138. 

  19139.      'NOTE_INSN_DELETED_LABEL'

  19140.           This marks what used to be a 'code_label', but was not used

  19141.           for other purposes than taking its address and was transformed

  19142.           to mark that no code jumps to it.

  19143. 

  19144.      'NOTE_INSN_BLOCK_BEG'

  19145.      'NOTE_INSN_BLOCK_END'

  19146.           These types of notes indicate the position of the beginning

  19147.           and end of a level of scoping of variable names.  They control

  19148.           the output of debugging information.

  19149. 

  19150.      'NOTE_INSN_EH_REGION_BEG'

  19151.      'NOTE_INSN_EH_REGION_END'

  19152.           These types of notes indicate the position of the beginning

  19153.           and end of a level of scoping for exception handling.

  19154.           'NOTE_EH_HANDLER' identifies which region is associated with

  19155.           these notes.

  19156. 

  19157.      'NOTE_INSN_FUNCTION_BEG'

  19158.           Appears at the start of the function body, after the function

  19159.           prologue.

  19160. 

  19161.      'NOTE_INSN_VAR_LOCATION'

  19162.           This note is used to generate variable location debugging

  19163.           information.  It indicates that the user variable in its

  19164.           'VAR_LOCATION' operand is at the location given in the RTL

  19165.           expression, or holds a value that can be computed by

  19166.           evaluating the RTL expression from that static point in the

  19167.           program up to the next such note for the same user variable.

  19168. 

  19169.      'NOTE_INSN_BEGIN_STMT'

  19170.           This note is used to generate 'is_stmt' markers in line number

  19171.           debuggign information.  It indicates the beginning of a user

  19172.           statement.

  19173. 

  19174.      'NOTE_INSN_INLINE_ENTRY'

  19175.           This note is used to generate 'entry_pc' for inlined

  19176.           subroutines in debugging information.  It indicates an

  19177.           inspection point at which all arguments for the inlined

  19178.           function have been bound, and before its first statement.

  19179. 

  19180.      These codes are printed symbolically when they appear in debugging

  19181.      dumps.

  19182. 

  19183. 'debug_insn'

  19184.      The expression code 'debug_insn' is used for pseudo-instructions

  19185.      that hold debugging information for variable tracking at

  19186.      assignments (see '-fvar-tracking-assignments' option).  They are

  19187.      the RTL representation of 'GIMPLE_DEBUG' statements (*note

  19188.      GIMPLE_DEBUG::), with a 'VAR_LOCATION' operand that binds a user

  19189.      variable tree to an RTL representation of the 'value' in the

  19190.      corresponding statement.  A 'DEBUG_EXPR' in it stands for the value

  19191.      bound to the corresponding 'DEBUG_EXPR_DECL'.

  19192. 

  19193.      'GIMPLE_DEBUG_BEGIN_STMT' and 'GIMPLE_DEBUG_INLINE_ENTRY' are

  19194.      expanded to RTL as a 'DEBUG_INSN' with a 'DEBUG_MARKER' 'PATTERN';

  19195.      the difference is the RTL mode: the former's 'DEBUG_MARKER' is

  19196.      'VOIDmode', whereas the latter is 'BLKmode'; information about the

  19197.      inlined function can be taken from the lexical block encoded in the

  19198.      'INSN_LOCATION'.  These 'DEBUG_INSN's, that do not carry

  19199.      'VAR_LOCATION' information, just 'DEBUG_MARKER's, can be detected

  19200.      by testing 'DEBUG_MARKER_INSN_P', whereas those that do can be

  19201.      recognized as 'DEBUG_BIND_INSN_P'.

  19202. 

  19203.      Throughout optimization passes, 'DEBUG_INSN's are not reordered

  19204.      with respect to each other, particularly during scheduling.

  19205.      Binding information is kept in pseudo-instruction form, so that,

  19206.      unlike notes, it gets the same treatment and adjustments that

  19207.      regular instructions would.  It is the variable tracking pass that

  19208.      turns these pseudo-instructions into 'NOTE_INSN_VAR_LOCATION',

  19209.      'NOTE_INSN_BEGIN_STMT' and 'NOTE_INSN_INLINE_ENTRY' notes,

  19210.      analyzing control flow, value equivalences and changes to registers

  19211.      and memory referenced in value expressions, propagating the values

  19212.      of debug temporaries and determining expressions that can be used

  19213.      to compute the value of each user variable at as many points

  19214.      (ranges, actually) in the program as possible.

  19215. 

  19216.      Unlike 'NOTE_INSN_VAR_LOCATION', the value expression in an

  19217.      'INSN_VAR_LOCATION' denotes a value at that specific point in the

  19218.      program, rather than an expression that can be evaluated at any

  19219.      later point before an overriding 'VAR_LOCATION' is encountered.

  19220.      E.g., if a user variable is bound to a 'REG' and then a subsequent

  19221.      insn modifies the 'REG', the note location would keep mapping the

  19222.      user variable to the register across the insn, whereas the insn

  19223.      location would keep the variable bound to the value, so that the

  19224.      variable tracking pass would emit another location note for the

  19225.      variable at the point in which the register is modified.

  19226. 

  19227.  The machine mode of an insn is normally 'VOIDmode', but some phases use

  19228. the mode for various purposes.

  19229. 

  19230.  The common subexpression elimination pass sets the mode of an insn to

  19231. 'QImode' when it is the first insn in a block that has already been

  19232. processed.

  19233. 

  19234.  The second Haifa scheduling pass, for targets that can multiple issue,

  19235. sets the mode of an insn to 'TImode' when it is believed that the

  19236. instruction begins an issue group.  That is, when the instruction cannot

  19237. issue simultaneously with the previous.  This may be relied on by later

  19238. passes, in particular machine-dependent reorg.

  19239. 

  19240.  Here is a table of the extra fields of 'insn', 'jump_insn' and

  19241. 'call_insn' insns:

  19242. 

  19243. 'PATTERN (I)'

  19244.      An expression for the side effect performed by this insn.  This

  19245.      must be one of the following codes: 'set', 'call', 'use',

  19246.      'clobber', 'return', 'simple_return', 'asm_input', 'asm_output',

  19247.      'addr_vec', 'addr_diff_vec', 'trap_if', 'unspec',

  19248.      'unspec_volatile', 'parallel', 'cond_exec', or 'sequence'.  If it

  19249.      is a 'parallel', each element of the 'parallel' must be one these

  19250.      codes, except that 'parallel' expressions cannot be nested and

  19251.      'addr_vec' and 'addr_diff_vec' are not permitted inside a

  19252.      'parallel' expression.

  19253. 

  19254. 'INSN_CODE (I)'

  19255.      An integer that says which pattern in the machine description

  19256.      matches this insn, or -1 if the matching has not yet been

  19257.      attempted.

  19258. 

  19259.      Such matching is never attempted and this field remains -1 on an

  19260.      insn whose pattern consists of a single 'use', 'clobber',

  19261.      'asm_input', 'addr_vec' or 'addr_diff_vec' expression.

  19262. 

  19263.      Matching is also never attempted on insns that result from an 'asm'

  19264.      statement.  These contain at least one 'asm_operands' expression.

  19265.      The function 'asm_noperands' returns a non-negative value for such

  19266.      insns.

  19267. 

  19268.      In the debugging output, this field is printed as a number followed

  19269.      by a symbolic representation that locates the pattern in the 'md'

  19270.      file as some small positive or negative offset from a named

  19271.      pattern.

  19272. 

  19273. 'LOG_LINKS (I)'

  19274.      A list (chain of 'insn_list' expressions) giving information about

  19275.      dependencies between instructions within a basic block.  Neither a

  19276.      jump nor a label may come between the related insns.  These are

  19277.      only used by the schedulers and by combine.  This is a deprecated

  19278.      data structure.  Def-use and use-def chains are now preferred.

  19279. 

  19280. 'REG_NOTES (I)'

  19281.      A list (chain of 'expr_list', 'insn_list' and 'int_list'

  19282.      expressions) giving miscellaneous information about the insn.  It

  19283.      is often information pertaining to the registers used in this insn.

  19284. 

  19285.  The 'LOG_LINKS' field of an insn is a chain of 'insn_list' expressions.

  19286. Each of these has two operands: the first is an insn, and the second is

  19287. another 'insn_list' expression (the next one in the chain).  The last

  19288. 'insn_list' in the chain has a null pointer as second operand.  The

  19289. significant thing about the chain is which insns appear in it (as first

  19290. operands of 'insn_list' expressions).  Their order is not significant.

  19291. 

  19292.  This list is originally set up by the flow analysis pass; it is a null

  19293. pointer until then.  Flow only adds links for those data dependencies

  19294. which can be used for instruction combination.  For each insn, the flow

  19295. analysis pass adds a link to insns which store into registers values

  19296. that are used for the first time in this insn.

  19297. 

  19298.  The 'REG_NOTES' field of an insn is a chain similar to the 'LOG_LINKS'

  19299. field but it includes 'expr_list' and 'int_list' expressions in addition

  19300. to 'insn_list' expressions.  There are several kinds of register notes,

  19301. which are distinguished by the machine mode, which in a register note is

  19302. really understood as being an 'enum reg_note'.  The first operand OP of

  19303. the note is data whose meaning depends on the kind of note.

  19304. 

  19305.  The macro 'REG_NOTE_KIND (X)' returns the kind of register note.  Its

  19306. counterpart, the macro 'PUT_REG_NOTE_KIND (X, NEWKIND)' sets the

  19307. register note type of X to be NEWKIND.

  19308. 

  19309.  Register notes are of three classes: They may say something about an

  19310. input to an insn, they may say something about an output of an insn, or

  19311. they may create a linkage between two insns.  There are also a set of

  19312. values that are only used in 'LOG_LINKS'.

  19313. 

  19314.  These register notes annotate inputs to an insn:

  19315. 

  19316. 'REG_DEAD'

  19317.      The value in OP dies in this insn; that is to say, altering the

  19318.      value immediately after this insn would not affect the future

  19319.      behavior of the program.

  19320. 

  19321.      It does not follow that the register OP has no useful value after

  19322.      this insn since OP is not necessarily modified by this insn.

  19323.      Rather, no subsequent instruction uses the contents of OP.

  19324. 

  19325. 'REG_UNUSED'

  19326.      The register OP being set by this insn will not be used in a

  19327.      subsequent insn.  This differs from a 'REG_DEAD' note, which

  19328.      indicates that the value in an input will not be used subsequently.

  19329.      These two notes are independent; both may be present for the same

  19330.      register.

  19331. 

  19332. 'REG_INC'

  19333.      The register OP is incremented (or decremented; at this level there

  19334.      is no distinction) by an embedded side effect inside this insn.

  19335.      This means it appears in a 'post_inc', 'pre_inc', 'post_dec' or

  19336.      'pre_dec' expression.

  19337. 

  19338. 'REG_NONNEG'

  19339.      The register OP is known to have a nonnegative value when this insn

  19340.      is reached.  This is used by special looping instructions that

  19341.      terminate when the register goes negative.

  19342. 

  19343.      The 'REG_NONNEG' note is added only to 'doloop_end' insns, if its

  19344.      pattern uses a 'ge' condition.

  19345. 

  19346. 'REG_LABEL_OPERAND'

  19347.      This insn uses OP, a 'code_label' or a 'note' of type

  19348.      'NOTE_INSN_DELETED_LABEL', but is not a 'jump_insn', or it is a

  19349.      'jump_insn' that refers to the operand as an ordinary operand.  The

  19350.      label may still eventually be a jump target, but if so in an

  19351.      indirect jump in a subsequent insn.  The presence of this note

  19352.      allows jump optimization to be aware that OP is, in fact, being

  19353.      used, and flow optimization to build an accurate flow graph.

  19354. 

  19355. 'REG_LABEL_TARGET'

  19356.      This insn is a 'jump_insn' but not an 'addr_vec' or

  19357.      'addr_diff_vec'.  It uses OP, a 'code_label' as a direct or

  19358.      indirect jump target.  Its purpose is similar to that of

  19359.      'REG_LABEL_OPERAND'.  This note is only present if the insn has

  19360.      multiple targets; the last label in the insn (in the highest

  19361.      numbered insn-field) goes into the 'JUMP_LABEL' field and does not

  19362.      have a 'REG_LABEL_TARGET' note.  *Note JUMP_LABEL: Insns.

  19363. 

  19364. 'REG_SETJMP'

  19365.      Appears attached to each 'CALL_INSN' to 'setjmp' or a related

  19366.      function.

  19367. 

  19368.  The following notes describe attributes of outputs of an insn:

  19369. 

  19370. 'REG_EQUIV'

  19371. 'REG_EQUAL'

  19372.      This note is only valid on an insn that sets only one register and

  19373.      indicates that that register will be equal to OP at run time; the

  19374.      scope of this equivalence differs between the two types of notes.

  19375.      The value which the insn explicitly copies into the register may

  19376.      look different from OP, but they will be equal at run time.  If the

  19377.      output of the single 'set' is a 'strict_low_part' or 'zero_extract'

  19378.      expression, the note refers to the register that is contained in

  19379.      its first operand.

  19380. 

  19381.      For 'REG_EQUIV', the register is equivalent to OP throughout the

  19382.      entire function, and could validly be replaced in all its

  19383.      occurrences by OP.  ("Validly" here refers to the data flow of the

  19384.      program; simple replacement may make some insns invalid.)  For

  19385.      example, when a constant is loaded into a register that is never

  19386.      assigned any other value, this kind of note is used.

  19387. 

  19388.      When a parameter is copied into a pseudo-register at entry to a

  19389.      function, a note of this kind records that the register is

  19390.      equivalent to the stack slot where the parameter was passed.

  19391.      Although in this case the register may be set by other insns, it is

  19392.      still valid to replace the register by the stack slot throughout

  19393.      the function.

  19394. 

  19395.      A 'REG_EQUIV' note is also used on an instruction which copies a

  19396.      register parameter into a pseudo-register at entry to a function,

  19397.      if there is a stack slot where that parameter could be stored.

  19398.      Although other insns may set the pseudo-register, it is valid for

  19399.      the compiler to replace the pseudo-register by stack slot

  19400.      throughout the function, provided the compiler ensures that the

  19401.      stack slot is properly initialized by making the replacement in the

  19402.      initial copy instruction as well.  This is used on machines for

  19403.      which the calling convention allocates stack space for register

  19404.      parameters.  See 'REG_PARM_STACK_SPACE' in *note Stack Arguments::.

  19405. 

  19406.      In the case of 'REG_EQUAL', the register that is set by this insn

  19407.      will be equal to OP at run time at the end of this insn but not

  19408.      necessarily elsewhere in the function.  In this case, OP is

  19409.      typically an arithmetic expression.  For example, when a sequence

  19410.      of insns such as a library call is used to perform an arithmetic

  19411.      operation, this kind of note is attached to the insn that produces

  19412.      or copies the final value.

  19413. 

  19414.      These two notes are used in different ways by the compiler passes.

  19415.      'REG_EQUAL' is used by passes prior to register allocation (such as

  19416.      common subexpression elimination and loop optimization) to tell

  19417.      them how to think of that value.  'REG_EQUIV' notes are used by

  19418.      register allocation to indicate that there is an available

  19419.      substitute expression (either a constant or a 'mem' expression for

  19420.      the location of a parameter on the stack) that may be used in place

  19421.      of a register if insufficient registers are available.

  19422. 

  19423.      Except for stack homes for parameters, which are indicated by a

  19424.      'REG_EQUIV' note and are not useful to the early optimization

  19425.      passes and pseudo registers that are equivalent to a memory

  19426.      location throughout their entire life, which is not detected until

  19427.      later in the compilation, all equivalences are initially indicated

  19428.      by an attached 'REG_EQUAL' note.  In the early stages of register

  19429.      allocation, a 'REG_EQUAL' note is changed into a 'REG_EQUIV' note

  19430.      if OP is a constant and the insn represents the only set of its

  19431.      destination register.

  19432. 

  19433.      Thus, compiler passes prior to register allocation need only check

  19434.      for 'REG_EQUAL' notes and passes subsequent to register allocation

  19435.      need only check for 'REG_EQUIV' notes.

  19436. 

  19437.  These notes describe linkages between insns.  They occur in pairs: one

  19438. insn has one of a pair of notes that points to a second insn, which has

  19439. the inverse note pointing back to the first insn.

  19440. 

  19441. 'REG_CC_SETTER'

  19442. 'REG_CC_USER'

  19443.      On machines that use 'cc0', the insns which set and use 'cc0' set

  19444.      and use 'cc0' are adjacent.  However, when branch delay slot

  19445.      filling is done, this may no longer be true.  In this case a

  19446.      'REG_CC_USER' note will be placed on the insn setting 'cc0' to

  19447.      point to the insn using 'cc0' and a 'REG_CC_SETTER' note will be

  19448.      placed on the insn using 'cc0' to point to the insn setting 'cc0'.

  19449. 

  19450.  These values are only used in the 'LOG_LINKS' field, and indicate the

  19451. type of dependency that each link represents.  Links which indicate a

  19452. data dependence (a read after write dependence) do not use any code,

  19453. they simply have mode 'VOIDmode', and are printed without any

  19454. descriptive text.

  19455. 

  19456. 'REG_DEP_TRUE'

  19457.      This indicates a true dependence (a read after write dependence).

  19458. 

  19459. 'REG_DEP_OUTPUT'

  19460.      This indicates an output dependence (a write after write

  19461.      dependence).

  19462. 

  19463. 'REG_DEP_ANTI'

  19464.      This indicates an anti dependence (a write after read dependence).

  19465. 

  19466.  These notes describe information gathered from gcov profile data.  They

  19467. are stored in the 'REG_NOTES' field of an insn.

  19468. 

  19469. 'REG_BR_PROB'

  19470.      This is used to specify the ratio of branches to non-branches of a

  19471.      branch insn according to the profile data.  The note is represented

  19472.      as an 'int_list' expression whose integer value is an encoding of

  19473.      'profile_probability' type.  'profile_probability' provide member

  19474.      function 'from_reg_br_prob_note' and 'to_reg_br_prob_note' to

  19475.      extract and store the probability into the RTL encoding.

  19476. 

  19477. 'REG_BR_PRED'

  19478.      These notes are found in JUMP insns after delayed branch scheduling

  19479.      has taken place.  They indicate both the direction and the

  19480.      likelihood of the JUMP.  The format is a bitmask of ATTR_FLAG_*

  19481.      values.

  19482. 

  19483. 'REG_FRAME_RELATED_EXPR'

  19484.      This is used on an RTX_FRAME_RELATED_P insn wherein the attached

  19485.      expression is used in place of the actual insn pattern.  This is

  19486.      done in cases where the pattern is either complex or misleading.

  19487. 

  19488.  The note 'REG_CALL_NOCF_CHECK' is used in conjunction with the

  19489. '-fcf-protection=branch' option.  The note is set if a 'nocf_check'

  19490. attribute is specified for a function type or a pointer to function

  19491. type.  The note is stored in the 'REG_NOTES' field of an insn.

  19492. 

  19493. 'REG_CALL_NOCF_CHECK'

  19494.      Users have control through the 'nocf_check' attribute to identify

  19495.      which calls to a function should be skipped from control-flow

  19496.      instrumentation when the option '-fcf-protection=branch' is

  19497.      specified.  The compiler puts a 'REG_CALL_NOCF_CHECK' note on each

  19498.      'CALL_INSN' instruction that has a function type marked with a

  19499.      'nocf_check' attribute.

  19500. 

  19501.  For convenience, the machine mode in an 'insn_list' or 'expr_list' is

  19502. printed using these symbolic codes in debugging dumps.

  19503. 

  19504.  The only difference between the expression codes 'insn_list' and

  19505. 'expr_list' is that the first operand of an 'insn_list' is assumed to be

  19506. an insn and is printed in debugging dumps as the insn's unique id; the

  19507. first operand of an 'expr_list' is printed in the ordinary way as an

  19508. expression.

  19509. 

  19510. 

  19511. File: gccint.info,  Node: Calls,  Next: Sharing,  Prev: Insns,  Up: RTL

  19512. 

  19513. 14.20 RTL Representation of Function-Call Insns

  19514. ===============================================

  19515. 

  19516. Insns that call subroutines have the RTL expression code 'call_insn'.

  19517. These insns must satisfy special rules, and their bodies must use a

  19518. special RTL expression code, 'call'.

  19519. 

  19520.  A 'call' expression has two operands, as follows:

  19521. 

  19522.      (call (mem:FM ADDR) NBYTES)

  19523. 

  19524. Here NBYTES is an operand that represents the number of bytes of

  19525. argument data being passed to the subroutine, FM is a machine mode

  19526. (which must equal as the definition of the 'FUNCTION_MODE' macro in the

  19527. machine description) and ADDR represents the address of the subroutine.

  19528. 

  19529.  For a subroutine that returns no value, the 'call' expression as shown

  19530. above is the entire body of the insn, except that the insn might also

  19531. contain 'use' or 'clobber' expressions.

  19532. 

  19533.  For a subroutine that returns a value whose mode is not 'BLKmode', the

  19534. value is returned in a hard register.  If this register's number is R,

  19535. then the body of the call insn looks like this:

  19536. 

  19537.      (set (reg:M R)

  19538.           (call (mem:FM ADDR) NBYTES))

  19539. 

  19540. This RTL expression makes it clear (to the optimizer passes) that the

  19541. appropriate register receives a useful value in this insn.

  19542. 

  19543.  When a subroutine returns a 'BLKmode' value, it is handled by passing

  19544. to the subroutine the address of a place to store the value.  So the

  19545. call insn itself does not "return" any value, and it has the same RTL

  19546. form as a call that returns nothing.

  19547. 

  19548.  On some machines, the call instruction itself clobbers some register,

  19549. for example to contain the return address.  'call_insn' insns on these

  19550. machines should have a body which is a 'parallel' that contains both the

  19551. 'call' expression and 'clobber' expressions that indicate which

  19552. registers are destroyed.  Similarly, if the call instruction requires

  19553. some register other than the stack pointer that is not explicitly

  19554. mentioned in its RTL, a 'use' subexpression should mention that

  19555. register.

  19556. 

  19557.  Functions that are called are assumed to modify all registers listed in

  19558. the configuration macro 'CALL_USED_REGISTERS' (*note Register Basics::)

  19559. and, with the exception of 'const' functions and library calls, to

  19560. modify all of memory.

  19561. 

  19562.  Insns containing just 'use' expressions directly precede the

  19563. 'call_insn' insn to indicate which registers contain inputs to the

  19564. function.  Similarly, if registers other than those in

  19565. 'CALL_USED_REGISTERS' are clobbered by the called function, insns

  19566. containing a single 'clobber' follow immediately after the call to

  19567. indicate which registers.

  19568. 

  19569. 

  19570. File: gccint.info,  Node: Sharing,  Next: Reading RTL,  Prev: Calls,  Up: RTL

  19571. 

  19572. 14.21 Structure Sharing Assumptions

  19573. ===================================

  19574. 

  19575. The compiler assumes that certain kinds of RTL expressions are unique;

  19576. there do not exist two distinct objects representing the same value.  In

  19577. other cases, it makes an opposite assumption: that no RTL expression

  19578. object of a certain kind appears in more than one place in the

  19579. containing structure.

  19580. 

  19581.  These assumptions refer to a single function; except for the RTL

  19582. objects that describe global variables and external functions, and a few

  19583. standard objects such as small integer constants, no RTL objects are

  19584. common to two functions.

  19585. 

  19586.    * Each pseudo-register has only a single 'reg' object to represent

  19587.      it, and therefore only a single machine mode.

  19588. 

  19589.    * For any symbolic label, there is only one 'symbol_ref' object

  19590.      referring to it.

  19591. 

  19592.    * All 'const_int' expressions with equal values are shared.

  19593. 

  19594.    * All 'const_poly_int' expressions with equal modes and values are

  19595.      shared.

  19596. 

  19597.    * There is only one 'pc' expression.

  19598. 

  19599.    * There is only one 'cc0' expression.

  19600. 

  19601.    * There is only one 'const_double' expression with value 0 for each

  19602.      floating point mode.  Likewise for values 1 and 2.

  19603. 

  19604.    * There is only one 'const_vector' expression with value 0 for each

  19605.      vector mode, be it an integer or a double constant vector.

  19606. 

  19607.    * No 'label_ref' or 'scratch' appears in more than one place in the

  19608.      RTL structure; in other words, it is safe to do a tree-walk of all

  19609.      the insns in the function and assume that each time a 'label_ref'

  19610.      or 'scratch' is seen it is distinct from all others that are seen.

  19611. 

  19612.    * Only one 'mem' object is normally created for each static variable

  19613.      or stack slot, so these objects are frequently shared in all the

  19614.      places they appear.  However, separate but equal objects for these

  19615.      variables are occasionally made.

  19616. 

  19617.    * When a single 'asm' statement has multiple output operands, a

  19618.      distinct 'asm_operands' expression is made for each output operand.

  19619.      However, these all share the vector which contains the sequence of

  19620.      input operands.  This sharing is used later on to test whether two

  19621.      'asm_operands' expressions come from the same statement, so all

  19622.      optimizations must carefully preserve the sharing if they copy the

  19623.      vector at all.

  19624. 

  19625.    * No RTL object appears in more than one place in the RTL structure

  19626.      except as described above.  Many passes of the compiler rely on

  19627.      this by assuming that they can modify RTL objects in place without

  19628.      unwanted side-effects on other insns.

  19629. 

  19630.    * During initial RTL generation, shared structure is freely

  19631.      introduced.  After all the RTL for a function has been generated,

  19632.      all shared structure is copied by 'unshare_all_rtl' in

  19633.      'emit-rtl.c', after which the above rules are guaranteed to be

  19634.      followed.

  19635. 

  19636.    * During the combiner pass, shared structure within an insn can exist

  19637.      temporarily.  However, the shared structure is copied before the

  19638.      combiner is finished with the insn.  This is done by calling

  19639.      'copy_rtx_if_shared', which is a subroutine of 'unshare_all_rtl'.

  19640. 

  19641. 

  19642. File: gccint.info,  Node: Reading RTL,  Prev: Sharing,  Up: RTL

  19643. 

  19644. 14.22 Reading RTL

  19645. =================

  19646. 

  19647. To read an RTL object from a file, call 'read_rtx'.  It takes one

  19648. argument, a stdio stream, and returns a single RTL object.  This routine

  19649. is defined in 'read-rtl.c'.  It is not available in the compiler itself,

  19650. only the various programs that generate the compiler back end from the

  19651. machine description.

  19652. 

  19653.  People frequently have the idea of using RTL stored as text in a file

  19654. as an interface between a language front end and the bulk of GCC.  This

  19655. idea is not feasible.

  19656. 

  19657.  GCC was designed to use RTL internally only.  Correct RTL for a given

  19658. program is very dependent on the particular target machine.  And the RTL

  19659. does not contain all the information about the program.

  19660. 

  19661.  The proper way to interface GCC to a new language front end is with the

  19662. "tree" data structure, described in the files 'tree.h' and 'tree.def'.

  19663. The documentation for this structure (*note GENERIC::) is incomplete.

  19664. 

  19665. 

  19666. File: gccint.info,  Node: Control Flow,  Next: Loop Analysis and Representation,  Prev: RTL,  Up: Top

  19667. 

  19668. 15 Control Flow Graph

  19669. *********************

  19670. 

  19671. A control flow graph (CFG) is a data structure built on top of the

  19672. intermediate code representation (the RTL or 'GIMPLE' instruction

  19673. stream) abstracting the control flow behavior of a function that is

  19674. being compiled.  The CFG is a directed graph where the vertices

  19675. represent basic blocks and edges represent possible transfer of control

  19676. flow from one basic block to another.  The data structures used to

  19677. represent the control flow graph are defined in 'basic-block.h'.

  19678. 

  19679.  In GCC, the representation of control flow is maintained throughout the

  19680. compilation process, from constructing the CFG early in 'pass_build_cfg'

  19681. to 'pass_free_cfg' (see 'passes.def').  The CFG takes various different

  19682. modes and may undergo extensive manipulations, but the graph is always

  19683. valid between its construction and its release.  This way, transfer of

  19684. information such as data flow, a measured profile, or the loop tree, can

  19685. be propagated through the passes pipeline, and even from 'GIMPLE' to

  19686. 'RTL'.

  19687. 

  19688.  Often the CFG may be better viewed as integral part of instruction

  19689. chain, than structure built on the top of it.  Updating the compiler's

  19690. intermediate representation for instructions cannot be easily done

  19691. without proper maintenance of the CFG simultaneously.

  19692. 

  19693. * Menu:

  19694. 

  19695. * Basic Blocks::           The definition and representation of basic blocks.

  19696. * Edges::                  Types of edges and their representation.

  19697. * Profile information::    Representation of frequencies and probabilities.

  19698. * Maintaining the CFG::    Keeping the control flow graph and up to date.

  19699. * Liveness information::   Using and maintaining liveness information.

  19700. 

  19701. 

  19702. File: gccint.info,  Node: Basic Blocks,  Next: Edges,  Up: Control Flow

  19703. 

  19704. 15.1 Basic Blocks

  19705. =================

  19706. 

  19707. A basic block is a straight-line sequence of code with only one entry

  19708. point and only one exit.  In GCC, basic blocks are represented using the

  19709. 'basic_block' data type.

  19710. 

  19711.  Special basic blocks represent possible entry and exit points of a

  19712. function.  These blocks are called 'ENTRY_BLOCK_PTR' and

  19713. 'EXIT_BLOCK_PTR'.  These blocks do not contain any code.

  19714. 

  19715.  The 'BASIC_BLOCK' array contains all basic blocks in an unspecified

  19716. order.  Each 'basic_block' structure has a field that holds a unique

  19717. integer identifier 'index' that is the index of the block in the

  19718. 'BASIC_BLOCK' array.  The total number of basic blocks in the function

  19719. is 'n_basic_blocks'.  Both the basic block indices and the total number

  19720. of basic blocks may vary during the compilation process, as passes

  19721. reorder, create, duplicate, and destroy basic blocks.  The index for any

  19722. block should never be greater than 'last_basic_block'.  The indices 0

  19723. and 1 are special codes reserved for 'ENTRY_BLOCK' and 'EXIT_BLOCK', the

  19724. indices of 'ENTRY_BLOCK_PTR' and 'EXIT_BLOCK_PTR'.

  19725. 

  19726.  Two pointer members of the 'basic_block' structure are the pointers

  19727. 'next_bb' and 'prev_bb'.  These are used to keep doubly linked chain of

  19728. basic blocks in the same order as the underlying instruction stream.

  19729. The chain of basic blocks is updated transparently by the provided API

  19730. for manipulating the CFG.  The macro 'FOR_EACH_BB' can be used to visit

  19731. all the basic blocks in lexicographical order, except 'ENTRY_BLOCK' and

  19732. 'EXIT_BLOCK'.  The macro 'FOR_ALL_BB' also visits all basic blocks in

  19733. lexicographical order, including 'ENTRY_BLOCK' and 'EXIT_BLOCK'.

  19734. 

  19735.  The functions 'post_order_compute' and 'inverted_post_order_compute'

  19736. can be used to compute topological orders of the CFG. The orders are

  19737. stored as vectors of basic block indices.  The 'BASIC_BLOCK' array can

  19738. be used to iterate each basic block by index.  Dominator traversals are

  19739. also possible using 'walk_dominator_tree'.  Given two basic blocks A and

  19740. B, block A dominates block B if A is _always_ executed before B.

  19741. 

  19742.  Each 'basic_block' also contains pointers to the first instruction (the

  19743. "head") and the last instruction (the "tail") or "end" of the

  19744. instruction stream contained in a basic block.  In fact, since the

  19745. 'basic_block' data type is used to represent blocks in both major

  19746. intermediate representations of GCC ('GIMPLE' and RTL), there are

  19747. pointers to the head and end of a basic block for both representations,

  19748. stored in intermediate representation specific data in the 'il' field of

  19749. 'struct basic_block_def'.

  19750. 

  19751.  For RTL, these pointers are 'BB_HEAD' and 'BB_END'.

  19752. 

  19753.  In the RTL representation of a function, the instruction stream

  19754. contains not only the "real" instructions, but also "notes" or "insn

  19755. notes" (to distinguish them from "reg notes").  Any function that moves

  19756. or duplicates the basic blocks needs to take care of updating of these

  19757. notes.  Many of these notes expect that the instruction stream consists

  19758. of linear regions, so updating can sometimes be tedious.  All types of

  19759. insn notes are defined in 'insn-notes.def'.

  19760. 

  19761.  In the RTL function representation, the instructions contained in a

  19762. basic block always follow a 'NOTE_INSN_BASIC_BLOCK', but zero or more

  19763. 'CODE_LABEL' nodes can precede the block note.  A basic block ends with

  19764. a control flow instruction or with the last instruction before the next

  19765. 'CODE_LABEL' or 'NOTE_INSN_BASIC_BLOCK'.  By definition, a 'CODE_LABEL'

  19766. cannot appear in the middle of the instruction stream of a basic block.

  19767. 

  19768.  In addition to notes, the jump table vectors are also represented as

  19769. "pseudo-instructions" inside the insn stream.  These vectors never

  19770. appear in the basic block and should always be placed just after the

  19771. table jump instructions referencing them.  After removing the table-jump

  19772. it is often difficult to eliminate the code computing the address and

  19773. referencing the vector, so cleaning up these vectors is postponed until

  19774. after liveness analysis.  Thus the jump table vectors may appear in the

  19775. insn stream unreferenced and without any purpose.  Before any edge is

  19776. made "fall-thru", the existence of such construct in the way needs to be

  19777. checked by calling 'can_fallthru' function.

  19778. 

  19779.  For the 'GIMPLE' representation, the PHI nodes and statements contained

  19780. in a basic block are in a 'gimple_seq' pointed to by the basic block

  19781. intermediate language specific pointers.  Abstract containers and

  19782. iterators are used to access the PHI nodes and statements in a basic

  19783. blocks.  These iterators are called "GIMPLE statement iterators" (GSIs).

  19784. Grep for '^gsi' in the various 'gimple-*' and 'tree-*' files.  There is

  19785. a 'gimple_stmt_iterator' type for iterating over all kinds of statement,

  19786. and a 'gphi_iterator' subclass for iterating over PHI nodes.  The

  19787. following snippet will pretty-print all PHI nodes the statements of the

  19788. current function in the GIMPLE representation.

  19789. 

  19790.      basic_block bb;

  19791. 

  19792.      FOR_EACH_BB (bb)

  19793.        {

  19794.         gphi_iterator pi;

  19795.         gimple_stmt_iterator si;

  19796. 

  19797.         for (pi = gsi_start_phis (bb); !gsi_end_p (pi); gsi_next (&pi))

  19798.           {

  19799.             gphi *phi = pi.phi ();

  19800.             print_gimple_stmt (dump_file, phi, 0, TDF_SLIM);

  19801.           }

  19802.         for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si))

  19803.           {

  19804.             gimple stmt = gsi_stmt (si);

  19805.             print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);

  19806.           }

  19807.        }

  19808. 

  19809. 

  19810. File: gccint.info,  Node: Edges,  Next: Profile information,  Prev: Basic Blocks,  Up: Control Flow

  19811. 

  19812. 15.2 Edges

  19813. ==========

  19814. 

  19815. Edges represent possible control flow transfers from the end of some

  19816. basic block A to the head of another basic block B.  We say that A is a

  19817. predecessor of B, and B is a successor of A.  Edges are represented in

  19818. GCC with the 'edge' data type.  Each 'edge' acts as a link between two

  19819. basic blocks: The 'src' member of an edge points to the predecessor

  19820. basic block of the 'dest' basic block.  The members 'preds' and 'succs'

  19821. of the 'basic_block' data type point to type-safe vectors of edges to

  19822. the predecessors and successors of the block.

  19823. 

  19824.  When walking the edges in an edge vector, "edge iterators" should be

  19825. used.  Edge iterators are constructed using the 'edge_iterator' data

  19826. structure and several methods are available to operate on them:

  19827. 

  19828. 'ei_start'

  19829.      This function initializes an 'edge_iterator' that points to the

  19830.      first edge in a vector of edges.

  19831. 

  19832. 'ei_last'

  19833.      This function initializes an 'edge_iterator' that points to the

  19834.      last edge in a vector of edges.

  19835. 

  19836. 'ei_end_p'

  19837.      This predicate is 'true' if an 'edge_iterator' represents the last

  19838.      edge in an edge vector.

  19839. 

  19840. 'ei_one_before_end_p'

  19841.      This predicate is 'true' if an 'edge_iterator' represents the

  19842.      second last edge in an edge vector.

  19843. 

  19844. 'ei_next'

  19845.      This function takes a pointer to an 'edge_iterator' and makes it

  19846.      point to the next edge in the sequence.

  19847. 

  19848. 'ei_prev'

  19849.      This function takes a pointer to an 'edge_iterator' and makes it

  19850.      point to the previous edge in the sequence.

  19851. 

  19852. 'ei_edge'

  19853.      This function returns the 'edge' currently pointed to by an

  19854.      'edge_iterator'.

  19855. 

  19856. 'ei_safe_safe'

  19857.      This function returns the 'edge' currently pointed to by an

  19858.      'edge_iterator', but returns 'NULL' if the iterator is pointing at

  19859.      the end of the sequence.  This function has been provided for

  19860.      existing code makes the assumption that a 'NULL' edge indicates the

  19861.      end of the sequence.

  19862. 

  19863.  The convenience macro 'FOR_EACH_EDGE' can be used to visit all of the

  19864. edges in a sequence of predecessor or successor edges.  It must not be

  19865. used when an element might be removed during the traversal, otherwise

  19866. elements will be missed.  Here is an example of how to use the macro:

  19867. 

  19868.      edge e;

  19869.      edge_iterator ei;

  19870. 

  19871.      FOR_EACH_EDGE (e, ei, bb->succs)

  19872.        {

  19873.           if (e->flags & EDGE_FALLTHRU)

  19874.             break;

  19875.        }

  19876. 

  19877.  There are various reasons why control flow may transfer from one block

  19878. to another.  One possibility is that some instruction, for example a

  19879. 'CODE_LABEL', in a linearized instruction stream just always starts a

  19880. new basic block.  In this case a "fall-thru" edge links the basic block

  19881. to the first following basic block.  But there are several other reasons

  19882. why edges may be created.  The 'flags' field of the 'edge' data type is

  19883. used to store information about the type of edge we are dealing with.

  19884. Each edge is of one of the following types:

  19885. 

  19886. _jump_

  19887.      No type flags are set for edges corresponding to jump instructions.

  19888.      These edges are used for unconditional or conditional jumps and in

  19889.      RTL also for table jumps.  They are the easiest to manipulate as

  19890.      they may be freely redirected when the flow graph is not in SSA

  19891.      form.

  19892. 

  19893. _fall-thru_

  19894.      Fall-thru edges are present in case where the basic block may

  19895.      continue execution to the following one without branching.  These

  19896.      edges have the 'EDGE_FALLTHRU' flag set.  Unlike other types of

  19897.      edges, these edges must come into the basic block immediately

  19898.      following in the instruction stream.  The function

  19899.      'force_nonfallthru' is available to insert an unconditional jump in

  19900.      the case that redirection is needed.  Note that this may require

  19901.      creation of a new basic block.

  19902. 

  19903. _exception handling_

  19904.      Exception handling edges represent possible control transfers from

  19905.      a trapping instruction to an exception handler.  The definition of

  19906.      "trapping" varies.  In C++, only function calls can throw, but for

  19907.      Ada exceptions like division by zero or segmentation fault are

  19908.      defined and thus each instruction possibly throwing this kind of

  19909.      exception needs to be handled as control flow instruction.

  19910.      Exception edges have the 'EDGE_ABNORMAL' and 'EDGE_EH' flags set.

  19911. 

  19912.      When updating the instruction stream it is easy to change possibly

  19913.      trapping instruction to non-trapping, by simply removing the

  19914.      exception edge.  The opposite conversion is difficult, but should

  19915.      not happen anyway.  The edges can be eliminated via

  19916.      'purge_dead_edges' call.

  19917. 

  19918.      In the RTL representation, the destination of an exception edge is

  19919.      specified by 'REG_EH_REGION' note attached to the insn.  In case of

  19920.      a trapping call the 'EDGE_ABNORMAL_CALL' flag is set too.  In the

  19921.      'GIMPLE' representation, this extra flag is not set.

  19922. 

  19923.      In the RTL representation, the predicate 'may_trap_p' may be used

  19924.      to check whether instruction still may trap or not.  For the tree

  19925.      representation, the 'tree_could_trap_p' predicate is available, but

  19926.      this predicate only checks for possible memory traps, as in

  19927.      dereferencing an invalid pointer location.

  19928. 

  19929. _sibling calls_

  19930.      Sibling calls or tail calls terminate the function in a

  19931.      non-standard way and thus an edge to the exit must be present.

  19932.      'EDGE_SIBCALL' and 'EDGE_ABNORMAL' are set in such case.  These

  19933.      edges only exist in the RTL representation.

  19934. 

  19935. _computed jumps_

  19936.      Computed jumps contain edges to all labels in the function

  19937.      referenced from the code.  All those edges have 'EDGE_ABNORMAL'

  19938.      flag set.  The edges used to represent computed jumps often cause

  19939.      compile time performance problems, since functions consisting of

  19940.      many taken labels and many computed jumps may have _very_ dense

  19941.      flow graphs, so these edges need to be handled with special care.

  19942.      During the earlier stages of the compilation process, GCC tries to

  19943.      avoid such dense flow graphs by factoring computed jumps.  For

  19944.      example, given the following series of jumps,

  19945. 

  19946.             goto *x;

  19947.             [ ... ]

  19948. 

  19949.             goto *x;

  19950.             [ ... ]

  19951. 

  19952.             goto *x;

  19953.             [ ... ]

  19954. 

  19955.      factoring the computed jumps results in the following code sequence

  19956.      which has a much simpler flow graph:

  19957. 

  19958.             goto y;

  19959.             [ ... ]

  19960. 

  19961.             goto y;

  19962.             [ ... ]

  19963. 

  19964.             goto y;

  19965.             [ ... ]

  19966. 

  19967.           y:

  19968.             goto *x;

  19969. 

  19970.      However, the classic problem with this transformation is that it

  19971.      has a runtime cost in there resulting code: An extra jump.

  19972.      Therefore, the computed jumps are un-factored in the later passes

  19973.      of the compiler (in the pass called

  19974.      'pass_duplicate_computed_gotos').  Be aware of that when you work

  19975.      on passes in that area.  There have been numerous examples already

  19976.      where the compile time for code with unfactored computed jumps

  19977.      caused some serious headaches.

  19978. 

  19979. _nonlocal goto handlers_

  19980.      GCC allows nested functions to return into caller using a 'goto' to

  19981.      a label passed to as an argument to the callee.  The labels passed

  19982.      to nested functions contain special code to cleanup after function

  19983.      call.  Such sections of code are referred to as "nonlocal goto

  19984.      receivers".  If a function contains such nonlocal goto receivers,

  19985.      an edge from the call to the label is created with the

  19986.      'EDGE_ABNORMAL' and 'EDGE_ABNORMAL_CALL' flags set.

  19987. 

  19988. _function entry points_

  19989.      By definition, execution of function starts at basic block 0, so

  19990.      there is always an edge from the 'ENTRY_BLOCK_PTR' to basic block

  19991.      0.  There is no 'GIMPLE' representation for alternate entry points

  19992.      at this moment.  In RTL, alternate entry points are specified by

  19993.      'CODE_LABEL' with 'LABEL_ALTERNATE_NAME' defined.  This feature is

  19994.      currently used for multiple entry point prologues and is limited to

  19995.      post-reload passes only.  This can be used by back-ends to emit

  19996.      alternate prologues for functions called from different contexts.

  19997.      In future full support for multiple entry functions defined by

  19998.      Fortran 90 needs to be implemented.

  19999. 

  20000. _function exits_

  20001.      In the pre-reload representation a function terminates after the

  20002.      last instruction in the insn chain and no explicit return

  20003.      instructions are used.  This corresponds to the fall-thru edge into

  20004.      exit block.  After reload, optimal RTL epilogues are used that use

  20005.      explicit (conditional) return instructions that are represented by

  20006.      edges with no flags set.

  20007. 

  20008. 

  20009. File: gccint.info,  Node: Profile information,  Next: Maintaining the CFG,  Prev: Edges,  Up: Control Flow

  20010. 

  20011. 15.3 Profile information

  20012. ========================

  20013. 

  20014. In many cases a compiler must make a choice whether to trade speed in

  20015. one part of code for speed in another, or to trade code size for code

  20016. speed.  In such cases it is useful to know information about how often

  20017. some given block will be executed.  That is the purpose for maintaining

  20018. profile within the flow graph.  GCC can handle profile information

  20019. obtained through "profile feedback", but it can also estimate branch

  20020. probabilities based on statics and heuristics.

  20021. 

  20022.  The feedback based profile is produced by compiling the program with

  20023. instrumentation, executing it on a train run and reading the numbers of

  20024. executions of basic blocks and edges back to the compiler while

  20025. re-compiling the program to produce the final executable.  This method

  20026. provides very accurate information about where a program spends most of

  20027. its time on the train run.  Whether it matches the average run of course

  20028. depends on the choice of train data set, but several studies have shown

  20029. that the behavior of a program usually changes just marginally over

  20030. different data sets.

  20031. 

  20032.  When profile feedback is not available, the compiler may be asked to

  20033. attempt to predict the behavior of each branch in the program using a

  20034. set of heuristics (see 'predict.def' for details) and compute estimated

  20035. frequencies of each basic block by propagating the probabilities over

  20036. the graph.

  20037. 

  20038.  Each 'basic_block' contains two integer fields to represent profile

  20039. information: 'frequency' and 'count'.  The 'frequency' is an estimation

  20040. how often is basic block executed within a function.  It is represented

  20041. as an integer scaled in the range from 0 to 'BB_FREQ_BASE'.  The most

  20042. frequently executed basic block in function is initially set to

  20043. 'BB_FREQ_BASE' and the rest of frequencies are scaled accordingly.

  20044. During optimization, the frequency of the most frequent basic block can

  20045. both decrease (for instance by loop unrolling) or grow (for instance by

  20046. cross-jumping optimization), so scaling sometimes has to be performed

  20047. multiple times.

  20048. 

  20049.  The 'count' contains hard-counted numbers of execution measured during

  20050. training runs and is nonzero only when profile feedback is available.

  20051. This value is represented as the host's widest integer (typically a 64

  20052. bit integer) of the special type 'gcov_type'.

  20053. 

  20054.  Most optimization passes can use only the frequency information of a

  20055. basic block, but a few passes may want to know hard execution counts.

  20056. The frequencies should always match the counts after scaling, however

  20057. during updating of the profile information numerical error may

  20058. accumulate into quite large errors.

  20059. 

  20060.  Each edge also contains a branch probability field: an integer in the

  20061. range from 0 to 'REG_BR_PROB_BASE'.  It represents probability of

  20062. passing control from the end of the 'src' basic block to the 'dest'

  20063. basic block, i.e. the probability that control will flow along this

  20064. edge.  The 'EDGE_FREQUENCY' macro is available to compute how frequently

  20065. a given edge is taken.  There is a 'count' field for each edge as well,

  20066. representing same information as for a basic block.

  20067. 

  20068.  The basic block frequencies are not represented in the instruction

  20069. stream, but in the RTL representation the edge frequencies are

  20070. represented for conditional jumps (via the 'REG_BR_PROB' macro) since

  20071. they are used when instructions are output to the assembly file and the

  20072. flow graph is no longer maintained.

  20073. 

  20074.  The probability that control flow arrives via a given edge to its

  20075. destination basic block is called "reverse probability" and is not

  20076. directly represented, but it may be easily computed from frequencies of

  20077. basic blocks.

  20078. 

  20079.  Updating profile information is a delicate task that can unfortunately

  20080. not be easily integrated with the CFG manipulation API.  Many of the

  20081. functions and hooks to modify the CFG, such as

  20082. 'redirect_edge_and_branch', do not have enough information to easily

  20083. update the profile, so updating it is in the majority of cases left up

  20084. to the caller.  It is difficult to uncover bugs in the profile updating

  20085. code, because they manifest themselves only by producing worse code, and

  20086. checking profile consistency is not possible because of numeric error

  20087. accumulation.  Hence special attention needs to be given to this issue

  20088. in each pass that modifies the CFG.

  20089. 

  20090.  It is important to point out that 'REG_BR_PROB_BASE' and 'BB_FREQ_BASE'

  20091. are both set low enough to be possible to compute second power of any

  20092. frequency or probability in the flow graph, it is not possible to even

  20093. square the 'count' field, as modern CPUs are fast enough to execute

  20094. $2^32$ operations quickly.

  20095. 

  20096. 

  20097. File: gccint.info,  Node: Maintaining the CFG,  Next: Liveness information,  Prev: Profile information,  Up: Control Flow

  20098. 

  20099. 15.4 Maintaining the CFG

  20100. ========================

  20101. 

  20102. An important task of each compiler pass is to keep both the control flow

  20103. graph and all profile information up-to-date.  Reconstruction of the

  20104. control flow graph after each pass is not an option, since it may be

  20105. very expensive and lost profile information cannot be reconstructed at

  20106. all.

  20107. 

  20108.  GCC has two major intermediate representations, and both use the

  20109. 'basic_block' and 'edge' data types to represent control flow.  Both

  20110. representations share as much of the CFG maintenance code as possible.

  20111. For each representation, a set of "hooks" is defined so that each

  20112. representation can provide its own implementation of CFG manipulation

  20113. routines when necessary.  These hooks are defined in 'cfghooks.h'.

  20114. There are hooks for almost all common CFG manipulations, including block

  20115. splitting and merging, edge redirection and creating and deleting basic

  20116. blocks.  These hooks should provide everything you need to maintain and

  20117. manipulate the CFG in both the RTL and 'GIMPLE' representation.

  20118. 

  20119.  At the moment, the basic block boundaries are maintained transparently

  20120. when modifying instructions, so there rarely is a need to move them

  20121. manually (such as in case someone wants to output instruction outside

  20122. basic block explicitly).

  20123. 

  20124.  In the RTL representation, each instruction has a 'BLOCK_FOR_INSN'

  20125. value that represents pointer to the basic block that contains the

  20126. instruction.  In the 'GIMPLE' representation, the function 'gimple_bb'

  20127. returns a pointer to the basic block containing the queried statement.

  20128. 

  20129.  When changes need to be applied to a function in its 'GIMPLE'

  20130. representation, "GIMPLE statement iterators" should be used.  These

  20131. iterators provide an integrated abstraction of the flow graph and the

  20132. instruction stream.  Block statement iterators are constructed using the

  20133. 'gimple_stmt_iterator' data structure and several modifiers are

  20134. available, including the following:

  20135. 

  20136. 'gsi_start'

  20137.      This function initializes a 'gimple_stmt_iterator' that points to

  20138.      the first non-empty statement in a basic block.

  20139. 

  20140. 'gsi_last'

  20141.      This function initializes a 'gimple_stmt_iterator' that points to

  20142.      the last statement in a basic block.

  20143. 

  20144. 'gsi_end_p'

  20145.      This predicate is 'true' if a 'gimple_stmt_iterator' represents the

  20146.      end of a basic block.

  20147. 

  20148. 'gsi_next'

  20149.      This function takes a 'gimple_stmt_iterator' and makes it point to

  20150.      its successor.

  20151. 

  20152. 'gsi_prev'

  20153.      This function takes a 'gimple_stmt_iterator' and makes it point to

  20154.      its predecessor.

  20155. 

  20156. 'gsi_insert_after'

  20157.      This function inserts a statement after the 'gimple_stmt_iterator'

  20158.      passed in.  The final parameter determines whether the statement

  20159.      iterator is updated to point to the newly inserted statement, or

  20160.      left pointing to the original statement.

  20161. 

  20162. 'gsi_insert_before'

  20163.      This function inserts a statement before the 'gimple_stmt_iterator'

  20164.      passed in.  The final parameter determines whether the statement

  20165.      iterator is updated to point to the newly inserted statement, or

  20166.      left pointing to the original statement.

  20167. 

  20168. 'gsi_remove'

  20169.      This function removes the 'gimple_stmt_iterator' passed in and

  20170.      rechains the remaining statements in a basic block, if any.

  20171. 

  20172.  In the RTL representation, the macros 'BB_HEAD' and 'BB_END' may be

  20173. used to get the head and end 'rtx' of a basic block.  No abstract

  20174. iterators are defined for traversing the insn chain, but you can just

  20175. use 'NEXT_INSN' and 'PREV_INSN' instead.  *Note Insns::.

  20176. 

  20177.  Usually a code manipulating pass simplifies the instruction stream and

  20178. the flow of control, possibly eliminating some edges.  This may for

  20179. example happen when a conditional jump is replaced with an unconditional

  20180. jump.  Updating of edges is not transparent and each optimization pass

  20181. is required to do so manually.  However only few cases occur in

  20182. practice.  The pass may call 'purge_dead_edges' on a given basic block

  20183. to remove superfluous edges, if any.

  20184. 

  20185.  Another common scenario is redirection of branch instructions, but this

  20186. is best modeled as redirection of edges in the control flow graph and

  20187. thus use of 'redirect_edge_and_branch' is preferred over more low level

  20188. functions, such as 'redirect_jump' that operate on RTL chain only.  The

  20189. CFG hooks defined in 'cfghooks.h' should provide the complete API

  20190. required for manipulating and maintaining the CFG.

  20191. 

  20192.  It is also possible that a pass has to insert control flow instruction

  20193. into the middle of a basic block, thus creating an entry point in the

  20194. middle of the basic block, which is impossible by definition: The block

  20195. must be split to make sure it only has one entry point, i.e. the head of

  20196. the basic block.  The CFG hook 'split_block' may be used when an

  20197. instruction in the middle of a basic block has to become the target of a

  20198. jump or branch instruction.

  20199. 

  20200.  For a global optimizer, a common operation is to split edges in the

  20201. flow graph and insert instructions on them.  In the RTL representation,

  20202. this can be easily done using the 'insert_insn_on_edge' function that

  20203. emits an instruction "on the edge", caching it for a later

  20204. 'commit_edge_insertions' call that will take care of moving the inserted

  20205. instructions off the edge into the instruction stream contained in a

  20206. basic block.  This includes the creation of new basic blocks where

  20207. needed.  In the 'GIMPLE' representation, the equivalent functions are

  20208. 'gsi_insert_on_edge' which inserts a block statement iterator on an

  20209. edge, and 'gsi_commit_edge_inserts' which flushes the instruction to

  20210. actual instruction stream.

  20211. 

  20212.  While debugging the optimization pass, the 'verify_flow_info' function

  20213. may be useful to find bugs in the control flow graph updating code.

  20214. 

  20215. 

  20216. File: gccint.info,  Node: Liveness information,  Prev: Maintaining the CFG,  Up: Control Flow

  20217. 

  20218. 15.5 Liveness information

  20219. =========================

  20220. 

  20221. Liveness information is useful to determine whether some register is

  20222. "live" at given point of program, i.e. that it contains a value that may

  20223. be used at a later point in the program.  This information is used, for

  20224. instance, during register allocation, as the pseudo registers only need

  20225. to be assigned to a unique hard register or to a stack slot if they are

  20226. live.  The hard registers and stack slots may be freely reused for other

  20227. values when a register is dead.

  20228. 

  20229.  Liveness information is available in the back end starting with

  20230. 'pass_df_initialize' and ending with 'pass_df_finish'.  Three flavors of

  20231. live analysis are available: With 'LR', it is possible to determine at

  20232. any point 'P' in the function if the register may be used on some path

  20233. from 'P' to the end of the function.  With 'UR', it is possible to

  20234. determine if there is a path from the beginning of the function to 'P'

  20235. that defines the variable.  'LIVE' is the intersection of the 'LR' and

  20236. 'UR' and a variable is live at 'P' if there is both an assignment that

  20237. reaches it from the beginning of the function and a use that can be

  20238. reached on some path from 'P' to the end of the function.

  20239. 

  20240.  In general 'LIVE' is the most useful of the three.  The macros

  20241. 'DF_[LR,UR,LIVE]_[IN,OUT]' can be used to access this information.  The

  20242. macros take a basic block number and return a bitmap that is indexed by

  20243. the register number.  This information is only guaranteed to be up to

  20244. date after calls are made to 'df_analyze'.  See the file 'df-core.c' for

  20245. details on using the dataflow.

  20246. 

  20247.  The liveness information is stored partly in the RTL instruction stream

  20248. and partly in the flow graph.  Local information is stored in the

  20249. instruction stream: Each instruction may contain 'REG_DEAD' notes

  20250. representing that the value of a given register is no longer needed, or

  20251. 'REG_UNUSED' notes representing that the value computed by the

  20252. instruction is never used.  The second is useful for instructions

  20253. computing multiple values at once.

  20254. 

  20255. 

  20256. File: gccint.info,  Node: Loop Analysis and Representation,  Next: Machine Desc,  Prev: Control Flow,  Up: Top

  20257. 

  20258. 16 Analysis and Representation of Loops

  20259. ***************************************

  20260. 

  20261. GCC provides extensive infrastructure for work with natural loops, i.e.,

  20262. strongly connected components of CFG with only one entry block.  This

  20263. chapter describes representation of loops in GCC, both on GIMPLE and in

  20264. RTL, as well as the interfaces to loop-related analyses (induction

  20265. variable analysis and number of iterations analysis).

  20266. 

  20267. * Menu:

  20268. 

  20269. * Loop representation::         Representation and analysis of loops.

  20270. * Loop querying::               Getting information about loops.

  20271. * Loop manipulation::           Loop manipulation functions.

  20272. * LCSSA::                       Loop-closed SSA form.

  20273. * Scalar evolutions::           Induction variables on GIMPLE.

  20274. * loop-iv::                     Induction variables on RTL.

  20275. * Number of iterations::        Number of iterations analysis.

  20276. * Dependency analysis::         Data dependency analysis.

  20277. 

  20278. 

  20279. File: gccint.info,  Node: Loop representation,  Next: Loop querying,  Up: Loop Analysis and Representation

  20280. 

  20281. 16.1 Loop representation

  20282. ========================

  20283. 

  20284. This chapter describes the representation of loops in GCC, and functions

  20285. that can be used to build, modify and analyze this representation.  Most

  20286. of the interfaces and data structures are declared in 'cfgloop.h'.  Loop

  20287. structures are analyzed and this information disposed or updated at the

  20288. discretion of individual passes.  Still most of the generic CFG

  20289. manipulation routines are aware of loop structures and try to keep them

  20290. up-to-date.  By this means an increasing part of the compilation

  20291. pipeline is setup to maintain loop structure across passes to allow

  20292. attaching meta information to individual loops for consumption by later

  20293. passes.

  20294. 

  20295.  In general, a natural loop has one entry block (header) and possibly

  20296. several back edges (latches) leading to the header from the inside of

  20297. the loop.  Loops with several latches may appear if several loops share

  20298. a single header, or if there is a branching in the middle of the loop.

  20299. The representation of loops in GCC however allows only loops with a

  20300. single latch.  During loop analysis, headers of such loops are split and

  20301. forwarder blocks are created in order to disambiguate their structures.

  20302. Heuristic based on profile information and structure of the induction

  20303. variables in the loops is used to determine whether the latches

  20304. correspond to sub-loops or to control flow in a single loop.  This means

  20305. that the analysis sometimes changes the CFG, and if you run it in the

  20306. middle of an optimization pass, you must be able to deal with the new

  20307. blocks.  You may avoid CFG changes by passing

  20308. 'LOOPS_MAY_HAVE_MULTIPLE_LATCHES' flag to the loop discovery, note

  20309. however that most other loop manipulation functions will not work

  20310. correctly for loops with multiple latch edges (the functions that only

  20311. query membership of blocks to loops and subloop relationships, or

  20312. enumerate and test loop exits, can be expected to work).

  20313. 

  20314.  Body of the loop is the set of blocks that are dominated by its header,

  20315. and reachable from its latch against the direction of edges in CFG.  The

  20316. loops are organized in a containment hierarchy (tree) such that all the

  20317. loops immediately contained inside loop L are the children of L in the

  20318. tree.  This tree is represented by the 'struct loops' structure.  The

  20319. root of this tree is a fake loop that contains all blocks in the

  20320. function.  Each of the loops is represented in a 'struct loop'

  20321. structure.  Each loop is assigned an index ('num' field of the 'struct

  20322. loop' structure), and the pointer to the loop is stored in the

  20323. corresponding field of the 'larray' vector in the loops structure.  The

  20324. indices do not have to be continuous, there may be empty ('NULL')

  20325. entries in the 'larray' created by deleting loops.  Also, there is no

  20326. guarantee on the relative order of a loop and its subloops in the

  20327. numbering.  The index of a loop never changes.

  20328. 

  20329.  The entries of the 'larray' field should not be accessed directly.  The

  20330. function 'get_loop' returns the loop description for a loop with the

  20331. given index.  'number_of_loops' function returns number of loops in the

  20332. function.  To traverse all loops, use 'FOR_EACH_LOOP' macro.  The

  20333. 'flags' argument of the macro is used to determine the direction of

  20334. traversal and the set of loops visited.  Each loop is guaranteed to be

  20335. visited exactly once, regardless of the changes to the loop tree, and

  20336. the loops may be removed during the traversal.  The newly created loops

  20337. are never traversed, if they need to be visited, this must be done

  20338. separately after their creation.

  20339. 

  20340.  Each basic block contains the reference to the innermost loop it

  20341. belongs to ('loop_father').  For this reason, it is only possible to

  20342. have one 'struct loops' structure initialized at the same time for each

  20343. CFG.  The global variable 'current_loops' contains the 'struct loops'

  20344. structure.  Many of the loop manipulation functions assume that

  20345. dominance information is up-to-date.

  20346. 

  20347.  The loops are analyzed through 'loop_optimizer_init' function.  The

  20348. argument of this function is a set of flags represented in an integer

  20349. bitmask.  These flags specify what other properties of the loop

  20350. structures should be calculated/enforced and preserved later:

  20351. 

  20352.    * 'LOOPS_MAY_HAVE_MULTIPLE_LATCHES': If this flag is set, no changes

  20353.      to CFG will be performed in the loop analysis, in particular, loops

  20354.      with multiple latch edges will not be disambiguated.  If a loop has

  20355.      multiple latches, its latch block is set to NULL.  Most of the loop

  20356.      manipulation functions will not work for loops in this shape.  No

  20357.      other flags that require CFG changes can be passed to

  20358.      loop_optimizer_init.

  20359.    * 'LOOPS_HAVE_PREHEADERS': Forwarder blocks are created in such a way

  20360.      that each loop has only one entry edge, and additionally, the

  20361.      source block of this entry edge has only one successor.  This

  20362.      creates a natural place where the code can be moved out of the

  20363.      loop, and ensures that the entry edge of the loop leads from its

  20364.      immediate super-loop.

  20365.    * 'LOOPS_HAVE_SIMPLE_LATCHES': Forwarder blocks are created to force

  20366.      the latch block of each loop to have only one successor.  This

  20367.      ensures that the latch of the loop does not belong to any of its

  20368.      sub-loops, and makes manipulation with the loops significantly

  20369.      easier.  Most of the loop manipulation functions assume that the

  20370.      loops are in this shape.  Note that with this flag, the "normal"

  20371.      loop without any control flow inside and with one exit consists of

  20372.      two basic blocks.

  20373.    * 'LOOPS_HAVE_MARKED_IRREDUCIBLE_REGIONS': Basic blocks and edges in

  20374.      the strongly connected components that are not natural loops (have

  20375.      more than one entry block) are marked with 'BB_IRREDUCIBLE_LOOP'

  20376.      and 'EDGE_IRREDUCIBLE_LOOP' flags.  The flag is not set for blocks

  20377.      and edges that belong to natural loops that are in such an

  20378.      irreducible region (but it is set for the entry and exit edges of

  20379.      such a loop, if they lead to/from this region).

  20380.    * 'LOOPS_HAVE_RECORDED_EXITS': The lists of exits are recorded and

  20381.      updated for each loop.  This makes some functions (e.g.,

  20382.      'get_loop_exit_edges') more efficient.  Some functions (e.g.,

  20383.      'single_exit') can be used only if the lists of exits are recorded.

  20384. 

  20385.  These properties may also be computed/enforced later, using functions

  20386. 'create_preheaders', 'force_single_succ_latches',

  20387. 'mark_irreducible_loops' and 'record_loop_exits'.  The properties can be

  20388. queried using 'loops_state_satisfies_p'.

  20389. 

  20390.  The memory occupied by the loops structures should be freed with

  20391. 'loop_optimizer_finalize' function.  When loop structures are setup to

  20392. be preserved across passes this function reduces the information to be

  20393. kept up-to-date to a minimum (only 'LOOPS_MAY_HAVE_MULTIPLE_LATCHES'

  20394. set).

  20395. 

  20396.  The CFG manipulation functions in general do not update loop

  20397. structures.  Specialized versions that additionally do so are provided

  20398. for the most common tasks.  On GIMPLE, 'cleanup_tree_cfg_loop' function

  20399. can be used to cleanup CFG while updating the loops structures if

  20400. 'current_loops' is set.

  20401. 

  20402.  At the moment loop structure is preserved from the start of GIMPLE loop

  20403. optimizations until the end of RTL loop optimizations.  During this time

  20404. a loop can be tracked by its 'struct loop' and number.

  20405. 

  20406. 

  20407. File: gccint.info,  Node: Loop querying,  Next: Loop manipulation,  Prev: Loop representation,  Up: Loop Analysis and Representation

  20408. 

  20409. 16.2 Loop querying

  20410. ==================

  20411. 

  20412. The functions to query the information about loops are declared in

  20413. 'cfgloop.h'.  Some of the information can be taken directly from the

  20414. structures.  'loop_father' field of each basic block contains the

  20415. innermost loop to that the block belongs.  The most useful fields of

  20416. loop structure (that are kept up-to-date at all times) are:

  20417. 

  20418.    * 'header', 'latch': Header and latch basic blocks of the loop.

  20419.    * 'num_nodes': Number of basic blocks in the loop (including the

  20420.      basic blocks of the sub-loops).

  20421.    * 'outer', 'inner', 'next': The super-loop, the first sub-loop, and

  20422.      the sibling of the loop in the loops tree.

  20423. 

  20424.  There are other fields in the loop structures, many of them used only

  20425. by some of the passes, or not updated during CFG changes; in general,

  20426. they should not be accessed directly.

  20427. 

  20428.  The most important functions to query loop structures are:

  20429. 

  20430.    * 'loop_depth': The depth of the loop in the loops tree, i.e., the

  20431.      number of super-loops of the loop.

  20432.    * 'flow_loops_dump': Dumps the information about loops to a file.

  20433.    * 'verify_loop_structure': Checks consistency of the loop structures.

  20434.    * 'loop_latch_edge': Returns the latch edge of a loop.

  20435.    * 'loop_preheader_edge': If loops have preheaders, returns the

  20436.      preheader edge of a loop.

  20437.    * 'flow_loop_nested_p': Tests whether loop is a sub-loop of another

  20438.      loop.

  20439.    * 'flow_bb_inside_loop_p': Tests whether a basic block belongs to a

  20440.      loop (including its sub-loops).

  20441.    * 'find_common_loop': Finds the common super-loop of two loops.

  20442.    * 'superloop_at_depth': Returns the super-loop of a loop with the

  20443.      given depth.

  20444.    * 'tree_num_loop_insns', 'num_loop_insns': Estimates the number of

  20445.      insns in the loop, on GIMPLE and on RTL.

  20446.    * 'loop_exit_edge_p': Tests whether edge is an exit from a loop.

  20447.    * 'mark_loop_exit_edges': Marks all exit edges of all loops with

  20448.      'EDGE_LOOP_EXIT' flag.

  20449.    * 'get_loop_body', 'get_loop_body_in_dom_order',

  20450.      'get_loop_body_in_bfs_order': Enumerates the basic blocks in the

  20451.      loop in depth-first search order in reversed CFG, ordered by

  20452.      dominance relation, and breath-first search order, respectively.

  20453.    * 'single_exit': Returns the single exit edge of the loop, or 'NULL'

  20454.      if the loop has more than one exit.  You can only use this function

  20455.      if LOOPS_HAVE_MARKED_SINGLE_EXITS property is used.

  20456.    * 'get_loop_exit_edges': Enumerates the exit edges of a loop.

  20457.    * 'just_once_each_iteration_p': Returns true if the basic block is

  20458.      executed exactly once during each iteration of a loop (that is, it

  20459.      does not belong to a sub-loop, and it dominates the latch of the

  20460.      loop).

  20461. 

  20462. 

  20463. File: gccint.info,  Node: Loop manipulation,  Next: LCSSA,  Prev: Loop querying,  Up: Loop Analysis and Representation

  20464. 

  20465. 16.3 Loop manipulation

  20466. ======================

  20467. 

  20468. The loops tree can be manipulated using the following functions:

  20469. 

  20470.    * 'flow_loop_tree_node_add': Adds a node to the tree.

  20471.    * 'flow_loop_tree_node_remove': Removes a node from the tree.

  20472.    * 'add_bb_to_loop': Adds a basic block to a loop.

  20473.    * 'remove_bb_from_loops': Removes a basic block from loops.

  20474. 

  20475.  Most low-level CFG functions update loops automatically.  The following

  20476. functions handle some more complicated cases of CFG manipulations:

  20477. 

  20478.    * 'remove_path': Removes an edge and all blocks it dominates.

  20479.    * 'split_loop_exit_edge': Splits exit edge of the loop, ensuring that

  20480.      PHI node arguments remain in the loop (this ensures that

  20481.      loop-closed SSA form is preserved).  Only useful on GIMPLE.

  20482. 

  20483.  Finally, there are some higher-level loop transformations implemented.

  20484. While some of them are written so that they should work on non-innermost

  20485. loops, they are mostly untested in that case, and at the moment, they

  20486. are only reliable for the innermost loops:

  20487. 

  20488.    * 'create_iv': Creates a new induction variable.  Only works on

  20489.      GIMPLE.  'standard_iv_increment_position' can be used to find a

  20490.      suitable place for the iv increment.

  20491.    * 'duplicate_loop_to_header_edge',

  20492.      'tree_duplicate_loop_to_header_edge': These functions (on RTL and

  20493.      on GIMPLE) duplicate the body of the loop prescribed number of

  20494.      times on one of the edges entering loop header, thus performing

  20495.      either loop unrolling or loop peeling.  'can_duplicate_loop_p'

  20496.      ('can_unroll_loop_p' on GIMPLE) must be true for the duplicated

  20497.      loop.

  20498.    * 'loop_version': This function creates a copy of a loop, and a

  20499.      branch before them that selects one of them depending on the

  20500.      prescribed condition.  This is useful for optimizations that need

  20501.      to verify some assumptions in runtime (one of the copies of the

  20502.      loop is usually left unchanged, while the other one is transformed

  20503.      in some way).

  20504.    * 'tree_unroll_loop': Unrolls the loop, including peeling the extra

  20505.      iterations to make the number of iterations divisible by unroll

  20506.      factor, updating the exit condition, and removing the exits that

  20507.      now cannot be taken.  Works only on GIMPLE.

  20508. 

  20509. 

  20510. File: gccint.info,  Node: LCSSA,  Next: Scalar evolutions,  Prev: Loop manipulation,  Up: Loop Analysis and Representation

  20511. 

  20512. 16.4 Loop-closed SSA form

  20513. =========================

  20514. 

  20515. Throughout the loop optimizations on tree level, one extra condition is

  20516. enforced on the SSA form: No SSA name is used outside of the loop in

  20517. that it is defined.  The SSA form satisfying this condition is called

  20518. "loop-closed SSA form" - LCSSA.  To enforce LCSSA, PHI nodes must be

  20519. created at the exits of the loops for the SSA names that are used

  20520. outside of them.  Only the real operands (not virtual SSA names) are

  20521. held in LCSSA, in order to save memory.

  20522. 

  20523.  There are various benefits of LCSSA:

  20524. 

  20525.    * Many optimizations (value range analysis, final value replacement)

  20526.      are interested in the values that are defined in the loop and used

  20527.      outside of it, i.e., exactly those for that we create new PHI

  20528.      nodes.

  20529.    * In induction variable analysis, it is not necessary to specify the

  20530.      loop in that the analysis should be performed - the scalar

  20531.      evolution analysis always returns the results with respect to the

  20532.      loop in that the SSA name is defined.

  20533.    * It makes updating of SSA form during loop transformations simpler.

  20534.      Without LCSSA, operations like loop unrolling may force creation of

  20535.      PHI nodes arbitrarily far from the loop, while in LCSSA, the SSA

  20536.      form can be updated locally.  However, since we only keep real

  20537.      operands in LCSSA, we cannot use this advantage (we could have

  20538.      local updating of real operands, but it is not much more efficient

  20539.      than to use generic SSA form updating for it as well; the amount of

  20540.      changes to SSA is the same).

  20541. 

  20542.  However, it also means LCSSA must be updated.  This is usually

  20543. straightforward, unless you create a new value in loop and use it

  20544. outside, or unless you manipulate loop exit edges (functions are

  20545. provided to make these manipulations simple).

  20546. 'rewrite_into_loop_closed_ssa' is used to rewrite SSA form to LCSSA, and

  20547. 'verify_loop_closed_ssa' to check that the invariant of LCSSA is

  20548. preserved.

  20549. 

  20550. 

  20551. File: gccint.info,  Node: Scalar evolutions,  Next: loop-iv,  Prev: LCSSA,  Up: Loop Analysis and Representation

  20552. 

  20553. 16.5 Scalar evolutions

  20554. ======================

  20555. 

  20556. Scalar evolutions (SCEV) are used to represent results of induction

  20557. variable analysis on GIMPLE.  They enable us to represent variables with

  20558. complicated behavior in a simple and consistent way (we only use it to

  20559. express values of polynomial induction variables, but it is possible to

  20560. extend it).  The interfaces to SCEV analysis are declared in

  20561. 'tree-scalar-evolution.h'.  To use scalar evolutions analysis,

  20562. 'scev_initialize' must be used.  To stop using SCEV, 'scev_finalize'

  20563. should be used.  SCEV analysis caches results in order to save time and

  20564. memory.  This cache however is made invalid by most of the loop

  20565. transformations, including removal of code.  If such a transformation is

  20566. performed, 'scev_reset' must be called to clean the caches.

  20567. 

  20568.  Given an SSA name, its behavior in loops can be analyzed using the

  20569. 'analyze_scalar_evolution' function.  The returned SCEV however does not

  20570. have to be fully analyzed and it may contain references to other SSA

  20571. names defined in the loop.  To resolve these (potentially recursive)

  20572. references, 'instantiate_parameters' or 'resolve_mixers' functions must

  20573. be used.  'instantiate_parameters' is useful when you use the results of

  20574. SCEV only for some analysis, and when you work with whole nest of loops

  20575. at once.  It will try replacing all SSA names by their SCEV in all

  20576. loops, including the super-loops of the current loop, thus providing a

  20577. complete information about the behavior of the variable in the loop

  20578. nest.  'resolve_mixers' is useful if you work with only one loop at a

  20579. time, and if you possibly need to create code based on the value of the

  20580. induction variable.  It will only resolve the SSA names defined in the

  20581. current loop, leaving the SSA names defined outside unchanged, even if

  20582. their evolution in the outer loops is known.

  20583. 

  20584.  The SCEV is a normal tree expression, except for the fact that it may

  20585. contain several special tree nodes.  One of them is 'SCEV_NOT_KNOWN',

  20586. used for SSA names whose value cannot be expressed.  The other one is

  20587. 'POLYNOMIAL_CHREC'.  Polynomial chrec has three arguments - base, step

  20588. and loop (both base and step may contain further polynomial chrecs).

  20589. Type of the expression and of base and step must be the same.  A

  20590. variable has evolution 'POLYNOMIAL_CHREC(base, step, loop)' if it is (in

  20591. the specified loop) equivalent to 'x_1' in the following example

  20592. 

  20593.      while (...)

  20594.        {

  20595.          x_1 = phi (base, x_2);

  20596.          x_2 = x_1 + step;

  20597.        }

  20598. 

  20599.  Note that this includes the language restrictions on the operations.

  20600. For example, if we compile C code and 'x' has signed type, then the

  20601. overflow in addition would cause undefined behavior, and we may assume

  20602. that this does not happen.  Hence, the value with this SCEV cannot

  20603. overflow (which restricts the number of iterations of such a loop).

  20604. 

  20605.  In many cases, one wants to restrict the attention just to affine

  20606. induction variables.  In this case, the extra expressive power of SCEV

  20607. is not useful, and may complicate the optimizations.  In this case,

  20608. 'simple_iv' function may be used to analyze a value - the result is a

  20609. loop-invariant base and step.

  20610. 

  20611. 

  20612. File: gccint.info,  Node: loop-iv,  Next: Number of iterations,  Prev: Scalar evolutions,  Up: Loop Analysis and Representation

  20613. 

  20614. 16.6 IV analysis on RTL

  20615. =======================

  20616. 

  20617. The induction variable on RTL is simple and only allows analysis of

  20618. affine induction variables, and only in one loop at once.  The interface

  20619. is declared in 'cfgloop.h'.  Before analyzing induction variables in a

  20620. loop L, 'iv_analysis_loop_init' function must be called on L. After the

  20621. analysis (possibly calling 'iv_analysis_loop_init' for several loops) is

  20622. finished, 'iv_analysis_done' should be called.  The following functions

  20623. can be used to access the results of the analysis:

  20624. 

  20625.    * 'iv_analyze': Analyzes a single register used in the given insn.

  20626.      If no use of the register in this insn is found, the following

  20627.      insns are scanned, so that this function can be called on the insn

  20628.      returned by get_condition.

  20629.    * 'iv_analyze_result': Analyzes result of the assignment in the given

  20630.      insn.

  20631.    * 'iv_analyze_expr': Analyzes a more complicated expression.  All its

  20632.      operands are analyzed by 'iv_analyze', and hence they must be used

  20633.      in the specified insn or one of the following insns.

  20634. 

  20635.  The description of the induction variable is provided in 'struct

  20636. rtx_iv'.  In order to handle subregs, the representation is a bit

  20637. complicated; if the value of the 'extend' field is not 'UNKNOWN', the

  20638. value of the induction variable in the i-th iteration is

  20639. 

  20640.      delta + mult * extend_{extend_mode} (subreg_{mode} (base + i * step)),

  20641. 

  20642.  with the following exception: if 'first_special' is true, then the

  20643. value in the first iteration (when 'i' is zero) is 'delta + mult *

  20644. base'.  However, if 'extend' is equal to 'UNKNOWN', then 'first_special'

  20645. must be false, 'delta' 0, 'mult' 1 and the value in the i-th iteration

  20646. is

  20647. 

  20648.      subreg_{mode} (base + i * step)

  20649. 

  20650.  The function 'get_iv_value' can be used to perform these calculations.

  20651. 

  20652. 

  20653. File: gccint.info,  Node: Number of iterations,  Next: Dependency analysis,  Prev: loop-iv,  Up: Loop Analysis and Representation

  20654. 

  20655. 16.7 Number of iterations analysis

  20656. ==================================

  20657. 

  20658. Both on GIMPLE and on RTL, there are functions available to determine

  20659. the number of iterations of a loop, with a similar interface.  The

  20660. number of iterations of a loop in GCC is defined as the number of

  20661. executions of the loop latch.  In many cases, it is not possible to

  20662. determine the number of iterations unconditionally - the determined

  20663. number is correct only if some assumptions are satisfied.  The analysis

  20664. tries to verify these conditions using the information contained in the

  20665. program; if it fails, the conditions are returned together with the

  20666. result.  The following information and conditions are provided by the

  20667. analysis:

  20668. 

  20669.    * 'assumptions': If this condition is false, the rest of the

  20670.      information is invalid.

  20671.    * 'noloop_assumptions' on RTL, 'may_be_zero' on GIMPLE: If this

  20672.      condition is true, the loop exits in the first iteration.

  20673.    * 'infinite': If this condition is true, the loop is infinite.  This

  20674.      condition is only available on RTL.  On GIMPLE, conditions for

  20675.      finiteness of the loop are included in 'assumptions'.

  20676.    * 'niter_expr' on RTL, 'niter' on GIMPLE: The expression that gives

  20677.      number of iterations.  The number of iterations is defined as the

  20678.      number of executions of the loop latch.

  20679. 

  20680.  Both on GIMPLE and on RTL, it necessary for the induction variable

  20681. analysis framework to be initialized (SCEV on GIMPLE, loop-iv on RTL).

  20682. On GIMPLE, the results are stored to 'struct tree_niter_desc' structure.

  20683. Number of iterations before the loop is exited through a given exit can

  20684. be determined using 'number_of_iterations_exit' function.  On RTL, the

  20685. results are returned in 'struct niter_desc' structure.  The

  20686. corresponding function is named 'check_simple_exit'.  There are also

  20687. functions that pass through all the exits of a loop and try to find one

  20688. with easy to determine number of iterations - 'find_loop_niter' on

  20689. GIMPLE and 'find_simple_exit' on RTL.  Finally, there are functions that

  20690. provide the same information, but additionally cache it, so that

  20691. repeated calls to number of iterations are not so costly -

  20692. 'number_of_latch_executions' on GIMPLE and 'get_simple_loop_desc' on

  20693. RTL.

  20694. 

  20695.  Note that some of these functions may behave slightly differently than

  20696. others - some of them return only the expression for the number of

  20697. iterations, and fail if there are some assumptions.  The function

  20698. 'number_of_latch_executions' works only for single-exit loops.  The

  20699. function 'number_of_cond_exit_executions' can be used to determine

  20700. number of executions of the exit condition of a single-exit loop (i.e.,

  20701. the 'number_of_latch_executions' increased by one).

  20702. 

  20703.  On GIMPLE, below constraint flags affect semantics of some APIs of

  20704. number of iterations analyzer:

  20705. 

  20706.    * 'LOOP_C_INFINITE': If this constraint flag is set, the loop is

  20707.      known to be infinite.  APIs like 'number_of_iterations_exit' can

  20708.      return false directly without doing any analysis.

  20709.    * 'LOOP_C_FINITE': If this constraint flag is set, the loop is known

  20710.      to be finite, in other words, loop's number of iterations can be

  20711.      computed with 'assumptions' be true.

  20712. 

  20713.  Generally, the constraint flags are set/cleared by consumers which are

  20714. loop optimizers.  It's also the consumers' responsibility to set/clear

  20715. constraints correctly.  Failing to do that might result in hard to track

  20716. down bugs in scev/niter consumers.  One typical use case is vectorizer:

  20717. it drives number of iterations analyzer by setting 'LOOP_C_FINITE' and

  20718. vectorizes possibly infinite loop by versioning loop with analysis

  20719. result.  In return, constraints set by consumers can also help number of

  20720. iterations analyzer in following optimizers.  For example, 'niter' of a

  20721. loop versioned under 'assumptions' is valid unconditionally.

  20722. 

  20723.  Other constraints may be added in the future, for example, a constraint

  20724. indicating that loops' latch must roll thus 'may_be_zero' would be false

  20725. unconditionally.

  20726. 

  20727. 

  20728. File: gccint.info,  Node: Dependency analysis,  Prev: Number of iterations,  Up: Loop Analysis and Representation

  20729. 

  20730. 16.8 Data Dependency Analysis

  20731. =============================

  20732. 

  20733. The code for the data dependence analysis can be found in

  20734. 'tree-data-ref.c' and its interface and data structures are described in

  20735. 'tree-data-ref.h'.  The function that computes the data dependences for

  20736. all the array and pointer references for a given loop is

  20737. 'compute_data_dependences_for_loop'.  This function is currently used by

  20738. the linear loop transform and the vectorization passes.  Before calling

  20739. this function, one has to allocate two vectors: a first vector will

  20740. contain the set of data references that are contained in the analyzed

  20741. loop body, and the second vector will contain the dependence relations

  20742. between the data references.  Thus if the vector of data references is

  20743. of size 'n', the vector containing the dependence relations will contain

  20744. 'n*n' elements.  However if the analyzed loop contains side effects,

  20745. such as calls that potentially can interfere with the data references in

  20746. the current analyzed loop, the analysis stops while scanning the loop

  20747. body for data references, and inserts a single 'chrec_dont_know' in the

  20748. dependence relation array.

  20749. 

  20750.  The data references are discovered in a particular order during the

  20751. scanning of the loop body: the loop body is analyzed in execution order,

  20752. and the data references of each statement are pushed at the end of the

  20753. data reference array.  Two data references syntactically occur in the

  20754. program in the same order as in the array of data references.  This

  20755. syntactic order is important in some classical data dependence tests,

  20756. and mapping this order to the elements of this array avoids costly

  20757. queries to the loop body representation.

  20758. 

  20759.  Three types of data references are currently handled: ARRAY_REF,

  20760. INDIRECT_REF and COMPONENT_REF.  The data structure for the data

  20761. reference is 'data_reference', where 'data_reference_p' is a name of a

  20762. pointer to the data reference structure.  The structure contains the

  20763. following elements:

  20764. 

  20765.    * 'base_object_info': Provides information about the base object of

  20766.      the data reference and its access functions.  These access

  20767.      functions represent the evolution of the data reference in the loop

  20768.      relative to its base, in keeping with the classical meaning of the

  20769.      data reference access function for the support of arrays.  For

  20770.      example, for a reference 'a.b[i][j]', the base object is 'a.b' and

  20771.      the access functions, one for each array subscript, are: '{i_init,

  20772.      + i_step}_1, {j_init, +, j_step}_2'.

  20773. 

  20774.    * 'first_location_in_loop': Provides information about the first

  20775.      location accessed by the data reference in the loop and about the

  20776.      access function used to represent evolution relative to this

  20777.      location.  This data is used to support pointers, and is not used

  20778.      for arrays (for which we have base objects).  Pointer accesses are

  20779.      represented as a one-dimensional access that starts from the first

  20780.      location accessed in the loop.  For example:

  20781. 

  20782.                 for1 i

  20783.                    for2 j

  20784.                     *((int *)p + i + j) = a[i][j];

  20785. 

  20786.      The access function of the pointer access is '{0, + 4B}_for2'

  20787.      relative to 'p + i'.  The access functions of the array are

  20788.      '{i_init, + i_step}_for1' and '{j_init, +, j_step}_for2' relative

  20789.      to 'a'.

  20790. 

  20791.      Usually, the object the pointer refers to is either unknown, or we

  20792.      cannot prove that the access is confined to the boundaries of a

  20793.      certain object.

  20794. 

  20795.      Two data references can be compared only if at least one of these

  20796.      two representations has all its fields filled for both data

  20797.      references.

  20798. 

  20799.      The current strategy for data dependence tests is as follows: If

  20800.      both 'a' and 'b' are represented as arrays, compare 'a.base_object'

  20801.      and 'b.base_object'; if they are equal, apply dependence tests (use

  20802.      access functions based on base_objects).  Else if both 'a' and 'b'

  20803.      are represented as pointers, compare 'a.first_location' and

  20804.      'b.first_location'; if they are equal, apply dependence tests (use

  20805.      access functions based on first location).  However, if 'a' and 'b'

  20806.      are represented differently, only try to prove that the bases are

  20807.      definitely different.

  20808. 

  20809.    * Aliasing information.

  20810.    * Alignment information.

  20811. 

  20812.  The structure describing the relation between two data references is

  20813. 'data_dependence_relation' and the shorter name for a pointer to such a

  20814. structure is 'ddr_p'.  This structure contains:

  20815. 

  20816.    * a pointer to each data reference,

  20817.    * a tree node 'are_dependent' that is set to 'chrec_known' if the

  20818.      analysis has proved that there is no dependence between these two

  20819.      data references, 'chrec_dont_know' if the analysis was not able to

  20820.      determine any useful result and potentially there could exist a

  20821.      dependence between these data references, and 'are_dependent' is

  20822.      set to 'NULL_TREE' if there exist a dependence relation between the

  20823.      data references, and the description of this dependence relation is

  20824.      given in the 'subscripts', 'dir_vects', and 'dist_vects' arrays,

  20825.    * a boolean that determines whether the dependence relation can be

  20826.      represented by a classical distance vector,

  20827.    * an array 'subscripts' that contains a description of each subscript

  20828.      of the data references.  Given two array accesses a subscript is

  20829.      the tuple composed of the access functions for a given dimension.

  20830.      For example, given 'A[f1][f2][f3]' and 'B[g1][g2][g3]', there are

  20831.      three subscripts: '(f1, g1), (f2, g2), (f3, g3)'.

  20832.    * two arrays 'dir_vects' and 'dist_vects' that contain classical

  20833.      representations of the data dependences under the form of direction

  20834.      and distance dependence vectors,

  20835.    * an array of loops 'loop_nest' that contains the loops to which the

  20836.      distance and direction vectors refer to.

  20837. 

  20838.  Several functions for pretty printing the information extracted by the

  20839. data dependence analysis are available: 'dump_ddrs' prints with a

  20840. maximum verbosity the details of a data dependence relations array,

  20841. 'dump_dist_dir_vectors' prints only the classical distance and direction

  20842. vectors for a data dependence relations array, and

  20843. 'dump_data_references' prints the details of the data references

  20844. contained in a data reference array.

  20845. 

  20846. 

  20847. File: gccint.info,  Node: Machine Desc,  Next: Target Macros,  Prev: Loop Analysis and Representation,  Up: Top

  20848. 

  20849. 17 Machine Descriptions

  20850. ***********************

  20851. 

  20852. A machine description has two parts: a file of instruction patterns

  20853. ('.md' file) and a C header file of macro definitions.

  20854. 

  20855.  The '.md' file for a target machine contains a pattern for each

  20856. instruction that the target machine supports (or at least each

  20857. instruction that is worth telling the compiler about).  It may also

  20858. contain comments.  A semicolon causes the rest of the line to be a

  20859. comment, unless the semicolon is inside a quoted string.

  20860. 

  20861.  See the next chapter for information on the C header file.

  20862. 

  20863. * Menu:

  20864. 

  20865. * Overview::            How the machine description is used.

  20866. * Patterns::            How to write instruction patterns.

  20867. * Example::             An explained example of a 'define_insn' pattern.

  20868. * RTL Template::        The RTL template defines what insns match a pattern.

  20869. * Output Template::     The output template says how to make assembler code

  20870.                         from such an insn.

  20871. * Output Statement::    For more generality, write C code to output

  20872.                         the assembler code.

  20873. * Predicates::          Controlling what kinds of operands can be used

  20874.                         for an insn.

  20875. * Constraints::         Fine-tuning operand selection.

  20876. * Standard Names::      Names mark patterns to use for code generation.

  20877. * Pattern Ordering::    When the order of patterns makes a difference.

  20878. * Dependent Patterns::  Having one pattern may make you need another.

  20879. * Jump Patterns::       Special considerations for patterns for jump insns.

  20880. * Looping Patterns::    How to define patterns for special looping insns.

  20881. * Insn Canonicalizations::Canonicalization of Instructions

  20882. * Expander Definitions::Generating a sequence of several RTL insns

  20883.                         for a standard operation.

  20884. * Insn Splitting::      Splitting Instructions into Multiple Instructions.

  20885. * Including Patterns::  Including Patterns in Machine Descriptions.

  20886. * Peephole Definitions::Defining machine-specific peephole optimizations.

  20887. * Insn Attributes::     Specifying the value of attributes for generated insns.

  20888. * Conditional Execution::Generating 'define_insn' patterns for

  20889.                          predication.

  20890. * Define Subst::	Generating 'define_insn' and 'define_expand'

  20891. 			patterns from other patterns.

  20892. * Constant Definitions::Defining symbolic constants that can be used in the

  20893.                         md file.

  20894. * Iterators::           Using iterators to generate patterns from a template.

  20895. 

  20896. 

  20897. File: gccint.info,  Node: Overview,  Next: Patterns,  Up: Machine Desc

  20898. 

  20899. 17.1 Overview of How the Machine Description is Used

  20900. ====================================================

  20901. 

  20902. There are three main conversions that happen in the compiler:

  20903. 

  20904.   1. The front end reads the source code and builds a parse tree.

  20905. 

  20906.   2. The parse tree is used to generate an RTL insn list based on named

  20907.      instruction patterns.

  20908. 

  20909.   3. The insn list is matched against the RTL templates to produce

  20910.      assembler code.

  20911. 

  20912.  For the generate pass, only the names of the insns matter, from either

  20913. a named 'define_insn' or a 'define_expand'.  The compiler will choose

  20914. the pattern with the right name and apply the operands according to the

  20915. documentation later in this chapter, without regard for the RTL template

  20916. or operand constraints.  Note that the names the compiler looks for are

  20917. hard-coded in the compiler--it will ignore unnamed patterns and patterns

  20918. with names it doesn't know about, but if you don't provide a named

  20919. pattern it needs, it will abort.

  20920. 

  20921.  If a 'define_insn' is used, the template given is inserted into the

  20922. insn list.  If a 'define_expand' is used, one of three things happens,

  20923. based on the condition logic.  The condition logic may manually create

  20924. new insns for the insn list, say via 'emit_insn()', and invoke 'DONE'.

  20925. For certain named patterns, it may invoke 'FAIL' to tell the compiler to

  20926. use an alternate way of performing that task.  If it invokes neither

  20927. 'DONE' nor 'FAIL', the template given in the pattern is inserted, as if

  20928. the 'define_expand' were a 'define_insn'.

  20929. 

  20930.  Once the insn list is generated, various optimization passes convert,

  20931. replace, and rearrange the insns in the insn list.  This is where the

  20932. 'define_split' and 'define_peephole' patterns get used, for example.

  20933. 

  20934.  Finally, the insn list's RTL is matched up with the RTL templates in

  20935. the 'define_insn' patterns, and those patterns are used to emit the

  20936. final assembly code.  For this purpose, each named 'define_insn' acts

  20937. like it's unnamed, since the names are ignored.

  20938. 

  20939. 

  20940. File: gccint.info,  Node: Patterns,  Next: Example,  Prev: Overview,  Up: Machine Desc

  20941. 

  20942. 17.2 Everything about Instruction Patterns

  20943. ==========================================

  20944. 

  20945. A 'define_insn' expression is used to define instruction patterns to

  20946. which insns may be matched.  A 'define_insn' expression contains an

  20947. incomplete RTL expression, with pieces to be filled in later, operand

  20948. constraints that restrict how the pieces can be filled in, and an output

  20949. template or C code to generate the assembler output.

  20950. 

  20951.  A 'define_insn' is an RTL expression containing four or five operands:

  20952. 

  20953.   1. An optional name N.  When a name is present, the compiler

  20954.      automically generates a C++ function 'gen_N' that takes the

  20955.      operands of the instruction as arguments and returns the

  20956.      instruction's rtx pattern.  The compiler also assigns the

  20957.      instruction a unique code 'CODE_FOR_N', with all such codes

  20958.      belonging to an enum called 'insn_code'.

  20959. 

  20960.      These names serve one of two purposes.  The first is to indicate

  20961.      that the instruction performs a certain standard job for the

  20962.      RTL-generation pass of the compiler, such as a move, an addition,

  20963.      or a conditional jump.  The second is to help the target generate

  20964.      certain target-specific operations, such as when implementing

  20965.      target-specific intrinsic functions.

  20966. 

  20967.      It is better to prefix target-specific names with the name of the

  20968.      target, to avoid any clash with current or future standard names.

  20969. 

  20970.      The absence of a name is indicated by writing an empty string where

  20971.      the name should go.  Nameless instruction patterns are never used

  20972.      for generating RTL code, but they may permit several simpler insns

  20973.      to be combined later on.

  20974. 

  20975.      For the purpose of debugging the compiler, you may also specify a

  20976.      name beginning with the '*' character.  Such a name is used only

  20977.      for identifying the instruction in RTL dumps; it is equivalent to

  20978.      having a nameless pattern for all other purposes.  Names beginning

  20979.      with the '*' character are not required to be unique.

  20980. 

  20981.      The name may also have the form '@N'.  This has the same effect as

  20982.      a name 'N', but in addition tells the compiler to generate further

  20983.      helper functions; see *note Parameterized Names:: for details.

  20984. 

  20985.   2. The "RTL template": This is a vector of incomplete RTL expressions

  20986.      which describe the semantics of the instruction (*note RTL

  20987.      Template::).  It is incomplete because it may contain

  20988.      'match_operand', 'match_operator', and 'match_dup' expressions that

  20989.      stand for operands of the instruction.

  20990. 

  20991.      If the vector has multiple elements, the RTL template is treated as

  20992.      a 'parallel' expression.

  20993. 

  20994.   3. The condition: This is a string which contains a C expression.

  20995.      When the compiler attempts to match RTL against a pattern, the

  20996.      condition is evaluated.  If the condition evaluates to 'true', the

  20997.      match is permitted.  The condition may be an empty string, which is

  20998.      treated as always 'true'.

  20999. 

  21000.      For a named pattern, the condition may not depend on the data in

  21001.      the insn being matched, but only the target-machine-type flags.

  21002.      The compiler needs to test these conditions during initialization

  21003.      in order to learn exactly which named instructions are available in

  21004.      a particular run.

  21005. 

  21006.      For nameless patterns, the condition is applied only when matching

  21007.      an individual insn, and only after the insn has matched the

  21008.      pattern's recognition template.  The insn's operands may be found

  21009.      in the vector 'operands'.

  21010. 

  21011.      An instruction condition cannot become more restrictive as

  21012.      compilation progresses.  If the condition accepts a particular RTL

  21013.      instruction at one stage of compilation, it must continue to accept

  21014.      that instruction until the final pass.  For example,

  21015.      '!reload_completed' and 'can_create_pseudo_p ()' are both invalid

  21016.      instruction conditions, because they are true during the earlier

  21017.      RTL passes and false during the later ones.  For the same reason,

  21018.      if a condition accepts an instruction before register allocation,

  21019.      it cannot later try to control register allocation by excluding

  21020.      certain register or value combinations.

  21021. 

  21022.      Although a condition cannot become more restrictive as compilation

  21023.      progresses, the condition for a nameless pattern _can_ become more

  21024.      permissive.  For example, a nameless instruction can require

  21025.      'reload_completed' to be true, in which case it only matches after

  21026.      register allocation.

  21027. 

  21028.   4. The "output template" or "output statement": This is either a

  21029.      string, or a fragment of C code which returns a string.

  21030. 

  21031.      When simple substitution isn't general enough, you can specify a

  21032.      piece of C code to compute the output.  *Note Output Statement::.

  21033. 

  21034.   5. The "insn attributes": This is an optional vector containing the

  21035.      values of attributes for insns matching this pattern (*note Insn

  21036.      Attributes::).

  21037. 

  21038. 

  21039. File: gccint.info,  Node: Example,  Next: RTL Template,  Prev: Patterns,  Up: Machine Desc

  21040. 

  21041. 17.3 Example of 'define_insn'

  21042. =============================

  21043. 

  21044. Here is an example of an instruction pattern, taken from the machine

  21045. description for the 68000/68020.

  21046. 

  21047.      (define_insn "tstsi"

  21048.        [(set (cc0)

  21049.              (match_operand:SI 0 "general_operand" "rm"))]

  21050.        ""

  21051.        "*

  21052.      {

  21053.        if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))

  21054.          return \"tstl %0\";

  21055.        return \"cmpl #0,%0\";

  21056.      }")

  21057. 

  21058. This can also be written using braced strings:

  21059. 

  21060.      (define_insn "tstsi"

  21061.        [(set (cc0)

  21062.              (match_operand:SI 0 "general_operand" "rm"))]

  21063.        ""

  21064.      {

  21065.        if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))

  21066.          return "tstl %0";

  21067.        return "cmpl #0,%0";

  21068.      })

  21069. 

  21070.  This describes an instruction which sets the condition codes based on

  21071. the value of a general operand.  It has no condition, so any insn with

  21072. an RTL description of the form shown may be matched to this pattern.

  21073. The name 'tstsi' means "test a 'SImode' value" and tells the RTL

  21074. generation pass that, when it is necessary to test such a value, an insn

  21075. to do so can be constructed using this pattern.

  21076. 

  21077.  The output control string is a piece of C code which chooses which

  21078. output template to return based on the kind of operand and the specific

  21079. type of CPU for which code is being generated.

  21080. 

  21081.  '"rm"' is an operand constraint.  Its meaning is explained below.

  21082. 

  21083. 

  21084. File: gccint.info,  Node: RTL Template,  Next: Output Template,  Prev: Example,  Up: Machine Desc

  21085. 

  21086. 17.4 RTL Template

  21087. =================

  21088. 

  21089. The RTL template is used to define which insns match the particular

  21090. pattern and how to find their operands.  For named patterns, the RTL

  21091. template also says how to construct an insn from specified operands.

  21092. 

  21093.  Construction involves substituting specified operands into a copy of

  21094. the template.  Matching involves determining the values that serve as

  21095. the operands in the insn being matched.  Both of these activities are

  21096. controlled by special expression types that direct matching and

  21097. substitution of the operands.

  21098. 

  21099. '(match_operand:M N PREDICATE CONSTRAINT)'

  21100.      This expression is a placeholder for operand number N of the insn.

  21101.      When constructing an insn, operand number N will be substituted at

  21102.      this point.  When matching an insn, whatever appears at this

  21103.      position in the insn will be taken as operand number N; but it must

  21104.      satisfy PREDICATE or this instruction pattern will not match at

  21105.      all.

  21106. 

  21107.      Operand numbers must be chosen consecutively counting from zero in

  21108.      each instruction pattern.  There may be only one 'match_operand'

  21109.      expression in the pattern for each operand number.  Usually

  21110.      operands are numbered in the order of appearance in 'match_operand'

  21111.      expressions.  In the case of a 'define_expand', any operand numbers

  21112.      used only in 'match_dup' expressions have higher values than all

  21113.      other operand numbers.

  21114. 

  21115.      PREDICATE is a string that is the name of a function that accepts

  21116.      two arguments, an expression and a machine mode.  *Note

  21117.      Predicates::.  During matching, the function will be called with

  21118.      the putative operand as the expression and M as the mode argument

  21119.      (if M is not specified, 'VOIDmode' will be used, which normally

  21120.      causes PREDICATE to accept any mode).  If it returns zero, this

  21121.      instruction pattern fails to match.  PREDICATE may be an empty

  21122.      string; then it means no test is to be done on the operand, so

  21123.      anything which occurs in this position is valid.

  21124. 

  21125.      Most of the time, PREDICATE will reject modes other than M--but not

  21126.      always.  For example, the predicate 'address_operand' uses M as the

  21127.      mode of memory ref that the address should be valid for.  Many

  21128.      predicates accept 'const_int' nodes even though their mode is

  21129.      'VOIDmode'.

  21130. 

  21131.      CONSTRAINT controls reloading and the choice of the best register

  21132.      class to use for a value, as explained later (*note Constraints::).

  21133.      If the constraint would be an empty string, it can be omitted.

  21134. 

  21135.      People are often unclear on the difference between the constraint

  21136.      and the predicate.  The predicate helps decide whether a given insn

  21137.      matches the pattern.  The constraint plays no role in this

  21138.      decision; instead, it controls various decisions in the case of an

  21139.      insn which does match.

  21140. 

  21141. '(match_scratch:M N CONSTRAINT)'

  21142.      This expression is also a placeholder for operand number N and

  21143.      indicates that operand must be a 'scratch' or 'reg' expression.

  21144. 

  21145.      When matching patterns, this is equivalent to

  21146. 

  21147.           (match_operand:M N "scratch_operand" CONSTRAINT)

  21148. 

  21149.      but, when generating RTL, it produces a ('scratch':M) expression.

  21150. 

  21151.      If the last few expressions in a 'parallel' are 'clobber'

  21152.      expressions whose operands are either a hard register or

  21153.      'match_scratch', the combiner can add or delete them when

  21154.      necessary.  *Note Side Effects::.

  21155. 

  21156. '(match_dup N)'

  21157.      This expression is also a placeholder for operand number N.  It is

  21158.      used when the operand needs to appear more than once in the insn.

  21159. 

  21160.      In construction, 'match_dup' acts just like 'match_operand': the

  21161.      operand is substituted into the insn being constructed.  But in

  21162.      matching, 'match_dup' behaves differently.  It assumes that operand

  21163.      number N has already been determined by a 'match_operand' appearing

  21164.      earlier in the recognition template, and it matches only an

  21165.      identical-looking expression.

  21166. 

  21167.      Note that 'match_dup' should not be used to tell the compiler that

  21168.      a particular register is being used for two operands (example:

  21169.      'add' that adds one register to another; the second register is

  21170.      both an input operand and the output operand).  Use a matching

  21171.      constraint (*note Simple Constraints::) for those.  'match_dup' is

  21172.      for the cases where one operand is used in two places in the

  21173.      template, such as an instruction that computes both a quotient and

  21174.      a remainder, where the opcode takes two input operands but the RTL

  21175.      template has to refer to each of those twice; once for the quotient

  21176.      pattern and once for the remainder pattern.

  21177. 

  21178. '(match_operator:M N PREDICATE [OPERANDS...])'

  21179.      This pattern is a kind of placeholder for a variable RTL expression

  21180.      code.

  21181. 

  21182.      When constructing an insn, it stands for an RTL expression whose

  21183.      expression code is taken from that of operand N, and whose operands

  21184.      are constructed from the patterns OPERANDS.

  21185. 

  21186.      When matching an expression, it matches an expression if the

  21187.      function PREDICATE returns nonzero on that expression _and_ the

  21188.      patterns OPERANDS match the operands of the expression.

  21189. 

  21190.      Suppose that the function 'commutative_operator' is defined as

  21191.      follows, to match any expression whose operator is one of the

  21192.      commutative arithmetic operators of RTL and whose mode is MODE:

  21193. 

  21194.           int

  21195.           commutative_integer_operator (x, mode)

  21196.                rtx x;

  21197.                machine_mode mode;

  21198.           {

  21199.             enum rtx_code code = GET_CODE (x);

  21200.             if (GET_MODE (x) != mode)

  21201.               return 0;

  21202.             return (GET_RTX_CLASS (code) == RTX_COMM_ARITH

  21203.                     || code == EQ || code == NE);

  21204.           }

  21205. 

  21206.      Then the following pattern will match any RTL expression consisting

  21207.      of a commutative operator applied to two general operands:

  21208. 

  21209.           (match_operator:SI 3 "commutative_operator"

  21210.             [(match_operand:SI 1 "general_operand" "g")

  21211.              (match_operand:SI 2 "general_operand" "g")])

  21212. 

  21213.      Here the vector '[OPERANDS...]' contains two patterns because the

  21214.      expressions to be matched all contain two operands.

  21215. 

  21216.      When this pattern does match, the two operands of the commutative

  21217.      operator are recorded as operands 1 and 2 of the insn.  (This is

  21218.      done by the two instances of 'match_operand'.)  Operand 3 of the

  21219.      insn will be the entire commutative expression: use 'GET_CODE

  21220.      (operands[3])' to see which commutative operator was used.

  21221. 

  21222.      The machine mode M of 'match_operator' works like that of

  21223.      'match_operand': it is passed as the second argument to the

  21224.      predicate function, and that function is solely responsible for

  21225.      deciding whether the expression to be matched "has" that mode.

  21226. 

  21227.      When constructing an insn, argument 3 of the gen-function will

  21228.      specify the operation (i.e. the expression code) for the expression

  21229.      to be made.  It should be an RTL expression, whose expression code

  21230.      is copied into a new expression whose operands are arguments 1 and

  21231.      2 of the gen-function.  The subexpressions of argument 3 are not

  21232.      used; only its expression code matters.

  21233. 

  21234.      When 'match_operator' is used in a pattern for matching an insn, it

  21235.      usually best if the operand number of the 'match_operator' is

  21236.      higher than that of the actual operands of the insn.  This improves

  21237.      register allocation because the register allocator often looks at

  21238.      operands 1 and 2 of insns to see if it can do register tying.

  21239. 

  21240.      There is no way to specify constraints in 'match_operator'.  The

  21241.      operand of the insn which corresponds to the 'match_operator' never

  21242.      has any constraints because it is never reloaded as a whole.

  21243.      However, if parts of its OPERANDS are matched by 'match_operand'

  21244.      patterns, those parts may have constraints of their own.

  21245. 

  21246. '(match_op_dup:M N[OPERANDS...])'

  21247.      Like 'match_dup', except that it applies to operators instead of

  21248.      operands.  When constructing an insn, operand number N will be

  21249.      substituted at this point.  But in matching, 'match_op_dup' behaves

  21250.      differently.  It assumes that operand number N has already been

  21251.      determined by a 'match_operator' appearing earlier in the

  21252.      recognition template, and it matches only an identical-looking

  21253.      expression.

  21254. 

  21255. '(match_parallel N PREDICATE [SUBPAT...])'

  21256.      This pattern is a placeholder for an insn that consists of a

  21257.      'parallel' expression with a variable number of elements.  This

  21258.      expression should only appear at the top level of an insn pattern.

  21259. 

  21260.      When constructing an insn, operand number N will be substituted at

  21261.      this point.  When matching an insn, it matches if the body of the

  21262.      insn is a 'parallel' expression with at least as many elements as

  21263.      the vector of SUBPAT expressions in the 'match_parallel', if each

  21264.      SUBPAT matches the corresponding element of the 'parallel', _and_

  21265.      the function PREDICATE returns nonzero on the 'parallel' that is

  21266.      the body of the insn.  It is the responsibility of the predicate to

  21267.      validate elements of the 'parallel' beyond those listed in the

  21268.      'match_parallel'.

  21269. 

  21270.      A typical use of 'match_parallel' is to match load and store

  21271.      multiple expressions, which can contain a variable number of

  21272.      elements in a 'parallel'.  For example,

  21273. 

  21274.           (define_insn ""

  21275.             [(match_parallel 0 "load_multiple_operation"

  21276.                [(set (match_operand:SI 1 "gpc_reg_operand" "=r")

  21277.                      (match_operand:SI 2 "memory_operand" "m"))

  21278.                 (use (reg:SI 179))

  21279.                 (clobber (reg:SI 179))])]

  21280.             ""

  21281.             "loadm 0,0,%1,%2")

  21282. 

  21283.      This example comes from 'a29k.md'.  The function

  21284.      'load_multiple_operation' is defined in 'a29k.c' and checks that

  21285.      subsequent elements in the 'parallel' are the same as the 'set' in

  21286.      the pattern, except that they are referencing subsequent registers

  21287.      and memory locations.

  21288. 

  21289.      An insn that matches this pattern might look like:

  21290. 

  21291.           (parallel

  21292.            [(set (reg:SI 20) (mem:SI (reg:SI 100)))

  21293.             (use (reg:SI 179))

  21294.             (clobber (reg:SI 179))

  21295.             (set (reg:SI 21)

  21296.                  (mem:SI (plus:SI (reg:SI 100)

  21297.                                   (const_int 4))))

  21298.             (set (reg:SI 22)

  21299.                  (mem:SI (plus:SI (reg:SI 100)

  21300.                                   (const_int 8))))])

  21301. 

  21302. '(match_par_dup N [SUBPAT...])'

  21303.      Like 'match_op_dup', but for 'match_parallel' instead of

  21304.      'match_operator'.

  21305. 

  21306. 

  21307. File: gccint.info,  Node: Output Template,  Next: Output Statement,  Prev: RTL Template,  Up: Machine Desc

  21308. 

  21309. 17.5 Output Templates and Operand Substitution

  21310. ==============================================

  21311. 

  21312. The "output template" is a string which specifies how to output the

  21313. assembler code for an instruction pattern.  Most of the template is a

  21314. fixed string which is output literally.  The character '%' is used to

  21315. specify where to substitute an operand; it can also be used to identify

  21316. places where different variants of the assembler require different

  21317. syntax.

  21318. 

  21319.  In the simplest case, a '%' followed by a digit N says to output

  21320. operand N at that point in the string.

  21321. 

  21322.  '%' followed by a letter and a digit says to output an operand in an

  21323. alternate fashion.  Four letters have standard, built-in meanings

  21324. described below.  The machine description macro 'PRINT_OPERAND' can

  21325. define additional letters with nonstandard meanings.

  21326. 

  21327.  '%cDIGIT' can be used to substitute an operand that is a constant value

  21328. without the syntax that normally indicates an immediate operand.

  21329. 

  21330.  '%nDIGIT' is like '%cDIGIT' except that the value of the constant is

  21331. negated before printing.

  21332. 

  21333.  '%aDIGIT' can be used to substitute an operand as if it were a memory

  21334. reference, with the actual operand treated as the address.  This may be

  21335. useful when outputting a "load address" instruction, because often the

  21336. assembler syntax for such an instruction requires you to write the

  21337. operand as if it were a memory reference.

  21338. 

  21339.  '%lDIGIT' is used to substitute a 'label_ref' into a jump instruction.

  21340. 

  21341.  '%=' outputs a number which is unique to each instruction in the entire

  21342. compilation.  This is useful for making local labels to be referred to

  21343. more than once in a single template that generates multiple assembler

  21344. instructions.

  21345. 

  21346.  '%' followed by a punctuation character specifies a substitution that

  21347. does not use an operand.  Only one case is standard: '%%' outputs a '%'

  21348. into the assembler code.  Other nonstandard cases can be defined in the

  21349. 'PRINT_OPERAND' macro.  You must also define which punctuation

  21350. characters are valid with the 'PRINT_OPERAND_PUNCT_VALID_P' macro.

  21351. 

  21352.  The template may generate multiple assembler instructions.  Write the

  21353. text for the instructions, with '\;' between them.

  21354. 

  21355.  When the RTL contains two operands which are required by constraint to

  21356. match each other, the output template must refer only to the

  21357. lower-numbered operand.  Matching operands are not always identical, and

  21358. the rest of the compiler arranges to put the proper RTL expression for

  21359. printing into the lower-numbered operand.

  21360. 

  21361.  One use of nonstandard letters or punctuation following '%' is to

  21362. distinguish between different assembler languages for the same machine;

  21363. for example, Motorola syntax versus MIT syntax for the 68000.  Motorola

  21364. syntax requires periods in most opcode names, while MIT syntax does not.

  21365. For example, the opcode 'movel' in MIT syntax is 'move.l' in Motorola

  21366. syntax.  The same file of patterns is used for both kinds of output

  21367. syntax, but the character sequence '%.' is used in each place where

  21368. Motorola syntax wants a period.  The 'PRINT_OPERAND' macro for Motorola

  21369. syntax defines the sequence to output a period; the macro for MIT syntax

  21370. defines it to do nothing.

  21371. 

  21372.  As a special case, a template consisting of the single character '#'

  21373. instructs the compiler to first split the insn, and then output the

  21374. resulting instructions separately.  This helps eliminate redundancy in

  21375. the output templates.  If you have a 'define_insn' that needs to emit

  21376. multiple assembler instructions, and there is a matching 'define_split'

  21377. already defined, then you can simply use '#' as the output template

  21378. instead of writing an output template that emits the multiple assembler

  21379. instructions.

  21380. 

  21381.  Note that '#' only has an effect while generating assembly code; it

  21382. does not affect whether a split occurs earlier.  An associated

  21383. 'define_split' must exist and it must be suitable for use after register

  21384. allocation.

  21385. 

  21386.  If the macro 'ASSEMBLER_DIALECT' is defined, you can use construct of

  21387. the form '{option0|option1|option2}' in the templates.  These describe

  21388. multiple variants of assembler language syntax.  *Note Instruction

  21389. Output::.

  21390. 

  21391. 

  21392. File: gccint.info,  Node: Output Statement,  Next: Predicates,  Prev: Output Template,  Up: Machine Desc

  21393. 

  21394. 17.6 C Statements for Assembler Output

  21395. ======================================

  21396. 

  21397. Often a single fixed template string cannot produce correct and

  21398. efficient assembler code for all the cases that are recognized by a

  21399. single instruction pattern.  For example, the opcodes may depend on the

  21400. kinds of operands; or some unfortunate combinations of operands may

  21401. require extra machine instructions.

  21402. 

  21403.  If the output control string starts with a '@', then it is actually a

  21404. series of templates, each on a separate line.  (Blank lines and leading

  21405. spaces and tabs are ignored.)  The templates correspond to the pattern's

  21406. constraint alternatives (*note Multi-Alternative::).  For example, if a

  21407. target machine has a two-address add instruction 'addr' to add into a

  21408. register and another 'addm' to add a register to memory, you might write

  21409. this pattern:

  21410. 

  21411.      (define_insn "addsi3"

  21412.        [(set (match_operand:SI 0 "general_operand" "=r,m")

  21413.              (plus:SI (match_operand:SI 1 "general_operand" "0,0")

  21414.                       (match_operand:SI 2 "general_operand" "g,r")))]

  21415.        ""

  21416.        "@

  21417.         addr %2,%0

  21418.         addm %2,%0")

  21419. 

  21420.  If the output control string starts with a '*', then it is not an

  21421. output template but rather a piece of C program that should compute a

  21422. template.  It should execute a 'return' statement to return the

  21423. template-string you want.  Most such templates use C string literals,

  21424. which require doublequote characters to delimit them.  To include these

  21425. doublequote characters in the string, prefix each one with '\'.

  21426. 

  21427.  If the output control string is written as a brace block instead of a

  21428. double-quoted string, it is automatically assumed to be C code.  In that

  21429. case, it is not necessary to put in a leading asterisk, or to escape the

  21430. doublequotes surrounding C string literals.

  21431. 

  21432.  The operands may be found in the array 'operands', whose C data type is

  21433. 'rtx []'.

  21434. 

  21435.  It is very common to select different ways of generating assembler code

  21436. based on whether an immediate operand is within a certain range.  Be

  21437. careful when doing this, because the result of 'INTVAL' is an integer on

  21438. the host machine.  If the host machine has more bits in an 'int' than

  21439. the target machine has in the mode in which the constant will be used,

  21440. then some of the bits you get from 'INTVAL' will be superfluous.  For

  21441. proper results, you must carefully disregard the values of those bits.

  21442. 

  21443.  It is possible to output an assembler instruction and then go on to

  21444. output or compute more of them, using the subroutine 'output_asm_insn'.

  21445. This receives two arguments: a template-string and a vector of operands.

  21446. The vector may be 'operands', or it may be another array of 'rtx' that

  21447. you declare locally and initialize yourself.

  21448. 

  21449.  When an insn pattern has multiple alternatives in its constraints,

  21450. often the appearance of the assembler code is determined mostly by which

  21451. alternative was matched.  When this is so, the C code can test the

  21452. variable 'which_alternative', which is the ordinal number of the

  21453. alternative that was actually satisfied (0 for the first, 1 for the

  21454. second alternative, etc.).

  21455. 

  21456.  For example, suppose there are two opcodes for storing zero, 'clrreg'

  21457. for registers and 'clrmem' for memory locations.  Here is how a pattern

  21458. could use 'which_alternative' to choose between them:

  21459. 

  21460.      (define_insn ""

  21461.        [(set (match_operand:SI 0 "general_operand" "=r,m")

  21462.              (const_int 0))]

  21463.        ""

  21464.        {

  21465.        return (which_alternative == 0

  21466.                ? "clrreg %0" : "clrmem %0");

  21467.        })

  21468. 

  21469.  The example above, where the assembler code to generate was _solely_

  21470. determined by the alternative, could also have been specified as

  21471. follows, having the output control string start with a '@':

  21472. 

  21473.      (define_insn ""

  21474.        [(set (match_operand:SI 0 "general_operand" "=r,m")

  21475.              (const_int 0))]

  21476.        ""

  21477.        "@

  21478.         clrreg %0

  21479.         clrmem %0")

  21480. 

  21481.  If you just need a little bit of C code in one (or a few) alternatives,

  21482. you can use '*' inside of a '@' multi-alternative template:

  21483. 

  21484.      (define_insn ""

  21485.        [(set (match_operand:SI 0 "general_operand" "=r,<,m")

  21486.              (const_int 0))]

  21487.        ""

  21488.        "@

  21489.         clrreg %0

  21490.         * return stack_mem_p (operands[0]) ? \"push 0\" : \"clrmem %0\";

  21491.         clrmem %0")

  21492. 

  21493. 

  21494. File: gccint.info,  Node: Predicates,  Next: Constraints,  Prev: Output Statement,  Up: Machine Desc

  21495. 

  21496. 17.7 Predicates

  21497. ===============

  21498. 

  21499. A predicate determines whether a 'match_operand' or 'match_operator'

  21500. expression matches, and therefore whether the surrounding instruction

  21501. pattern will be used for that combination of operands.  GCC has a number

  21502. of machine-independent predicates, and you can define machine-specific

  21503. predicates as needed.  By convention, predicates used with

  21504. 'match_operand' have names that end in '_operand', and those used with

  21505. 'match_operator' have names that end in '_operator'.

  21506. 

  21507.  All predicates are boolean functions (in the mathematical sense) of two

  21508. arguments: the RTL expression that is being considered at that position

  21509. in the instruction pattern, and the machine mode that the

  21510. 'match_operand' or 'match_operator' specifies.  In this section, the

  21511. first argument is called OP and the second argument MODE.  Predicates

  21512. can be called from C as ordinary two-argument functions; this can be

  21513. useful in output templates or other machine-specific code.

  21514. 

  21515.  Operand predicates can allow operands that are not actually acceptable

  21516. to the hardware, as long as the constraints give reload the ability to

  21517. fix them up (*note Constraints::).  However, GCC will usually generate

  21518. better code if the predicates specify the requirements of the machine

  21519. instructions as closely as possible.  Reload cannot fix up operands that

  21520. must be constants ("immediate operands"); you must use a predicate that

  21521. allows only constants, or else enforce the requirement in the extra

  21522. condition.

  21523. 

  21524.  Most predicates handle their MODE argument in a uniform manner.  If

  21525. MODE is 'VOIDmode' (unspecified), then OP can have any mode.  If MODE is

  21526. anything else, then OP must have the same mode, unless OP is a

  21527. 'CONST_INT' or integer 'CONST_DOUBLE'.  These RTL expressions always

  21528. have 'VOIDmode', so it would be counterproductive to check that their

  21529. mode matches.  Instead, predicates that accept 'CONST_INT' and/or

  21530. integer 'CONST_DOUBLE' check that the value stored in the constant will

  21531. fit in the requested mode.

  21532. 

  21533.  Predicates with this behavior are called "normal".  'genrecog' can

  21534. optimize the instruction recognizer based on knowledge of how normal

  21535. predicates treat modes.  It can also diagnose certain kinds of common

  21536. errors in the use of normal predicates; for instance, it is almost

  21537. always an error to use a normal predicate without specifying a mode.

  21538. 

  21539.  Predicates that do something different with their MODE argument are

  21540. called "special".  The generic predicates 'address_operand' and

  21541. 'pmode_register_operand' are special predicates.  'genrecog' does not do

  21542. any optimizations or diagnosis when special predicates are used.

  21543. 

  21544. * Menu:

  21545. 

  21546. * Machine-Independent Predicates::  Predicates available to all back ends.

  21547. * Defining Predicates::             How to write machine-specific predicate

  21548.                                     functions.

  21549. 

  21550. 

  21551. File: gccint.info,  Node: Machine-Independent Predicates,  Next: Defining Predicates,  Up: Predicates

  21552. 

  21553. 17.7.1 Machine-Independent Predicates

  21554. -------------------------------------

  21555. 

  21556. These are the generic predicates available to all back ends.  They are

  21557. defined in 'recog.c'.  The first category of predicates allow only

  21558. constant, or "immediate", operands.

  21559. 

  21560.  -- Function: immediate_operand

  21561.      This predicate allows any sort of constant that fits in MODE.  It

  21562.      is an appropriate choice for instructions that take operands that

  21563.      must be constant.

  21564. 

  21565.  -- Function: const_int_operand

  21566.      This predicate allows any 'CONST_INT' expression that fits in MODE.

  21567.      It is an appropriate choice for an immediate operand that does not

  21568.      allow a symbol or label.

  21569. 

  21570.  -- Function: const_double_operand

  21571.      This predicate accepts any 'CONST_DOUBLE' expression that has

  21572.      exactly MODE.  If MODE is 'VOIDmode', it will also accept

  21573.      'CONST_INT'.  It is intended for immediate floating point

  21574.      constants.

  21575. 

  21576. The second category of predicates allow only some kind of machine

  21577. register.

  21578. 

  21579.  -- Function: register_operand

  21580.      This predicate allows any 'REG' or 'SUBREG' expression that is

  21581.      valid for MODE.  It is often suitable for arithmetic instruction

  21582.      operands on a RISC machine.

  21583. 

  21584.  -- Function: pmode_register_operand

  21585.      This is a slight variant on 'register_operand' which works around a

  21586.      limitation in the machine-description reader.

  21587. 

  21588.           (match_operand N "pmode_register_operand" CONSTRAINT)

  21589. 

  21590.      means exactly what

  21591. 

  21592.           (match_operand:P N "register_operand" CONSTRAINT)

  21593. 

  21594.      would mean, if the machine-description reader accepted ':P' mode

  21595.      suffixes.  Unfortunately, it cannot, because 'Pmode' is an alias

  21596.      for some other mode, and might vary with machine-specific options.

  21597.      *Note Misc::.

  21598. 

  21599.  -- Function: scratch_operand

  21600.      This predicate allows hard registers and 'SCRATCH' expressions, but

  21601.      not pseudo-registers.  It is used internally by 'match_scratch'; it

  21602.      should not be used directly.

  21603. 

  21604. The third category of predicates allow only some kind of memory

  21605. reference.

  21606. 

  21607.  -- Function: memory_operand

  21608.      This predicate allows any valid reference to a quantity of mode

  21609.      MODE in memory, as determined by the weak form of

  21610.      'GO_IF_LEGITIMATE_ADDRESS' (*note Addressing Modes::).

  21611. 

  21612.  -- Function: address_operand

  21613.      This predicate is a little unusual; it allows any operand that is a

  21614.      valid expression for the _address_ of a quantity of mode MODE,

  21615.      again determined by the weak form of 'GO_IF_LEGITIMATE_ADDRESS'.

  21616.      To first order, if '(mem:MODE (EXP))' is acceptable to

  21617.      'memory_operand', then EXP is acceptable to 'address_operand'.

  21618.      Note that EXP does not necessarily have the mode MODE.

  21619. 

  21620.  -- Function: indirect_operand

  21621.      This is a stricter form of 'memory_operand' which allows only

  21622.      memory references with a 'general_operand' as the address

  21623.      expression.  New uses of this predicate are discouraged, because

  21624.      'general_operand' is very permissive, so it's hard to tell what an

  21625.      'indirect_operand' does or does not allow.  If a target has

  21626.      different requirements for memory operands for different

  21627.      instructions, it is better to define target-specific predicates

  21628.      which enforce the hardware's requirements explicitly.

  21629. 

  21630.  -- Function: push_operand

  21631.      This predicate allows a memory reference suitable for pushing a

  21632.      value onto the stack.  This will be a 'MEM' which refers to

  21633.      'stack_pointer_rtx', with a side effect in its address expression

  21634.      (*note Incdec::); which one is determined by the 'STACK_PUSH_CODE'

  21635.      macro (*note Frame Layout::).

  21636. 

  21637.  -- Function: pop_operand

  21638.      This predicate allows a memory reference suitable for popping a

  21639.      value off the stack.  Again, this will be a 'MEM' referring to

  21640.      'stack_pointer_rtx', with a side effect in its address expression.

  21641.      However, this time 'STACK_POP_CODE' is expected.

  21642. 

  21643. The fourth category of predicates allow some combination of the above

  21644. operands.

  21645. 

  21646.  -- Function: nonmemory_operand

  21647.      This predicate allows any immediate or register operand valid for

  21648.      MODE.

  21649. 

  21650.  -- Function: nonimmediate_operand

  21651.      This predicate allows any register or memory operand valid for

  21652.      MODE.

  21653. 

  21654.  -- Function: general_operand

  21655.      This predicate allows any immediate, register, or memory operand

  21656.      valid for MODE.

  21657. 

  21658. Finally, there are two generic operator predicates.

  21659. 

  21660.  -- Function: comparison_operator

  21661.      This predicate matches any expression which performs an arithmetic

  21662.      comparison in MODE; that is, 'COMPARISON_P' is true for the

  21663.      expression code.

  21664. 

  21665.  -- Function: ordered_comparison_operator

  21666.      This predicate matches any expression which performs an arithmetic

  21667.      comparison in MODE and whose expression code is valid for integer

  21668.      modes; that is, the expression code will be one of 'eq', 'ne',

  21669.      'lt', 'ltu', 'le', 'leu', 'gt', 'gtu', 'ge', 'geu'.

  21670. 

  21671. 

  21672. File: gccint.info,  Node: Defining Predicates,  Prev: Machine-Independent Predicates,  Up: Predicates

  21673. 

  21674. 17.7.2 Defining Machine-Specific Predicates

  21675. -------------------------------------------

  21676. 

  21677. Many machines have requirements for their operands that cannot be

  21678. expressed precisely using the generic predicates.  You can define

  21679. additional predicates using 'define_predicate' and

  21680. 'define_special_predicate' expressions.  These expressions have three

  21681. operands:

  21682. 

  21683.    * The name of the predicate, as it will be referred to in

  21684.      'match_operand' or 'match_operator' expressions.

  21685. 

  21686.    * An RTL expression which evaluates to true if the predicate allows

  21687.      the operand OP, false if it does not.  This expression can only use

  21688.      the following RTL codes:

  21689. 

  21690.      'MATCH_OPERAND'

  21691.           When written inside a predicate expression, a 'MATCH_OPERAND'

  21692.           expression evaluates to true if the predicate it names would

  21693.           allow OP.  The operand number and constraint are ignored.  Due

  21694.           to limitations in 'genrecog', you can only refer to generic

  21695.           predicates and predicates that have already been defined.

  21696. 

  21697.      'MATCH_CODE'

  21698.           This expression evaluates to true if OP or a specified

  21699.           subexpression of OP has one of a given list of RTX codes.

  21700. 

  21701.           The first operand of this expression is a string constant

  21702.           containing a comma-separated list of RTX code names (in lower

  21703.           case).  These are the codes for which the 'MATCH_CODE' will be

  21704.           true.

  21705. 

  21706.           The second operand is a string constant which indicates what

  21707.           subexpression of OP to examine.  If it is absent or the empty

  21708.           string, OP itself is examined.  Otherwise, the string constant

  21709.           must be a sequence of digits and/or lowercase letters.  Each

  21710.           character indicates a subexpression to extract from the

  21711.           current expression; for the first character this is OP, for

  21712.           the second and subsequent characters it is the result of the

  21713.           previous character.  A digit N extracts 'XEXP (E, N)'; a

  21714.           letter L extracts 'XVECEXP (E, 0, N)' where N is the

  21715.           alphabetic ordinal of L (0 for 'a', 1 for 'b', and so on).

  21716.           The 'MATCH_CODE' then examines the RTX code of the

  21717.           subexpression extracted by the complete string.  It is not

  21718.           possible to extract components of an 'rtvec' that is not at

  21719.           position 0 within its RTX object.

  21720. 

  21721.      'MATCH_TEST'

  21722.           This expression has one operand, a string constant containing

  21723.           a C expression.  The predicate's arguments, OP and MODE, are

  21724.           available with those names in the C expression.  The

  21725.           'MATCH_TEST' evaluates to true if the C expression evaluates

  21726.           to a nonzero value.  'MATCH_TEST' expressions must not have

  21727.           side effects.

  21728. 

  21729.      'AND'

  21730.      'IOR'

  21731.      'NOT'

  21732.      'IF_THEN_ELSE'

  21733.           The basic 'MATCH_' expressions can be combined using these

  21734.           logical operators, which have the semantics of the C operators

  21735.           '&&', '||', '!', and '? :' respectively.  As in Common Lisp,

  21736.           you may give an 'AND' or 'IOR' expression an arbitrary number

  21737.           of arguments; this has exactly the same effect as writing a

  21738.           chain of two-argument 'AND' or 'IOR' expressions.

  21739. 

  21740.    * An optional block of C code, which should execute 'return true' if

  21741.      the predicate is found to match and 'return false' if it does not.

  21742.      It must not have any side effects.  The predicate arguments, OP and

  21743.      MODE, are available with those names.

  21744. 

  21745.      If a code block is present in a predicate definition, then the RTL

  21746.      expression must evaluate to true _and_ the code block must execute

  21747.      'return true' for the predicate to allow the operand.  The RTL

  21748.      expression is evaluated first; do not re-check anything in the code

  21749.      block that was checked in the RTL expression.

  21750. 

  21751.  The program 'genrecog' scans 'define_predicate' and

  21752. 'define_special_predicate' expressions to determine which RTX codes are

  21753. possibly allowed.  You should always make this explicit in the RTL

  21754. predicate expression, using 'MATCH_OPERAND' and 'MATCH_CODE'.

  21755. 

  21756.  Here is an example of a simple predicate definition, from the IA64

  21757. machine description:

  21758. 

  21759.      ;; True if OP is a 'SYMBOL_REF' which refers to the sdata section.

  21760.      (define_predicate "small_addr_symbolic_operand"

  21761.        (and (match_code "symbol_ref")

  21762.             (match_test "SYMBOL_REF_SMALL_ADDR_P (op)")))

  21763. 

  21764. And here is another, showing the use of the C block.

  21765. 

  21766.      ;; True if OP is a register operand that is (or could be) a GR reg.

  21767.      (define_predicate "gr_register_operand"

  21768.        (match_operand 0 "register_operand")

  21769.      {

  21770.        unsigned int regno;

  21771.        if (GET_CODE (op) == SUBREG)

  21772.          op = SUBREG_REG (op);

  21773. 

  21774.        regno = REGNO (op);

  21775.        return (regno >= FIRST_PSEUDO_REGISTER || GENERAL_REGNO_P (regno));

  21776.      })

  21777. 

  21778.  Predicates written with 'define_predicate' automatically include a test

  21779. that MODE is 'VOIDmode', or OP has the same mode as MODE, or OP is a

  21780. 'CONST_INT' or 'CONST_DOUBLE'.  They do _not_ check specifically for

  21781. integer 'CONST_DOUBLE', nor do they test that the value of either kind

  21782. of constant fits in the requested mode.  This is because target-specific

  21783. predicates that take constants usually have to do more stringent value

  21784. checks anyway.  If you need the exact same treatment of 'CONST_INT' or

  21785. 'CONST_DOUBLE' that the generic predicates provide, use a

  21786. 'MATCH_OPERAND' subexpression to call 'const_int_operand',

  21787. 'const_double_operand', or 'immediate_operand'.

  21788. 

  21789.  Predicates written with 'define_special_predicate' do not get any

  21790. automatic mode checks, and are treated as having special mode handling

  21791. by 'genrecog'.

  21792. 

  21793.  The program 'genpreds' is responsible for generating code to test

  21794. predicates.  It also writes a header file containing function

  21795. declarations for all machine-specific predicates.  It is not necessary

  21796. to declare these predicates in 'CPU-protos.h'.

  21797. 

  21798. 

  21799. File: gccint.info,  Node: Constraints,  Next: Standard Names,  Prev: Predicates,  Up: Machine Desc

  21800. 

  21801. 17.8 Operand Constraints

  21802. ========================

  21803. 

  21804. Each 'match_operand' in an instruction pattern can specify constraints

  21805. for the operands allowed.  The constraints allow you to fine-tune

  21806. matching within the set of operands allowed by the predicate.

  21807. 

  21808.  Constraints can say whether an operand may be in a register, and which

  21809. kinds of register; whether the operand can be a memory reference, and

  21810. which kinds of address; whether the operand may be an immediate

  21811. constant, and which possible values it may have.  Constraints can also

  21812. require two operands to match.  Side-effects aren't allowed in operands

  21813. of inline 'asm', unless '<' or '>' constraints are used, because there

  21814. is no guarantee that the side effects will happen exactly once in an

  21815. instruction that can update the addressing register.

  21816. 

  21817. * Menu:

  21818. 

  21819. * Simple Constraints::  Basic use of constraints.

  21820. * Multi-Alternative::   When an insn has two alternative constraint-patterns.

  21821. * Class Preferences::   Constraints guide which hard register to put things in.

  21822. * Modifiers::           More precise control over effects of constraints.

  21823. * Machine Constraints:: Existing constraints for some particular machines.

  21824. * Disable Insn Alternatives:: Disable insn alternatives using attributes.

  21825. * Define Constraints::  How to define machine-specific constraints.

  21826. * C Constraint Interface:: How to test constraints from C code.

  21827. 

  21828. 

  21829. File: gccint.info,  Node: Simple Constraints,  Next: Multi-Alternative,  Up: Constraints

  21830. 

  21831. 17.8.1 Simple Constraints

  21832. -------------------------

  21833. 

  21834. The simplest kind of constraint is a string full of letters, each of

  21835. which describes one kind of operand that is permitted.  Here are the

  21836. letters that are allowed:

  21837. 

  21838. whitespace

  21839.      Whitespace characters are ignored and can be inserted at any

  21840.      position except the first.  This enables each alternative for

  21841.      different operands to be visually aligned in the machine

  21842.      description even if they have different number of constraints and

  21843.      modifiers.

  21844. 

  21845. 'm'

  21846.      A memory operand is allowed, with any kind of address that the

  21847.      machine supports in general.  Note that the letter used for the

  21848.      general memory constraint can be re-defined by a back end using the

  21849.      'TARGET_MEM_CONSTRAINT' macro.

  21850. 

  21851. 'o'

  21852.      A memory operand is allowed, but only if the address is

  21853.      "offsettable".  This means that adding a small integer (actually,

  21854.      the width in bytes of the operand, as determined by its machine

  21855.      mode) may be added to the address and the result is also a valid

  21856.      memory address.

  21857. 

  21858.      For example, an address which is constant is offsettable; so is an

  21859.      address that is the sum of a register and a constant (as long as a

  21860.      slightly larger constant is also within the range of

  21861.      address-offsets supported by the machine); but an autoincrement or

  21862.      autodecrement address is not offsettable.  More complicated

  21863.      indirect/indexed addresses may or may not be offsettable depending

  21864.      on the other addressing modes that the machine supports.

  21865. 

  21866.      Note that in an output operand which can be matched by another

  21867.      operand, the constraint letter 'o' is valid only when accompanied

  21868.      by both '<' (if the target machine has predecrement addressing) and

  21869.      '>' (if the target machine has preincrement addressing).

  21870. 

  21871. 'V'

  21872.      A memory operand that is not offsettable.  In other words, anything

  21873.      that would fit the 'm' constraint but not the 'o' constraint.

  21874. 

  21875. '<'

  21876.      A memory operand with autodecrement addressing (either predecrement

  21877.      or postdecrement) is allowed.  In inline 'asm' this constraint is

  21878.      only allowed if the operand is used exactly once in an instruction

  21879.      that can handle the side effects.  Not using an operand with '<' in

  21880.      constraint string in the inline 'asm' pattern at all or using it in

  21881.      multiple instructions isn't valid, because the side effects

  21882.      wouldn't be performed or would be performed more than once.

  21883.      Furthermore, on some targets the operand with '<' in constraint

  21884.      string must be accompanied by special instruction suffixes like

  21885.      '%U0' instruction suffix on PowerPC or '%P0' on IA-64.

  21886. 

  21887. '>'

  21888.      A memory operand with autoincrement addressing (either preincrement

  21889.      or postincrement) is allowed.  In inline 'asm' the same

  21890.      restrictions as for '<' apply.

  21891. 

  21892. 'r'

  21893.      A register operand is allowed provided that it is in a general

  21894.      register.

  21895. 

  21896. 'i'

  21897.      An immediate integer operand (one with constant value) is allowed.

  21898.      This includes symbolic constants whose values will be known only at

  21899.      assembly time or later.

  21900. 

  21901. 'n'

  21902.      An immediate integer operand with a known numeric value is allowed.

  21903.      Many systems cannot support assembly-time constants for operands

  21904.      less than a word wide.  Constraints for these operands should use

  21905.      'n' rather than 'i'.

  21906. 

  21907. 'I', 'J', 'K', ... 'P'

  21908.      Other letters in the range 'I' through 'P' may be defined in a

  21909.      machine-dependent fashion to permit immediate integer operands with

  21910.      explicit integer values in specified ranges.  For example, on the

  21911.      68000, 'I' is defined to stand for the range of values 1 to 8.

  21912.      This is the range permitted as a shift count in the shift

  21913.      instructions.

  21914. 

  21915. 'E'

  21916.      An immediate floating operand (expression code 'const_double') is

  21917.      allowed, but only if the target floating point format is the same

  21918.      as that of the host machine (on which the compiler is running).

  21919. 

  21920. 'F'

  21921.      An immediate floating operand (expression code 'const_double' or

  21922.      'const_vector') is allowed.

  21923. 

  21924. 'G', 'H'

  21925.      'G' and 'H' may be defined in a machine-dependent fashion to permit

  21926.      immediate floating operands in particular ranges of values.

  21927. 

  21928. 's'

  21929.      An immediate integer operand whose value is not an explicit integer

  21930.      is allowed.

  21931. 

  21932.      This might appear strange; if an insn allows a constant operand

  21933.      with a value not known at compile time, it certainly must allow any

  21934.      known value.  So why use 's' instead of 'i'?  Sometimes it allows

  21935.      better code to be generated.

  21936. 

  21937.      For example, on the 68000 in a fullword instruction it is possible

  21938.      to use an immediate operand; but if the immediate value is between

  21939.      -128 and 127, better code results from loading the value into a

  21940.      register and using the register.  This is because the load into the

  21941.      register can be done with a 'moveq' instruction.  We arrange for

  21942.      this to happen by defining the letter 'K' to mean "any integer

  21943.      outside the range -128 to 127", and then specifying 'Ks' in the

  21944.      operand constraints.

  21945. 

  21946. 'g'

  21947.      Any register, memory or immediate integer operand is allowed,

  21948.      except for registers that are not general registers.

  21949. 

  21950. 'X'

  21951.      Any operand whatsoever is allowed, even if it does not satisfy

  21952.      'general_operand'.  This is normally used in the constraint of a

  21953.      'match_scratch' when certain alternatives will not actually require

  21954.      a scratch register.

  21955. 

  21956. '0', '1', '2', ... '9'

  21957.      An operand that matches the specified operand number is allowed.

  21958.      If a digit is used together with letters within the same

  21959.      alternative, the digit should come last.

  21960. 

  21961.      This number is allowed to be more than a single digit.  If multiple

  21962.      digits are encountered consecutively, they are interpreted as a

  21963.      single decimal integer.  There is scant chance for ambiguity, since

  21964.      to-date it has never been desirable that '10' be interpreted as

  21965.      matching either operand 1 _or_ operand 0.  Should this be desired,

  21966.      one can use multiple alternatives instead.

  21967. 

  21968.      This is called a "matching constraint" and what it really means is

  21969.      that the assembler has only a single operand that fills two roles

  21970.      considered separate in the RTL insn.  For example, an add insn has

  21971.      two input operands and one output operand in the RTL, but on most

  21972.      CISC machines an add instruction really has only two operands, one

  21973.      of them an input-output operand:

  21974. 

  21975.           addl #35,r12

  21976. 

  21977.      Matching constraints are used in these circumstances.  More

  21978.      precisely, the two operands that match must include one input-only

  21979.      operand and one output-only operand.  Moreover, the digit must be a

  21980.      smaller number than the number of the operand that uses it in the

  21981.      constraint.

  21982. 

  21983.      For operands to match in a particular case usually means that they

  21984.      are identical-looking RTL expressions.  But in a few special cases

  21985.      specific kinds of dissimilarity are allowed.  For example, '*x' as

  21986.      an input operand will match '*x++' as an output operand.  For

  21987.      proper results in such cases, the output template should always use

  21988.      the output-operand's number when printing the operand.

  21989. 

  21990. 'p'

  21991.      An operand that is a valid memory address is allowed.  This is for

  21992.      "load address" and "push address" instructions.

  21993. 

  21994.      'p' in the constraint must be accompanied by 'address_operand' as

  21995.      the predicate in the 'match_operand'.  This predicate interprets

  21996.      the mode specified in the 'match_operand' as the mode of the memory

  21997.      reference for which the address would be valid.

  21998. 

  21999. OTHER-LETTERS

  22000.      Other letters can be defined in machine-dependent fashion to stand

  22001.      for particular classes of registers or other arbitrary operand

  22002.      types.  'd', 'a' and 'f' are defined on the 68000/68020 to stand

  22003.      for data, address and floating point registers.

  22004. 

  22005.  In order to have valid assembler code, each operand must satisfy its

  22006. constraint.  But a failure to do so does not prevent the pattern from

  22007. applying to an insn.  Instead, it directs the compiler to modify the

  22008. code so that the constraint will be satisfied.  Usually this is done by

  22009. copying an operand into a register.

  22010. 

  22011.  Contrast, therefore, the two instruction patterns that follow:

  22012. 

  22013.      (define_insn ""

  22014.        [(set (match_operand:SI 0 "general_operand" "=r")

  22015.              (plus:SI (match_dup 0)

  22016.                       (match_operand:SI 1 "general_operand" "r")))]

  22017.        ""

  22018.        "...")

  22019. 

  22020. which has two operands, one of which must appear in two places, and

  22021. 

  22022.      (define_insn ""

  22023.        [(set (match_operand:SI 0 "general_operand" "=r")

  22024.              (plus:SI (match_operand:SI 1 "general_operand" "0")

  22025.                       (match_operand:SI 2 "general_operand" "r")))]

  22026.        ""

  22027.        "...")

  22028. 

  22029. which has three operands, two of which are required by a constraint to

  22030. be identical.  If we are considering an insn of the form

  22031. 

  22032.      (insn N PREV NEXT

  22033.        (set (reg:SI 3)

  22034.             (plus:SI (reg:SI 6) (reg:SI 109)))

  22035.        ...)

  22036. 

  22037. the first pattern would not apply at all, because this insn does not

  22038. contain two identical subexpressions in the right place.  The pattern

  22039. would say, "That does not look like an add instruction; try other

  22040. patterns".  The second pattern would say, "Yes, that's an add

  22041. instruction, but there is something wrong with it".  It would direct the

  22042. reload pass of the compiler to generate additional insns to make the

  22043. constraint true.  The results might look like this:

  22044. 

  22045.      (insn N2 PREV N

  22046.        (set (reg:SI 3) (reg:SI 6))

  22047.        ...)

  22048. 

  22049.      (insn N N2 NEXT

  22050.        (set (reg:SI 3)

  22051.             (plus:SI (reg:SI 3) (reg:SI 109)))

  22052.        ...)

  22053. 

  22054.  It is up to you to make sure that each operand, in each pattern, has

  22055. constraints that can handle any RTL expression that could be present for

  22056. that operand.  (When multiple alternatives are in use, each pattern

  22057. must, for each possible combination of operand expressions, have at

  22058. least one alternative which can handle that combination of operands.)

  22059. The constraints don't need to _allow_ any possible operand--when this is

  22060. the case, they do not constrain--but they must at least point the way to

  22061. reloading any possible operand so that it will fit.

  22062. 

  22063.    * If the constraint accepts whatever operands the predicate permits,

  22064.      there is no problem: reloading is never necessary for this operand.

  22065. 

  22066.      For example, an operand whose constraints permit everything except

  22067.      registers is safe provided its predicate rejects registers.

  22068. 

  22069.      An operand whose predicate accepts only constant values is safe

  22070.      provided its constraints include the letter 'i'.  If any possible

  22071.      constant value is accepted, then nothing less than 'i' will do; if

  22072.      the predicate is more selective, then the constraints may also be

  22073.      more selective.

  22074. 

  22075.    * Any operand expression can be reloaded by copying it into a

  22076.      register.  So if an operand's constraints allow some kind of

  22077.      register, it is certain to be safe.  It need not permit all classes

  22078.      of registers; the compiler knows how to copy a register into

  22079.      another register of the proper class in order to make an

  22080.      instruction valid.

  22081. 

  22082.    * A nonoffsettable memory reference can be reloaded by copying the

  22083.      address into a register.  So if the constraint uses the letter 'o',

  22084.      all memory references are taken care of.

  22085. 

  22086.    * A constant operand can be reloaded by allocating space in memory to

  22087.      hold it as preinitialized data.  Then the memory reference can be

  22088.      used in place of the constant.  So if the constraint uses the

  22089.      letters 'o' or 'm', constant operands are not a problem.

  22090. 

  22091.    * If the constraint permits a constant and a pseudo register used in

  22092.      an insn was not allocated to a hard register and is equivalent to a

  22093.      constant, the register will be replaced with the constant.  If the

  22094.      predicate does not permit a constant and the insn is re-recognized

  22095.      for some reason, the compiler will crash.  Thus the predicate must

  22096.      always recognize any objects allowed by the constraint.

  22097. 

  22098.  If the operand's predicate can recognize registers, but the constraint

  22099. does not permit them, it can make the compiler crash.  When this operand

  22100. happens to be a register, the reload pass will be stymied, because it

  22101. does not know how to copy a register temporarily into memory.

  22102. 

  22103.  If the predicate accepts a unary operator, the constraint applies to

  22104. the operand.  For example, the MIPS processor at ISA level 3 supports an

  22105. instruction which adds two registers in 'SImode' to produce a 'DImode'

  22106. result, but only if the registers are correctly sign extended.  This

  22107. predicate for the input operands accepts a 'sign_extend' of an 'SImode'

  22108. register.  Write the constraint to indicate the type of register that is

  22109. required for the operand of the 'sign_extend'.

  22110. 

  22111. 

  22112. File: gccint.info,  Node: Multi-Alternative,  Next: Class Preferences,  Prev: Simple Constraints,  Up: Constraints

  22113. 

  22114. 17.8.2 Multiple Alternative Constraints

  22115. ---------------------------------------

  22116. 

  22117. Sometimes a single instruction has multiple alternative sets of possible

  22118. operands.  For example, on the 68000, a logical-or instruction can

  22119. combine register or an immediate value into memory, or it can combine

  22120. any kind of operand into a register; but it cannot combine one memory

  22121. location into another.

  22122. 

  22123.  These constraints are represented as multiple alternatives.  An

  22124. alternative can be described by a series of letters for each operand.

  22125. The overall constraint for an operand is made from the letters for this

  22126. operand from the first alternative, a comma, the letters for this

  22127. operand from the second alternative, a comma, and so on until the last

  22128. alternative.  All operands for a single instruction must have the same

  22129. number of alternatives.  Here is how it is done for fullword logical-or

  22130. on the 68000:

  22131. 

  22132.      (define_insn "iorsi3"

  22133.        [(set (match_operand:SI 0 "general_operand" "=m,d")

  22134.              (ior:SI (match_operand:SI 1 "general_operand" "%0,0")

  22135.                      (match_operand:SI 2 "general_operand" "dKs,dmKs")))]

  22136.        ...)

  22137. 

  22138.  The first alternative has 'm' (memory) for operand 0, '0' for operand 1

  22139. (meaning it must match operand 0), and 'dKs' for operand 2.  The second

  22140. alternative has 'd' (data register) for operand 0, '0' for operand 1,

  22141. and 'dmKs' for operand 2.  The '=' and '%' in the constraints apply to

  22142. all the alternatives; their meaning is explained in the next section

  22143. (*note Class Preferences::).

  22144. 

  22145.  If all the operands fit any one alternative, the instruction is valid.

  22146. Otherwise, for each alternative, the compiler counts how many

  22147. instructions must be added to copy the operands so that that alternative

  22148. applies.  The alternative requiring the least copying is chosen.  If two

  22149. alternatives need the same amount of copying, the one that comes first

  22150. is chosen.  These choices can be altered with the '?' and '!'

  22151. characters:

  22152. 

  22153. '?'

  22154.      Disparage slightly the alternative that the '?' appears in, as a

  22155.      choice when no alternative applies exactly.  The compiler regards

  22156.      this alternative as one unit more costly for each '?' that appears

  22157.      in it.

  22158. 

  22159. '!'

  22160.      Disparage severely the alternative that the '!' appears in.  This

  22161.      alternative can still be used if it fits without reloading, but if

  22162.      reloading is needed, some other alternative will be used.

  22163. 

  22164. '^'

  22165.      This constraint is analogous to '?' but it disparages slightly the

  22166.      alternative only if the operand with the '^' needs a reload.

  22167. 

  22168. '$'

  22169.      This constraint is analogous to '!' but it disparages severely the

  22170.      alternative only if the operand with the '$' needs a reload.

  22171. 

  22172.  When an insn pattern has multiple alternatives in its constraints,

  22173. often the appearance of the assembler code is determined mostly by which

  22174. alternative was matched.  When this is so, the C code for writing the

  22175. assembler code can use the variable 'which_alternative', which is the

  22176. ordinal number of the alternative that was actually satisfied (0 for the

  22177. first, 1 for the second alternative, etc.).  *Note Output Statement::.

  22178. 

  22179. 

  22180. File: gccint.info,  Node: Class Preferences,  Next: Modifiers,  Prev: Multi-Alternative,  Up: Constraints

  22181. 

  22182. 17.8.3 Register Class Preferences

  22183. ---------------------------------

  22184. 

  22185. The operand constraints have another function: they enable the compiler

  22186. to decide which kind of hardware register a pseudo register is best

  22187. allocated to.  The compiler examines the constraints that apply to the

  22188. insns that use the pseudo register, looking for the machine-dependent

  22189. letters such as 'd' and 'a' that specify classes of registers.  The

  22190. pseudo register is put in whichever class gets the most "votes".  The

  22191. constraint letters 'g' and 'r' also vote: they vote in favor of a

  22192. general register.  The machine description says which registers are

  22193. considered general.

  22194. 

  22195.  Of course, on some machines all registers are equivalent, and no

  22196. register classes are defined.  Then none of this complexity is relevant.

  22197. 

  22198. 

  22199. File: gccint.info,  Node: Modifiers,  Next: Machine Constraints,  Prev: Class Preferences,  Up: Constraints

  22200. 

  22201. 17.8.4 Constraint Modifier Characters

  22202. -------------------------------------

  22203. 

  22204. Here are constraint modifier characters.

  22205. 

  22206. '='

  22207.      Means that this operand is written to by this instruction: the

  22208.      previous value is discarded and replaced by new data.

  22209. 

  22210. '+'

  22211.      Means that this operand is both read and written by the

  22212.      instruction.

  22213. 

  22214.      When the compiler fixes up the operands to satisfy the constraints,

  22215.      it needs to know which operands are read by the instruction and

  22216.      which are written by it.  '=' identifies an operand which is only

  22217.      written; '+' identifies an operand that is both read and written;

  22218.      all other operands are assumed to only be read.

  22219. 

  22220.      If you specify '=' or '+' in a constraint, you put it in the first

  22221.      character of the constraint string.

  22222. 

  22223. '&'

  22224.      Means (in a particular alternative) that this operand is an

  22225.      "earlyclobber" operand, which is written before the instruction is

  22226.      finished using the input operands.  Therefore, this operand may not

  22227.      lie in a register that is read by the instruction or as part of any

  22228.      memory address.

  22229. 

  22230.      '&' applies only to the alternative in which it is written.  In

  22231.      constraints with multiple alternatives, sometimes one alternative

  22232.      requires '&' while others do not.  See, for example, the 'movdf'

  22233.      insn of the 68000.

  22234. 

  22235.      A operand which is read by the instruction can be tied to an

  22236.      earlyclobber operand if its only use as an input occurs before the

  22237.      early result is written.  Adding alternatives of this form often

  22238.      allows GCC to produce better code when only some of the read

  22239.      operands can be affected by the earlyclobber.  See, for example,

  22240.      the 'mulsi3' insn of the ARM.

  22241. 

  22242.      Furthermore, if the "earlyclobber" operand is also a read/write

  22243.      operand, then that operand is written only after it's used.

  22244. 

  22245.      '&' does not obviate the need to write '=' or '+'.  As

  22246.      "earlyclobber" operands are always written, a read-only

  22247.      "earlyclobber" operand is ill-formed and will be rejected by the

  22248.      compiler.

  22249. 

  22250. '%'

  22251.      Declares the instruction to be commutative for this operand and the

  22252.      following operand.  This means that the compiler may interchange

  22253.      the two operands if that is the cheapest way to make all operands

  22254.      fit the constraints.  '%' applies to all alternatives and must

  22255.      appear as the first character in the constraint.  Only read-only

  22256.      operands can use '%'.

  22257. 

  22258.      This is often used in patterns for addition instructions that

  22259.      really have only two operands: the result must go in one of the

  22260.      arguments.  Here for example, is how the 68000 halfword-add

  22261.      instruction is defined:

  22262. 

  22263.           (define_insn "addhi3"

  22264.             [(set (match_operand:HI 0 "general_operand" "=m,r")

  22265.                (plus:HI (match_operand:HI 1 "general_operand" "%0,0")

  22266.                         (match_operand:HI 2 "general_operand" "di,g")))]

  22267.             ...)

  22268.      GCC can only handle one commutative pair in an asm; if you use

  22269.      more, the compiler may fail.  Note that you need not use the

  22270.      modifier if the two alternatives are strictly identical; this would

  22271.      only waste time in the reload pass.  The modifier is not

  22272.      operational after register allocation, so the result of

  22273.      'define_peephole2' and 'define_split's performed after reload

  22274.      cannot rely on '%' to make the intended insn match.

  22275. 

  22276. '#'

  22277.      Says that all following characters, up to the next comma, are to be

  22278.      ignored as a constraint.  They are significant only for choosing

  22279.      register preferences.

  22280. 

  22281. '*'

  22282.      Says that the following character should be ignored when choosing

  22283.      register preferences.  '*' has no effect on the meaning of the

  22284.      constraint as a constraint, and no effect on reloading.  For LRA

  22285.      '*' additionally disparages slightly the alternative if the

  22286.      following character matches the operand.

  22287. 

  22288.      Here is an example: the 68000 has an instruction to sign-extend a

  22289.      halfword in a data register, and can also sign-extend a value by

  22290.      copying it into an address register.  While either kind of register

  22291.      is acceptable, the constraints on an address-register destination

  22292.      are less strict, so it is best if register allocation makes an

  22293.      address register its goal.  Therefore, '*' is used so that the 'd'

  22294.      constraint letter (for data register) is ignored when computing

  22295.      register preferences.

  22296. 

  22297.           (define_insn "extendhisi2"

  22298.             [(set (match_operand:SI 0 "general_operand" "=*d,a")

  22299.                   (sign_extend:SI

  22300.                    (match_operand:HI 1 "general_operand" "0,g")))]

  22301.             ...)

  22302. 

  22303. 

  22304. File: gccint.info,  Node: Machine Constraints,  Next: Disable Insn Alternatives,  Prev: Modifiers,  Up: Constraints

  22305. 

  22306. 17.8.5 Constraints for Particular Machines

  22307. ------------------------------------------

  22308. 

  22309. Whenever possible, you should use the general-purpose constraint letters

  22310. in 'asm' arguments, since they will convey meaning more readily to

  22311. people reading your code.  Failing that, use the constraint letters that

  22312. usually have very similar meanings across architectures.  The most

  22313. commonly used constraints are 'm' and 'r' (for memory and

  22314. general-purpose registers respectively; *note Simple Constraints::), and

  22315. 'I', usually the letter indicating the most common immediate-constant

  22316. format.

  22317. 

  22318.  Each architecture defines additional constraints.  These constraints

  22319. are used by the compiler itself for instruction generation, as well as

  22320. for 'asm' statements; therefore, some of the constraints are not

  22321. particularly useful for 'asm'.  Here is a summary of some of the

  22322. machine-dependent constraints available on some particular machines; it

  22323. includes both constraints that are useful for 'asm' and constraints that

  22324. aren't.  The compiler source file mentioned in the table heading for

  22325. each architecture is the definitive reference for the meanings of that

  22326. architecture's constraints.

  22327. 

  22328. _AArch64 family--'config/aarch64/constraints.md'_

  22329.      'k'

  22330.           The stack pointer register ('SP')

  22331. 

  22332.      'w'

  22333.           Floating point register, Advanced SIMD vector register or SVE

  22334.           vector register

  22335. 

  22336.      'x'

  22337.           Like 'w', but restricted to registers 0 to 15 inclusive.

  22338. 

  22339.      'y'

  22340.           Like 'w', but restricted to registers 0 to 7 inclusive.

  22341. 

  22342.      'Upl'

  22343.           One of the low eight SVE predicate registers ('P0' to 'P7')

  22344. 

  22345.      'Upa'

  22346.           Any of the SVE predicate registers ('P0' to 'P15')

  22347. 

  22348.      'I'

  22349.           Integer constant that is valid as an immediate operand in an

  22350.           'ADD' instruction

  22351. 

  22352.      'J'

  22353.           Integer constant that is valid as an immediate operand in a

  22354.           'SUB' instruction (once negated)

  22355. 

  22356.      'K'

  22357.           Integer constant that can be used with a 32-bit logical

  22358.           instruction

  22359. 

  22360.      'L'

  22361.           Integer constant that can be used with a 64-bit logical

  22362.           instruction

  22363. 

  22364.      'M'

  22365.           Integer constant that is valid as an immediate operand in a

  22366.           32-bit 'MOV' pseudo instruction.  The 'MOV' may be assembled

  22367.           to one of several different machine instructions depending on

  22368.           the value

  22369. 

  22370.      'N'

  22371.           Integer constant that is valid as an immediate operand in a

  22372.           64-bit 'MOV' pseudo instruction

  22373. 

  22374.      'S'

  22375.           An absolute symbolic address or a label reference

  22376. 

  22377.      'Y'

  22378.           Floating point constant zero

  22379. 

  22380.      'Z'

  22381.           Integer constant zero

  22382. 

  22383.      'Ush'

  22384.           The high part (bits 12 and upwards) of the pc-relative address

  22385.           of a symbol within 4GB of the instruction

  22386. 

  22387.      'Q'

  22388.           A memory address which uses a single base register with no

  22389.           offset

  22390. 

  22391.      'Ump'

  22392.           A memory address suitable for a load/store pair instruction in

  22393.           SI, DI, SF and DF modes

  22394. 

  22395. _AMD GCN --'config/gcn/constraints.md'_

  22396.      'I'

  22397.           Immediate integer in the range -16 to 64

  22398. 

  22399.      'J'

  22400.           Immediate 16-bit signed integer

  22401. 

  22402.      'Kf'

  22403.           Immediate constant -1

  22404. 

  22405.      'L'

  22406.           Immediate 15-bit unsigned integer

  22407. 

  22408.      'A'

  22409.           Immediate constant that can be inlined in an instruction

  22410.           encoding: integer -16..64, or float 0.0, +/-0.5, +/-1.0,

  22411.           +/-2.0, +/-4.0, 1.0/(2.0*PI)

  22412. 

  22413.      'B'

  22414.           Immediate 32-bit signed integer that can be attached to an

  22415.           instruction encoding

  22416. 

  22417.      'C'

  22418.           Immediate 32-bit integer in range -16..4294967295 (i.e.

  22419.           32-bit unsigned integer or 'A' constraint)

  22420. 

  22421.      'DA'

  22422.           Immediate 64-bit constant that can be split into two 'A'

  22423.           constants

  22424. 

  22425.      'DB'

  22426.           Immediate 64-bit constant that can be split into two 'B'

  22427.           constants

  22428. 

  22429.      'U'

  22430.           Any 'unspec'

  22431. 

  22432.      'Y'

  22433.           Any 'symbol_ref' or 'label_ref'

  22434. 

  22435.      'v'

  22436.           VGPR register

  22437. 

  22438.      'Sg'

  22439.           SGPR register

  22440. 

  22441.      'SD'

  22442.           SGPR registers valid for instruction destinations, including

  22443.           VCC, M0 and EXEC

  22444. 

  22445.      'SS'

  22446.           SGPR registers valid for instruction sources, including VCC,

  22447.           M0, EXEC and SCC

  22448. 

  22449.      'Sm'

  22450.           SGPR registers valid as a source for scalar memory

  22451.           instructions (excludes M0 and EXEC)

  22452. 

  22453.      'Sv'

  22454.           SGPR registers valid as a source or destination for vector

  22455.           instructions (excludes EXEC)

  22456. 

  22457.      'ca'

  22458.           All condition registers: SCC, VCCZ, EXECZ

  22459. 

  22460.      'cs'

  22461.           Scalar condition register: SCC

  22462. 

  22463.      'cV'

  22464.           Vector condition register: VCC, VCC_LO, VCC_HI

  22465. 

  22466.      'e'

  22467.           EXEC register (EXEC_LO and EXEC_HI)

  22468. 

  22469.      'RB'

  22470.           Memory operand with address space suitable for 'buffer_*'

  22471.           instructions

  22472. 

  22473.      'RF'

  22474.           Memory operand with address space suitable for 'flat_*'

  22475.           instructions

  22476. 

  22477.      'RS'

  22478.           Memory operand with address space suitable for 's_*'

  22479.           instructions

  22480. 

  22481.      'RL'

  22482.           Memory operand with address space suitable for 'ds_*' LDS

  22483.           instructions

  22484. 

  22485.      'RG'

  22486.           Memory operand with address space suitable for 'ds_*' GDS

  22487.           instructions

  22488. 

  22489.      'RD'

  22490.           Memory operand with address space suitable for any 'ds_*'

  22491.           instructions

  22492. 

  22493.      'RM'

  22494.           Memory operand with address space suitable for 'global_*'

  22495.           instructions

  22496. 

  22497. _ARC --'config/arc/constraints.md'_

  22498.      'q'

  22499.           Registers usable in ARCompact 16-bit instructions: 'r0'-'r3',

  22500.           'r12'-'r15'.  This constraint can only match when the '-mq'

  22501.           option is in effect.

  22502. 

  22503.      'e'

  22504.           Registers usable as base-regs of memory addresses in ARCompact

  22505.           16-bit memory instructions: 'r0'-'r3', 'r12'-'r15', 'sp'.

  22506.           This constraint can only match when the '-mq' option is in

  22507.           effect.

  22508.      'D'

  22509.           ARC FPX (dpfp) 64-bit registers.  'D0', 'D1'.

  22510. 

  22511.      'I'

  22512.           A signed 12-bit integer constant.

  22513. 

  22514.      'Cal'

  22515.           constant for arithmetic/logical operations.  This might be any

  22516.           constant that can be put into a long immediate by the assmbler

  22517.           or linker without involving a PIC relocation.

  22518. 

  22519.      'K'

  22520.           A 3-bit unsigned integer constant.

  22521. 

  22522.      'L'

  22523.           A 6-bit unsigned integer constant.

  22524. 

  22525.      'CnL'

  22526.           One's complement of a 6-bit unsigned integer constant.

  22527. 

  22528.      'CmL'

  22529.           Two's complement of a 6-bit unsigned integer constant.

  22530. 

  22531.      'M'

  22532.           A 5-bit unsigned integer constant.

  22533. 

  22534.      'O'

  22535.           A 7-bit unsigned integer constant.

  22536. 

  22537.      'P'

  22538.           A 8-bit unsigned integer constant.

  22539. 

  22540.      'H'

  22541.           Any const_double value.

  22542. 

  22543. _ARM family--'config/arm/constraints.md'_

  22544. 

  22545.      'h'

  22546.           In Thumb state, the core registers 'r8'-'r15'.

  22547. 

  22548.      'k'

  22549.           The stack pointer register.

  22550. 

  22551.      'l'

  22552.           In Thumb State the core registers 'r0'-'r7'.  In ARM state

  22553.           this is an alias for the 'r' constraint.

  22554. 

  22555.      't'

  22556.           VFP floating-point registers 's0'-'s31'.  Used for 32 bit

  22557.           values.

  22558. 

  22559.      'w'

  22560.           VFP floating-point registers 'd0'-'d31' and the appropriate

  22561.           subset 'd0'-'d15' based on command line options.  Used for 64

  22562.           bit values only.  Not valid for Thumb1.

  22563. 

  22564.      'y'

  22565.           The iWMMX co-processor registers.

  22566. 

  22567.      'z'

  22568.           The iWMMX GR registers.

  22569. 

  22570.      'G'

  22571.           The floating-point constant 0.0

  22572. 

  22573.      'I'

  22574.           Integer that is valid as an immediate operand in a data

  22575.           processing instruction.  That is, an integer in the range 0 to

  22576.           255 rotated by a multiple of 2

  22577. 

  22578.      'J'

  22579.           Integer in the range -4095 to 4095

  22580. 

  22581.      'K'

  22582.           Integer that satisfies constraint 'I' when inverted (ones

  22583.           complement)

  22584. 

  22585.      'L'

  22586.           Integer that satisfies constraint 'I' when negated (twos

  22587.           complement)

  22588. 

  22589.      'M'

  22590.           Integer in the range 0 to 32

  22591. 

  22592.      'Q'

  22593.           A memory reference where the exact address is in a single

  22594.           register (''m'' is preferable for 'asm' statements)

  22595. 

  22596.      'R'

  22597.           An item in the constant pool

  22598. 

  22599.      'S'

  22600.           A symbol in the text segment of the current file

  22601. 

  22602.      'Uv'

  22603.           A memory reference suitable for VFP load/store insns

  22604.           (reg+constant offset)

  22605. 

  22606.      'Uy'

  22607.           A memory reference suitable for iWMMXt load/store

  22608.           instructions.

  22609. 

  22610.      'Uq'

  22611.           A memory reference suitable for the ARMv4 ldrsb instruction.

  22612. 

  22613. _AVR family--'config/avr/constraints.md'_

  22614.      'l'

  22615.           Registers from r0 to r15

  22616. 

  22617.      'a'

  22618.           Registers from r16 to r23

  22619. 

  22620.      'd'

  22621.           Registers from r16 to r31

  22622. 

  22623.      'w'

  22624.           Registers from r24 to r31.  These registers can be used in

  22625.           'adiw' command

  22626. 

  22627.      'e'

  22628.           Pointer register (r26-r31)

  22629. 

  22630.      'b'

  22631.           Base pointer register (r28-r31)

  22632. 

  22633.      'q'

  22634.           Stack pointer register (SPH:SPL)

  22635. 

  22636.      't'

  22637.           Temporary register r0

  22638. 

  22639.      'x'

  22640.           Register pair X (r27:r26)

  22641. 

  22642.      'y'

  22643.           Register pair Y (r29:r28)

  22644. 

  22645.      'z'

  22646.           Register pair Z (r31:r30)

  22647. 

  22648.      'I'

  22649.           Constant greater than -1, less than 64

  22650. 

  22651.      'J'

  22652.           Constant greater than -64, less than 1

  22653. 

  22654.      'K'

  22655.           Constant integer 2

  22656. 

  22657.      'L'

  22658.           Constant integer 0

  22659. 

  22660.      'M'

  22661.           Constant that fits in 8 bits

  22662. 

  22663.      'N'

  22664.           Constant integer -1

  22665. 

  22666.      'O'

  22667.           Constant integer 8, 16, or 24

  22668. 

  22669.      'P'

  22670.           Constant integer 1

  22671. 

  22672.      'G'

  22673.           A floating point constant 0.0

  22674. 

  22675.      'Q'

  22676.           A memory address based on Y or Z pointer with displacement.

  22677. 

  22678. _Blackfin family--'config/bfin/constraints.md'_

  22679.      'a'

  22680.           P register

  22681. 

  22682.      'd'

  22683.           D register

  22684. 

  22685.      'z'

  22686.           A call clobbered P register.

  22687. 

  22688.      'qN'

  22689.           A single register.  If N is in the range 0 to 7, the

  22690.           corresponding D register.  If it is 'A', then the register P0.

  22691. 

  22692.      'D'

  22693.           Even-numbered D register

  22694. 

  22695.      'W'

  22696.           Odd-numbered D register

  22697. 

  22698.      'e'

  22699.           Accumulator register.

  22700. 

  22701.      'A'

  22702.           Even-numbered accumulator register.

  22703. 

  22704.      'B'

  22705.           Odd-numbered accumulator register.

  22706. 

  22707.      'b'

  22708.           I register

  22709. 

  22710.      'v'

  22711.           B register

  22712. 

  22713.      'f'

  22714.           M register

  22715. 

  22716.      'c'

  22717.           Registers used for circular buffering, i.e. I, B, or L

  22718.           registers.

  22719. 

  22720.      'C'

  22721.           The CC register.

  22722. 

  22723.      't'

  22724.           LT0 or LT1.

  22725. 

  22726.      'k'

  22727.           LC0 or LC1.

  22728. 

  22729.      'u'

  22730.           LB0 or LB1.

  22731. 

  22732.      'x'

  22733.           Any D, P, B, M, I or L register.

  22734. 

  22735.      'y'

  22736.           Additional registers typically used only in prologues and

  22737.           epilogues: RETS, RETN, RETI, RETX, RETE, ASTAT, SEQSTAT and

  22738.           USP.

  22739. 

  22740.      'w'

  22741.           Any register except accumulators or CC.

  22742. 

  22743.      'Ksh'

  22744.           Signed 16 bit integer (in the range -32768 to 32767)

  22745. 

  22746.      'Kuh'

  22747.           Unsigned 16 bit integer (in the range 0 to 65535)

  22748. 

  22749.      'Ks7'

  22750.           Signed 7 bit integer (in the range -64 to 63)

  22751. 

  22752.      'Ku7'

  22753.           Unsigned 7 bit integer (in the range 0 to 127)

  22754. 

  22755.      'Ku5'

  22756.           Unsigned 5 bit integer (in the range 0 to 31)

  22757. 

  22758.      'Ks4'

  22759.           Signed 4 bit integer (in the range -8 to 7)

  22760. 

  22761.      'Ks3'

  22762.           Signed 3 bit integer (in the range -3 to 4)

  22763. 

  22764.      'Ku3'

  22765.           Unsigned 3 bit integer (in the range 0 to 7)

  22766. 

  22767.      'PN'

  22768.           Constant N, where N is a single-digit constant in the range 0

  22769.           to 4.

  22770. 

  22771.      'PA'

  22772.           An integer equal to one of the MACFLAG_XXX constants that is

  22773.           suitable for use with either accumulator.

  22774. 

  22775.      'PB'

  22776.           An integer equal to one of the MACFLAG_XXX constants that is

  22777.           suitable for use only with accumulator A1.

  22778. 

  22779.      'M1'

  22780.           Constant 255.

  22781. 

  22782.      'M2'

  22783.           Constant 65535.

  22784. 

  22785.      'J'

  22786.           An integer constant with exactly a single bit set.

  22787. 

  22788.      'L'

  22789.           An integer constant with all bits set except exactly one.

  22790. 

  22791.      'H'

  22792. 

  22793.      'Q'

  22794.           Any SYMBOL_REF.

  22795. 

  22796. _CR16 Architecture--'config/cr16/cr16.h'_

  22797. 

  22798.      'b'

  22799.           Registers from r0 to r14 (registers without stack pointer)

  22800. 

  22801.      't'

  22802.           Register from r0 to r11 (all 16-bit registers)

  22803. 

  22804.      'p'

  22805.           Register from r12 to r15 (all 32-bit registers)

  22806. 

  22807.      'I'

  22808.           Signed constant that fits in 4 bits

  22809. 

  22810.      'J'

  22811.           Signed constant that fits in 5 bits

  22812. 

  22813.      'K'

  22814.           Signed constant that fits in 6 bits

  22815. 

  22816.      'L'

  22817.           Unsigned constant that fits in 4 bits

  22818. 

  22819.      'M'

  22820.           Signed constant that fits in 32 bits

  22821. 

  22822.      'N'

  22823.           Check for 64 bits wide constants for add/sub instructions

  22824. 

  22825.      'G'

  22826.           Floating point constant that is legal for store immediate

  22827. 

  22828. _C-SKY--'config/csky/constraints.md'_

  22829. 

  22830.      'a'

  22831.           The mini registers r0 - r7.

  22832. 

  22833.      'b'

  22834.           The low registers r0 - r15.

  22835. 

  22836.      'c'

  22837.           C register.

  22838. 

  22839.      'y'

  22840.           HI and LO registers.

  22841. 

  22842.      'l'

  22843.           LO register.

  22844. 

  22845.      'h'

  22846.           HI register.

  22847. 

  22848.      'v'

  22849.           Vector registers.

  22850. 

  22851.      'z'

  22852.           Stack pointer register (SP).

  22853. 

  22854.      The C-SKY back end supports a large set of additional constraints

  22855.      that are only useful for instruction selection or splitting rather

  22856.      than inline asm, such as constraints representing constant integer

  22857.      ranges accepted by particular instruction encodings.  Refer to the

  22858.      source code for details.

  22859. 

  22860. _Epiphany--'config/epiphany/constraints.md'_

  22861.      'U16'

  22862.           An unsigned 16-bit constant.

  22863. 

  22864.      'K'

  22865.           An unsigned 5-bit constant.

  22866. 

  22867.      'L'

  22868.           A signed 11-bit constant.

  22869. 

  22870.      'Cm1'

  22871.           A signed 11-bit constant added to -1.  Can only match when the

  22872.           '-m1reg-REG' option is active.

  22873. 

  22874.      'Cl1'

  22875.           Left-shift of -1, i.e., a bit mask with a block of leading

  22876.           ones, the rest being a block of trailing zeroes.  Can only

  22877.           match when the '-m1reg-REG' option is active.

  22878. 

  22879.      'Cr1'

  22880.           Right-shift of -1, i.e., a bit mask with a trailing block of

  22881.           ones, the rest being zeroes.  Or to put it another way, one

  22882.           less than a power of two.  Can only match when the

  22883.           '-m1reg-REG' option is active.

  22884. 

  22885.      'Cal'

  22886.           Constant for arithmetic/logical operations.  This is like 'i',

  22887.           except that for position independent code, no symbols /

  22888.           expressions needing relocations are allowed.

  22889. 

  22890.      'Csy'

  22891.           Symbolic constant for call/jump instruction.

  22892. 

  22893.      'Rcs'

  22894.           The register class usable in short insns.  This is a register

  22895.           class constraint, and can thus drive register allocation.

  22896.           This constraint won't match unless '-mprefer-short-insn-regs'

  22897.           is in effect.

  22898. 

  22899.      'Rsc'

  22900.           The the register class of registers that can be used to hold a

  22901.           sibcall call address.  I.e., a caller-saved register.

  22902. 

  22903.      'Rct'

  22904.           Core control register class.

  22905. 

  22906.      'Rgs'

  22907.           The register group usable in short insns.  This constraint

  22908.           does not use a register class, so that it only passively

  22909.           matches suitable registers, and doesn't drive register

  22910.           allocation.

  22911. 

  22912.      'Car'

  22913.           Constant suitable for the addsi3_r pattern.  This is a valid

  22914.           offset For byte, halfword, or word addressing.

  22915. 

  22916.      'Rra'

  22917.           Matches the return address if it can be replaced with the link

  22918.           register.

  22919. 

  22920.      'Rcc'

  22921.           Matches the integer condition code register.

  22922. 

  22923.      'Sra'

  22924.           Matches the return address if it is in a stack slot.

  22925. 

  22926.      'Cfm'

  22927.           Matches control register values to switch fp mode, which are

  22928.           encapsulated in 'UNSPEC_FP_MODE'.

  22929. 

  22930. _FRV--'config/frv/frv.h'_

  22931.      'a'

  22932.           Register in the class 'ACC_REGS' ('acc0' to 'acc7').

  22933. 

  22934.      'b'

  22935.           Register in the class 'EVEN_ACC_REGS' ('acc0' to 'acc7').

  22936. 

  22937.      'c'

  22938.           Register in the class 'CC_REGS' ('fcc0' to 'fcc3' and 'icc0'

  22939.           to 'icc3').

  22940. 

  22941.      'd'

  22942.           Register in the class 'GPR_REGS' ('gr0' to 'gr63').

  22943. 

  22944.      'e'

  22945.           Register in the class 'EVEN_REGS' ('gr0' to 'gr63').  Odd

  22946.           registers are excluded not in the class but through the use of

  22947.           a machine mode larger than 4 bytes.

  22948. 

  22949.      'f'

  22950.           Register in the class 'FPR_REGS' ('fr0' to 'fr63').

  22951. 

  22952.      'h'

  22953.           Register in the class 'FEVEN_REGS' ('fr0' to 'fr63').  Odd

  22954.           registers are excluded not in the class but through the use of

  22955.           a machine mode larger than 4 bytes.

  22956. 

  22957.      'l'

  22958.           Register in the class 'LR_REG' (the 'lr' register).

  22959. 

  22960.      'q'

  22961.           Register in the class 'QUAD_REGS' ('gr2' to 'gr63').  Register

  22962.           numbers not divisible by 4 are excluded not in the class but

  22963.           through the use of a machine mode larger than 8 bytes.

  22964. 

  22965.      't'

  22966.           Register in the class 'ICC_REGS' ('icc0' to 'icc3').

  22967. 

  22968.      'u'

  22969.           Register in the class 'FCC_REGS' ('fcc0' to 'fcc3').

  22970. 

  22971.      'v'

  22972.           Register in the class 'ICR_REGS' ('cc4' to 'cc7').

  22973. 

  22974.      'w'

  22975.           Register in the class 'FCR_REGS' ('cc0' to 'cc3').

  22976. 

  22977.      'x'

  22978.           Register in the class 'QUAD_FPR_REGS' ('fr0' to 'fr63').

  22979.           Register numbers not divisible by 4 are excluded not in the

  22980.           class but through the use of a machine mode larger than 8

  22981.           bytes.

  22982. 

  22983.      'z'

  22984.           Register in the class 'SPR_REGS' ('lcr' and 'lr').

  22985. 

  22986.      'A'

  22987.           Register in the class 'QUAD_ACC_REGS' ('acc0' to 'acc7').

  22988. 

  22989.      'B'

  22990.           Register in the class 'ACCG_REGS' ('accg0' to 'accg7').

  22991. 

  22992.      'C'

  22993.           Register in the class 'CR_REGS' ('cc0' to 'cc7').

  22994. 

  22995.      'G'

  22996.           Floating point constant zero

  22997. 

  22998.      'I'

  22999.           6-bit signed integer constant

  23000. 

  23001.      'J'

  23002.           10-bit signed integer constant

  23003. 

  23004.      'L'

  23005.           16-bit signed integer constant

  23006. 

  23007.      'M'

  23008.           16-bit unsigned integer constant

  23009. 

  23010.      'N'

  23011.           12-bit signed integer constant that is negative--i.e. in the

  23012.           range of -2048 to -1

  23013. 

  23014.      'O'

  23015.           Constant zero

  23016. 

  23017.      'P'

  23018.           12-bit signed integer constant that is greater than zero--i.e.

  23019.           in the range of 1 to 2047.

  23020. 

  23021. _FT32--'config/ft32/constraints.md'_

  23022.      'A'

  23023.           An absolute address

  23024. 

  23025.      'B'

  23026.           An offset address

  23027. 

  23028.      'W'

  23029.           A register indirect memory operand

  23030. 

  23031.      'e'

  23032.           An offset address.

  23033. 

  23034.      'f'

  23035.           An offset address.

  23036. 

  23037.      'O'

  23038.           The constant zero or one

  23039. 

  23040.      'I'

  23041.           A 16-bit signed constant (-32768 ... 32767)

  23042. 

  23043.      'w'

  23044.           A bitfield mask suitable for bext or bins

  23045. 

  23046.      'x'

  23047.           An inverted bitfield mask suitable for bext or bins

  23048. 

  23049.      'L'

  23050.           A 16-bit unsigned constant, multiple of 4 (0 ... 65532)

  23051. 

  23052.      'S'

  23053.           A 20-bit signed constant (-524288 ... 524287)

  23054. 

  23055.      'b'

  23056.           A constant for a bitfield width (1 ... 16)

  23057. 

  23058.      'KA'

  23059.           A 10-bit signed constant (-512 ... 511)

  23060. 

  23061. _Hewlett-Packard PA-RISC--'config/pa/pa.h'_

  23062.      'a'

  23063.           General register 1

  23064. 

  23065.      'f'

  23066.           Floating point register

  23067. 

  23068.      'q'

  23069.           Shift amount register

  23070. 

  23071.      'x'

  23072.           Floating point register (deprecated)

  23073. 

  23074.      'y'

  23075.           Upper floating point register (32-bit), floating point

  23076.           register (64-bit)

  23077. 

  23078.      'Z'

  23079.           Any register

  23080. 

  23081.      'I'

  23082.           Signed 11-bit integer constant

  23083. 

  23084.      'J'

  23085.           Signed 14-bit integer constant

  23086. 

  23087.      'K'

  23088.           Integer constant that can be deposited with a 'zdepi'

  23089.           instruction

  23090. 

  23091.      'L'

  23092.           Signed 5-bit integer constant

  23093. 

  23094.      'M'

  23095.           Integer constant 0

  23096. 

  23097.      'N'

  23098.           Integer constant that can be loaded with a 'ldil' instruction

  23099. 

  23100.      'O'

  23101.           Integer constant whose value plus one is a power of 2

  23102. 

  23103.      'P'

  23104.           Integer constant that can be used for 'and' operations in

  23105.           'depi' and 'extru' instructions

  23106. 

  23107.      'S'

  23108.           Integer constant 31

  23109. 

  23110.      'U'

  23111.           Integer constant 63

  23112. 

  23113.      'G'

  23114.           Floating-point constant 0.0

  23115. 

  23116.      'A'

  23117.           A 'lo_sum' data-linkage-table memory operand

  23118. 

  23119.      'Q'

  23120.           A memory operand that can be used as the destination operand

  23121.           of an integer store instruction

  23122. 

  23123.      'R'

  23124.           A scaled or unscaled indexed memory operand

  23125. 

  23126.      'T'

  23127.           A memory operand for floating-point loads and stores

  23128. 

  23129.      'W'

  23130.           A register indirect memory operand

  23131. 

  23132. _Intel IA-64--'config/ia64/ia64.h'_

  23133.      'a'

  23134.           General register 'r0' to 'r3' for 'addl' instruction

  23135. 

  23136.      'b'

  23137.           Branch register

  23138. 

  23139.      'c'

  23140.           Predicate register ('c' as in "conditional")

  23141. 

  23142.      'd'

  23143.           Application register residing in M-unit

  23144. 

  23145.      'e'

  23146.           Application register residing in I-unit

  23147. 

  23148.      'f'

  23149.           Floating-point register

  23150. 

  23151.      'm'

  23152.           Memory operand.  If used together with '<' or '>', the operand

  23153.           can have postincrement and postdecrement which require

  23154.           printing with '%Pn' on IA-64.

  23155. 

  23156.      'G'

  23157.           Floating-point constant 0.0 or 1.0

  23158. 

  23159.      'I'

  23160.           14-bit signed integer constant

  23161. 

  23162.      'J'

  23163.           22-bit signed integer constant

  23164. 

  23165.      'K'

  23166.           8-bit signed integer constant for logical instructions

  23167. 

  23168.      'L'

  23169.           8-bit adjusted signed integer constant for compare pseudo-ops

  23170. 

  23171.      'M'

  23172.           6-bit unsigned integer constant for shift counts

  23173. 

  23174.      'N'

  23175.           9-bit signed integer constant for load and store

  23176.           postincrements

  23177. 

  23178.      'O'

  23179.           The constant zero

  23180. 

  23181.      'P'

  23182.           0 or -1 for 'dep' instruction

  23183. 

  23184.      'Q'

  23185.           Non-volatile memory for floating-point loads and stores

  23186. 

  23187.      'R'

  23188.           Integer constant in the range 1 to 4 for 'shladd' instruction

  23189. 

  23190.      'S'

  23191.           Memory operand except postincrement and postdecrement.  This

  23192.           is now roughly the same as 'm' when not used together with '<'

  23193.           or '>'.

  23194. 

  23195. _M32C--'config/m32c/m32c.c'_

  23196.      'Rsp'

  23197.      'Rfb'

  23198.      'Rsb'

  23199.           '$sp', '$fb', '$sb'.

  23200. 

  23201.      'Rcr'

  23202.           Any control register, when they're 16 bits wide (nothing if

  23203.           control registers are 24 bits wide)

  23204. 

  23205.      'Rcl'

  23206.           Any control register, when they're 24 bits wide.

  23207. 

  23208.      'R0w'

  23209.      'R1w'

  23210.      'R2w'

  23211.      'R3w'

  23212.           $r0, $r1, $r2, $r3.

  23213. 

  23214.      'R02'

  23215.           $r0 or $r2, or $r2r0 for 32 bit values.

  23216. 

  23217.      'R13'

  23218.           $r1 or $r3, or $r3r1 for 32 bit values.

  23219. 

  23220.      'Rdi'

  23221.           A register that can hold a 64 bit value.

  23222. 

  23223.      'Rhl'

  23224.           $r0 or $r1 (registers with addressable high/low bytes)

  23225. 

  23226.      'R23'

  23227.           $r2 or $r3

  23228. 

  23229.      'Raa'

  23230.           Address registers

  23231. 

  23232.      'Raw'

  23233.           Address registers when they're 16 bits wide.

  23234. 

  23235.      'Ral'

  23236.           Address registers when they're 24 bits wide.

  23237. 

  23238.      'Rqi'

  23239.           Registers that can hold QI values.

  23240. 

  23241.      'Rad'

  23242.           Registers that can be used with displacements ($a0, $a1, $sb).

  23243. 

  23244.      'Rsi'

  23245.           Registers that can hold 32 bit values.

  23246. 

  23247.      'Rhi'

  23248.           Registers that can hold 16 bit values.

  23249. 

  23250.      'Rhc'

  23251.           Registers chat can hold 16 bit values, including all control

  23252.           registers.

  23253. 

  23254.      'Rra'

  23255.           $r0 through R1, plus $a0 and $a1.

  23256. 

  23257.      'Rfl'

  23258.           The flags register.

  23259. 

  23260.      'Rmm'

  23261.           The memory-based pseudo-registers $mem0 through $mem15.

  23262. 

  23263.      'Rpi'

  23264.           Registers that can hold pointers (16 bit registers for r8c,

  23265.           m16c; 24 bit registers for m32cm, m32c).

  23266. 

  23267.      'Rpa'

  23268.           Matches multiple registers in a PARALLEL to form a larger

  23269.           register.  Used to match function return values.

  23270. 

  23271.      'Is3'

  23272.           -8 ... 7

  23273. 

  23274.      'IS1'

  23275.           -128 ... 127

  23276. 

  23277.      'IS2'

  23278.           -32768 ... 32767

  23279. 

  23280.      'IU2'

  23281.           0 ... 65535

  23282. 

  23283.      'In4'

  23284.           -8 ... -1 or 1 ... 8

  23285. 

  23286.      'In5'

  23287.           -16 ... -1 or 1 ... 16

  23288. 

  23289.      'In6'

  23290.           -32 ... -1 or 1 ... 32

  23291. 

  23292.      'IM2'

  23293.           -65536 ... -1

  23294. 

  23295.      'Ilb'

  23296.           An 8 bit value with exactly one bit set.

  23297. 

  23298.      'Ilw'

  23299.           A 16 bit value with exactly one bit set.

  23300. 

  23301.      'Sd'

  23302.           The common src/dest memory addressing modes.

  23303. 

  23304.      'Sa'

  23305.           Memory addressed using $a0 or $a1.

  23306. 

  23307.      'Si'

  23308.           Memory addressed with immediate addresses.

  23309. 

  23310.      'Ss'

  23311.           Memory addressed using the stack pointer ($sp).

  23312. 

  23313.      'Sf'

  23314.           Memory addressed using the frame base register ($fb).

  23315. 

  23316.      'Ss'

  23317.           Memory addressed using the small base register ($sb).

  23318. 

  23319.      'S1'

  23320.           $r1h

  23321. 

  23322. _MicroBlaze--'config/microblaze/constraints.md'_

  23323.      'd'

  23324.           A general register ('r0' to 'r31').

  23325. 

  23326.      'z'

  23327.           A status register ('rmsr', '$fcc1' to '$fcc7').

  23328. 

  23329. _MIPS--'config/mips/constraints.md'_

  23330.      'd'

  23331.           A general-purpose register.  This is equivalent to 'r' unless

  23332.           generating MIPS16 code, in which case the MIPS16 register set

  23333.           is used.

  23334. 

  23335.      'f'

  23336.           A floating-point register (if available).

  23337. 

  23338.      'h'

  23339.           Formerly the 'hi' register.  This constraint is no longer

  23340.           supported.

  23341. 

  23342.      'l'

  23343.           The 'lo' register.  Use this register to store values that are

  23344.           no bigger than a word.

  23345. 

  23346.      'x'

  23347.           The concatenated 'hi' and 'lo' registers.  Use this register

  23348.           to store doubleword values.

  23349. 

  23350.      'c'

  23351.           A register suitable for use in an indirect jump.  This will

  23352.           always be '$25' for '-mabicalls'.

  23353. 

  23354.      'v'

  23355.           Register '$3'.  Do not use this constraint in new code; it is

  23356.           retained only for compatibility with glibc.

  23357. 

  23358.      'y'

  23359.           Equivalent to 'r'; retained for backwards compatibility.

  23360. 

  23361.      'z'

  23362.           A floating-point condition code register.

  23363. 

  23364.      'I'

  23365.           A signed 16-bit constant (for arithmetic instructions).

  23366. 

  23367.      'J'

  23368.           Integer zero.

  23369. 

  23370.      'K'

  23371.           An unsigned 16-bit constant (for logic instructions).

  23372. 

  23373.      'L'

  23374.           A signed 32-bit constant in which the lower 16 bits are zero.

  23375.           Such constants can be loaded using 'lui'.

  23376. 

  23377.      'M'

  23378.           A constant that cannot be loaded using 'lui', 'addiu' or

  23379.           'ori'.

  23380. 

  23381.      'N'

  23382.           A constant in the range -65535 to -1 (inclusive).

  23383. 

  23384.      'O'

  23385.           A signed 15-bit constant.

  23386. 

  23387.      'P'

  23388.           A constant in the range 1 to 65535 (inclusive).

  23389. 

  23390.      'G'

  23391.           Floating-point zero.

  23392. 

  23393.      'R'

  23394.           An address that can be used in a non-macro load or store.

  23395. 

  23396.      'ZC'

  23397.           A memory operand whose address is formed by a base register

  23398.           and offset that is suitable for use in instructions with the

  23399.           same addressing mode as 'll' and 'sc'.

  23400. 

  23401.      'ZD'

  23402.           An address suitable for a 'prefetch' instruction, or for any

  23403.           other instruction with the same addressing mode as 'prefetch'.

  23404. 

  23405. _Motorola 680x0--'config/m68k/constraints.md'_

  23406.      'a'

  23407.           Address register

  23408. 

  23409.      'd'

  23410.           Data register

  23411. 

  23412.      'f'

  23413.           68881 floating-point register, if available

  23414. 

  23415.      'I'

  23416.           Integer in the range 1 to 8

  23417. 

  23418.      'J'

  23419.           16-bit signed number

  23420. 

  23421.      'K'

  23422.           Signed number whose magnitude is greater than 0x80

  23423. 

  23424.      'L'

  23425.           Integer in the range -8 to -1

  23426. 

  23427.      'M'

  23428.           Signed number whose magnitude is greater than 0x100

  23429. 

  23430.      'N'

  23431.           Range 24 to 31, rotatert:SI 8 to 1 expressed as rotate

  23432. 

  23433.      'O'

  23434.           16 (for rotate using swap)

  23435. 

  23436.      'P'

  23437.           Range 8 to 15, rotatert:HI 8 to 1 expressed as rotate

  23438. 

  23439.      'R'

  23440.           Numbers that mov3q can handle

  23441. 

  23442.      'G'

  23443.           Floating point constant that is not a 68881 constant

  23444. 

  23445.      'S'

  23446.           Operands that satisfy 'm' when -mpcrel is in effect

  23447. 

  23448.      'T'

  23449.           Operands that satisfy 's' when -mpcrel is not in effect

  23450. 

  23451.      'Q'

  23452.           Address register indirect addressing mode

  23453. 

  23454.      'U'

  23455.           Register offset addressing

  23456. 

  23457.      'W'

  23458.           const_call_operand

  23459. 

  23460.      'Cs'

  23461.           symbol_ref or const

  23462. 

  23463.      'Ci'

  23464.           const_int

  23465. 

  23466.      'C0'

  23467.           const_int 0

  23468. 

  23469.      'Cj'

  23470.           Range of signed numbers that don't fit in 16 bits

  23471. 

  23472.      'Cmvq'

  23473.           Integers valid for mvq

  23474. 

  23475.      'Capsw'

  23476.           Integers valid for a moveq followed by a swap

  23477. 

  23478.      'Cmvz'

  23479.           Integers valid for mvz

  23480. 

  23481.      'Cmvs'

  23482.           Integers valid for mvs

  23483. 

  23484.      'Ap'

  23485.           push_operand

  23486. 

  23487.      'Ac'

  23488.           Non-register operands allowed in clr

  23489. 

  23490. _Moxie--'config/moxie/constraints.md'_

  23491.      'A'

  23492.           An absolute address

  23493. 

  23494.      'B'

  23495.           An offset address

  23496. 

  23497.      'W'

  23498.           A register indirect memory operand

  23499. 

  23500.      'I'

  23501.           A constant in the range of 0 to 255.

  23502. 

  23503.      'N'

  23504.           A constant in the range of 0 to -255.

  23505. 

  23506. _MSP430-'config/msp430/constraints.md'_

  23507. 

  23508.      'R12'

  23509.           Register R12.

  23510. 

  23511.      'R13'

  23512.           Register R13.

  23513. 

  23514.      'K'

  23515.           Integer constant 1.

  23516. 

  23517.      'L'

  23518.           Integer constant -1^20..1^19.

  23519. 

  23520.      'M'

  23521.           Integer constant 1-4.

  23522. 

  23523.      'Ya'

  23524.           Memory references which do not require an extended MOVX

  23525.           instruction.

  23526. 

  23527.      'Yl'

  23528.           Memory reference, labels only.

  23529. 

  23530.      'Ys'

  23531.           Memory reference, stack only.

  23532. 

  23533. _NDS32--'config/nds32/constraints.md'_

  23534.      'w'

  23535.           LOW register class $r0 to $r7 constraint for V3/V3M ISA.

  23536.      'l'

  23537.           LOW register class $r0 to $r7.

  23538.      'd'

  23539.           MIDDLE register class $r0 to $r11, $r16 to $r19.

  23540.      'h'

  23541.           HIGH register class $r12 to $r14, $r20 to $r31.

  23542.      't'

  23543.           Temporary assist register $ta (i.e. $r15).

  23544.      'k'

  23545.           Stack register $sp.

  23546.      'Iu03'

  23547.           Unsigned immediate 3-bit value.

  23548.      'In03'

  23549.           Negative immediate 3-bit value in the range of -7-0.

  23550.      'Iu04'

  23551.           Unsigned immediate 4-bit value.

  23552.      'Is05'

  23553.           Signed immediate 5-bit value.

  23554.      'Iu05'

  23555.           Unsigned immediate 5-bit value.

  23556.      'In05'

  23557.           Negative immediate 5-bit value in the range of -31-0.

  23558.      'Ip05'

  23559.           Unsigned immediate 5-bit value for movpi45 instruction with

  23560.           range 16-47.

  23561.      'Iu06'

  23562.           Unsigned immediate 6-bit value constraint for addri36.sp

  23563.           instruction.

  23564.      'Iu08'

  23565.           Unsigned immediate 8-bit value.

  23566.      'Iu09'

  23567.           Unsigned immediate 9-bit value.

  23568.      'Is10'

  23569.           Signed immediate 10-bit value.

  23570.      'Is11'

  23571.           Signed immediate 11-bit value.

  23572.      'Is15'

  23573.           Signed immediate 15-bit value.

  23574.      'Iu15'

  23575.           Unsigned immediate 15-bit value.

  23576.      'Ic15'

  23577.           A constant which is not in the range of imm15u but ok for bclr

  23578.           instruction.

  23579.      'Ie15'

  23580.           A constant which is not in the range of imm15u but ok for bset

  23581.           instruction.

  23582.      'It15'

  23583.           A constant which is not in the range of imm15u but ok for btgl

  23584.           instruction.

  23585.      'Ii15'

  23586.           A constant whose compliment value is in the range of imm15u

  23587.           and ok for bitci instruction.

  23588.      'Is16'

  23589.           Signed immediate 16-bit value.

  23590.      'Is17'

  23591.           Signed immediate 17-bit value.

  23592.      'Is19'

  23593.           Signed immediate 19-bit value.

  23594.      'Is20'

  23595.           Signed immediate 20-bit value.

  23596.      'Ihig'

  23597.           The immediate value that can be simply set high 20-bit.

  23598.      'Izeb'

  23599.           The immediate value 0xff.

  23600.      'Izeh'

  23601.           The immediate value 0xffff.

  23602.      'Ixls'

  23603.           The immediate value 0x01.

  23604.      'Ix11'

  23605.           The immediate value 0x7ff.

  23606.      'Ibms'

  23607.           The immediate value with power of 2.

  23608.      'Ifex'

  23609.           The immediate value with power of 2 minus 1.

  23610.      'U33'

  23611.           Memory constraint for 333 format.

  23612.      'U45'

  23613.           Memory constraint for 45 format.

  23614.      'U37'

  23615.           Memory constraint for 37 format.

  23616. 

  23617. _Nios II family--'config/nios2/constraints.md'_

  23618. 

  23619.      'I'

  23620.           Integer that is valid as an immediate operand in an

  23621.           instruction taking a signed 16-bit number.  Range -32768 to

  23622.           32767.

  23623. 

  23624.      'J'

  23625.           Integer that is valid as an immediate operand in an

  23626.           instruction taking an unsigned 16-bit number.  Range 0 to

  23627.           65535.

  23628. 

  23629.      'K'

  23630.           Integer that is valid as an immediate operand in an

  23631.           instruction taking only the upper 16-bits of a 32-bit number.

  23632.           Range 32-bit numbers with the lower 16-bits being 0.

  23633. 

  23634.      'L'

  23635.           Integer that is valid as an immediate operand for a shift

  23636.           instruction.  Range 0 to 31.

  23637. 

  23638.      'M'

  23639.           Integer that is valid as an immediate operand for only the

  23640.           value 0.  Can be used in conjunction with the format modifier

  23641.           'z' to use 'r0' instead of '0' in the assembly output.

  23642. 

  23643.      'N'

  23644.           Integer that is valid as an immediate operand for a custom

  23645.           instruction opcode.  Range 0 to 255.

  23646. 

  23647.      'P'

  23648.           An immediate operand for R2 andchi/andci instructions.

  23649. 

  23650.      'S'

  23651.           Matches immediates which are addresses in the small data

  23652.           section and therefore can be added to 'gp' as a 16-bit

  23653.           immediate to re-create their 32-bit value.

  23654. 

  23655.      'U'

  23656.           Matches constants suitable as an operand for the rdprs and

  23657.           cache instructions.

  23658. 

  23659.      'v'

  23660.           A memory operand suitable for Nios II R2 load/store exclusive

  23661.           instructions.

  23662. 

  23663.      'w'

  23664.           A memory operand suitable for load/store IO and cache

  23665.           instructions.

  23666. 

  23667.      'T'

  23668.           A 'const' wrapped 'UNSPEC' expression, representing a

  23669.           supported PIC or TLS relocation.

  23670. 

  23671. _OpenRISC--'config/or1k/constraints.md'_

  23672.      'I'

  23673.           Integer that is valid as an immediate operand in an

  23674.           instruction taking a signed 16-bit number.  Range -32768 to

  23675.           32767.

  23676. 

  23677.      'K'

  23678.           Integer that is valid as an immediate operand in an

  23679.           instruction taking an unsigned 16-bit number.  Range 0 to

  23680.           65535.

  23681. 

  23682.      'M'

  23683.           Signed 16-bit constant shifted left 16 bits.  (Used with

  23684.           'l.movhi')

  23685. 

  23686.      'O'

  23687.           Zero

  23688. 

  23689.      'c'

  23690.           Register usable for sibcalls.

  23691. 

  23692. _PDP-11--'config/pdp11/constraints.md'_

  23693.      'a'

  23694.           Floating point registers AC0 through AC3.  These can be loaded

  23695.           from/to memory with a single instruction.

  23696. 

  23697.      'd'

  23698.           Odd numbered general registers (R1, R3, R5).  These are used

  23699.           for 16-bit multiply operations.

  23700. 

  23701.      'D'

  23702.           A memory reference that is encoded within the opcode, but not

  23703.           auto-increment or auto-decrement.

  23704. 

  23705.      'f'

  23706.           Any of the floating point registers (AC0 through AC5).

  23707. 

  23708.      'G'

  23709.           Floating point constant 0.

  23710. 

  23711.      'h'

  23712.           Floating point registers AC4 and AC5.  These cannot be loaded

  23713.           from/to memory with a single instruction.

  23714. 

  23715.      'I'

  23716.           An integer constant that fits in 16 bits.

  23717. 

  23718.      'J'

  23719.           An integer constant whose low order 16 bits are zero.

  23720. 

  23721.      'K'

  23722.           An integer constant that does not meet the constraints for

  23723.           codes 'I' or 'J'.

  23724. 

  23725.      'L'

  23726.           The integer constant 1.

  23727. 

  23728.      'M'

  23729.           The integer constant -1.

  23730. 

  23731.      'N'

  23732.           The integer constant 0.

  23733. 

  23734.      'O'

  23735.           Integer constants 0 through 3; shifts by these amounts are

  23736.           handled as multiple single-bit shifts rather than a single

  23737.           variable-length shift.

  23738. 

  23739.      'Q'

  23740.           A memory reference which requires an additional word (address

  23741.           or offset) after the opcode.

  23742. 

  23743.      'R'

  23744.           A memory reference that is encoded within the opcode.

  23745. 

  23746. _PowerPC and IBM RS6000--'config/rs6000/constraints.md'_

  23747.      'r'

  23748.           A general purpose register (GPR), 'r0'...'r31'.

  23749. 

  23750.      'b'

  23751.           A base register.  Like 'r', but 'r0' is not allowed, so

  23752.           'r1'...'r31'.

  23753. 

  23754.      'f'

  23755.           A floating point register (FPR), 'f0'...'f31'.

  23756. 

  23757.      'd'

  23758.           A floating point register.  This is the same as 'f' nowadays;

  23759.           historically 'f' was for single-precision and 'd' was for

  23760.           double-precision floating point.

  23761. 

  23762.      'v'

  23763.           An Altivec vector register (VR), 'v0'...'v31'.

  23764. 

  23765.      'wa'

  23766.           A VSX register (VSR), 'vs0'...'vs63'.  This is either an FPR

  23767.           ('vs0'...'vs31' are 'f0'...'f31') or a VR ('vs32'...'vs63' are

  23768.           'v0'...'v31').

  23769. 

  23770.           When using 'wa', you should use the '%x' output modifier, so

  23771.           that the correct register number is printed.  For example:

  23772. 

  23773.                asm ("xvadddp %x0,%x1,%x2"

  23774.                     : "=wa" (v1)

  23775.                     : "wa" (v2), "wa" (v3));

  23776. 

  23777.           You should not use '%x' for 'v' operands:

  23778. 

  23779.                asm ("xsaddqp %0,%1,%2"

  23780.                     : "=v" (v1)

  23781.                     : "v" (v2), "v" (v3));

  23782. 

  23783.      'h'

  23784.           A special register ('vrsave', 'ctr', or 'lr').

  23785. 

  23786.      'c'

  23787.           The count register, 'ctr'.

  23788. 

  23789.      'l'

  23790.           The link register, 'lr'.

  23791. 

  23792.      'x'

  23793.           Condition register field 0, 'cr0'.

  23794. 

  23795.      'y'

  23796.           Any condition register field, 'cr0'...'cr7'.

  23797. 

  23798.      'z'

  23799.           The carry bit, 'XER[CA]'.

  23800. 

  23801.      'we'

  23802.           Like 'wa', if '-mpower9-vector' and '-m64' are used;

  23803.           otherwise, 'NO_REGS'.

  23804. 

  23805.      'wn'

  23806.           No register ('NO_REGS').

  23807. 

  23808.      'wr'

  23809.           Like 'r', if '-mpowerpc64' is used; otherwise, 'NO_REGS'.

  23810. 

  23811.      'wx'

  23812.           Like 'd', if '-mpowerpc-gfxopt' is used; otherwise, 'NO_REGS'.

  23813. 

  23814.      'wA'

  23815.           Like 'b', if '-mpowerpc64' is used; otherwise, 'NO_REGS'.

  23816. 

  23817.      'wB'

  23818.           Signed 5-bit constant integer that can be loaded into an

  23819.           Altivec register.

  23820. 

  23821.      'wD'

  23822.           Int constant that is the element number of the 64-bit scalar

  23823.           in a vector.

  23824. 

  23825.      'wE'

  23826.           Vector constant that can be loaded with the XXSPLTIB

  23827.           instruction.

  23828. 

  23829.      'wF'

  23830.           Memory operand suitable for power8 GPR load fusion.

  23831. 

  23832.      'wL'

  23833.           Int constant that is the element number mfvsrld accesses in a

  23834.           vector.

  23835. 

  23836.      'wM'

  23837.           Match vector constant with all 1's if the XXLORC instruction

  23838.           is available.

  23839. 

  23840.      'wO'

  23841.           Memory operand suitable for the ISA 3.0 vector d-form

  23842.           instructions.

  23843. 

  23844.      'wQ'

  23845.           Memory operand suitable for the load/store quad instructions.

  23846. 

  23847.      'wS'

  23848.           Vector constant that can be loaded with XXSPLTIB & sign

  23849.           extension.

  23850. 

  23851.      'wY'

  23852.           A memory operand for a DS-form instruction.

  23853. 

  23854.      'wZ'

  23855.           An indexed or indirect memory operand, ignoring the bottom 4

  23856.           bits.

  23857. 

  23858.      'I'

  23859.           A signed 16-bit constant.

  23860. 

  23861.      'J'

  23862.           An unsigned 16-bit constant shifted left 16 bits (use 'L'

  23863.           instead for 'SImode' constants).

  23864. 

  23865.      'K'

  23866.           An unsigned 16-bit constant.

  23867. 

  23868.      'L'

  23869.           A signed 16-bit constant shifted left 16 bits.

  23870. 

  23871.      'M'

  23872.           An integer constant greater than 31.

  23873. 

  23874.      'N'

  23875.           An exact power of 2.

  23876. 

  23877.      'O'

  23878.           The integer constant zero.

  23879. 

  23880.      'P'

  23881.           A constant whose negation is a signed 16-bit constant.

  23882. 

  23883.      'eI'

  23884.           A signed 34-bit integer constant if prefixed instructions are

  23885.           supported.

  23886. 

  23887.      'G'

  23888.           A floating point constant that can be loaded into a register

  23889.           with one instruction per word.

  23890. 

  23891.      'H'

  23892.           A floating point constant that can be loaded into a register

  23893.           using three instructions.

  23894. 

  23895.      'm'

  23896.           A memory operand.  Normally, 'm' does not allow addresses that

  23897.           update the base register.  If the '<' or '>' constraint is

  23898.           also used, they are allowed and therefore on PowerPC targets

  23899.           in that case it is only safe to use 'm<>' in an 'asm'

  23900.           statement if that 'asm' statement accesses the operand exactly

  23901.           once.  The 'asm' statement must also use '%U<OPNO>' as a

  23902.           placeholder for the "update" flag in the corresponding load or

  23903.           store instruction.  For example:

  23904. 

  23905.                asm ("st%U0 %1,%0" : "=m<>" (mem) : "r" (val));

  23906. 

  23907.           is correct but:

  23908. 

  23909.                asm ("st %1,%0" : "=m<>" (mem) : "r" (val));

  23910. 

  23911.           is not.

  23912. 

  23913.      'es'

  23914.           A "stable" memory operand; that is, one which does not include

  23915.           any automodification of the base register.  This used to be

  23916.           useful when 'm' allowed automodification of the base register,

  23917.           but as those are now only allowed when '<' or '>' is used,

  23918.           'es' is basically the same as 'm' without '<' and '>'.

  23919. 

  23920.      'Q'

  23921.           A memory operand addressed by just a base register.

  23922. 

  23923.      'Y'

  23924.           A memory operand for a DQ-form instruction.

  23925. 

  23926.      'Z'

  23927.           A memory operand accessed with indexed or indirect addressing.

  23928. 

  23929.      'R'

  23930.           An AIX TOC entry.

  23931. 

  23932.      'a'

  23933.           An indexed or indirect address.

  23934. 

  23935.      'U'

  23936.           A V.4 small data reference.

  23937. 

  23938.      'W'

  23939.           A vector constant that does not require memory.

  23940. 

  23941.      'j'

  23942.           The zero vector constant.

  23943. 

  23944. _PRU--'config/pru/constraints.md'_

  23945.      'I'

  23946.           An unsigned 8-bit integer constant.

  23947. 

  23948.      'J'

  23949.           An unsigned 16-bit integer constant.

  23950. 

  23951.      'L'

  23952.           An unsigned 5-bit integer constant (for shift counts).

  23953. 

  23954.      'T'

  23955.           A text segment (program memory) constant label.

  23956. 

  23957.      'Z'

  23958.           Integer constant zero.

  23959. 

  23960. _RL78--'config/rl78/constraints.md'_

  23961. 

  23962.      'Int3'

  23963.           An integer constant in the range 1 ... 7.

  23964.      'Int8'

  23965.           An integer constant in the range 0 ... 255.

  23966.      'J'

  23967.           An integer constant in the range -255 ... 0

  23968.      'K'

  23969.           The integer constant 1.

  23970.      'L'

  23971.           The integer constant -1.

  23972.      'M'

  23973.           The integer constant 0.

  23974.      'N'

  23975.           The integer constant 2.

  23976.      'O'

  23977.           The integer constant -2.

  23978.      'P'

  23979.           An integer constant in the range 1 ... 15.

  23980.      'Qbi'

  23981.           The built-in compare types-eq, ne, gtu, ltu, geu, and leu.

  23982.      'Qsc'

  23983.           The synthetic compare types-gt, lt, ge, and le.

  23984.      'Wab'

  23985.           A memory reference with an absolute address.

  23986.      'Wbc'

  23987.           A memory reference using 'BC' as a base register, with an

  23988.           optional offset.

  23989.      'Wca'

  23990.           A memory reference using 'AX', 'BC', 'DE', or 'HL' for the

  23991.           address, for calls.

  23992.      'Wcv'

  23993.           A memory reference using any 16-bit register pair for the

  23994.           address, for calls.

  23995.      'Wd2'

  23996.           A memory reference using 'DE' as a base register, with an

  23997.           optional offset.

  23998.      'Wde'

  23999.           A memory reference using 'DE' as a base register, without any

  24000.           offset.

  24001.      'Wfr'

  24002.           Any memory reference to an address in the far address space.

  24003.      'Wh1'

  24004.           A memory reference using 'HL' as a base register, with an

  24005.           optional one-byte offset.

  24006.      'Whb'

  24007.           A memory reference using 'HL' as a base register, with 'B' or

  24008.           'C' as the index register.

  24009.      'Whl'

  24010.           A memory reference using 'HL' as a base register, without any

  24011.           offset.

  24012.      'Ws1'

  24013.           A memory reference using 'SP' as a base register, with an

  24014.           optional one-byte offset.

  24015.      'Y'

  24016.           Any memory reference to an address in the near address space.

  24017.      'A'

  24018.           The 'AX' register.

  24019.      'B'

  24020.           The 'BC' register.

  24021.      'D'

  24022.           The 'DE' register.

  24023.      'R'

  24024.           'A' through 'L' registers.

  24025.      'S'

  24026.           The 'SP' register.

  24027.      'T'

  24028.           The 'HL' register.

  24029.      'Z08W'

  24030.           The 16-bit 'R8' register.

  24031.      'Z10W'

  24032.           The 16-bit 'R10' register.

  24033.      'Zint'

  24034.           The registers reserved for interrupts ('R24' to 'R31').

  24035.      'a'

  24036.           The 'A' register.

  24037.      'b'

  24038.           The 'B' register.

  24039.      'c'

  24040.           The 'C' register.

  24041.      'd'

  24042.           The 'D' register.

  24043.      'e'

  24044.           The 'E' register.

  24045.      'h'

  24046.           The 'H' register.

  24047.      'l'

  24048.           The 'L' register.

  24049.      'v'

  24050.           The virtual registers.

  24051.      'w'

  24052.           The 'PSW' register.

  24053.      'x'

  24054.           The 'X' register.

  24055. 

  24056. _RISC-V--'config/riscv/constraints.md'_

  24057. 

  24058.      'f'

  24059.           A floating-point register (if available).

  24060. 

  24061.      'I'

  24062.           An I-type 12-bit signed immediate.

  24063. 

  24064.      'J'

  24065.           Integer zero.

  24066. 

  24067.      'K'

  24068.           A 5-bit unsigned immediate for CSR access instructions.

  24069. 

  24070.      'A'

  24071.           An address that is held in a general-purpose register.

  24072. 

  24073. _RX--'config/rx/constraints.md'_

  24074.      'Q'

  24075.           An address which does not involve register indirect addressing

  24076.           or pre/post increment/decrement addressing.

  24077. 

  24078.      'Symbol'

  24079.           A symbol reference.

  24080. 

  24081.      'Int08'

  24082.           A constant in the range -256 to 255, inclusive.

  24083. 

  24084.      'Sint08'

  24085.           A constant in the range -128 to 127, inclusive.

  24086. 

  24087.      'Sint16'

  24088.           A constant in the range -32768 to 32767, inclusive.

  24089. 

  24090.      'Sint24'

  24091.           A constant in the range -8388608 to 8388607, inclusive.

  24092. 

  24093.      'Uint04'

  24094.           A constant in the range 0 to 15, inclusive.

  24095. 

  24096. _S/390 and zSeries--'config/s390/s390.h'_

  24097.      'a'

  24098.           Address register (general purpose register except r0)

  24099. 

  24100.      'c'

  24101.           Condition code register

  24102. 

  24103.      'd'

  24104.           Data register (arbitrary general purpose register)

  24105. 

  24106.      'f'

  24107.           Floating-point register

  24108. 

  24109.      'I'

  24110.           Unsigned 8-bit constant (0-255)

  24111. 

  24112.      'J'

  24113.           Unsigned 12-bit constant (0-4095)

  24114. 

  24115.      'K'

  24116.           Signed 16-bit constant (-32768-32767)

  24117. 

  24118.      'L'

  24119.           Value appropriate as displacement.

  24120.           '(0..4095)'

  24121.                for short displacement

  24122.           '(-524288..524287)'

  24123.                for long displacement

  24124. 

  24125.      'M'

  24126.           Constant integer with a value of 0x7fffffff.

  24127. 

  24128.      'N'

  24129.           Multiple letter constraint followed by 4 parameter letters.

  24130.           '0..9:'

  24131.                number of the part counting from most to least

  24132.                significant

  24133.           'H,Q:'

  24134.                mode of the part

  24135.           'D,S,H:'

  24136.                mode of the containing operand

  24137.           '0,F:'

  24138.                value of the other parts (F--all bits set)

  24139.           The constraint matches if the specified part of a constant has

  24140.           a value different from its other parts.

  24141. 

  24142.      'Q'

  24143.           Memory reference without index register and with short

  24144.           displacement.

  24145. 

  24146.      'R'

  24147.           Memory reference with index register and short displacement.

  24148. 

  24149.      'S'

  24150.           Memory reference without index register but with long

  24151.           displacement.

  24152. 

  24153.      'T'

  24154.           Memory reference with index register and long displacement.

  24155. 

  24156.      'U'

  24157.           Pointer with short displacement.

  24158. 

  24159.      'W'

  24160.           Pointer with long displacement.

  24161. 

  24162.      'Y'

  24163.           Shift count operand.

  24164. 

  24165. _SPARC--'config/sparc/sparc.h'_

  24166.      'f'

  24167.           Floating-point register on the SPARC-V8 architecture and lower

  24168.           floating-point register on the SPARC-V9 architecture.

  24169. 

  24170.      'e'

  24171.           Floating-point register.  It is equivalent to 'f' on the

  24172.           SPARC-V8 architecture and contains both lower and upper

  24173.           floating-point registers on the SPARC-V9 architecture.

  24174. 

  24175.      'c'

  24176.           Floating-point condition code register.

  24177. 

  24178.      'd'

  24179.           Lower floating-point register.  It is only valid on the

  24180.           SPARC-V9 architecture when the Visual Instruction Set is

  24181.           available.

  24182. 

  24183.      'b'

  24184.           Floating-point register.  It is only valid on the SPARC-V9

  24185.           architecture when the Visual Instruction Set is available.

  24186. 

  24187.      'h'

  24188.           64-bit global or out register for the SPARC-V8+ architecture.

  24189. 

  24190.      'C'

  24191.           The constant all-ones, for floating-point.

  24192. 

  24193.      'A'

  24194.           Signed 5-bit constant

  24195. 

  24196.      'D'

  24197.           A vector constant

  24198. 

  24199.      'I'

  24200.           Signed 13-bit constant

  24201. 

  24202.      'J'

  24203.           Zero

  24204. 

  24205.      'K'

  24206.           32-bit constant with the low 12 bits clear (a constant that

  24207.           can be loaded with the 'sethi' instruction)

  24208. 

  24209.      'L'

  24210.           A constant in the range supported by 'movcc' instructions

  24211.           (11-bit signed immediate)

  24212. 

  24213.      'M'

  24214.           A constant in the range supported by 'movrcc' instructions

  24215.           (10-bit signed immediate)

  24216. 

  24217.      'N'

  24218.           Same as 'K', except that it verifies that bits that are not in

  24219.           the lower 32-bit range are all zero.  Must be used instead of

  24220.           'K' for modes wider than 'SImode'

  24221. 

  24222.      'O'

  24223.           The constant 4096

  24224. 

  24225.      'G'

  24226.           Floating-point zero

  24227. 

  24228.      'H'

  24229.           Signed 13-bit constant, sign-extended to 32 or 64 bits

  24230. 

  24231.      'P'

  24232.           The constant -1

  24233. 

  24234.      'Q'

  24235.           Floating-point constant whose integral representation can be

  24236.           moved into an integer register using a single sethi

  24237.           instruction

  24238. 

  24239.      'R'

  24240.           Floating-point constant whose integral representation can be

  24241.           moved into an integer register using a single mov instruction

  24242. 

  24243.      'S'

  24244.           Floating-point constant whose integral representation can be

  24245.           moved into an integer register using a high/lo_sum instruction

  24246.           sequence

  24247. 

  24248.      'T'

  24249.           Memory address aligned to an 8-byte boundary

  24250. 

  24251.      'U'

  24252.           Even register

  24253. 

  24254.      'W'

  24255.           Memory address for 'e' constraint registers

  24256. 

  24257.      'w'

  24258.           Memory address with only a base register

  24259. 

  24260.      'Y'

  24261.           Vector zero

  24262. 

  24263. _TI C6X family--'config/c6x/constraints.md'_

  24264.      'a'

  24265.           Register file A (A0-A31).

  24266. 

  24267.      'b'

  24268.           Register file B (B0-B31).

  24269. 

  24270.      'A'

  24271.           Predicate registers in register file A (A0-A2 on C64X and

  24272.           higher, A1 and A2 otherwise).

  24273. 

  24274.      'B'

  24275.           Predicate registers in register file B (B0-B2).

  24276. 

  24277.      'C'

  24278.           A call-used register in register file B (B0-B9, B16-B31).

  24279. 

  24280.      'Da'

  24281.           Register file A, excluding predicate registers (A3-A31, plus

  24282.           A0 if not C64X or higher).

  24283. 

  24284.      'Db'

  24285.           Register file B, excluding predicate registers (B3-B31).

  24286. 

  24287.      'Iu4'

  24288.           Integer constant in the range 0 ... 15.

  24289. 

  24290.      'Iu5'

  24291.           Integer constant in the range 0 ... 31.

  24292. 

  24293.      'In5'

  24294.           Integer constant in the range -31 ... 0.

  24295. 

  24296.      'Is5'

  24297.           Integer constant in the range -16 ... 15.

  24298. 

  24299.      'I5x'

  24300.           Integer constant that can be the operand of an ADDA or a SUBA

  24301.           insn.

  24302. 

  24303.      'IuB'

  24304.           Integer constant in the range 0 ... 65535.

  24305. 

  24306.      'IsB'

  24307.           Integer constant in the range -32768 ... 32767.

  24308. 

  24309.      'IsC'

  24310.           Integer constant in the range -2^{20} ... 2^{20} - 1.

  24311. 

  24312.      'Jc'

  24313.           Integer constant that is a valid mask for the clr instruction.

  24314. 

  24315.      'Js'

  24316.           Integer constant that is a valid mask for the set instruction.

  24317. 

  24318.      'Q'

  24319.           Memory location with A base register.

  24320. 

  24321.      'R'

  24322.           Memory location with B base register.

  24323. 

  24324.      'S0'

  24325.           On C64x+ targets, a GP-relative small data reference.

  24326. 

  24327.      'S1'

  24328.           Any kind of 'SYMBOL_REF', for use in a call address.

  24329. 

  24330.      'Si'

  24331.           Any kind of immediate operand, unless it matches the S0

  24332.           constraint.

  24333. 

  24334.      'T'

  24335.           Memory location with B base register, but not using a long

  24336.           offset.

  24337. 

  24338.      'W'

  24339.           A memory operand with an address that cannot be used in an

  24340.           unaligned access.

  24341. 

  24342.      'Z'

  24343.           Register B14 (aka DP).

  24344. 

  24345. _TILE-Gx--'config/tilegx/constraints.md'_

  24346.      'R00'

  24347.      'R01'

  24348.      'R02'

  24349.      'R03'

  24350.      'R04'

  24351.      'R05'

  24352.      'R06'

  24353.      'R07'

  24354.      'R08'

  24355.      'R09'

  24356.      'R10'

  24357.           Each of these represents a register constraint for an

  24358.           individual register, from r0 to r10.

  24359. 

  24360.      'I'

  24361.           Signed 8-bit integer constant.

  24362. 

  24363.      'J'

  24364.           Signed 16-bit integer constant.

  24365. 

  24366.      'K'

  24367.           Unsigned 16-bit integer constant.

  24368. 

  24369.      'L'

  24370.           Integer constant that fits in one signed byte when incremented

  24371.           by one (-129 ... 126).

  24372. 

  24373.      'm'

  24374.           Memory operand.  If used together with '<' or '>', the operand

  24375.           can have postincrement which requires printing with '%In' and

  24376.           '%in' on TILE-Gx.  For example:

  24377. 

  24378.                asm ("st_add %I0,%1,%i0" : "=m<>" (*mem) : "r" (val));

  24379. 

  24380.      'M'

  24381.           A bit mask suitable for the BFINS instruction.

  24382. 

  24383.      'N'

  24384.           Integer constant that is a byte tiled out eight times.

  24385. 

  24386.      'O'

  24387.           The integer zero constant.

  24388. 

  24389.      'P'

  24390.           Integer constant that is a sign-extended byte tiled out as

  24391.           four shorts.

  24392. 

  24393.      'Q'

  24394.           Integer constant that fits in one signed byte when incremented

  24395.           (-129 ... 126), but excluding -1.

  24396. 

  24397.      'S'

  24398.           Integer constant that has all 1 bits consecutive and starting

  24399.           at bit 0.

  24400. 

  24401.      'T'

  24402.           A 16-bit fragment of a got, tls, or pc-relative reference.

  24403. 

  24404.      'U'

  24405.           Memory operand except postincrement.  This is roughly the same

  24406.           as 'm' when not used together with '<' or '>'.

  24407. 

  24408.      'W'

  24409.           An 8-element vector constant with identical elements.

  24410. 

  24411.      'Y'

  24412.           A 4-element vector constant with identical elements.

  24413. 

  24414.      'Z0'

  24415.           The integer constant 0xffffffff.

  24416. 

  24417.      'Z1'

  24418.           The integer constant 0xffffffff00000000.

  24419. 

  24420. _TILEPro--'config/tilepro/constraints.md'_

  24421.      'R00'

  24422.      'R01'

  24423.      'R02'

  24424.      'R03'

  24425.      'R04'

  24426.      'R05'

  24427.      'R06'

  24428.      'R07'

  24429.      'R08'

  24430.      'R09'

  24431.      'R10'

  24432.           Each of these represents a register constraint for an

  24433.           individual register, from r0 to r10.

  24434. 

  24435.      'I'

  24436.           Signed 8-bit integer constant.

  24437. 

  24438.      'J'

  24439.           Signed 16-bit integer constant.

  24440. 

  24441.      'K'

  24442.           Nonzero integer constant with low 16 bits zero.

  24443. 

  24444.      'L'

  24445.           Integer constant that fits in one signed byte when incremented

  24446.           by one (-129 ... 126).

  24447. 

  24448.      'm'

  24449.           Memory operand.  If used together with '<' or '>', the operand

  24450.           can have postincrement which requires printing with '%In' and

  24451.           '%in' on TILEPro.  For example:

  24452. 

  24453.                asm ("swadd %I0,%1,%i0" : "=m<>" (mem) : "r" (val));

  24454. 

  24455.      'M'

  24456.           A bit mask suitable for the MM instruction.

  24457. 

  24458.      'N'

  24459.           Integer constant that is a byte tiled out four times.

  24460. 

  24461.      'O'

  24462.           The integer zero constant.

  24463. 

  24464.      'P'

  24465.           Integer constant that is a sign-extended byte tiled out as two

  24466.           shorts.

  24467. 

  24468.      'Q'

  24469.           Integer constant that fits in one signed byte when incremented

  24470.           (-129 ... 126), but excluding -1.

  24471. 

  24472.      'T'

  24473.           A symbolic operand, or a 16-bit fragment of a got, tls, or

  24474.           pc-relative reference.

  24475. 

  24476.      'U'

  24477.           Memory operand except postincrement.  This is roughly the same

  24478.           as 'm' when not used together with '<' or '>'.

  24479. 

  24480.      'W'

  24481.           A 4-element vector constant with identical elements.

  24482. 

  24483.      'Y'

  24484.           A 2-element vector constant with identical elements.

  24485. 

  24486. _Visium--'config/visium/constraints.md'_

  24487.      'b'

  24488.           EAM register 'mdb'

  24489. 

  24490.      'c'

  24491.           EAM register 'mdc'

  24492. 

  24493.      'f'

  24494.           Floating point register

  24495. 

  24496.      'k'

  24497.           Register for sibcall optimization

  24498. 

  24499.      'l'

  24500.           General register, but not 'r29', 'r30' and 'r31'

  24501. 

  24502.      't'

  24503.           Register 'r1'

  24504. 

  24505.      'u'

  24506.           Register 'r2'

  24507. 

  24508.      'v'

  24509.           Register 'r3'

  24510. 

  24511.      'G'

  24512.           Floating-point constant 0.0

  24513. 

  24514.      'J'

  24515.           Integer constant in the range 0 ..  65535 (16-bit immediate)

  24516. 

  24517.      'K'

  24518.           Integer constant in the range 1 ..  31 (5-bit immediate)

  24519. 

  24520.      'L'

  24521.           Integer constant in the range -65535 ..  -1 (16-bit negative

  24522.           immediate)

  24523. 

  24524.      'M'

  24525.           Integer constant -1

  24526. 

  24527.      'O'

  24528.           Integer constant 0

  24529. 

  24530.      'P'

  24531.           Integer constant 32

  24532. 

  24533. _x86 family--'config/i386/constraints.md'_

  24534.      'R'

  24535.           Legacy register--the eight integer registers available on all

  24536.           i386 processors ('a', 'b', 'c', 'd', 'si', 'di', 'bp', 'sp').

  24537. 

  24538.      'q'

  24539.           Any register accessible as 'Rl'.  In 32-bit mode, 'a', 'b',

  24540.           'c', and 'd'; in 64-bit mode, any integer register.

  24541. 

  24542.      'Q'

  24543.           Any register accessible as 'Rh': 'a', 'b', 'c', and 'd'.

  24544. 

  24545.      'l'

  24546.           Any register that can be used as the index in a base+index

  24547.           memory access: that is, any general register except the stack

  24548.           pointer.

  24549. 

  24550.      'a'

  24551.           The 'a' register.

  24552. 

  24553.      'b'

  24554.           The 'b' register.

  24555. 

  24556.      'c'

  24557.           The 'c' register.

  24558. 

  24559.      'd'

  24560.           The 'd' register.

  24561. 

  24562.      'S'

  24563.           The 'si' register.

  24564. 

  24565.      'D'

  24566.           The 'di' register.

  24567. 

  24568.      'A'

  24569.           The 'a' and 'd' registers.  This class is used for

  24570.           instructions that return double word results in the 'ax:dx'

  24571.           register pair.  Single word values will be allocated either in

  24572.           'ax' or 'dx'.  For example on i386 the following implements

  24573.           'rdtsc':

  24574. 

  24575.                unsigned long long rdtsc (void)

  24576.                {

  24577.                  unsigned long long tick;

  24578.                  __asm__ __volatile__("rdtsc":"=A"(tick));

  24579.                  return tick;

  24580.                }

  24581. 

  24582.           This is not correct on x86-64 as it would allocate tick in

  24583.           either 'ax' or 'dx'.  You have to use the following variant

  24584.           instead:

  24585. 

  24586.                unsigned long long rdtsc (void)

  24587.                {

  24588.                  unsigned int tickl, tickh;

  24589.                  __asm__ __volatile__("rdtsc":"=a"(tickl),"=d"(tickh));

  24590.                  return ((unsigned long long)tickh << 32)|tickl;

  24591.                }

  24592. 

  24593.      'U'

  24594.           The call-clobbered integer registers.

  24595. 

  24596.      'f'

  24597.           Any 80387 floating-point (stack) register.

  24598. 

  24599.      't'

  24600.           Top of 80387 floating-point stack ('%st(0)').

  24601. 

  24602.      'u'

  24603.           Second from top of 80387 floating-point stack ('%st(1)').

  24604. 

  24605.      'Yk'

  24606.           Any mask register that can be used as a predicate, i.e.

  24607.           'k1-k7'.

  24608. 

  24609.      'k'

  24610.           Any mask register.

  24611. 

  24612.      'y'

  24613.           Any MMX register.

  24614. 

  24615.      'x'

  24616.           Any SSE register.

  24617. 

  24618.      'v'

  24619.           Any EVEX encodable SSE register ('%xmm0-%xmm31').

  24620. 

  24621.      'w'

  24622.           Any bound register.

  24623. 

  24624.      'Yz'

  24625.           First SSE register ('%xmm0').

  24626. 

  24627.      'Yi'

  24628.           Any SSE register, when SSE2 and inter-unit moves are enabled.

  24629. 

  24630.      'Yj'

  24631.           Any SSE register, when SSE2 and inter-unit moves from vector

  24632.           registers are enabled.

  24633. 

  24634.      'Ym'

  24635.           Any MMX register, when inter-unit moves are enabled.

  24636. 

  24637.      'Yn'

  24638.           Any MMX register, when inter-unit moves from vector registers

  24639.           are enabled.

  24640. 

  24641.      'Yp'

  24642.           Any integer register when 'TARGET_PARTIAL_REG_STALL' is

  24643.           disabled.

  24644. 

  24645.      'Ya'

  24646.           Any integer register when zero extensions with 'AND' are

  24647.           disabled.

  24648. 

  24649.      'Yb'

  24650.           Any register that can be used as the GOT base when calling

  24651.           '___tls_get_addr': that is, any general register except 'a'

  24652.           and 'sp' registers, for '-fno-plt' if linker supports it.

  24653.           Otherwise, 'b' register.

  24654. 

  24655.      'Yf'

  24656.           Any x87 register when 80387 floating-point arithmetic is

  24657.           enabled.

  24658. 

  24659.      'Yr'

  24660.           Lower SSE register when avoiding REX prefix and all SSE

  24661.           registers otherwise.

  24662. 

  24663.      'Yv'

  24664.           For AVX512VL, any EVEX-encodable SSE register

  24665.           ('%xmm0-%xmm31'), otherwise any SSE register.

  24666. 

  24667.      'Yh'

  24668.           Any EVEX-encodable SSE register, that has number factor of

  24669.           four.

  24670. 

  24671.      'Bf'

  24672.           Flags register operand.

  24673. 

  24674.      'Bg'

  24675.           GOT memory operand.

  24676. 

  24677.      'Bm'

  24678.           Vector memory operand.

  24679. 

  24680.      'Bc'

  24681.           Constant memory operand.

  24682. 

  24683.      'Bn'

  24684.           Memory operand without REX prefix.

  24685. 

  24686.      'Bs'

  24687.           Sibcall memory operand.

  24688. 

  24689.      'Bw'

  24690.           Call memory operand.

  24691. 

  24692.      'Bz'

  24693.           Constant call address operand.

  24694. 

  24695.      'BC'

  24696.           SSE constant -1 operand.

  24697. 

  24698.      'I'

  24699.           Integer constant in the range 0 ... 31, for 32-bit shifts.

  24700. 

  24701.      'J'

  24702.           Integer constant in the range 0 ... 63, for 64-bit shifts.

  24703. 

  24704.      'K'

  24705.           Signed 8-bit integer constant.

  24706. 

  24707.      'L'

  24708.           '0xFF' or '0xFFFF', for andsi as a zero-extending move.

  24709. 

  24710.      'M'

  24711.           0, 1, 2, or 3 (shifts for the 'lea' instruction).

  24712. 

  24713.      'N'

  24714.           Unsigned 8-bit integer constant (for 'in' and 'out'

  24715.           instructions).

  24716. 

  24717.      'O'

  24718.           Integer constant in the range 0 ... 127, for 128-bit shifts.

  24719. 

  24720.      'G'

  24721.           Standard 80387 floating point constant.

  24722. 

  24723.      'C'

  24724.           SSE constant zero operand.

  24725. 

  24726.      'e'

  24727.           32-bit signed integer constant, or a symbolic reference known

  24728.           to fit that range (for immediate operands in sign-extending

  24729.           x86-64 instructions).

  24730. 

  24731.      'We'

  24732.           32-bit signed integer constant, or a symbolic reference known

  24733.           to fit that range (for sign-extending conversion operations

  24734.           that require non-'VOIDmode' immediate operands).

  24735. 

  24736.      'Wz'

  24737.           32-bit unsigned integer constant, or a symbolic reference

  24738.           known to fit that range (for zero-extending conversion

  24739.           operations that require non-'VOIDmode' immediate operands).

  24740. 

  24741.      'Wd'

  24742.           128-bit integer constant where both the high and low 64-bit

  24743.           word satisfy the 'e' constraint.

  24744. 

  24745.      'Z'

  24746.           32-bit unsigned integer constant, or a symbolic reference

  24747.           known to fit that range (for immediate operands in

  24748.           zero-extending x86-64 instructions).

  24749. 

  24750.      'Tv'

  24751.           VSIB address operand.

  24752. 

  24753.      'Ts'

  24754.           Address operand without segment register.

  24755. 

  24756. _Xstormy16--'config/stormy16/stormy16.h'_

  24757.      'a'

  24758.           Register r0.

  24759. 

  24760.      'b'

  24761.           Register r1.

  24762. 

  24763.      'c'

  24764.           Register r2.

  24765. 

  24766.      'd'

  24767.           Register r8.

  24768. 

  24769.      'e'

  24770.           Registers r0 through r7.

  24771. 

  24772.      't'

  24773.           Registers r0 and r1.

  24774. 

  24775.      'y'

  24776.           The carry register.

  24777. 

  24778.      'z'

  24779.           Registers r8 and r9.

  24780. 

  24781.      'I'

  24782.           A constant between 0 and 3 inclusive.

  24783. 

  24784.      'J'

  24785.           A constant that has exactly one bit set.

  24786. 

  24787.      'K'

  24788.           A constant that has exactly one bit clear.

  24789. 

  24790.      'L'

  24791.           A constant between 0 and 255 inclusive.

  24792. 

  24793.      'M'

  24794.           A constant between -255 and 0 inclusive.

  24795. 

  24796.      'N'

  24797.           A constant between -3 and 0 inclusive.

  24798. 

  24799.      'O'

  24800.           A constant between 1 and 4 inclusive.

  24801. 

  24802.      'P'

  24803.           A constant between -4 and -1 inclusive.

  24804. 

  24805.      'Q'

  24806.           A memory reference that is a stack push.

  24807. 

  24808.      'R'

  24809.           A memory reference that is a stack pop.

  24810. 

  24811.      'S'

  24812.           A memory reference that refers to a constant address of known

  24813.           value.

  24814. 

  24815.      'T'

  24816.           The register indicated by Rx (not implemented yet).

  24817. 

  24818.      'U'

  24819.           A constant that is not between 2 and 15 inclusive.

  24820. 

  24821.      'Z'

  24822.           The constant 0.

  24823. 

  24824. _Xtensa--'config/xtensa/constraints.md'_

  24825.      'a'

  24826.           General-purpose 32-bit register

  24827. 

  24828.      'b'

  24829.           One-bit boolean register

  24830. 

  24831.      'A'

  24832.           MAC16 40-bit accumulator register

  24833. 

  24834.      'I'

  24835.           Signed 12-bit integer constant, for use in MOVI instructions

  24836. 

  24837.      'J'

  24838.           Signed 8-bit integer constant, for use in ADDI instructions

  24839. 

  24840.      'K'

  24841.           Integer constant valid for BccI instructions

  24842. 

  24843.      'L'

  24844.           Unsigned constant valid for BccUI instructions

  24845. 

  24846. 

  24847. File: gccint.info,  Node: Disable Insn Alternatives,  Next: Define Constraints,  Prev: Machine Constraints,  Up: Constraints

  24848. 

  24849. 17.8.6 Disable insn alternatives using the 'enabled' attribute

  24850. --------------------------------------------------------------

  24851. 

  24852. There are three insn attributes that may be used to selectively disable

  24853. instruction alternatives:

  24854. 

  24855. 'enabled'

  24856.      Says whether an alternative is available on the current subtarget.

  24857. 

  24858. 'preferred_for_size'

  24859.      Says whether an enabled alternative should be used in code that is

  24860.      optimized for size.

  24861. 

  24862. 'preferred_for_speed'

  24863.      Says whether an enabled alternative should be used in code that is

  24864.      optimized for speed.

  24865. 

  24866.  All these attributes should use '(const_int 1)' to allow an alternative

  24867. or '(const_int 0)' to disallow it.  The attributes must be a static

  24868. property of the subtarget; they cannot for example depend on the current

  24869. operands, on the current optimization level, on the location of the insn

  24870. within the body of a loop, on whether register allocation has finished,

  24871. or on the current compiler pass.

  24872. 

  24873.  The 'enabled' attribute is a correctness property.  It tells GCC to act

  24874. as though the disabled alternatives were never defined in the first

  24875. place.  This is useful when adding new instructions to an existing

  24876. pattern in cases where the new instructions are only available for

  24877. certain cpu architecture levels (typically mapped to the '-march='

  24878. command-line option).

  24879. 

  24880.  In contrast, the 'preferred_for_size' and 'preferred_for_speed'

  24881. attributes are strong optimization hints rather than correctness

  24882. properties.  'preferred_for_size' tells GCC which alternatives to

  24883. consider when adding or modifying an instruction that GCC wants to

  24884. optimize for size.  'preferred_for_speed' does the same thing for speed.

  24885. Note that things like code motion can lead to cases where code optimized

  24886. for size uses alternatives that are not preferred for size, and

  24887. similarly for speed.

  24888. 

  24889.  Although 'define_insn's can in principle specify the 'enabled'

  24890. attribute directly, it is often clearer to have subsiduary attributes

  24891. for each architectural feature of interest.  The 'define_insn's can then

  24892. use these subsiduary attributes to say which alternatives require which

  24893. features.  The example below does this for 'cpu_facility'.

  24894. 

  24895.  E.g.  the following two patterns could easily be merged using the

  24896. 'enabled' attribute:

  24897. 

  24898. 

  24899.      (define_insn "*movdi_old"

  24900.        [(set (match_operand:DI 0 "register_operand" "=d")

  24901.              (match_operand:DI 1 "register_operand" " d"))]

  24902.        "!TARGET_NEW"

  24903.        "lgr %0,%1")

  24904. 

  24905.      (define_insn "*movdi_new"

  24906.        [(set (match_operand:DI 0 "register_operand" "=d,f,d")

  24907.              (match_operand:DI 1 "register_operand" " d,d,f"))]

  24908.        "TARGET_NEW"

  24909.        "@

  24910.         lgr  %0,%1

  24911.         ldgr %0,%1

  24912.         lgdr %0,%1")

  24913. 

  24914. 

  24915.  to:

  24916. 

  24917. 

  24918.      (define_insn "*movdi_combined"

  24919.        [(set (match_operand:DI 0 "register_operand" "=d,f,d")

  24920.              (match_operand:DI 1 "register_operand" " d,d,f"))]

  24921.        ""

  24922.        "@

  24923.         lgr  %0,%1

  24924.         ldgr %0,%1

  24925.         lgdr %0,%1"

  24926.        [(set_attr "cpu_facility" "*,new,new")])

  24927. 

  24928. 

  24929.  with the 'enabled' attribute defined like this:

  24930. 

  24931. 

  24932.      (define_attr "cpu_facility" "standard,new" (const_string "standard"))

  24933. 

  24934.      (define_attr "enabled" ""

  24935.        (cond [(eq_attr "cpu_facility" "standard") (const_int 1)

  24936.               (and (eq_attr "cpu_facility" "new")

  24937.                    (ne (symbol_ref "TARGET_NEW") (const_int 0)))

  24938.               (const_int 1)]

  24939.              (const_int 0)))

  24940. 

  24941. 

  24942. 

  24943. File: gccint.info,  Node: Define Constraints,  Next: C Constraint Interface,  Prev: Disable Insn Alternatives,  Up: Constraints

  24944. 

  24945. 17.8.7 Defining Machine-Specific Constraints

  24946. --------------------------------------------

  24947. 

  24948. Machine-specific constraints fall into two categories: register and

  24949. non-register constraints.  Within the latter category, constraints which

  24950. allow subsets of all possible memory or address operands should be

  24951. specially marked, to give 'reload' more information.

  24952. 

  24953.  Machine-specific constraints can be given names of arbitrary length,

  24954. but they must be entirely composed of letters, digits, underscores

  24955. ('_'), and angle brackets ('< >').  Like C identifiers, they must begin

  24956. with a letter or underscore.

  24957. 

  24958.  In order to avoid ambiguity in operand constraint strings, no

  24959. constraint can have a name that begins with any other constraint's name.

  24960. For example, if 'x' is defined as a constraint name, 'xy' may not be,

  24961. and vice versa.  As a consequence of this rule, no constraint may begin

  24962. with one of the generic constraint letters: 'E F V X g i m n o p r s'.

  24963. 

  24964.  Register constraints correspond directly to register classes.  *Note

  24965. Register Classes::.  There is thus not much flexibility in their

  24966. definitions.

  24967. 

  24968.  -- MD Expression: define_register_constraint name regclass docstring

  24969.      All three arguments are string constants.  NAME is the name of the

  24970.      constraint, as it will appear in 'match_operand' expressions.  If

  24971.      NAME is a multi-letter constraint its length shall be the same for

  24972.      all constraints starting with the same letter.  REGCLASS can be

  24973.      either the name of the corresponding register class (*note Register

  24974.      Classes::), or a C expression which evaluates to the appropriate

  24975.      register class.  If it is an expression, it must have no side

  24976.      effects, and it cannot look at the operand.  The usual use of

  24977.      expressions is to map some register constraints to 'NO_REGS' when

  24978.      the register class is not available on a given subarchitecture.

  24979. 

  24980.      DOCSTRING is a sentence documenting the meaning of the constraint.

  24981.      Docstrings are explained further below.

  24982. 

  24983.  Non-register constraints are more like predicates: the constraint

  24984. definition gives a boolean expression which indicates whether the

  24985. constraint matches.

  24986. 

  24987.  -- MD Expression: define_constraint name docstring exp

  24988.      The NAME and DOCSTRING arguments are the same as for

  24989.      'define_register_constraint', but note that the docstring comes

  24990.      immediately after the name for these expressions.  EXP is an RTL

  24991.      expression, obeying the same rules as the RTL expressions in

  24992.      predicate definitions.  *Note Defining Predicates::, for details.

  24993.      If it evaluates true, the constraint matches; if it evaluates

  24994.      false, it doesn't.  Constraint expressions should indicate which

  24995.      RTL codes they might match, just like predicate expressions.

  24996. 

  24997.      'match_test' C expressions have access to the following variables:

  24998. 

  24999.      OP

  25000.           The RTL object defining the operand.

  25001.      MODE

  25002.           The machine mode of OP.

  25003.      IVAL

  25004.           'INTVAL (OP)', if OP is a 'const_int'.

  25005.      HVAL

  25006.           'CONST_DOUBLE_HIGH (OP)', if OP is an integer 'const_double'.

  25007.      LVAL

  25008.           'CONST_DOUBLE_LOW (OP)', if OP is an integer 'const_double'.

  25009.      RVAL

  25010.           'CONST_DOUBLE_REAL_VALUE (OP)', if OP is a floating-point

  25011.           'const_double'.

  25012. 

  25013.      The *VAL variables should only be used once another piece of the

  25014.      expression has verified that OP is the appropriate kind of RTL

  25015.      object.

  25016. 

  25017.  Most non-register constraints should be defined with

  25018. 'define_constraint'.  The remaining two definition expressions are only

  25019. appropriate for constraints that should be handled specially by 'reload'

  25020. if they fail to match.

  25021. 

  25022.  -- MD Expression: define_memory_constraint name docstring exp

  25023.      Use this expression for constraints that match a subset of all

  25024.      memory operands: that is, 'reload' can make them match by

  25025.      converting the operand to the form '(mem (reg X))', where X is a

  25026.      base register (from the register class specified by

  25027.      'BASE_REG_CLASS', *note Register Classes::).

  25028. 

  25029.      For example, on the S/390, some instructions do not accept

  25030.      arbitrary memory references, but only those that do not make use of

  25031.      an index register.  The constraint letter 'Q' is defined to

  25032.      represent a memory address of this type.  If 'Q' is defined with

  25033.      'define_memory_constraint', a 'Q' constraint can handle any memory

  25034.      operand, because 'reload' knows it can simply copy the memory

  25035.      address into a base register if required.  This is analogous to the

  25036.      way an 'o' constraint can handle any memory operand.

  25037. 

  25038.      The syntax and semantics are otherwise identical to

  25039.      'define_constraint'.

  25040. 

  25041.  -- MD Expression: define_special_memory_constraint name docstring exp

  25042.      Use this expression for constraints that match a subset of all

  25043.      memory operands: that is, 'reload' cannot make them match by

  25044.      reloading the address as it is described for

  25045.      'define_memory_constraint' or such address reload is undesirable

  25046.      with the performance point of view.

  25047. 

  25048.      For example, 'define_special_memory_constraint' can be useful if

  25049.      specifically aligned memory is necessary or desirable for some insn

  25050.      operand.

  25051. 

  25052.      The syntax and semantics are otherwise identical to

  25053.      'define_constraint'.

  25054. 

  25055.  -- MD Expression: define_address_constraint name docstring exp

  25056.      Use this expression for constraints that match a subset of all

  25057.      address operands: that is, 'reload' can make the constraint match

  25058.      by converting the operand to the form '(reg X)', again with X a

  25059.      base register.

  25060. 

  25061.      Constraints defined with 'define_address_constraint' can only be

  25062.      used with the 'address_operand' predicate, or machine-specific

  25063.      predicates that work the same way.  They are treated analogously to

  25064.      the generic 'p' constraint.

  25065. 

  25066.      The syntax and semantics are otherwise identical to

  25067.      'define_constraint'.

  25068. 

  25069.  For historical reasons, names beginning with the letters 'G H' are

  25070. reserved for constraints that match only 'const_double's, and names

  25071. beginning with the letters 'I J K L M N O P' are reserved for

  25072. constraints that match only 'const_int's.  This may change in the

  25073. future.  For the time being, constraints with these names must be

  25074. written in a stylized form, so that 'genpreds' can tell you did it

  25075. correctly:

  25076. 

  25077.      (define_constraint "[GHIJKLMNOP]..."

  25078.        "DOC..."

  25079.        (and (match_code "const_int")  ; 'const_double' for G/H

  25080.             CONDITION...))            ; usually a 'match_test'

  25081. 

  25082.  It is fine to use names beginning with other letters for constraints

  25083. that match 'const_double's or 'const_int's.

  25084. 

  25085.  Each docstring in a constraint definition should be one or more

  25086. complete sentences, marked up in Texinfo format.  _They are currently

  25087. unused._  In the future they will be copied into the GCC manual, in

  25088. *note Machine Constraints::, replacing the hand-maintained tables

  25089. currently found in that section.  Also, in the future the compiler may

  25090. use this to give more helpful diagnostics when poor choice of 'asm'

  25091. constraints causes a reload failure.

  25092. 

  25093.  If you put the pseudo-Texinfo directive '@internal' at the beginning of

  25094. a docstring, then (in the future) it will appear only in the internals

  25095. manual's version of the machine-specific constraint tables.  Use this

  25096. for constraints that should not appear in 'asm' statements.

  25097. 

  25098. 

  25099. File: gccint.info,  Node: C Constraint Interface,  Prev: Define Constraints,  Up: Constraints

  25100. 

  25101. 17.8.8 Testing constraints from C

  25102. ---------------------------------

  25103. 

  25104. It is occasionally useful to test a constraint from C code rather than

  25105. implicitly via the constraint string in a 'match_operand'.  The

  25106. generated file 'tm_p.h' declares a few interfaces for working with

  25107. constraints.  At present these are defined for all constraints except

  25108. 'g' (which is equivalent to 'general_operand').

  25109. 

  25110.  Some valid constraint names are not valid C identifiers, so there is a

  25111. mangling scheme for referring to them from C.  Constraint names that do

  25112. not contain angle brackets or underscores are left unchanged.

  25113. Underscores are doubled, each '<' is replaced with '_l', and each '>'

  25114. with '_g'.  Here are some examples:

  25115. 

  25116.      *Original* *Mangled*

  25117.      x          x

  25118.      P42x       P42x

  25119.      P4_x       P4__x

  25120.      P4>x       P4_gx

  25121.      P4>>       P4_g_g

  25122.      P4_g>      P4__g_g

  25123. 

  25124.  Throughout this section, the variable C is either a constraint in the

  25125. abstract sense, or a constant from 'enum constraint_num'; the variable M

  25126. is a mangled constraint name (usually as part of a larger identifier).

  25127. 

  25128.  -- Enum: constraint_num

  25129.      For each constraint except 'g', there is a corresponding

  25130.      enumeration constant: 'CONSTRAINT_' plus the mangled name of the

  25131.      constraint.  Functions that take an 'enum constraint_num' as an

  25132.      argument expect one of these constants.

  25133. 

  25134.  -- Function: inline bool satisfies_constraint_M (rtx EXP)

  25135.      For each non-register constraint M except 'g', there is one of

  25136.      these functions; it returns 'true' if EXP satisfies the constraint.

  25137.      These functions are only visible if 'rtl.h' was included before

  25138.      'tm_p.h'.

  25139. 

  25140.  -- Function: bool constraint_satisfied_p (rtx EXP, enum constraint_num

  25141.           C)

  25142.      Like the 'satisfies_constraint_M' functions, but the constraint to

  25143.      test is given as an argument, C.  If C specifies a register

  25144.      constraint, this function will always return 'false'.

  25145. 

  25146.  -- Function: enum reg_class reg_class_for_constraint (enum

  25147.           constraint_num C)

  25148.      Returns the register class associated with C.  If C is not a

  25149.      register constraint, or those registers are not available for the

  25150.      currently selected subtarget, returns 'NO_REGS'.

  25151. 

  25152.  Here is an example use of 'satisfies_constraint_M'.  In peephole

  25153. optimizations (*note Peephole Definitions::), operand constraint strings

  25154. are ignored, so if there are relevant constraints, they must be tested

  25155. in the C condition.  In the example, the optimization is applied if

  25156. operand 2 does _not_ satisfy the 'K' constraint.  (This is a simplified

  25157. version of a peephole definition from the i386 machine description.)

  25158. 

  25159.      (define_peephole2

  25160.        [(match_scratch:SI 3 "r")

  25161.         (set (match_operand:SI 0 "register_operand" "")

  25162.              (mult:SI (match_operand:SI 1 "memory_operand" "")

  25163.                       (match_operand:SI 2 "immediate_operand" "")))]

  25164. 

  25165.        "!satisfies_constraint_K (operands[2])"

  25166. 

  25167.        [(set (match_dup 3) (match_dup 1))

  25168.         (set (match_dup 0) (mult:SI (match_dup 3) (match_dup 2)))]

  25169. 

  25170.        "")

  25171. 

  25172. 

  25173. File: gccint.info,  Node: Standard Names,  Next: Pattern Ordering,  Prev: Constraints,  Up: Machine Desc

  25174. 

  25175. 17.9 Standard Pattern Names For Generation

  25176. ==========================================

  25177. 

  25178. Here is a table of the instruction names that are meaningful in the RTL

  25179. generation pass of the compiler.  Giving one of these names to an

  25180. instruction pattern tells the RTL generation pass that it can use the

  25181. pattern to accomplish a certain task.

  25182. 

  25183. 'movM'

  25184.      Here M stands for a two-letter machine mode name, in lowercase.

  25185.      This instruction pattern moves data with that machine mode from

  25186.      operand 1 to operand 0.  For example, 'movsi' moves full-word data.

  25187. 

  25188.      If operand 0 is a 'subreg' with mode M of a register whose own mode

  25189.      is wider than M, the effect of this instruction is to store the

  25190.      specified value in the part of the register that corresponds to

  25191.      mode M.  Bits outside of M, but which are within the same target

  25192.      word as the 'subreg' are undefined.  Bits which are outside the

  25193.      target word are left unchanged.

  25194. 

  25195.      This class of patterns is special in several ways.  First of all,

  25196.      each of these names up to and including full word size _must_ be

  25197.      defined, because there is no other way to copy a datum from one

  25198.      place to another.  If there are patterns accepting operands in

  25199.      larger modes, 'movM' must be defined for integer modes of those

  25200.      sizes.

  25201. 

  25202.      Second, these patterns are not used solely in the RTL generation

  25203.      pass.  Even the reload pass can generate move insns to copy values

  25204.      from stack slots into temporary registers.  When it does so, one of

  25205.      the operands is a hard register and the other is an operand that

  25206.      can need to be reloaded into a register.

  25207. 

  25208.      Therefore, when given such a pair of operands, the pattern must

  25209.      generate RTL which needs no reloading and needs no temporary

  25210.      registers--no registers other than the operands.  For example, if

  25211.      you support the pattern with a 'define_expand', then in such a case

  25212.      the 'define_expand' mustn't call 'force_reg' or any other such

  25213.      function which might generate new pseudo registers.

  25214. 

  25215.      This requirement exists even for subword modes on a RISC machine

  25216.      where fetching those modes from memory normally requires several

  25217.      insns and some temporary registers.

  25218. 

  25219.      During reload a memory reference with an invalid address may be

  25220.      passed as an operand.  Such an address will be replaced with a

  25221.      valid address later in the reload pass.  In this case, nothing may

  25222.      be done with the address except to use it as it stands.  If it is

  25223.      copied, it will not be replaced with a valid address.  No attempt

  25224.      should be made to make such an address into a valid address and no

  25225.      routine (such as 'change_address') that will do so may be called.

  25226.      Note that 'general_operand' will fail when applied to such an

  25227.      address.

  25228. 

  25229.      The global variable 'reload_in_progress' (which must be explicitly

  25230.      declared if required) can be used to determine whether such special

  25231.      handling is required.

  25232. 

  25233.      The variety of operands that have reloads depends on the rest of

  25234.      the machine description, but typically on a RISC machine these can

  25235.      only be pseudo registers that did not get hard registers, while on

  25236.      other machines explicit memory references will get optional

  25237.      reloads.

  25238. 

  25239.      If a scratch register is required to move an object to or from

  25240.      memory, it can be allocated using 'gen_reg_rtx' prior to life

  25241.      analysis.

  25242. 

  25243.      If there are cases which need scratch registers during or after

  25244.      reload, you must provide an appropriate secondary_reload target

  25245.      hook.

  25246. 

  25247.      The macro 'can_create_pseudo_p' can be used to determine if it is

  25248.      unsafe to create new pseudo registers.  If this variable is

  25249.      nonzero, then it is unsafe to call 'gen_reg_rtx' to allocate a new

  25250.      pseudo.

  25251. 

  25252.      The constraints on a 'movM' must permit moving any hard register to

  25253.      any other hard register provided that 'TARGET_HARD_REGNO_MODE_OK'

  25254.      permits mode M in both registers and 'TARGET_REGISTER_MOVE_COST'

  25255.      applied to their classes returns a value of 2.

  25256. 

  25257.      It is obligatory to support floating point 'movM' instructions into

  25258.      and out of any registers that can hold fixed point values, because

  25259.      unions and structures (which have modes 'SImode' or 'DImode') can

  25260.      be in those registers and they may have floating point members.

  25261. 

  25262.      There may also be a need to support fixed point 'movM' instructions

  25263.      in and out of floating point registers.  Unfortunately, I have

  25264.      forgotten why this was so, and I don't know whether it is still

  25265.      true.  If 'TARGET_HARD_REGNO_MODE_OK' rejects fixed point values in

  25266.      floating point registers, then the constraints of the fixed point

  25267.      'movM' instructions must be designed to avoid ever trying to reload

  25268.      into a floating point register.

  25269. 

  25270. 'reload_inM'

  25271. 'reload_outM'

  25272.      These named patterns have been obsoleted by the target hook

  25273.      'secondary_reload'.

  25274. 

  25275.      Like 'movM', but used when a scratch register is required to move

  25276.      between operand 0 and operand 1.  Operand 2 describes the scratch

  25277.      register.  See the discussion of the 'SECONDARY_RELOAD_CLASS' macro

  25278.      in *note Register Classes::.

  25279. 

  25280.      There are special restrictions on the form of the 'match_operand's

  25281.      used in these patterns.  First, only the predicate for the reload

  25282.      operand is examined, i.e., 'reload_in' examines operand 1, but not

  25283.      the predicates for operand 0 or 2.  Second, there may be only one

  25284.      alternative in the constraints.  Third, only a single register

  25285.      class letter may be used for the constraint; subsequent constraint

  25286.      letters are ignored.  As a special exception, an empty constraint

  25287.      string matches the 'ALL_REGS' register class.  This may relieve

  25288.      ports of the burden of defining an 'ALL_REGS' constraint letter

  25289.      just for these patterns.

  25290. 

  25291. 'movstrictM'

  25292.      Like 'movM' except that if operand 0 is a 'subreg' with mode M of a

  25293.      register whose natural mode is wider, the 'movstrictM' instruction

  25294.      is guaranteed not to alter any of the register except the part

  25295.      which belongs to mode M.

  25296. 

  25297. 'movmisalignM'

  25298.      This variant of a move pattern is designed to load or store a value

  25299.      from a memory address that is not naturally aligned for its mode.

  25300.      For a store, the memory will be in operand 0; for a load, the

  25301.      memory will be in operand 1.  The other operand is guaranteed not

  25302.      to be a memory, so that it's easy to tell whether this is a load or

  25303.      store.

  25304. 

  25305.      This pattern is used by the autovectorizer, and when expanding a

  25306.      'MISALIGNED_INDIRECT_REF' expression.

  25307. 

  25308. 'load_multiple'

  25309.      Load several consecutive memory locations into consecutive

  25310.      registers.  Operand 0 is the first of the consecutive registers,

  25311.      operand 1 is the first memory location, and operand 2 is a

  25312.      constant: the number of consecutive registers.

  25313. 

  25314.      Define this only if the target machine really has such an

  25315.      instruction; do not define this if the most efficient way of

  25316.      loading consecutive registers from memory is to do them one at a

  25317.      time.

  25318. 

  25319.      On some machines, there are restrictions as to which consecutive

  25320.      registers can be stored into memory, such as particular starting or

  25321.      ending register numbers or only a range of valid counts.  For those

  25322.      machines, use a 'define_expand' (*note Expander Definitions::) and

  25323.      make the pattern fail if the restrictions are not met.

  25324. 

  25325.      Write the generated insn as a 'parallel' with elements being a

  25326.      'set' of one register from the appropriate memory location (you may

  25327.      also need 'use' or 'clobber' elements).  Use a 'match_parallel'

  25328.      (*note RTL Template::) to recognize the insn.  See 'rs6000.md' for

  25329.      examples of the use of this insn pattern.

  25330. 

  25331. 'store_multiple'

  25332.      Similar to 'load_multiple', but store several consecutive registers

  25333.      into consecutive memory locations.  Operand 0 is the first of the

  25334.      consecutive memory locations, operand 1 is the first register, and

  25335.      operand 2 is a constant: the number of consecutive registers.

  25336. 

  25337. 'vec_load_lanesMN'

  25338.      Perform an interleaved load of several vectors from memory operand

  25339.      1 into register operand 0.  Both operands have mode M.  The

  25340.      register operand is viewed as holding consecutive vectors of mode

  25341.      N, while the memory operand is a flat array that contains the same

  25342.      number of elements.  The operation is equivalent to:

  25343. 

  25344.           int c = GET_MODE_SIZE (M) / GET_MODE_SIZE (N);

  25345.           for (j = 0; j < GET_MODE_NUNITS (N); j++)

  25346.             for (i = 0; i < c; i++)

  25347.               operand0[i][j] = operand1[j * c + i];

  25348. 

  25349.      For example, 'vec_load_lanestiv4hi' loads 8 16-bit values from

  25350.      memory into a register of mode 'TI'.  The register contains two

  25351.      consecutive vectors of mode 'V4HI'.

  25352. 

  25353.      This pattern can only be used if:

  25354.           TARGET_ARRAY_MODE_SUPPORTED_P (N, C)

  25355.      is true.  GCC assumes that, if a target supports this kind of

  25356.      instruction for some mode N, it also supports unaligned loads for

  25357.      vectors of mode N.

  25358. 

  25359.      This pattern is not allowed to 'FAIL'.

  25360. 

  25361. 'vec_mask_load_lanesMN'

  25362.      Like 'vec_load_lanesMN', but takes an additional mask operand

  25363.      (operand 2) that specifies which elements of the destination

  25364.      vectors should be loaded.  Other elements of the destination

  25365.      vectors are set to zero.  The operation is equivalent to:

  25366. 

  25367.           int c = GET_MODE_SIZE (M) / GET_MODE_SIZE (N);

  25368.           for (j = 0; j < GET_MODE_NUNITS (N); j++)

  25369.             if (operand2[j])

  25370.               for (i = 0; i < c; i++)

  25371.                 operand0[i][j] = operand1[j * c + i];

  25372.             else

  25373.               for (i = 0; i < c; i++)

  25374.                 operand0[i][j] = 0;

  25375. 

  25376.      This pattern is not allowed to 'FAIL'.

  25377. 

  25378. 'vec_store_lanesMN'

  25379.      Equivalent to 'vec_load_lanesMN', with the memory and register

  25380.      operands reversed.  That is, the instruction is equivalent to:

  25381. 

  25382.           int c = GET_MODE_SIZE (M) / GET_MODE_SIZE (N);

  25383.           for (j = 0; j < GET_MODE_NUNITS (N); j++)

  25384.             for (i = 0; i < c; i++)

  25385.               operand0[j * c + i] = operand1[i][j];

  25386. 

  25387.      for a memory operand 0 and register operand 1.

  25388. 

  25389.      This pattern is not allowed to 'FAIL'.

  25390. 

  25391. 'vec_mask_store_lanesMN'

  25392.      Like 'vec_store_lanesMN', but takes an additional mask operand

  25393.      (operand 2) that specifies which elements of the source vectors

  25394.      should be stored.  The operation is equivalent to:

  25395. 

  25396.           int c = GET_MODE_SIZE (M) / GET_MODE_SIZE (N);

  25397.           for (j = 0; j < GET_MODE_NUNITS (N); j++)

  25398.             if (operand2[j])

  25399.               for (i = 0; i < c; i++)

  25400.                 operand0[j * c + i] = operand1[i][j];

  25401. 

  25402.      This pattern is not allowed to 'FAIL'.

  25403. 

  25404. 'gather_loadMN'

  25405.      Load several separate memory locations into a vector of mode M.

  25406.      Operand 1 is a scalar base address and operand 2 is a vector of

  25407.      mode N containing offsets from that base.  Operand 0 is a

  25408.      destination vector with the same number of elements as N.  For each

  25409.      element index I:

  25410. 

  25411.         * extend the offset element I to address width, using zero

  25412.           extension if operand 3 is 1 and sign extension if operand 3 is

  25413.           zero;

  25414.         * multiply the extended offset by operand 4;

  25415.         * add the result to the base; and

  25416.         * load the value at that address into element I of operand 0.

  25417. 

  25418.      The value of operand 3 does not matter if the offsets are already

  25419.      address width.

  25420. 

  25421. 'mask_gather_loadMN'

  25422.      Like 'gather_loadMN', but takes an extra mask operand as operand 5.

  25423.      Bit I of the mask is set if element I of the result should be

  25424.      loaded from memory and clear if element I of the result should be

  25425.      set to zero.

  25426. 

  25427. 'scatter_storeMN'

  25428.      Store a vector of mode M into several distinct memory locations.

  25429.      Operand 0 is a scalar base address and operand 1 is a vector of

  25430.      mode N containing offsets from that base.  Operand 4 is the vector

  25431.      of values that should be stored, which has the same number of

  25432.      elements as N.  For each element index I:

  25433. 

  25434.         * extend the offset element I to address width, using zero

  25435.           extension if operand 2 is 1 and sign extension if operand 2 is

  25436.           zero;

  25437.         * multiply the extended offset by operand 3;

  25438.         * add the result to the base; and

  25439.         * store element I of operand 4 to that address.

  25440. 

  25441.      The value of operand 2 does not matter if the offsets are already

  25442.      address width.

  25443. 

  25444. 'mask_scatter_storeMN'

  25445.      Like 'scatter_storeMN', but takes an extra mask operand as operand

  25446.      5.  Bit I of the mask is set if element I of the result should be

  25447.      stored to memory.

  25448. 

  25449. 'vec_setM'

  25450.      Set given field in the vector value.  Operand 0 is the vector to

  25451.      modify, operand 1 is new value of field and operand 2 specify the

  25452.      field index.

  25453. 

  25454. 'vec_extractMN'

  25455.      Extract given field from the vector value.  Operand 1 is the

  25456.      vector, operand 2 specify field index and operand 0 place to store

  25457.      value into.  The N mode is the mode of the field or vector of

  25458.      fields that should be extracted, should be either element mode of

  25459.      the vector mode M, or a vector mode with the same element mode and

  25460.      smaller number of elements.  If N is a vector mode, the index is

  25461.      counted in units of that mode.

  25462. 

  25463. 'vec_initMN'

  25464.      Initialize the vector to given values.  Operand 0 is the vector to

  25465.      initialize and operand 1 is parallel containing values for

  25466.      individual fields.  The N mode is the mode of the elements, should

  25467.      be either element mode of the vector mode M, or a vector mode with

  25468.      the same element mode and smaller number of elements.

  25469. 

  25470. 'vec_duplicateM'

  25471.      Initialize vector output operand 0 so that each element has the

  25472.      value given by scalar input operand 1.  The vector has mode M and

  25473.      the scalar has the mode appropriate for one element of M.

  25474. 

  25475.      This pattern only handles duplicates of non-constant inputs.

  25476.      Constant vectors go through the 'movM' pattern instead.

  25477. 

  25478.      This pattern is not allowed to 'FAIL'.

  25479. 

  25480. 'vec_seriesM'

  25481.      Initialize vector output operand 0 so that element I is equal to

  25482.      operand 1 plus I times operand 2.  In other words, create a linear

  25483.      series whose base value is operand 1 and whose step is operand 2.

  25484. 

  25485.      The vector output has mode M and the scalar inputs have the mode

  25486.      appropriate for one element of M.  This pattern is not used for

  25487.      floating-point vectors, in order to avoid having to specify the

  25488.      rounding behavior for I > 1.

  25489. 

  25490.      This pattern is not allowed to 'FAIL'.

  25491. 

  25492. 'while_ultMN'

  25493.      Set operand 0 to a mask that is true while incrementing operand 1

  25494.      gives a value that is less than operand 2.  Operand 0 has mode N

  25495.      and operands 1 and 2 are scalar integers of mode M.  The operation

  25496.      is equivalent to:

  25497. 

  25498.           operand0[0] = operand1 < operand2;

  25499.           for (i = 1; i < GET_MODE_NUNITS (N); i++)

  25500.             operand0[i] = operand0[i - 1] && (operand1 + i < operand2);

  25501. 

  25502. 'check_raw_ptrsM'

  25503.      Check whether, given two pointers A and B and a length LEN, a write

  25504.      of LEN bytes at A followed by a read of LEN bytes at B can be split

  25505.      into interleaved byte accesses 'A[0], B[0], A[1], B[1], ...'

  25506.      without affecting the dependencies between the bytes.  Set operand

  25507.      0 to true if the split is possible and false otherwise.

  25508. 

  25509.      Operands 1, 2 and 3 provide the values of A, B and LEN

  25510.      respectively.  Operand 4 is a constant integer that provides the

  25511.      known common alignment of A and B.  All inputs have mode M.

  25512. 

  25513.      This split is possible if:

  25514. 

  25515.           A == B || A + LEN <= B || B + LEN <= A

  25516. 

  25517.      You should only define this pattern if the target has a way of

  25518.      accelerating the test without having to do the individual

  25519.      comparisons.

  25520. 

  25521. 'check_war_ptrsM'

  25522.      Like 'check_raw_ptrsM', but with the read and write swapped round.

  25523.      The split is possible in this case if:

  25524. 

  25525.           B <= A || A + LEN <= B

  25526. 

  25527. 'vec_cmpMN'

  25528.      Output a vector comparison.  Operand 0 of mode N is the destination

  25529.      for predicate in operand 1 which is a signed vector comparison with

  25530.      operands of mode M in operands 2 and 3.  Predicate is computed by

  25531.      element-wise evaluation of the vector comparison with a truth value

  25532.      of all-ones and a false value of all-zeros.

  25533. 

  25534. 'vec_cmpuMN'

  25535.      Similar to 'vec_cmpMN' but perform unsigned vector comparison.

  25536. 

  25537. 'vec_cmpeqMN'

  25538.      Similar to 'vec_cmpMN' but perform equality or non-equality vector

  25539.      comparison only.  If 'vec_cmpMN' or 'vec_cmpuMN' instruction

  25540.      pattern is supported, it will be preferred over 'vec_cmpeqMN', so

  25541.      there is no need to define this instruction pattern if the others

  25542.      are supported.

  25543. 

  25544. 'vcondMN'

  25545.      Output a conditional vector move.  Operand 0 is the destination to

  25546.      receive a combination of operand 1 and operand 2, which are of mode

  25547.      M, dependent on the outcome of the predicate in operand 3 which is

  25548.      a signed vector comparison with operands of mode N in operands 4

  25549.      and 5.  The modes M and N should have the same size.  Operand 0

  25550.      will be set to the value OP1 & MSK | OP2 & ~MSK where MSK is

  25551.      computed by element-wise evaluation of the vector comparison with a

  25552.      truth value of all-ones and a false value of all-zeros.

  25553. 

  25554. 'vconduMN'

  25555.      Similar to 'vcondMN' but performs unsigned vector comparison.

  25556. 

  25557. 'vcondeqMN'

  25558.      Similar to 'vcondMN' but performs equality or non-equality vector

  25559.      comparison only.  If 'vcondMN' or 'vconduMN' instruction pattern is

  25560.      supported, it will be preferred over 'vcondeqMN', so there is no

  25561.      need to define this instruction pattern if the others are

  25562.      supported.

  25563. 

  25564. 'vcond_mask_MN'

  25565.      Similar to 'vcondMN' but operand 3 holds a pre-computed result of

  25566.      vector comparison.

  25567. 

  25568. 'maskloadMN'

  25569.      Perform a masked load of vector from memory operand 1 of mode M

  25570.      into register operand 0.  Mask is provided in register operand 2 of

  25571.      mode N.

  25572. 

  25573.      This pattern is not allowed to 'FAIL'.

  25574. 

  25575. 'maskstoreMN'

  25576.      Perform a masked store of vector from register operand 1 of mode M

  25577.      into memory operand 0.  Mask is provided in register operand 2 of

  25578.      mode N.

  25579. 

  25580.      This pattern is not allowed to 'FAIL'.

  25581. 

  25582. 'vec_permM'

  25583.      Output a (variable) vector permutation.  Operand 0 is the

  25584.      destination to receive elements from operand 1 and operand 2, which

  25585.      are of mode M.  Operand 3 is the "selector".  It is an integral

  25586.      mode vector of the same width and number of elements as mode M.

  25587. 

  25588.      The input elements are numbered from 0 in operand 1 through 2*N-1

  25589.      in operand 2.  The elements of the selector must be computed modulo

  25590.      2*N.  Note that if 'rtx_equal_p(operand1, operand2)', this can be

  25591.      implemented with just operand 1 and selector elements modulo N.

  25592. 

  25593.      In order to make things easy for a number of targets, if there is

  25594.      no 'vec_perm' pattern for mode M, but there is for mode Q where Q

  25595.      is a vector of 'QImode' of the same width as M, the middle-end will

  25596.      lower the mode M 'VEC_PERM_EXPR' to mode Q.

  25597. 

  25598.      See also 'TARGET_VECTORIZER_VEC_PERM_CONST', which performs the

  25599.      analogous operation for constant selectors.

  25600. 

  25601. 'pushM1'

  25602.      Output a push instruction.  Operand 0 is value to push.  Used only

  25603.      when 'PUSH_ROUNDING' is defined.  For historical reason, this

  25604.      pattern may be missing and in such case an 'mov' expander is used

  25605.      instead, with a 'MEM' expression forming the push operation.  The

  25606.      'mov' expander method is deprecated.

  25607. 

  25608. 'addM3'

  25609.      Add operand 2 and operand 1, storing the result in operand 0.  All

  25610.      operands must have mode M.  This can be used even on two-address

  25611.      machines, by means of constraints requiring operands 1 and 0 to be

  25612.      the same location.

  25613. 

  25614. 'ssaddM3', 'usaddM3'

  25615. 'subM3', 'sssubM3', 'ussubM3'

  25616. 'mulM3', 'ssmulM3', 'usmulM3'

  25617. 'divM3', 'ssdivM3'

  25618. 'udivM3', 'usdivM3'

  25619. 'modM3', 'umodM3'

  25620. 'uminM3', 'umaxM3'

  25621. 'andM3', 'iorM3', 'xorM3'

  25622.      Similar, for other arithmetic operations.

  25623. 

  25624. 'addvM4'

  25625.      Like 'addM3' but takes a 'code_label' as operand 3 and emits code

  25626.      to jump to it if signed overflow occurs during the addition.  This

  25627.      pattern is used to implement the built-in functions performing

  25628.      signed integer addition with overflow checking.

  25629. 

  25630. 'subvM4', 'mulvM4'

  25631.      Similar, for other signed arithmetic operations.

  25632. 

  25633. 'uaddvM4'

  25634.      Like 'addvM4' but for unsigned addition.  That is to say, the

  25635.      operation is the same as signed addition but the jump is taken only

  25636.      on unsigned overflow.

  25637. 

  25638. 'usubvM4', 'umulvM4'

  25639.      Similar, for other unsigned arithmetic operations.

  25640. 

  25641. 'addptrM3'

  25642.      Like 'addM3' but is guaranteed to only be used for address

  25643.      calculations.  The expanded code is not allowed to clobber the

  25644.      condition code.  It only needs to be defined if 'addM3' sets the

  25645.      condition code.  If adds used for address calculations and normal

  25646.      adds are not compatible it is required to expand a distinct pattern

  25647.      (e.g. using an unspec).  The pattern is used by LRA to emit address

  25648.      calculations.  'addM3' is used if 'addptrM3' is not defined.

  25649. 

  25650. 'fmaM4'

  25651.      Multiply operand 2 and operand 1, then add operand 3, storing the

  25652.      result in operand 0 without doing an intermediate rounding step.

  25653.      All operands must have mode M.  This pattern is used to implement

  25654.      the 'fma', 'fmaf', and 'fmal' builtin functions from the ISO C99

  25655.      standard.

  25656. 

  25657. 'fmsM4'

  25658.      Like 'fmaM4', except operand 3 subtracted from the product instead

  25659.      of added to the product.  This is represented in the rtl as

  25660. 

  25661.           (fma:M OP1 OP2 (neg:M OP3))

  25662. 

  25663. 'fnmaM4'

  25664.      Like 'fmaM4' except that the intermediate product is negated before

  25665.      being added to operand 3.  This is represented in the rtl as

  25666. 

  25667.           (fma:M (neg:M OP1) OP2 OP3)

  25668. 

  25669. 'fnmsM4'

  25670.      Like 'fmsM4' except that the intermediate product is negated before

  25671.      subtracting operand 3.  This is represented in the rtl as

  25672. 

  25673.           (fma:M (neg:M OP1) OP2 (neg:M OP3))

  25674. 

  25675. 'sminM3', 'smaxM3'

  25676.      Signed minimum and maximum operations.  When used with floating

  25677.      point, if both operands are zeros, or if either operand is 'NaN',

  25678.      then it is unspecified which of the two operands is returned as the

  25679.      result.

  25680. 

  25681. 'fminM3', 'fmaxM3'

  25682.      IEEE-conformant minimum and maximum operations.  If one operand is

  25683.      a quiet 'NaN', then the other operand is returned.  If both

  25684.      operands are quiet 'NaN', then a quiet 'NaN' is returned.  In the

  25685.      case when gcc supports signaling 'NaN' (-fsignaling-nans) an

  25686.      invalid floating point exception is raised and a quiet 'NaN' is

  25687.      returned.

  25688. 

  25689.      All operands have mode M, which is a scalar or vector

  25690.      floating-point mode.  These patterns are not allowed to 'FAIL'.

  25691. 

  25692. 'reduc_smin_scal_M', 'reduc_smax_scal_M'

  25693.      Find the signed minimum/maximum of the elements of a vector.  The

  25694.      vector is operand 1, and operand 0 is the scalar result, with mode

  25695.      equal to the mode of the elements of the input vector.

  25696. 

  25697. 'reduc_umin_scal_M', 'reduc_umax_scal_M'

  25698.      Find the unsigned minimum/maximum of the elements of a vector.  The

  25699.      vector is operand 1, and operand 0 is the scalar result, with mode

  25700.      equal to the mode of the elements of the input vector.

  25701. 

  25702. 'reduc_plus_scal_M'

  25703.      Compute the sum of the elements of a vector.  The vector is operand

  25704.      1, and operand 0 is the scalar result, with mode equal to the mode

  25705.      of the elements of the input vector.

  25706. 

  25707. 'reduc_and_scal_M'

  25708. 'reduc_ior_scal_M'

  25709. 'reduc_xor_scal_M'

  25710.      Compute the bitwise 'AND'/'IOR'/'XOR' reduction of the elements of

  25711.      a vector of mode M.  Operand 1 is the vector input and operand 0 is

  25712.      the scalar result.  The mode of the scalar result is the same as

  25713.      one element of M.

  25714. 

  25715. 'extract_last_M'

  25716.      Find the last set bit in mask operand 1 and extract the associated

  25717.      element of vector operand 2.  Store the result in scalar operand 0.

  25718.      Operand 2 has vector mode M while operand 0 has the mode

  25719.      appropriate for one element of M.  Operand 1 has the usual mask

  25720.      mode for vectors of mode M; see 'TARGET_VECTORIZE_GET_MASK_MODE'.

  25721. 

  25722. 'fold_extract_last_M'

  25723.      If any bits of mask operand 2 are set, find the last set bit,

  25724.      extract the associated element from vector operand 3, and store the

  25725.      result in operand 0.  Store operand 1 in operand 0 otherwise.

  25726.      Operand 3 has mode M and operands 0 and 1 have the mode appropriate

  25727.      for one element of M.  Operand 2 has the usual mask mode for

  25728.      vectors of mode M; see 'TARGET_VECTORIZE_GET_MASK_MODE'.

  25729. 

  25730. 'fold_left_plus_M'

  25731.      Take scalar operand 1 and successively add each element from vector

  25732.      operand 2.  Store the result in scalar operand 0.  The vector has

  25733.      mode M and the scalars have the mode appropriate for one element of

  25734.      M.  The operation is strictly in-order: there is no reassociation.

  25735. 

  25736. 'mask_fold_left_plus_M'

  25737.      Like 'fold_left_plus_M', but takes an additional mask operand

  25738.      (operand 3) that specifies which elements of the source vector

  25739.      should be added.

  25740. 

  25741. 'sdot_prodM'

  25742. 'udot_prodM'

  25743.      Compute the sum of the products of two signed/unsigned elements.

  25744.      Operand 1 and operand 2 are of the same mode.  Their product, which

  25745.      is of a wider mode, is computed and added to operand 3.  Operand 3

  25746.      is of a mode equal or wider than the mode of the product.  The

  25747.      result is placed in operand 0, which is of the same mode as operand

  25748.      3.

  25749. 

  25750. 'ssadM'

  25751. 'usadM'

  25752.      Compute the sum of absolute differences of two signed/unsigned

  25753.      elements.  Operand 1 and operand 2 are of the same mode.  Their

  25754.      absolute difference, which is of a wider mode, is computed and

  25755.      added to operand 3.  Operand 3 is of a mode equal or wider than the

  25756.      mode of the absolute difference.  The result is placed in operand

  25757.      0, which is of the same mode as operand 3.

  25758. 

  25759. 'widen_ssumM3'

  25760. 'widen_usumM3'

  25761.      Operands 0 and 2 are of the same mode, which is wider than the mode

  25762.      of operand 1.  Add operand 1 to operand 2 and place the widened

  25763.      result in operand 0.  (This is used express accumulation of

  25764.      elements into an accumulator of a wider mode.)

  25765. 

  25766. 'smulhsM3'

  25767. 'umulhsM3'

  25768.      Signed/unsigned multiply high with scale.  This is equivalent to

  25769.      the C code:

  25770.           narrow op0, op1, op2;

  25771.           ...

  25772.           op0 = (narrow) (((wide) op1 * (wide) op2) >> (N / 2 - 1));

  25773.      where the sign of 'narrow' determines whether this is a signed or

  25774.      unsigned operation, and N is the size of 'wide' in bits.

  25775. 

  25776. 'smulhrsM3'

  25777. 'umulhrsM3'

  25778.      Signed/unsigned multiply high with round and scale.  This is

  25779.      equivalent to the C code:

  25780.           narrow op0, op1, op2;

  25781.           ...

  25782.           op0 = (narrow) (((((wide) op1 * (wide) op2) >> (N / 2 - 2)) + 1) >> 1);

  25783.      where the sign of 'narrow' determines whether this is a signed or

  25784.      unsigned operation, and N is the size of 'wide' in bits.

  25785. 

  25786. 'sdiv_pow2M3'

  25787. 'sdiv_pow2M3'

  25788.      Signed division by power-of-2 immediate.  Equivalent to:

  25789.           signed op0, op1;

  25790.           ...

  25791.           op0 = op1 / (1 << imm);

  25792. 

  25793. 'vec_shl_insert_M'

  25794.      Shift the elements in vector input operand 1 left one element (i.e.

  25795.      away from element 0) and fill the vacated element 0 with the scalar

  25796.      in operand 2.  Store the result in vector output operand 0.

  25797.      Operands 0 and 1 have mode M and operand 2 has the mode appropriate

  25798.      for one element of M.

  25799. 

  25800. 'vec_shl_M'

  25801.      Whole vector left shift in bits, i.e. away from element 0.  Operand

  25802.      1 is a vector to be shifted.  Operand 2 is an integer shift amount

  25803.      in bits.  Operand 0 is where the resulting shifted vector is

  25804.      stored.  The output and input vectors should have the same modes.

  25805. 

  25806. 'vec_shr_M'

  25807.      Whole vector right shift in bits, i.e. towards element 0.  Operand

  25808.      1 is a vector to be shifted.  Operand 2 is an integer shift amount

  25809.      in bits.  Operand 0 is where the resulting shifted vector is

  25810.      stored.  The output and input vectors should have the same modes.

  25811. 

  25812. 'vec_pack_trunc_M'

  25813.      Narrow (demote) and merge the elements of two vectors.  Operands 1

  25814.      and 2 are vectors of the same mode having N integral or floating

  25815.      point elements of size S.  Operand 0 is the resulting vector in

  25816.      which 2*N elements of size S/2 are concatenated after narrowing

  25817.      them down using truncation.

  25818. 

  25819. 'vec_pack_sbool_trunc_M'

  25820.      Narrow and merge the elements of two vectors.  Operands 1 and 2 are

  25821.      vectors of the same type having N boolean elements.  Operand 0 is

  25822.      the resulting vector in which 2*N elements are concatenated.  The

  25823.      last operand (operand 3) is the number of elements in the output

  25824.      vector 2*N as a 'CONST_INT'.  This instruction pattern is used when

  25825.      all the vector input and output operands have the same scalar mode

  25826.      M and thus using 'vec_pack_trunc_M' would be ambiguous.

  25827. 

  25828. 'vec_pack_ssat_M', 'vec_pack_usat_M'

  25829.      Narrow (demote) and merge the elements of two vectors.  Operands 1

  25830.      and 2 are vectors of the same mode having N integral elements of

  25831.      size S. Operand 0 is the resulting vector in which the elements of

  25832.      the two input vectors are concatenated after narrowing them down

  25833.      using signed/unsigned saturating arithmetic.

  25834. 

  25835. 'vec_pack_sfix_trunc_M', 'vec_pack_ufix_trunc_M'

  25836.      Narrow, convert to signed/unsigned integral type and merge the

  25837.      elements of two vectors.  Operands 1 and 2 are vectors of the same

  25838.      mode having N floating point elements of size S.  Operand 0 is the

  25839.      resulting vector in which 2*N elements of size S/2 are

  25840.      concatenated.

  25841. 

  25842. 'vec_packs_float_M', 'vec_packu_float_M'

  25843.      Narrow, convert to floating point type and merge the elements of

  25844.      two vectors.  Operands 1 and 2 are vectors of the same mode having

  25845.      N signed/unsigned integral elements of size S.  Operand 0 is the

  25846.      resulting vector in which 2*N elements of size S/2 are

  25847.      concatenated.

  25848. 

  25849. 'vec_unpacks_hi_M', 'vec_unpacks_lo_M'

  25850.      Extract and widen (promote) the high/low part of a vector of signed

  25851.      integral or floating point elements.  The input vector (operand 1)

  25852.      has N elements of size S.  Widen (promote) the high/low elements of

  25853.      the vector using signed or floating point extension and place the

  25854.      resulting N/2 values of size 2*S in the output vector (operand 0).

  25855. 

  25856. 'vec_unpacku_hi_M', 'vec_unpacku_lo_M'

  25857.      Extract and widen (promote) the high/low part of a vector of

  25858.      unsigned integral elements.  The input vector (operand 1) has N

  25859.      elements of size S. Widen (promote) the high/low elements of the

  25860.      vector using zero extension and place the resulting N/2 values of

  25861.      size 2*S in the output vector (operand 0).

  25862. 

  25863. 'vec_unpacks_sbool_hi_M', 'vec_unpacks_sbool_lo_M'

  25864.      Extract the high/low part of a vector of boolean elements that have

  25865.      scalar mode M.  The input vector (operand 1) has N elements, the

  25866.      output vector (operand 0) has N/2 elements.  The last operand

  25867.      (operand 2) is the number of elements of the input vector N as a

  25868.      'CONST_INT'.  These patterns are used if both the input and output

  25869.      vectors have the same scalar mode M and thus using

  25870.      'vec_unpacks_hi_M' or 'vec_unpacks_lo_M' would be ambiguous.

  25871. 

  25872. 'vec_unpacks_float_hi_M', 'vec_unpacks_float_lo_M'

  25873. 'vec_unpacku_float_hi_M', 'vec_unpacku_float_lo_M'

  25874.      Extract, convert to floating point type and widen the high/low part

  25875.      of a vector of signed/unsigned integral elements.  The input vector

  25876.      (operand 1) has N elements of size S.  Convert the high/low

  25877.      elements of the vector using floating point conversion and place

  25878.      the resulting N/2 values of size 2*S in the output vector (operand

  25879.      0).

  25880. 

  25881. 'vec_unpack_sfix_trunc_hi_M',

  25882. 'vec_unpack_sfix_trunc_lo_M'

  25883. 'vec_unpack_ufix_trunc_hi_M'

  25884. 'vec_unpack_ufix_trunc_lo_M'

  25885.      Extract, convert to signed/unsigned integer type and widen the

  25886.      high/low part of a vector of floating point elements.  The input

  25887.      vector (operand 1) has N elements of size S.  Convert the high/low

  25888.      elements of the vector to integers and place the resulting N/2

  25889.      values of size 2*S in the output vector (operand 0).

  25890. 

  25891. 'vec_widen_umult_hi_M', 'vec_widen_umult_lo_M'

  25892. 'vec_widen_smult_hi_M', 'vec_widen_smult_lo_M'

  25893. 'vec_widen_umult_even_M', 'vec_widen_umult_odd_M'

  25894. 'vec_widen_smult_even_M', 'vec_widen_smult_odd_M'

  25895.      Signed/Unsigned widening multiplication.  The two inputs (operands

  25896.      1 and 2) are vectors with N signed/unsigned elements of size S.

  25897.      Multiply the high/low or even/odd elements of the two vectors, and

  25898.      put the N/2 products of size 2*S in the output vector (operand 0).

  25899.      A target shouldn't implement even/odd pattern pair if it is less

  25900.      efficient than lo/hi one.

  25901. 

  25902. 'vec_widen_ushiftl_hi_M', 'vec_widen_ushiftl_lo_M'

  25903. 'vec_widen_sshiftl_hi_M', 'vec_widen_sshiftl_lo_M'

  25904.      Signed/Unsigned widening shift left.  The first input (operand 1)

  25905.      is a vector with N signed/unsigned elements of size S.  Operand 2

  25906.      is a constant.  Shift the high/low elements of operand 1, and put

  25907.      the N/2 results of size 2*S in the output vector (operand 0).

  25908. 

  25909. 'mulhisi3'

  25910.      Multiply operands 1 and 2, which have mode 'HImode', and store a

  25911.      'SImode' product in operand 0.

  25912. 

  25913. 'mulqihi3', 'mulsidi3'

  25914.      Similar widening-multiplication instructions of other widths.

  25915. 

  25916. 'umulqihi3', 'umulhisi3', 'umulsidi3'

  25917.      Similar widening-multiplication instructions that do unsigned

  25918.      multiplication.

  25919. 

  25920. 'usmulqihi3', 'usmulhisi3', 'usmulsidi3'

  25921.      Similar widening-multiplication instructions that interpret the

  25922.      first operand as unsigned and the second operand as signed, then do

  25923.      a signed multiplication.

  25924. 

  25925. 'smulM3_highpart'

  25926.      Perform a signed multiplication of operands 1 and 2, which have

  25927.      mode M, and store the most significant half of the product in

  25928.      operand 0.  The least significant half of the product is discarded.

  25929. 

  25930. 'umulM3_highpart'

  25931.      Similar, but the multiplication is unsigned.

  25932. 

  25933. 'maddMN4'

  25934.      Multiply operands 1 and 2, sign-extend them to mode N, add operand

  25935.      3, and store the result in operand 0.  Operands 1 and 2 have mode M

  25936.      and operands 0 and 3 have mode N.  Both modes must be integer or

  25937.      fixed-point modes and N must be twice the size of M.

  25938. 

  25939.      In other words, 'maddMN4' is like 'mulMN3' except that it also adds

  25940.      operand 3.

  25941. 

  25942.      These instructions are not allowed to 'FAIL'.

  25943. 

  25944. 'umaddMN4'

  25945.      Like 'maddMN4', but zero-extend the multiplication operands instead

  25946.      of sign-extending them.

  25947. 

  25948. 'ssmaddMN4'

  25949.      Like 'maddMN4', but all involved operations must be

  25950.      signed-saturating.

  25951. 

  25952. 'usmaddMN4'

  25953.      Like 'umaddMN4', but all involved operations must be

  25954.      unsigned-saturating.

  25955. 

  25956. 'msubMN4'

  25957.      Multiply operands 1 and 2, sign-extend them to mode N, subtract the

  25958.      result from operand 3, and store the result in operand 0.  Operands

  25959.      1 and 2 have mode M and operands 0 and 3 have mode N.  Both modes

  25960.      must be integer or fixed-point modes and N must be twice the size

  25961.      of M.

  25962. 

  25963.      In other words, 'msubMN4' is like 'mulMN3' except that it also

  25964.      subtracts the result from operand 3.

  25965. 

  25966.      These instructions are not allowed to 'FAIL'.

  25967. 

  25968. 'umsubMN4'

  25969.      Like 'msubMN4', but zero-extend the multiplication operands instead

  25970.      of sign-extending them.

  25971. 

  25972. 'ssmsubMN4'

  25973.      Like 'msubMN4', but all involved operations must be

  25974.      signed-saturating.

  25975. 

  25976. 'usmsubMN4'

  25977.      Like 'umsubMN4', but all involved operations must be

  25978.      unsigned-saturating.

  25979. 

  25980. 'divmodM4'

  25981.      Signed division that produces both a quotient and a remainder.

  25982.      Operand 1 is divided by operand 2 to produce a quotient stored in

  25983.      operand 0 and a remainder stored in operand 3.

  25984. 

  25985.      For machines with an instruction that produces both a quotient and

  25986.      a remainder, provide a pattern for 'divmodM4' but do not provide

  25987.      patterns for 'divM3' and 'modM3'.  This allows optimization in the

  25988.      relatively common case when both the quotient and remainder are

  25989.      computed.

  25990. 

  25991.      If an instruction that just produces a quotient or just a remainder

  25992.      exists and is more efficient than the instruction that produces

  25993.      both, write the output routine of 'divmodM4' to call

  25994.      'find_reg_note' and look for a 'REG_UNUSED' note on the quotient or

  25995.      remainder and generate the appropriate instruction.

  25996. 

  25997. 'udivmodM4'

  25998.      Similar, but does unsigned division.

  25999. 

  26000. 'ashlM3', 'ssashlM3', 'usashlM3'

  26001.      Arithmetic-shift operand 1 left by a number of bits specified by

  26002.      operand 2, and store the result in operand 0.  Here M is the mode

  26003.      of operand 0 and operand 1; operand 2's mode is specified by the

  26004.      instruction pattern, and the compiler will convert the operand to

  26005.      that mode before generating the instruction.  The shift or rotate

  26006.      expander or instruction pattern should explicitly specify the mode

  26007.      of the operand 2, it should never be 'VOIDmode'.  The meaning of

  26008.      out-of-range shift counts can optionally be specified by

  26009.      'TARGET_SHIFT_TRUNCATION_MASK'.  *Note

  26010.      TARGET_SHIFT_TRUNCATION_MASK::.  Operand 2 is always a scalar type.

  26011. 

  26012. 'ashrM3', 'lshrM3', 'rotlM3', 'rotrM3'

  26013.      Other shift and rotate instructions, analogous to the 'ashlM3'

  26014.      instructions.  Operand 2 is always a scalar type.

  26015. 

  26016. 'vashlM3', 'vashrM3', 'vlshrM3', 'vrotlM3', 'vrotrM3'

  26017.      Vector shift and rotate instructions that take vectors as operand 2

  26018.      instead of a scalar type.

  26019. 

  26020. 'avgM3_floor'

  26021. 'uavgM3_floor'

  26022.      Signed and unsigned average instructions.  These instructions add

  26023.      operands 1 and 2 without truncation, divide the result by 2, round

  26024.      towards -Inf, and store the result in operand 0.  This is

  26025.      equivalent to the C code:

  26026.           narrow op0, op1, op2;

  26027.           ...

  26028.           op0 = (narrow) (((wide) op1 + (wide) op2) >> 1);

  26029.      where the sign of 'narrow' determines whether this is a signed or

  26030.      unsigned operation.

  26031. 

  26032. 'avgM3_ceil'

  26033. 'uavgM3_ceil'

  26034.      Like 'avgM3_floor' and 'uavgM3_floor', but round towards +Inf.

  26035.      This is equivalent to the C code:

  26036.           narrow op0, op1, op2;

  26037.           ...

  26038.           op0 = (narrow) (((wide) op1 + (wide) op2 + 1) >> 1);

  26039. 

  26040. 'bswapM2'

  26041.      Reverse the order of bytes of operand 1 and store the result in

  26042.      operand 0.

  26043. 

  26044. 'negM2', 'ssnegM2', 'usnegM2'

  26045.      Negate operand 1 and store the result in operand 0.

  26046. 

  26047. 'negvM3'

  26048.      Like 'negM2' but takes a 'code_label' as operand 2 and emits code

  26049.      to jump to it if signed overflow occurs during the negation.

  26050. 

  26051. 'absM2'

  26052.      Store the absolute value of operand 1 into operand 0.

  26053. 

  26054. 'sqrtM2'

  26055.      Store the square root of operand 1 into operand 0.  Both operands

  26056.      have mode M, which is a scalar or vector floating-point mode.

  26057. 

  26058.      This pattern is not allowed to 'FAIL'.

  26059. 

  26060. 'rsqrtM2'

  26061.      Store the reciprocal of the square root of operand 1 into operand

  26062.      0.  Both operands have mode M, which is a scalar or vector

  26063.      floating-point mode.

  26064. 

  26065.      On most architectures this pattern is only approximate, so either

  26066.      its C condition or the 'TARGET_OPTAB_SUPPORTED_P' hook should check

  26067.      for the appropriate math flags.  (Using the C condition is more

  26068.      direct, but using 'TARGET_OPTAB_SUPPORTED_P' can be useful if a

  26069.      target-specific built-in also uses the 'rsqrtM2' pattern.)

  26070. 

  26071.      This pattern is not allowed to 'FAIL'.

  26072. 

  26073. 'fmodM3'

  26074.      Store the remainder of dividing operand 1 by operand 2 into operand

  26075.      0, rounded towards zero to an integer.  All operands have mode M,

  26076.      which is a scalar or vector floating-point mode.

  26077. 

  26078.      This pattern is not allowed to 'FAIL'.

  26079. 

  26080. 'remainderM3'

  26081.      Store the remainder of dividing operand 1 by operand 2 into operand

  26082.      0, rounded to the nearest integer.  All operands have mode M, which

  26083.      is a scalar or vector floating-point mode.

  26084. 

  26085.      This pattern is not allowed to 'FAIL'.

  26086. 

  26087. 'scalbM3'

  26088.      Raise 'FLT_RADIX' to the power of operand 2, multiply it by operand

  26089.      1, and store the result in operand 0.  All operands have mode M,

  26090.      which is a scalar or vector floating-point mode.

  26091. 

  26092.      This pattern is not allowed to 'FAIL'.

  26093. 

  26094. 'ldexpM3'

  26095.      Raise 2 to the power of operand 2, multiply it by operand 1, and

  26096.      store the result in operand 0.  Operands 0 and 1 have mode M, which

  26097.      is a scalar or vector floating-point mode.  Operand 2's mode has

  26098.      the same number of elements as M and each element is wide enough to

  26099.      store an 'int'.  The integers are signed.

  26100. 

  26101.      This pattern is not allowed to 'FAIL'.

  26102. 

  26103. 'cosM2'

  26104.      Store the cosine of operand 1 into operand 0.  Both operands have

  26105.      mode M, which is a scalar or vector floating-point mode.

  26106. 

  26107.      This pattern is not allowed to 'FAIL'.

  26108. 

  26109. 'sinM2'

  26110.      Store the sine of operand 1 into operand 0.  Both operands have

  26111.      mode M, which is a scalar or vector floating-point mode.

  26112. 

  26113.      This pattern is not allowed to 'FAIL'.

  26114. 

  26115. 'sincosM3'

  26116.      Store the cosine of operand 2 into operand 0 and the sine of

  26117.      operand 2 into operand 1.  All operands have mode M, which is a

  26118.      scalar or vector floating-point mode.

  26119. 

  26120.      Targets that can calculate the sine and cosine simultaneously can

  26121.      implement this pattern as opposed to implementing individual

  26122.      'sinM2' and 'cosM2' patterns.  The 'sin' and 'cos' built-in

  26123.      functions will then be expanded to the 'sincosM3' pattern, with one

  26124.      of the output values left unused.

  26125. 

  26126. 'tanM2'

  26127.      Store the tangent of operand 1 into operand 0.  Both operands have

  26128.      mode M, which is a scalar or vector floating-point mode.

  26129. 

  26130.      This pattern is not allowed to 'FAIL'.

  26131. 

  26132. 'asinM2'

  26133.      Store the arc sine of operand 1 into operand 0.  Both operands have

  26134.      mode M, which is a scalar or vector floating-point mode.

  26135. 

  26136.      This pattern is not allowed to 'FAIL'.

  26137. 

  26138. 'acosM2'

  26139.      Store the arc cosine of operand 1 into operand 0.  Both operands

  26140.      have mode M, which is a scalar or vector floating-point mode.

  26141. 

  26142.      This pattern is not allowed to 'FAIL'.

  26143. 

  26144. 'atanM2'

  26145.      Store the arc tangent of operand 1 into operand 0.  Both operands

  26146.      have mode M, which is a scalar or vector floating-point mode.

  26147. 

  26148.      This pattern is not allowed to 'FAIL'.

  26149. 

  26150. 'expM2'

  26151.      Raise e (the base of natural logarithms) to the power of operand 1

  26152.      and store the result in operand 0.  Both operands have mode M,

  26153.      which is a scalar or vector floating-point mode.

  26154. 

  26155.      This pattern is not allowed to 'FAIL'.

  26156. 

  26157. 'expm1M2'

  26158.      Raise e (the base of natural logarithms) to the power of operand 1,

  26159.      subtract 1, and store the result in operand 0.  Both operands have

  26160.      mode M, which is a scalar or vector floating-point mode.

  26161. 

  26162.      For inputs close to zero, the pattern is expected to be more

  26163.      accurate than a separate 'expM2' and 'subM3' would be.

  26164. 

  26165.      This pattern is not allowed to 'FAIL'.

  26166. 

  26167. 'exp10M2'

  26168.      Raise 10 to the power of operand 1 and store the result in operand

  26169.      0.  Both operands have mode M, which is a scalar or vector

  26170.      floating-point mode.

  26171. 

  26172.      This pattern is not allowed to 'FAIL'.

  26173. 

  26174. 'exp2M2'

  26175.      Raise 2 to the power of operand 1 and store the result in operand

  26176.      0.  Both operands have mode M, which is a scalar or vector

  26177.      floating-point mode.

  26178. 

  26179.      This pattern is not allowed to 'FAIL'.

  26180. 

  26181. 'logM2'

  26182.      Store the natural logarithm of operand 1 into operand 0.  Both

  26183.      operands have mode M, which is a scalar or vector floating-point

  26184.      mode.

  26185. 

  26186.      This pattern is not allowed to 'FAIL'.

  26187. 

  26188. 'log1pM2'

  26189.      Add 1 to operand 1, compute the natural logarithm, and store the

  26190.      result in operand 0.  Both operands have mode M, which is a scalar

  26191.      or vector floating-point mode.

  26192. 

  26193.      For inputs close to zero, the pattern is expected to be more

  26194.      accurate than a separate 'addM3' and 'logM2' would be.

  26195. 

  26196.      This pattern is not allowed to 'FAIL'.

  26197. 

  26198. 'log10M2'

  26199.      Store the base-10 logarithm of operand 1 into operand 0.  Both

  26200.      operands have mode M, which is a scalar or vector floating-point

  26201.      mode.

  26202. 

  26203.      This pattern is not allowed to 'FAIL'.

  26204. 

  26205. 'log2M2'

  26206.      Store the base-2 logarithm of operand 1 into operand 0.  Both

  26207.      operands have mode M, which is a scalar or vector floating-point

  26208.      mode.

  26209. 

  26210.      This pattern is not allowed to 'FAIL'.

  26211. 

  26212. 'logbM2'

  26213.      Store the base-'FLT_RADIX' logarithm of operand 1 into operand 0.

  26214.      Both operands have mode M, which is a scalar or vector

  26215.      floating-point mode.

  26216. 

  26217.      This pattern is not allowed to 'FAIL'.

  26218. 

  26219. 'significandM2'

  26220.      Store the significand of floating-point operand 1 in operand 0.

  26221.      Both operands have mode M, which is a scalar or vector

  26222.      floating-point mode.

  26223. 

  26224.      This pattern is not allowed to 'FAIL'.

  26225. 

  26226. 'powM3'

  26227.      Store the value of operand 1 raised to the exponent operand 2 into

  26228.      operand 0.  All operands have mode M, which is a scalar or vector

  26229.      floating-point mode.

  26230. 

  26231.      This pattern is not allowed to 'FAIL'.

  26232. 

  26233. 'atan2M3'

  26234.      Store the arc tangent (inverse tangent) of operand 1 divided by

  26235.      operand 2 into operand 0, using the signs of both arguments to

  26236.      determine the quadrant of the result.  All operands have mode M,

  26237.      which is a scalar or vector floating-point mode.

  26238. 

  26239.      This pattern is not allowed to 'FAIL'.

  26240. 

  26241. 'floorM2'

  26242.      Store the largest integral value not greater than operand 1 in

  26243.      operand 0.  Both operands have mode M, which is a scalar or vector

  26244.      floating-point mode.  If '-ffp-int-builtin-inexact' is in effect,

  26245.      the "inexact" exception may be raised for noninteger operands;

  26246.      otherwise, it may not.

  26247. 

  26248.      This pattern is not allowed to 'FAIL'.

  26249. 

  26250. 'btruncM2'

  26251.      Round operand 1 to an integer, towards zero, and store the result

  26252.      in operand 0.  Both operands have mode M, which is a scalar or

  26253.      vector floating-point mode.  If '-ffp-int-builtin-inexact' is in

  26254.      effect, the "inexact" exception may be raised for noninteger

  26255.      operands; otherwise, it may not.

  26256. 

  26257.      This pattern is not allowed to 'FAIL'.

  26258. 

  26259. 'roundM2'

  26260.      Round operand 1 to the nearest integer, rounding away from zero in

  26261.      the event of a tie, and store the result in operand 0.  Both

  26262.      operands have mode M, which is a scalar or vector floating-point

  26263.      mode.  If '-ffp-int-builtin-inexact' is in effect, the "inexact"

  26264.      exception may be raised for noninteger operands; otherwise, it may

  26265.      not.

  26266. 

  26267.      This pattern is not allowed to 'FAIL'.

  26268. 

  26269. 'ceilM2'

  26270.      Store the smallest integral value not less than operand 1 in

  26271.      operand 0.  Both operands have mode M, which is a scalar or vector

  26272.      floating-point mode.  If '-ffp-int-builtin-inexact' is in effect,

  26273.      the "inexact" exception may be raised for noninteger operands;

  26274.      otherwise, it may not.

  26275. 

  26276.      This pattern is not allowed to 'FAIL'.

  26277. 

  26278. 'nearbyintM2'

  26279.      Round operand 1 to an integer, using the current rounding mode, and

  26280.      store the result in operand 0.  Do not raise an inexact condition

  26281.      when the result is different from the argument.  Both operands have

  26282.      mode M, which is a scalar or vector floating-point mode.

  26283. 

  26284.      This pattern is not allowed to 'FAIL'.

  26285. 

  26286. 'rintM2'

  26287.      Round operand 1 to an integer, using the current rounding mode, and

  26288.      store the result in operand 0.  Raise an inexact condition when the

  26289.      result is different from the argument.  Both operands have mode M,

  26290.      which is a scalar or vector floating-point mode.

  26291. 

  26292.      This pattern is not allowed to 'FAIL'.

  26293. 

  26294. 'lrintMN2'

  26295.      Convert operand 1 (valid for floating point mode M) to fixed point

  26296.      mode N as a signed number according to the current rounding mode

  26297.      and store in operand 0 (which has mode N).

  26298. 

  26299. 'lroundMN2'

  26300.      Convert operand 1 (valid for floating point mode M) to fixed point

  26301.      mode N as a signed number rounding to nearest and away from zero

  26302.      and store in operand 0 (which has mode N).

  26303. 

  26304. 'lfloorMN2'

  26305.      Convert operand 1 (valid for floating point mode M) to fixed point

  26306.      mode N as a signed number rounding down and store in operand 0

  26307.      (which has mode N).

  26308. 

  26309. 'lceilMN2'

  26310.      Convert operand 1 (valid for floating point mode M) to fixed point

  26311.      mode N as a signed number rounding up and store in operand 0 (which

  26312.      has mode N).

  26313. 

  26314. 'copysignM3'

  26315.      Store a value with the magnitude of operand 1 and the sign of

  26316.      operand 2 into operand 0.  All operands have mode M, which is a

  26317.      scalar or vector floating-point mode.

  26318. 

  26319.      This pattern is not allowed to 'FAIL'.

  26320. 

  26321. 'xorsignM3'

  26322.      Equivalent to 'op0 = op1 * copysign (1.0, op2)': store a value with

  26323.      the magnitude of operand 1 and the sign of operand 2 into operand

  26324.      0.  All operands have mode M, which is a scalar or vector

  26325.      floating-point mode.

  26326. 

  26327.      This pattern is not allowed to 'FAIL'.

  26328. 

  26329. 'ffsM2'

  26330.      Store into operand 0 one plus the index of the least significant

  26331.      1-bit of operand 1.  If operand 1 is zero, store zero.

  26332. 

  26333.      M is either a scalar or vector integer mode.  When it is a scalar,

  26334.      operand 1 has mode M but operand 0 can have whatever scalar integer

  26335.      mode is suitable for the target.  The compiler will insert

  26336.      conversion instructions as necessary (typically to convert the

  26337.      result to the same width as 'int').  When M is a vector, both

  26338.      operands must have mode M.

  26339. 

  26340.      This pattern is not allowed to 'FAIL'.

  26341. 

  26342. 'clrsbM2'

  26343.      Count leading redundant sign bits.  Store into operand 0 the number

  26344.      of redundant sign bits in operand 1, starting at the most

  26345.      significant bit position.  A redundant sign bit is defined as any

  26346.      sign bit after the first.  As such, this count will be one less

  26347.      than the count of leading sign bits.

  26348. 

  26349.      M is either a scalar or vector integer mode.  When it is a scalar,

  26350.      operand 1 has mode M but operand 0 can have whatever scalar integer

  26351.      mode is suitable for the target.  The compiler will insert

  26352.      conversion instructions as necessary (typically to convert the

  26353.      result to the same width as 'int').  When M is a vector, both

  26354.      operands must have mode M.

  26355. 

  26356.      This pattern is not allowed to 'FAIL'.

  26357. 

  26358. 'clzM2'

  26359.      Store into operand 0 the number of leading 0-bits in operand 1,

  26360.      starting at the most significant bit position.  If operand 1 is 0,

  26361.      the 'CLZ_DEFINED_VALUE_AT_ZERO' (*note Misc::) macro defines if the

  26362.      result is undefined or has a useful value.

  26363. 

  26364.      M is either a scalar or vector integer mode.  When it is a scalar,

  26365.      operand 1 has mode M but operand 0 can have whatever scalar integer

  26366.      mode is suitable for the target.  The compiler will insert

  26367.      conversion instructions as necessary (typically to convert the

  26368.      result to the same width as 'int').  When M is a vector, both

  26369.      operands must have mode M.

  26370. 

  26371.      This pattern is not allowed to 'FAIL'.

  26372. 

  26373. 'ctzM2'

  26374.      Store into operand 0 the number of trailing 0-bits in operand 1,

  26375.      starting at the least significant bit position.  If operand 1 is 0,

  26376.      the 'CTZ_DEFINED_VALUE_AT_ZERO' (*note Misc::) macro defines if the

  26377.      result is undefined or has a useful value.

  26378. 

  26379.      M is either a scalar or vector integer mode.  When it is a scalar,

  26380.      operand 1 has mode M but operand 0 can have whatever scalar integer

  26381.      mode is suitable for the target.  The compiler will insert

  26382.      conversion instructions as necessary (typically to convert the

  26383.      result to the same width as 'int').  When M is a vector, both

  26384.      operands must have mode M.

  26385. 

  26386.      This pattern is not allowed to 'FAIL'.

  26387. 

  26388. 'popcountM2'

  26389.      Store into operand 0 the number of 1-bits in operand 1.

  26390. 

  26391.      M is either a scalar or vector integer mode.  When it is a scalar,

  26392.      operand 1 has mode M but operand 0 can have whatever scalar integer

  26393.      mode is suitable for the target.  The compiler will insert

  26394.      conversion instructions as necessary (typically to convert the

  26395.      result to the same width as 'int').  When M is a vector, both

  26396.      operands must have mode M.

  26397. 

  26398.      This pattern is not allowed to 'FAIL'.

  26399. 

  26400. 'parityM2'

  26401.      Store into operand 0 the parity of operand 1, i.e. the number of

  26402.      1-bits in operand 1 modulo 2.

  26403. 

  26404.      M is either a scalar or vector integer mode.  When it is a scalar,

  26405.      operand 1 has mode M but operand 0 can have whatever scalar integer

  26406.      mode is suitable for the target.  The compiler will insert

  26407.      conversion instructions as necessary (typically to convert the

  26408.      result to the same width as 'int').  When M is a vector, both

  26409.      operands must have mode M.

  26410. 

  26411.      This pattern is not allowed to 'FAIL'.

  26412. 

  26413. 'one_cmplM2'

  26414.      Store the bitwise-complement of operand 1 into operand 0.

  26415. 

  26416. 'cpymemM'

  26417.      Block copy instruction.  The destination and source blocks of

  26418.      memory are the first two operands, and both are 'mem:BLK's with an

  26419.      address in mode 'Pmode'.

  26420. 

  26421.      The number of bytes to copy is the third operand, in mode M.

  26422.      Usually, you specify 'Pmode' for M.  However, if you can generate

  26423.      better code knowing the range of valid lengths is smaller than

  26424.      those representable in a full Pmode pointer, you should provide a

  26425.      pattern with a mode corresponding to the range of values you can

  26426.      handle efficiently (e.g., 'QImode' for values in the range 0-127;

  26427.      note we avoid numbers that appear negative) and also a pattern with

  26428.      'Pmode'.

  26429. 

  26430.      The fourth operand is the known shared alignment of the source and

  26431.      destination, in the form of a 'const_int' rtx.  Thus, if the

  26432.      compiler knows that both source and destination are word-aligned,

  26433.      it may provide the value 4 for this operand.

  26434. 

  26435.      Optional operands 5 and 6 specify expected alignment and size of

  26436.      block respectively.  The expected alignment differs from alignment

  26437.      in operand 4 in a way that the blocks are not required to be

  26438.      aligned according to it in all cases.  This expected alignment is

  26439.      also in bytes, just like operand 4.  Expected size, when unknown,

  26440.      is set to '(const_int -1)'.

  26441. 

  26442.      Descriptions of multiple 'cpymemM' patterns can only be beneficial

  26443.      if the patterns for smaller modes have fewer restrictions on their

  26444.      first, second and fourth operands.  Note that the mode M in

  26445.      'cpymemM' does not impose any restriction on the mode of

  26446.      individually copied data units in the block.

  26447. 

  26448.      The 'cpymemM' patterns need not give special consideration to the

  26449.      possibility that the source and destination strings might overlap.

  26450.      These patterns are used to do inline expansion of

  26451.      '__builtin_memcpy'.

  26452. 

  26453. 'movmemM'

  26454.      Block move instruction.  The destination and source blocks of

  26455.      memory are the first two operands, and both are 'mem:BLK's with an

  26456.      address in mode 'Pmode'.

  26457. 

  26458.      The number of bytes to copy is the third operand, in mode M.

  26459.      Usually, you specify 'Pmode' for M.  However, if you can generate

  26460.      better code knowing the range of valid lengths is smaller than

  26461.      those representable in a full Pmode pointer, you should provide a

  26462.      pattern with a mode corresponding to the range of values you can

  26463.      handle efficiently (e.g., 'QImode' for values in the range 0-127;

  26464.      note we avoid numbers that appear negative) and also a pattern with

  26465.      'Pmode'.

  26466. 

  26467.      The fourth operand is the known shared alignment of the source and

  26468.      destination, in the form of a 'const_int' rtx.  Thus, if the

  26469.      compiler knows that both source and destination are word-aligned,

  26470.      it may provide the value 4 for this operand.

  26471. 

  26472.      Optional operands 5 and 6 specify expected alignment and size of

  26473.      block respectively.  The expected alignment differs from alignment

  26474.      in operand 4 in a way that the blocks are not required to be

  26475.      aligned according to it in all cases.  This expected alignment is

  26476.      also in bytes, just like operand 4.  Expected size, when unknown,

  26477.      is set to '(const_int -1)'.

  26478. 

  26479.      Descriptions of multiple 'movmemM' patterns can only be beneficial

  26480.      if the patterns for smaller modes have fewer restrictions on their

  26481.      first, second and fourth operands.  Note that the mode M in

  26482.      'movmemM' does not impose any restriction on the mode of

  26483.      individually copied data units in the block.

  26484. 

  26485.      The 'movmemM' patterns must correctly handle the case where the

  26486.      source and destination strings overlap.  These patterns are used to

  26487.      do inline expansion of '__builtin_memmove'.

  26488. 

  26489. 'movstr'

  26490.      String copy instruction, with 'stpcpy' semantics.  Operand 0 is an

  26491.      output operand in mode 'Pmode'.  The addresses of the destination

  26492.      and source strings are operands 1 and 2, and both are 'mem:BLK's

  26493.      with addresses in mode 'Pmode'.  The execution of the expansion of

  26494.      this pattern should store in operand 0 the address in which the

  26495.      'NUL' terminator was stored in the destination string.

  26496. 

  26497.      This pattern has also several optional operands that are same as in

  26498.      'setmem'.

  26499. 

  26500. 'setmemM'

  26501.      Block set instruction.  The destination string is the first

  26502.      operand, given as a 'mem:BLK' whose address is in mode 'Pmode'.

  26503.      The number of bytes to set is the second operand, in mode M.  The

  26504.      value to initialize the memory with is the third operand.  Targets

  26505.      that only support the clearing of memory should reject any value

  26506.      that is not the constant 0.  See 'cpymemM' for a discussion of the

  26507.      choice of mode.

  26508. 

  26509.      The fourth operand is the known alignment of the destination, in

  26510.      the form of a 'const_int' rtx.  Thus, if the compiler knows that

  26511.      the destination is word-aligned, it may provide the value 4 for

  26512.      this operand.

  26513. 

  26514.      Optional operands 5 and 6 specify expected alignment and size of

  26515.      block respectively.  The expected alignment differs from alignment

  26516.      in operand 4 in a way that the blocks are not required to be

  26517.      aligned according to it in all cases.  This expected alignment is

  26518.      also in bytes, just like operand 4.  Expected size, when unknown,

  26519.      is set to '(const_int -1)'.  Operand 7 is the minimal size of the

  26520.      block and operand 8 is the maximal size of the block (NULL if it

  26521.      cannot be represented as CONST_INT). Operand 9 is the probable

  26522.      maximal size (i.e. we cannot rely on it for correctness, but it can

  26523.      be used for choosing proper code sequence for a given size).

  26524. 

  26525.      The use for multiple 'setmemM' is as for 'cpymemM'.

  26526. 

  26527. 'cmpstrnM'

  26528.      String compare instruction, with five operands.  Operand 0 is the

  26529.      output; it has mode M.  The remaining four operands are like the

  26530.      operands of 'cpymemM'.  The two memory blocks specified are

  26531.      compared byte by byte in lexicographic order starting at the

  26532.      beginning of each string.  The instruction is not allowed to

  26533.      prefetch more than one byte at a time since either string may end

  26534.      in the first byte and reading past that may access an invalid page

  26535.      or segment and cause a fault.  The comparison terminates early if

  26536.      the fetched bytes are different or if they are equal to zero.  The

  26537.      effect of the instruction is to store a value in operand 0 whose

  26538.      sign indicates the result of the comparison.

  26539. 

  26540. 'cmpstrM'

  26541.      String compare instruction, without known maximum length.  Operand

  26542.      0 is the output; it has mode M.  The second and third operand are

  26543.      the blocks of memory to be compared; both are 'mem:BLK' with an

  26544.      address in mode 'Pmode'.

  26545. 

  26546.      The fourth operand is the known shared alignment of the source and

  26547.      destination, in the form of a 'const_int' rtx.  Thus, if the

  26548.      compiler knows that both source and destination are word-aligned,

  26549.      it may provide the value 4 for this operand.

  26550. 

  26551.      The two memory blocks specified are compared byte by byte in

  26552.      lexicographic order starting at the beginning of each string.  The

  26553.      instruction is not allowed to prefetch more than one byte at a time

  26554.      since either string may end in the first byte and reading past that

  26555.      may access an invalid page or segment and cause a fault.  The

  26556.      comparison will terminate when the fetched bytes are different or

  26557.      if they are equal to zero.  The effect of the instruction is to

  26558.      store a value in operand 0 whose sign indicates the result of the

  26559.      comparison.

  26560. 

  26561. 'cmpmemM'

  26562.      Block compare instruction, with five operands like the operands of

  26563.      'cmpstrM'.  The two memory blocks specified are compared byte by

  26564.      byte in lexicographic order starting at the beginning of each

  26565.      block.  Unlike 'cmpstrM' the instruction can prefetch any bytes in

  26566.      the two memory blocks.  Also unlike 'cmpstrM' the comparison will

  26567.      not stop if both bytes are zero.  The effect of the instruction is

  26568.      to store a value in operand 0 whose sign indicates the result of

  26569.      the comparison.

  26570. 

  26571. 'strlenM'

  26572.      Compute the length of a string, with three operands.  Operand 0 is

  26573.      the result (of mode M), operand 1 is a 'mem' referring to the first

  26574.      character of the string, operand 2 is the character to search for

  26575.      (normally zero), and operand 3 is a constant describing the known

  26576.      alignment of the beginning of the string.

  26577. 

  26578. 'floatMN2'

  26579.      Convert signed integer operand 1 (valid for fixed point mode M) to

  26580.      floating point mode N and store in operand 0 (which has mode N).

  26581. 

  26582. 'floatunsMN2'

  26583.      Convert unsigned integer operand 1 (valid for fixed point mode M)

  26584.      to floating point mode N and store in operand 0 (which has mode N).

  26585. 

  26586. 'fixMN2'

  26587.      Convert operand 1 (valid for floating point mode M) to fixed point

  26588.      mode N as a signed number and store in operand 0 (which has mode

  26589.      N).  This instruction's result is defined only when the value of

  26590.      operand 1 is an integer.

  26591. 

  26592.      If the machine description defines this pattern, it also needs to

  26593.      define the 'ftrunc' pattern.

  26594. 

  26595. 'fixunsMN2'

  26596.      Convert operand 1 (valid for floating point mode M) to fixed point

  26597.      mode N as an unsigned number and store in operand 0 (which has mode

  26598.      N).  This instruction's result is defined only when the value of

  26599.      operand 1 is an integer.

  26600. 

  26601. 'ftruncM2'

  26602.      Convert operand 1 (valid for floating point mode M) to an integer

  26603.      value, still represented in floating point mode M, and store it in

  26604.      operand 0 (valid for floating point mode M).

  26605. 

  26606. 'fix_truncMN2'

  26607.      Like 'fixMN2' but works for any floating point value of mode M by

  26608.      converting the value to an integer.

  26609. 

  26610. 'fixuns_truncMN2'

  26611.      Like 'fixunsMN2' but works for any floating point value of mode M

  26612.      by converting the value to an integer.

  26613. 

  26614. 'truncMN2'

  26615.      Truncate operand 1 (valid for mode M) to mode N and store in

  26616.      operand 0 (which has mode N).  Both modes must be fixed point or

  26617.      both floating point.

  26618. 

  26619. 'extendMN2'

  26620.      Sign-extend operand 1 (valid for mode M) to mode N and store in

  26621.      operand 0 (which has mode N).  Both modes must be fixed point or

  26622.      both floating point.

  26623. 

  26624. 'zero_extendMN2'

  26625.      Zero-extend operand 1 (valid for mode M) to mode N and store in

  26626.      operand 0 (which has mode N).  Both modes must be fixed point.

  26627. 

  26628. 'fractMN2'

  26629.      Convert operand 1 of mode M to mode N and store in operand 0 (which

  26630.      has mode N).  Mode M and mode N could be fixed-point to

  26631.      fixed-point, signed integer to fixed-point, fixed-point to signed

  26632.      integer, floating-point to fixed-point, or fixed-point to

  26633.      floating-point.  When overflows or underflows happen, the results

  26634.      are undefined.

  26635. 

  26636. 'satfractMN2'

  26637.      Convert operand 1 of mode M to mode N and store in operand 0 (which

  26638.      has mode N).  Mode M and mode N could be fixed-point to

  26639.      fixed-point, signed integer to fixed-point, or floating-point to

  26640.      fixed-point.  When overflows or underflows happen, the instruction

  26641.      saturates the results to the maximum or the minimum.

  26642. 

  26643. 'fractunsMN2'

  26644.      Convert operand 1 of mode M to mode N and store in operand 0 (which

  26645.      has mode N).  Mode M and mode N could be unsigned integer to

  26646.      fixed-point, or fixed-point to unsigned integer.  When overflows or

  26647.      underflows happen, the results are undefined.

  26648. 

  26649. 'satfractunsMN2'

  26650.      Convert unsigned integer operand 1 of mode M to fixed-point mode N

  26651.      and store in operand 0 (which has mode N).  When overflows or

  26652.      underflows happen, the instruction saturates the results to the

  26653.      maximum or the minimum.

  26654. 

  26655. 'extvM'

  26656.      Extract a bit-field from register operand 1, sign-extend it, and

  26657.      store it in operand 0.  Operand 2 specifies the width of the field

  26658.      in bits and operand 3 the starting bit, which counts from the most

  26659.      significant bit if 'BITS_BIG_ENDIAN' is true and from the least

  26660.      significant bit otherwise.

  26661. 

  26662.      Operands 0 and 1 both have mode M.  Operands 2 and 3 have a

  26663.      target-specific mode.

  26664. 

  26665. 'extvmisalignM'

  26666.      Extract a bit-field from memory operand 1, sign extend it, and

  26667.      store it in operand 0.  Operand 2 specifies the width in bits and

  26668.      operand 3 the starting bit.  The starting bit is always somewhere

  26669.      in the first byte of operand 1; it counts from the most significant

  26670.      bit if 'BITS_BIG_ENDIAN' is true and from the least significant bit

  26671.      otherwise.

  26672. 

  26673.      Operand 0 has mode M while operand 1 has 'BLK' mode.  Operands 2

  26674.      and 3 have a target-specific mode.

  26675. 

  26676.      The instruction must not read beyond the last byte of the

  26677.      bit-field.

  26678. 

  26679. 'extzvM'

  26680.      Like 'extvM' except that the bit-field value is zero-extended.

  26681. 

  26682. 'extzvmisalignM'

  26683.      Like 'extvmisalignM' except that the bit-field value is

  26684.      zero-extended.

  26685. 

  26686. 'insvM'

  26687.      Insert operand 3 into a bit-field of register operand 0.  Operand 1

  26688.      specifies the width of the field in bits and operand 2 the starting

  26689.      bit, which counts from the most significant bit if

  26690.      'BITS_BIG_ENDIAN' is true and from the least significant bit

  26691.      otherwise.

  26692. 

  26693.      Operands 0 and 3 both have mode M.  Operands 1 and 2 have a

  26694.      target-specific mode.

  26695. 

  26696. 'insvmisalignM'

  26697.      Insert operand 3 into a bit-field of memory operand 0.  Operand 1

  26698.      specifies the width of the field in bits and operand 2 the starting

  26699.      bit.  The starting bit is always somewhere in the first byte of

  26700.      operand 0; it counts from the most significant bit if

  26701.      'BITS_BIG_ENDIAN' is true and from the least significant bit

  26702.      otherwise.

  26703. 

  26704.      Operand 3 has mode M while operand 0 has 'BLK' mode.  Operands 1

  26705.      and 2 have a target-specific mode.

  26706. 

  26707.      The instruction must not read or write beyond the last byte of the

  26708.      bit-field.

  26709. 

  26710. 'extv'

  26711.      Extract a bit-field from operand 1 (a register or memory operand),

  26712.      where operand 2 specifies the width in bits and operand 3 the

  26713.      starting bit, and store it in operand 0.  Operand 0 must have mode

  26714.      'word_mode'.  Operand 1 may have mode 'byte_mode' or 'word_mode';

  26715.      often 'word_mode' is allowed only for registers.  Operands 2 and 3

  26716.      must be valid for 'word_mode'.

  26717. 

  26718.      The RTL generation pass generates this instruction only with

  26719.      constants for operands 2 and 3 and the constant is never zero for

  26720.      operand 2.

  26721. 

  26722.      The bit-field value is sign-extended to a full word integer before

  26723.      it is stored in operand 0.

  26724. 

  26725.      This pattern is deprecated; please use 'extvM' and 'extvmisalignM'

  26726.      instead.

  26727. 

  26728. 'extzv'

  26729.      Like 'extv' except that the bit-field value is zero-extended.

  26730. 

  26731.      This pattern is deprecated; please use 'extzvM' and

  26732.      'extzvmisalignM' instead.

  26733. 

  26734. 'insv'

  26735.      Store operand 3 (which must be valid for 'word_mode') into a

  26736.      bit-field in operand 0, where operand 1 specifies the width in bits

  26737.      and operand 2 the starting bit.  Operand 0 may have mode

  26738.      'byte_mode' or 'word_mode'; often 'word_mode' is allowed only for

  26739.      registers.  Operands 1 and 2 must be valid for 'word_mode'.

  26740. 

  26741.      The RTL generation pass generates this instruction only with

  26742.      constants for operands 1 and 2 and the constant is never zero for

  26743.      operand 1.

  26744. 

  26745.      This pattern is deprecated; please use 'insvM' and 'insvmisalignM'

  26746.      instead.

  26747. 

  26748. 'movMODEcc'

  26749.      Conditionally move operand 2 or operand 3 into operand 0 according

  26750.      to the comparison in operand 1.  If the comparison is true, operand

  26751.      2 is moved into operand 0, otherwise operand 3 is moved.

  26752. 

  26753.      The mode of the operands being compared need not be the same as the

  26754.      operands being moved.  Some machines, sparc64 for example, have

  26755.      instructions that conditionally move an integer value based on the

  26756.      floating point condition codes and vice versa.

  26757. 

  26758.      If the machine does not have conditional move instructions, do not

  26759.      define these patterns.

  26760. 

  26761. 'addMODEcc'

  26762.      Similar to 'movMODEcc' but for conditional addition.  Conditionally

  26763.      move operand 2 or (operands 2 + operand 3) into operand 0 according

  26764.      to the comparison in operand 1.  If the comparison is false,

  26765.      operand 2 is moved into operand 0, otherwise (operand 2 + operand

  26766.      3) is moved.

  26767. 

  26768. 'cond_addMODE'

  26769. 'cond_subMODE'

  26770. 'cond_mulMODE'

  26771. 'cond_divMODE'

  26772. 'cond_udivMODE'

  26773. 'cond_modMODE'

  26774. 'cond_umodMODE'

  26775. 'cond_andMODE'

  26776. 'cond_iorMODE'

  26777. 'cond_xorMODE'

  26778. 'cond_sminMODE'

  26779. 'cond_smaxMODE'

  26780. 'cond_uminMODE'

  26781. 'cond_umaxMODE'

  26782.      When operand 1 is true, perform an operation on operands 2 and 3

  26783.      and store the result in operand 0, otherwise store operand 4 in

  26784.      operand 0.  The operation works elementwise if the operands are

  26785.      vectors.

  26786. 

  26787.      The scalar case is equivalent to:

  26788. 

  26789.           op0 = op1 ? op2 OP op3 : op4;

  26790. 

  26791.      while the vector case is equivalent to:

  26792. 

  26793.           for (i = 0; i < GET_MODE_NUNITS (M); i++)

  26794.             op0[i] = op1[i] ? op2[i] OP op3[i] : op4[i];

  26795. 

  26796.      where, for example, OP is '+' for 'cond_addMODE'.

  26797. 

  26798.      When defined for floating-point modes, the contents of 'op3[i]' are

  26799.      not interpreted if 'op1[i]' is false, just like they would not be

  26800.      in a normal C '?:' condition.

  26801. 

  26802.      Operands 0, 2, 3 and 4 all have mode M.  Operand 1 is a scalar

  26803.      integer if M is scalar, otherwise it has the mode returned by

  26804.      'TARGET_VECTORIZE_GET_MASK_MODE'.

  26805. 

  26806. 'cond_fmaMODE'

  26807. 'cond_fmsMODE'

  26808. 'cond_fnmaMODE'

  26809. 'cond_fnmsMODE'

  26810.      Like 'cond_addM', except that the conditional operation takes 3

  26811.      operands rather than two.  For example, the vector form of

  26812.      'cond_fmaMODE' is equivalent to:

  26813. 

  26814.           for (i = 0; i < GET_MODE_NUNITS (M); i++)

  26815.             op0[i] = op1[i] ? fma (op2[i], op3[i], op4[i]) : op5[i];

  26816. 

  26817. 'negMODEcc'

  26818.      Similar to 'movMODEcc' but for conditional negation.  Conditionally

  26819.      move the negation of operand 2 or the unchanged operand 3 into

  26820.      operand 0 according to the comparison in operand 1.  If the

  26821.      comparison is true, the negation of operand 2 is moved into operand

  26822.      0, otherwise operand 3 is moved.

  26823. 

  26824. 'notMODEcc'

  26825.      Similar to 'negMODEcc' but for conditional complement.

  26826.      Conditionally move the bitwise complement of operand 2 or the

  26827.      unchanged operand 3 into operand 0 according to the comparison in

  26828.      operand 1.  If the comparison is true, the complement of operand 2

  26829.      is moved into operand 0, otherwise operand 3 is moved.

  26830. 

  26831. 'cstoreMODE4'

  26832.      Store zero or nonzero in operand 0 according to whether a

  26833.      comparison is true.  Operand 1 is a comparison operator.  Operand 2

  26834.      and operand 3 are the first and second operand of the comparison,

  26835.      respectively.  You specify the mode that operand 0 must have when

  26836.      you write the 'match_operand' expression.  The compiler

  26837.      automatically sees which mode you have used and supplies an operand

  26838.      of that mode.

  26839. 

  26840.      The value stored for a true condition must have 1 as its low bit,

  26841.      or else must be negative.  Otherwise the instruction is not

  26842.      suitable and you should omit it from the machine description.  You

  26843.      describe to the compiler exactly which value is stored by defining

  26844.      the macro 'STORE_FLAG_VALUE' (*note Misc::).  If a description

  26845.      cannot be found that can be used for all the possible comparison

  26846.      operators, you should pick one and use a 'define_expand' to map all

  26847.      results onto the one you chose.

  26848. 

  26849.      These operations may 'FAIL', but should do so only in relatively

  26850.      uncommon cases; if they would 'FAIL' for common cases involving

  26851.      integer comparisons, it is best to restrict the predicates to not

  26852.      allow these operands.  Likewise if a given comparison operator will

  26853.      always fail, independent of the operands (for floating-point modes,

  26854.      the 'ordered_comparison_operator' predicate is often useful in this

  26855.      case).

  26856. 

  26857.      If this pattern is omitted, the compiler will generate a

  26858.      conditional branch--for example, it may copy a constant one to the

  26859.      target and branching around an assignment of zero to the target--or

  26860.      a libcall.  If the predicate for operand 1 only rejects some

  26861.      operators, it will also try reordering the operands and/or

  26862.      inverting the result value (e.g. by an exclusive OR). These

  26863.      possibilities could be cheaper or equivalent to the instructions

  26864.      used for the 'cstoreMODE4' pattern followed by those required to

  26865.      convert a positive result from 'STORE_FLAG_VALUE' to 1; in this

  26866.      case, you can and should make operand 1's predicate reject some

  26867.      operators in the 'cstoreMODE4' pattern, or remove the pattern

  26868.      altogether from the machine description.

  26869. 

  26870. 'cbranchMODE4'

  26871.      Conditional branch instruction combined with a compare instruction.

  26872.      Operand 0 is a comparison operator.  Operand 1 and operand 2 are

  26873.      the first and second operands of the comparison, respectively.

  26874.      Operand 3 is the 'code_label' to jump to.

  26875. 

  26876. 'jump'

  26877.      A jump inside a function; an unconditional branch.  Operand 0 is

  26878.      the 'code_label' to jump to.  This pattern name is mandatory on all

  26879.      machines.

  26880. 

  26881. 'call'

  26882.      Subroutine call instruction returning no value.  Operand 0 is the

  26883.      function to call; operand 1 is the number of bytes of arguments

  26884.      pushed as a 'const_int'; operand 2 is the number of registers used

  26885.      as operands.

  26886. 

  26887.      On most machines, operand 2 is not actually stored into the RTL

  26888.      pattern.  It is supplied for the sake of some RISC machines which

  26889.      need to put this information into the assembler code; they can put

  26890.      it in the RTL instead of operand 1.

  26891. 

  26892.      Operand 0 should be a 'mem' RTX whose address is the address of the

  26893.      function.  Note, however, that this address can be a 'symbol_ref'

  26894.      expression even if it would not be a legitimate memory address on

  26895.      the target machine.  If it is also not a valid argument for a call

  26896.      instruction, the pattern for this operation should be a

  26897.      'define_expand' (*note Expander Definitions::) that places the

  26898.      address into a register and uses that register in the call

  26899.      instruction.

  26900. 

  26901. 'call_value'

  26902.      Subroutine call instruction returning a value.  Operand 0 is the

  26903.      hard register in which the value is returned.  There are three more

  26904.      operands, the same as the three operands of the 'call' instruction

  26905.      (but with numbers increased by one).

  26906. 

  26907.      Subroutines that return 'BLKmode' objects use the 'call' insn.

  26908. 

  26909. 'call_pop', 'call_value_pop'

  26910.      Similar to 'call' and 'call_value', except used if defined and if

  26911.      'RETURN_POPS_ARGS' is nonzero.  They should emit a 'parallel' that

  26912.      contains both the function call and a 'set' to indicate the

  26913.      adjustment made to the frame pointer.

  26914. 

  26915.      For machines where 'RETURN_POPS_ARGS' can be nonzero, the use of

  26916.      these patterns increases the number of functions for which the

  26917.      frame pointer can be eliminated, if desired.

  26918. 

  26919. 'untyped_call'

  26920.      Subroutine call instruction returning a value of any type.  Operand

  26921.      0 is the function to call; operand 1 is a memory location where the

  26922.      result of calling the function is to be stored; operand 2 is a

  26923.      'parallel' expression where each element is a 'set' expression that

  26924.      indicates the saving of a function return value into the result

  26925.      block.

  26926. 

  26927.      This instruction pattern should be defined to support

  26928.      '__builtin_apply' on machines where special instructions are needed

  26929.      to call a subroutine with arbitrary arguments or to save the value

  26930.      returned.  This instruction pattern is required on machines that

  26931.      have multiple registers that can hold a return value (i.e.

  26932.      'FUNCTION_VALUE_REGNO_P' is true for more than one register).

  26933. 

  26934. 'return'

  26935.      Subroutine return instruction.  This instruction pattern name

  26936.      should be defined only if a single instruction can do all the work

  26937.      of returning from a function.

  26938. 

  26939.      Like the 'movM' patterns, this pattern is also used after the RTL

  26940.      generation phase.  In this case it is to support machines where

  26941.      multiple instructions are usually needed to return from a function,

  26942.      but some class of functions only requires one instruction to

  26943.      implement a return.  Normally, the applicable functions are those

  26944.      which do not need to save any registers or allocate stack space.

  26945. 

  26946.      It is valid for this pattern to expand to an instruction using

  26947.      'simple_return' if no epilogue is required.

  26948. 

  26949. 'simple_return'

  26950.      Subroutine return instruction.  This instruction pattern name

  26951.      should be defined only if a single instruction can do all the work

  26952.      of returning from a function on a path where no epilogue is

  26953.      required.  This pattern is very similar to the 'return' instruction

  26954.      pattern, but it is emitted only by the shrink-wrapping optimization

  26955.      on paths where the function prologue has not been executed, and a

  26956.      function return should occur without any of the effects of the

  26957.      epilogue.  Additional uses may be introduced on paths where both

  26958.      the prologue and the epilogue have executed.

  26959. 

  26960.      For such machines, the condition specified in this pattern should

  26961.      only be true when 'reload_completed' is nonzero and the function's

  26962.      epilogue would only be a single instruction.  For machines with

  26963.      register windows, the routine 'leaf_function_p' may be used to

  26964.      determine if a register window push is required.

  26965. 

  26966.      Machines that have conditional return instructions should define

  26967.      patterns such as

  26968. 

  26969.           (define_insn ""

  26970.             [(set (pc)

  26971.                   (if_then_else (match_operator

  26972.                                    0 "comparison_operator"

  26973.                                    [(cc0) (const_int 0)])

  26974.                                 (return)

  26975.                                 (pc)))]

  26976.             "CONDITION"

  26977.             "...")

  26978. 

  26979.      where CONDITION would normally be the same condition specified on

  26980.      the named 'return' pattern.

  26981. 

  26982. 'untyped_return'

  26983.      Untyped subroutine return instruction.  This instruction pattern

  26984.      should be defined to support '__builtin_return' on machines where

  26985.      special instructions are needed to return a value of any type.

  26986. 

  26987.      Operand 0 is a memory location where the result of calling a

  26988.      function with '__builtin_apply' is stored; operand 1 is a

  26989.      'parallel' expression where each element is a 'set' expression that

  26990.      indicates the restoring of a function return value from the result

  26991.      block.

  26992. 

  26993. 'nop'

  26994.      No-op instruction.  This instruction pattern name should always be

  26995.      defined to output a no-op in assembler code.  '(const_int 0)' will

  26996.      do as an RTL pattern.

  26997. 

  26998. 'indirect_jump'

  26999.      An instruction to jump to an address which is operand zero.  This

  27000.      pattern name is mandatory on all machines.

  27001. 

  27002. 'casesi'

  27003.      Instruction to jump through a dispatch table, including bounds

  27004.      checking.  This instruction takes five operands:

  27005. 

  27006.        1. The index to dispatch on, which has mode 'SImode'.

  27007. 

  27008.        2. The lower bound for indices in the table, an integer constant.

  27009. 

  27010.        3. The total range of indices in the table--the largest index

  27011.           minus the smallest one (both inclusive).

  27012. 

  27013.        4. A label that precedes the table itself.

  27014. 

  27015.        5. A label to jump to if the index has a value outside the

  27016.           bounds.

  27017. 

  27018.      The table is an 'addr_vec' or 'addr_diff_vec' inside of a

  27019.      'jump_table_data'.  The number of elements in the table is one plus

  27020.      the difference between the upper bound and the lower bound.

  27021. 

  27022. 'tablejump'

  27023.      Instruction to jump to a variable address.  This is a low-level

  27024.      capability which can be used to implement a dispatch table when

  27025.      there is no 'casesi' pattern.

  27026. 

  27027.      This pattern requires two operands: the address or offset, and a

  27028.      label which should immediately precede the jump table.  If the

  27029.      macro 'CASE_VECTOR_PC_RELATIVE' evaluates to a nonzero value then

  27030.      the first operand is an offset which counts from the address of the

  27031.      table; otherwise, it is an absolute address to jump to.  In either

  27032.      case, the first operand has mode 'Pmode'.

  27033. 

  27034.      The 'tablejump' insn is always the last insn before the jump table

  27035.      it uses.  Its assembler code normally has no need to use the second

  27036.      operand, but you should incorporate it in the RTL pattern so that

  27037.      the jump optimizer will not delete the table as unreachable code.

  27038. 

  27039. 'doloop_end'

  27040.      Conditional branch instruction that decrements a register and jumps

  27041.      if the register is nonzero.  Operand 0 is the register to decrement

  27042.      and test; operand 1 is the label to jump to if the register is

  27043.      nonzero.  *Note Looping Patterns::.

  27044. 

  27045.      This optional instruction pattern should be defined for machines

  27046.      with low-overhead looping instructions as the loop optimizer will

  27047.      try to modify suitable loops to utilize it.  The target hook

  27048.      'TARGET_CAN_USE_DOLOOP_P' controls the conditions under which

  27049.      low-overhead loops can be used.

  27050. 

  27051. 'doloop_begin'

  27052.      Companion instruction to 'doloop_end' required for machines that

  27053.      need to perform some initialization, such as loading a special

  27054.      counter register.  Operand 1 is the associated 'doloop_end' pattern

  27055.      and operand 0 is the register that it decrements.

  27056. 

  27057.      If initialization insns do not always need to be emitted, use a

  27058.      'define_expand' (*note Expander Definitions::) and make it fail.

  27059. 

  27060. 'canonicalize_funcptr_for_compare'

  27061.      Canonicalize the function pointer in operand 1 and store the result

  27062.      into operand 0.

  27063. 

  27064.      Operand 0 is always a 'reg' and has mode 'Pmode'; operand 1 may be

  27065.      a 'reg', 'mem', 'symbol_ref', 'const_int', etc and also has mode

  27066.      'Pmode'.

  27067. 

  27068.      Canonicalization of a function pointer usually involves computing

  27069.      the address of the function which would be called if the function

  27070.      pointer were used in an indirect call.

  27071. 

  27072.      Only define this pattern if function pointers on the target machine

  27073.      can have different values but still call the same function when

  27074.      used in an indirect call.

  27075. 

  27076. 'save_stack_block'

  27077. 'save_stack_function'

  27078. 'save_stack_nonlocal'

  27079. 'restore_stack_block'

  27080. 'restore_stack_function'

  27081. 'restore_stack_nonlocal'

  27082.      Most machines save and restore the stack pointer by copying it to

  27083.      or from an object of mode 'Pmode'.  Do not define these patterns on

  27084.      such machines.

  27085. 

  27086.      Some machines require special handling for stack pointer saves and

  27087.      restores.  On those machines, define the patterns corresponding to

  27088.      the non-standard cases by using a 'define_expand' (*note Expander

  27089.      Definitions::) that produces the required insns.  The three types

  27090.      of saves and restores are:

  27091. 

  27092.        1. 'save_stack_block' saves the stack pointer at the start of a

  27093.           block that allocates a variable-sized object, and

  27094.           'restore_stack_block' restores the stack pointer when the

  27095.           block is exited.

  27096. 

  27097.        2. 'save_stack_function' and 'restore_stack_function' do a

  27098.           similar job for the outermost block of a function and are used

  27099.           when the function allocates variable-sized objects or calls

  27100.           'alloca'.  Only the epilogue uses the restored stack pointer,

  27101.           allowing a simpler save or restore sequence on some machines.

  27102. 

  27103.        3. 'save_stack_nonlocal' is used in functions that contain labels

  27104.           branched to by nested functions.  It saves the stack pointer

  27105.           in such a way that the inner function can use

  27106.           'restore_stack_nonlocal' to restore the stack pointer.  The

  27107.           compiler generates code to restore the frame and argument

  27108.           pointer registers, but some machines require saving and

  27109.           restoring additional data such as register window information

  27110.           or stack backchains.  Place insns in these patterns to save

  27111.           and restore any such required data.

  27112. 

  27113.      When saving the stack pointer, operand 0 is the save area and

  27114.      operand 1 is the stack pointer.  The mode used to allocate the save

  27115.      area defaults to 'Pmode' but you can override that choice by

  27116.      defining the 'STACK_SAVEAREA_MODE' macro (*note Storage Layout::).

  27117.      You must specify an integral mode, or 'VOIDmode' if no save area is

  27118.      needed for a particular type of save (either because no save is

  27119.      needed or because a machine-specific save area can be used).

  27120.      Operand 0 is the stack pointer and operand 1 is the save area for

  27121.      restore operations.  If 'save_stack_block' is defined, operand 0

  27122.      must not be 'VOIDmode' since these saves can be arbitrarily nested.

  27123. 

  27124.      A save area is a 'mem' that is at a constant offset from

  27125.      'virtual_stack_vars_rtx' when the stack pointer is saved for use by

  27126.      nonlocal gotos and a 'reg' in the other two cases.

  27127. 

  27128. 'allocate_stack'

  27129.      Subtract (or add if 'STACK_GROWS_DOWNWARD' is undefined) operand 1

  27130.      from the stack pointer to create space for dynamically allocated

  27131.      data.

  27132. 

  27133.      Store the resultant pointer to this space into operand 0.  If you

  27134.      are allocating space from the main stack, do this by emitting a

  27135.      move insn to copy 'virtual_stack_dynamic_rtx' to operand 0.  If you

  27136.      are allocating the space elsewhere, generate code to copy the

  27137.      location of the space to operand 0.  In the latter case, you must

  27138.      ensure this space gets freed when the corresponding space on the

  27139.      main stack is free.

  27140. 

  27141.      Do not define this pattern if all that must be done is the

  27142.      subtraction.  Some machines require other operations such as stack

  27143.      probes or maintaining the back chain.  Define this pattern to emit

  27144.      those operations in addition to updating the stack pointer.

  27145. 

  27146. 'check_stack'

  27147.      If stack checking (*note Stack Checking::) cannot be done on your

  27148.      system by probing the stack, define this pattern to perform the

  27149.      needed check and signal an error if the stack has overflowed.  The

  27150.      single operand is the address in the stack farthest from the

  27151.      current stack pointer that you need to validate.  Normally, on

  27152.      platforms where this pattern is needed, you would obtain the stack

  27153.      limit from a global or thread-specific variable or register.

  27154. 

  27155. 'probe_stack_address'

  27156.      If stack checking (*note Stack Checking::) can be done on your

  27157.      system by probing the stack but without the need to actually access

  27158.      it, define this pattern and signal an error if the stack has

  27159.      overflowed.  The single operand is the memory address in the stack

  27160.      that needs to be probed.

  27161. 

  27162. 'probe_stack'

  27163.      If stack checking (*note Stack Checking::) can be done on your

  27164.      system by probing the stack but doing it with a "store zero"

  27165.      instruction is not valid or optimal, define this pattern to do the

  27166.      probing differently and signal an error if the stack has

  27167.      overflowed.  The single operand is the memory reference in the

  27168.      stack that needs to be probed.

  27169. 

  27170. 'nonlocal_goto'

  27171.      Emit code to generate a non-local goto, e.g., a jump from one

  27172.      function to a label in an outer function.  This pattern has four

  27173.      arguments, each representing a value to be used in the jump.  The

  27174.      first argument is to be loaded into the frame pointer, the second

  27175.      is the address to branch to (code to dispatch to the actual label),

  27176.      the third is the address of a location where the stack is saved,

  27177.      and the last is the address of the label, to be placed in the

  27178.      location for the incoming static chain.

  27179. 

  27180.      On most machines you need not define this pattern, since GCC will

  27181.      already generate the correct code, which is to load the frame

  27182.      pointer and static chain, restore the stack (using the

  27183.      'restore_stack_nonlocal' pattern, if defined), and jump indirectly

  27184.      to the dispatcher.  You need only define this pattern if this code

  27185.      will not work on your machine.

  27186. 

  27187. 'nonlocal_goto_receiver'

  27188.      This pattern, if defined, contains code needed at the target of a

  27189.      nonlocal goto after the code already generated by GCC.  You will

  27190.      not normally need to define this pattern.  A typical reason why you

  27191.      might need this pattern is if some value, such as a pointer to a

  27192.      global table, must be restored when the frame pointer is restored.

  27193.      Note that a nonlocal goto only occurs within a unit-of-translation,

  27194.      so a global table pointer that is shared by all functions of a

  27195.      given module need not be restored.  There are no arguments.

  27196. 

  27197. 'exception_receiver'

  27198.      This pattern, if defined, contains code needed at the site of an

  27199.      exception handler that isn't needed at the site of a nonlocal goto.

  27200.      You will not normally need to define this pattern.  A typical

  27201.      reason why you might need this pattern is if some value, such as a

  27202.      pointer to a global table, must be restored after control flow is

  27203.      branched to the handler of an exception.  There are no arguments.

  27204. 

  27205. 'builtin_setjmp_setup'

  27206.      This pattern, if defined, contains additional code needed to

  27207.      initialize the 'jmp_buf'.  You will not normally need to define

  27208.      this pattern.  A typical reason why you might need this pattern is

  27209.      if some value, such as a pointer to a global table, must be

  27210.      restored.  Though it is preferred that the pointer value be

  27211.      recalculated if possible (given the address of a label for

  27212.      instance).  The single argument is a pointer to the 'jmp_buf'.

  27213.      Note that the buffer is five words long and that the first three

  27214.      are normally used by the generic mechanism.

  27215. 

  27216. 'builtin_setjmp_receiver'

  27217.      This pattern, if defined, contains code needed at the site of a

  27218.      built-in setjmp that isn't needed at the site of a nonlocal goto.

  27219.      You will not normally need to define this pattern.  A typical

  27220.      reason why you might need this pattern is if some value, such as a

  27221.      pointer to a global table, must be restored.  It takes one

  27222.      argument, which is the label to which builtin_longjmp transferred

  27223.      control; this pattern may be emitted at a small offset from that

  27224.      label.

  27225. 

  27226. 'builtin_longjmp'

  27227.      This pattern, if defined, performs the entire action of the

  27228.      longjmp.  You will not normally need to define this pattern unless

  27229.      you also define 'builtin_setjmp_setup'.  The single argument is a

  27230.      pointer to the 'jmp_buf'.

  27231. 

  27232. 'eh_return'

  27233.      This pattern, if defined, affects the way '__builtin_eh_return',

  27234.      and thence the call frame exception handling library routines, are

  27235.      built.  It is intended to handle non-trivial actions needed along

  27236.      the abnormal return path.

  27237. 

  27238.      The address of the exception handler to which the function should

  27239.      return is passed as operand to this pattern.  It will normally need

  27240.      to copied by the pattern to some special register or memory

  27241.      location.  If the pattern needs to determine the location of the

  27242.      target call frame in order to do so, it may use

  27243.      'EH_RETURN_STACKADJ_RTX', if defined; it will have already been

  27244.      assigned.

  27245. 

  27246.      If this pattern is not defined, the default action will be to

  27247.      simply copy the return address to 'EH_RETURN_HANDLER_RTX'.  Either

  27248.      that macro or this pattern needs to be defined if call frame

  27249.      exception handling is to be used.

  27250. 

  27251. 'prologue'

  27252.      This pattern, if defined, emits RTL for entry to a function.  The

  27253.      function entry is responsible for setting up the stack frame,

  27254.      initializing the frame pointer register, saving callee saved

  27255.      registers, etc.

  27256. 

  27257.      Using a prologue pattern is generally preferred over defining

  27258.      'TARGET_ASM_FUNCTION_PROLOGUE' to emit assembly code for the

  27259.      prologue.

  27260. 

  27261.      The 'prologue' pattern is particularly useful for targets which

  27262.      perform instruction scheduling.

  27263. 

  27264. 'window_save'

  27265.      This pattern, if defined, emits RTL for a register window save.  It

  27266.      should be defined if the target machine has register windows but

  27267.      the window events are decoupled from calls to subroutines.  The

  27268.      canonical example is the SPARC architecture.

  27269. 

  27270. 'epilogue'

  27271.      This pattern emits RTL for exit from a function.  The function exit

  27272.      is responsible for deallocating the stack frame, restoring callee

  27273.      saved registers and emitting the return instruction.

  27274. 

  27275.      Using an epilogue pattern is generally preferred over defining

  27276.      'TARGET_ASM_FUNCTION_EPILOGUE' to emit assembly code for the

  27277.      epilogue.

  27278. 

  27279.      The 'epilogue' pattern is particularly useful for targets which

  27280.      perform instruction scheduling or which have delay slots for their

  27281.      return instruction.

  27282. 

  27283. 'sibcall_epilogue'

  27284.      This pattern, if defined, emits RTL for exit from a function

  27285.      without the final branch back to the calling function.  This

  27286.      pattern will be emitted before any sibling call (aka tail call)

  27287.      sites.

  27288. 

  27289.      The 'sibcall_epilogue' pattern must not clobber any arguments used

  27290.      for parameter passing or any stack slots for arguments passed to

  27291.      the current function.

  27292. 

  27293. 'trap'

  27294.      This pattern, if defined, signals an error, typically by causing

  27295.      some kind of signal to be raised.

  27296. 

  27297. 'ctrapMM4'

  27298.      Conditional trap instruction.  Operand 0 is a piece of RTL which

  27299.      performs a comparison, and operands 1 and 2 are the arms of the

  27300.      comparison.  Operand 3 is the trap code, an integer.

  27301. 

  27302.      A typical 'ctrap' pattern looks like

  27303. 

  27304.           (define_insn "ctrapsi4"

  27305.             [(trap_if (match_operator 0 "trap_operator"

  27306.                        [(match_operand 1 "register_operand")

  27307.                         (match_operand 2 "immediate_operand")])

  27308.                       (match_operand 3 "const_int_operand" "i"))]

  27309.             ""

  27310.             "...")

  27311. 

  27312. 'prefetch'

  27313.      This pattern, if defined, emits code for a non-faulting data

  27314.      prefetch instruction.  Operand 0 is the address of the memory to

  27315.      prefetch.  Operand 1 is a constant 1 if the prefetch is preparing

  27316.      for a write to the memory address, or a constant 0 otherwise.

  27317.      Operand 2 is the expected degree of temporal locality of the data

  27318.      and is a value between 0 and 3, inclusive; 0 means that the data

  27319.      has no temporal locality, so it need not be left in the cache after

  27320.      the access; 3 means that the data has a high degree of temporal

  27321.      locality and should be left in all levels of cache possible; 1 and

  27322.      2 mean, respectively, a low or moderate degree of temporal

  27323.      locality.

  27324. 

  27325.      Targets that do not support write prefetches or locality hints can

  27326.      ignore the values of operands 1 and 2.

  27327. 

  27328. 'blockage'

  27329.      This pattern defines a pseudo insn that prevents the instruction

  27330.      scheduler and other passes from moving instructions and using

  27331.      register equivalences across the boundary defined by the blockage

  27332.      insn.  This needs to be an UNSPEC_VOLATILE pattern or a volatile

  27333.      ASM.

  27334. 

  27335. 'memory_blockage'

  27336.      This pattern, if defined, represents a compiler memory barrier, and

  27337.      will be placed at points across which RTL passes may not propagate

  27338.      memory accesses.  This instruction needs to read and write volatile

  27339.      BLKmode memory.  It does not need to generate any machine

  27340.      instruction.  If this pattern is not defined, the compiler falls

  27341.      back to emitting an instruction corresponding to 'asm volatile (""

  27342.      ::: "memory")'.

  27343. 

  27344. 'memory_barrier'

  27345.      If the target memory model is not fully synchronous, then this

  27346.      pattern should be defined to an instruction that orders both loads

  27347.      and stores before the instruction with respect to loads and stores

  27348.      after the instruction.  This pattern has no operands.

  27349. 

  27350. 'speculation_barrier'

  27351.      If the target can support speculative execution, then this pattern

  27352.      should be defined to an instruction that will block subsequent

  27353.      execution until any prior speculation conditions has been resolved.

  27354.      The pattern must also ensure that the compiler cannot move memory

  27355.      operations past the barrier, so it needs to be an UNSPEC_VOLATILE

  27356.      pattern.  The pattern has no operands.

  27357. 

  27358.      If this pattern is not defined then the default expansion of

  27359.      '__builtin_speculation_safe_value' will emit a warning.  You can

  27360.      suppress this warning by defining this pattern with a final

  27361.      condition of '0' (zero), which tells the compiler that a

  27362.      speculation barrier is not needed for this target.

  27363. 

  27364. 'sync_compare_and_swapMODE'

  27365.      This pattern, if defined, emits code for an atomic compare-and-swap

  27366.      operation.  Operand 1 is the memory on which the atomic operation

  27367.      is performed.  Operand 2 is the "old" value to be compared against

  27368.      the current contents of the memory location.  Operand 3 is the

  27369.      "new" value to store in the memory if the compare succeeds.

  27370.      Operand 0 is the result of the operation; it should contain the

  27371.      contents of the memory before the operation.  If the compare

  27372.      succeeds, this should obviously be a copy of operand 2.

  27373. 

  27374.      This pattern must show that both operand 0 and operand 1 are

  27375.      modified.

  27376. 

  27377.      This pattern must issue any memory barrier instructions such that

  27378.      all memory operations before the atomic operation occur before the

  27379.      atomic operation and all memory operations after the atomic

  27380.      operation occur after the atomic operation.

  27381. 

  27382.      For targets where the success or failure of the compare-and-swap

  27383.      operation is available via the status flags, it is possible to

  27384.      avoid a separate compare operation and issue the subsequent branch

  27385.      or store-flag operation immediately after the compare-and-swap.  To

  27386.      this end, GCC will look for a 'MODE_CC' set in the output of

  27387.      'sync_compare_and_swapMODE'; if the machine description includes

  27388.      such a set, the target should also define special 'cbranchcc4'

  27389.      and/or 'cstorecc4' instructions.  GCC will then be able to take the

  27390.      destination of the 'MODE_CC' set and pass it to the 'cbranchcc4' or

  27391.      'cstorecc4' pattern as the first operand of the comparison (the

  27392.      second will be '(const_int 0)').

  27393. 

  27394.      For targets where the operating system may provide support for this

  27395.      operation via library calls, the 'sync_compare_and_swap_optab' may

  27396.      be initialized to a function with the same interface as the

  27397.      '__sync_val_compare_and_swap_N' built-in.  If the entire set of

  27398.      __SYNC builtins are supported via library calls, the target can

  27399.      initialize all of the optabs at once with 'init_sync_libfuncs'.

  27400.      For the purposes of C++11 'std::atomic::is_lock_free', it is

  27401.      assumed that these library calls do _not_ use any kind of

  27402.      interruptable locking.

  27403. 

  27404. 'sync_addMODE', 'sync_subMODE'

  27405. 'sync_iorMODE', 'sync_andMODE'

  27406. 'sync_xorMODE', 'sync_nandMODE'

  27407.      These patterns emit code for an atomic operation on memory.

  27408.      Operand 0 is the memory on which the atomic operation is performed.

  27409.      Operand 1 is the second operand to the binary operator.

  27410. 

  27411.      This pattern must issue any memory barrier instructions such that

  27412.      all memory operations before the atomic operation occur before the

  27413.      atomic operation and all memory operations after the atomic

  27414.      operation occur after the atomic operation.

  27415. 

  27416.      If these patterns are not defined, the operation will be

  27417.      constructed from a compare-and-swap operation, if defined.

  27418. 

  27419. 'sync_old_addMODE', 'sync_old_subMODE'

  27420. 'sync_old_iorMODE', 'sync_old_andMODE'

  27421. 'sync_old_xorMODE', 'sync_old_nandMODE'

  27422.      These patterns emit code for an atomic operation on memory, and

  27423.      return the value that the memory contained before the operation.

  27424.      Operand 0 is the result value, operand 1 is the memory on which the

  27425.      atomic operation is performed, and operand 2 is the second operand

  27426.      to the binary operator.

  27427. 

  27428.      This pattern must issue any memory barrier instructions such that

  27429.      all memory operations before the atomic operation occur before the

  27430.      atomic operation and all memory operations after the atomic

  27431.      operation occur after the atomic operation.

  27432. 

  27433.      If these patterns are not defined, the operation will be

  27434.      constructed from a compare-and-swap operation, if defined.

  27435. 

  27436. 'sync_new_addMODE', 'sync_new_subMODE'

  27437. 'sync_new_iorMODE', 'sync_new_andMODE'

  27438. 'sync_new_xorMODE', 'sync_new_nandMODE'

  27439.      These patterns are like their 'sync_old_OP' counterparts, except

  27440.      that they return the value that exists in the memory location after

  27441.      the operation, rather than before the operation.

  27442. 

  27443. 'sync_lock_test_and_setMODE'

  27444.      This pattern takes two forms, based on the capabilities of the

  27445.      target.  In either case, operand 0 is the result of the operand,

  27446.      operand 1 is the memory on which the atomic operation is performed,

  27447.      and operand 2 is the value to set in the lock.

  27448. 

  27449.      In the ideal case, this operation is an atomic exchange operation,

  27450.      in which the previous value in memory operand is copied into the

  27451.      result operand, and the value operand is stored in the memory

  27452.      operand.

  27453. 

  27454.      For less capable targets, any value operand that is not the

  27455.      constant 1 should be rejected with 'FAIL'.  In this case the target

  27456.      may use an atomic test-and-set bit operation.  The result operand

  27457.      should contain 1 if the bit was previously set and 0 if the bit was

  27458.      previously clear.  The true contents of the memory operand are

  27459.      implementation defined.

  27460. 

  27461.      This pattern must issue any memory barrier instructions such that

  27462.      the pattern as a whole acts as an acquire barrier, that is all

  27463.      memory operations after the pattern do not occur until the lock is

  27464.      acquired.

  27465. 

  27466.      If this pattern is not defined, the operation will be constructed

  27467.      from a compare-and-swap operation, if defined.

  27468. 

  27469. 'sync_lock_releaseMODE'

  27470.      This pattern, if defined, releases a lock set by

  27471.      'sync_lock_test_and_setMODE'.  Operand 0 is the memory that

  27472.      contains the lock; operand 1 is the value to store in the lock.

  27473. 

  27474.      If the target doesn't implement full semantics for

  27475.      'sync_lock_test_and_setMODE', any value operand which is not the

  27476.      constant 0 should be rejected with 'FAIL', and the true contents of

  27477.      the memory operand are implementation defined.

  27478. 

  27479.      This pattern must issue any memory barrier instructions such that

  27480.      the pattern as a whole acts as a release barrier, that is the lock

  27481.      is released only after all previous memory operations have

  27482.      completed.

  27483. 

  27484.      If this pattern is not defined, then a 'memory_barrier' pattern

  27485.      will be emitted, followed by a store of the value to the memory

  27486.      operand.

  27487. 

  27488. 'atomic_compare_and_swapMODE'

  27489.      This pattern, if defined, emits code for an atomic compare-and-swap

  27490.      operation with memory model semantics.  Operand 2 is the memory on

  27491.      which the atomic operation is performed.  Operand 0 is an output

  27492.      operand which is set to true or false based on whether the

  27493.      operation succeeded.  Operand 1 is an output operand which is set

  27494.      to the contents of the memory before the operation was attempted.

  27495.      Operand 3 is the value that is expected to be in memory.  Operand 4

  27496.      is the value to put in memory if the expected value is found there.

  27497.      Operand 5 is set to 1 if this compare and swap is to be treated as

  27498.      a weak operation.  Operand 6 is the memory model to be used if the

  27499.      operation is a success.  Operand 7 is the memory model to be used

  27500.      if the operation fails.

  27501. 

  27502.      If memory referred to in operand 2 contains the value in operand 3,

  27503.      then operand 4 is stored in memory pointed to by operand 2 and

  27504.      fencing based on the memory model in operand 6 is issued.

  27505. 

  27506.      If memory referred to in operand 2 does not contain the value in

  27507.      operand 3, then fencing based on the memory model in operand 7 is

  27508.      issued.

  27509. 

  27510.      If a target does not support weak compare-and-swap operations, or

  27511.      the port elects not to implement weak operations, the argument in

  27512.      operand 5 can be ignored.  Note a strong implementation must be

  27513.      provided.

  27514. 

  27515.      If this pattern is not provided, the '__atomic_compare_exchange'

  27516.      built-in functions will utilize the legacy 'sync_compare_and_swap'

  27517.      pattern with an '__ATOMIC_SEQ_CST' memory model.

  27518. 

  27519. 'atomic_loadMODE'

  27520.      This pattern implements an atomic load operation with memory model

  27521.      semantics.  Operand 1 is the memory address being loaded from.

  27522.      Operand 0 is the result of the load.  Operand 2 is the memory model

  27523.      to be used for the load operation.

  27524. 

  27525.      If not present, the '__atomic_load' built-in function will either

  27526.      resort to a normal load with memory barriers, or a compare-and-swap

  27527.      operation if a normal load would not be atomic.

  27528. 

  27529. 'atomic_storeMODE'

  27530.      This pattern implements an atomic store operation with memory model

  27531.      semantics.  Operand 0 is the memory address being stored to.

  27532.      Operand 1 is the value to be written.  Operand 2 is the memory

  27533.      model to be used for the operation.

  27534. 

  27535.      If not present, the '__atomic_store' built-in function will attempt

  27536.      to perform a normal store and surround it with any required memory

  27537.      fences.  If the store would not be atomic, then an

  27538.      '__atomic_exchange' is attempted with the result being ignored.

  27539. 

  27540. 'atomic_exchangeMODE'

  27541.      This pattern implements an atomic exchange operation with memory

  27542.      model semantics.  Operand 1 is the memory location the operation is

  27543.      performed on.  Operand 0 is an output operand which is set to the

  27544.      original value contained in the memory pointed to by operand 1.

  27545.      Operand 2 is the value to be stored.  Operand 3 is the memory model

  27546.      to be used.

  27547. 

  27548.      If this pattern is not present, the built-in function

  27549.      '__atomic_exchange' will attempt to preform the operation with a

  27550.      compare and swap loop.

  27551. 

  27552. 'atomic_addMODE', 'atomic_subMODE'

  27553. 'atomic_orMODE', 'atomic_andMODE'

  27554. 'atomic_xorMODE', 'atomic_nandMODE'

  27555.      These patterns emit code for an atomic operation on memory with

  27556.      memory model semantics.  Operand 0 is the memory on which the

  27557.      atomic operation is performed.  Operand 1 is the second operand to

  27558.      the binary operator.  Operand 2 is the memory model to be used by

  27559.      the operation.

  27560. 

  27561.      If these patterns are not defined, attempts will be made to use

  27562.      legacy 'sync' patterns, or equivalent patterns which return a

  27563.      result.  If none of these are available a compare-and-swap loop

  27564.      will be used.

  27565. 

  27566. 'atomic_fetch_addMODE', 'atomic_fetch_subMODE'

  27567. 'atomic_fetch_orMODE', 'atomic_fetch_andMODE'

  27568. 'atomic_fetch_xorMODE', 'atomic_fetch_nandMODE'

  27569.      These patterns emit code for an atomic operation on memory with

  27570.      memory model semantics, and return the original value.  Operand 0

  27571.      is an output operand which contains the value of the memory

  27572.      location before the operation was performed.  Operand 1 is the

  27573.      memory on which the atomic operation is performed.  Operand 2 is

  27574.      the second operand to the binary operator.  Operand 3 is the memory

  27575.      model to be used by the operation.

  27576. 

  27577.      If these patterns are not defined, attempts will be made to use

  27578.      legacy 'sync' patterns.  If none of these are available a

  27579.      compare-and-swap loop will be used.

  27580. 

  27581. 'atomic_add_fetchMODE', 'atomic_sub_fetchMODE'

  27582. 'atomic_or_fetchMODE', 'atomic_and_fetchMODE'

  27583. 'atomic_xor_fetchMODE', 'atomic_nand_fetchMODE'

  27584.      These patterns emit code for an atomic operation on memory with

  27585.      memory model semantics and return the result after the operation is

  27586.      performed.  Operand 0 is an output operand which contains the value

  27587.      after the operation.  Operand 1 is the memory on which the atomic

  27588.      operation is performed.  Operand 2 is the second operand to the

  27589.      binary operator.  Operand 3 is the memory model to be used by the

  27590.      operation.

  27591. 

  27592.      If these patterns are not defined, attempts will be made to use

  27593.      legacy 'sync' patterns, or equivalent patterns which return the

  27594.      result before the operation followed by the arithmetic operation

  27595.      required to produce the result.  If none of these are available a

  27596.      compare-and-swap loop will be used.

  27597. 

  27598. 'atomic_test_and_set'

  27599.      This pattern emits code for '__builtin_atomic_test_and_set'.

  27600.      Operand 0 is an output operand which is set to true if the previous

  27601.      previous contents of the byte was "set", and false otherwise.

  27602.      Operand 1 is the 'QImode' memory to be modified.  Operand 2 is the

  27603.      memory model to be used.

  27604. 

  27605.      The specific value that defines "set" is implementation defined,

  27606.      and is normally based on what is performed by the native atomic

  27607.      test and set instruction.

  27608. 

  27609. 'atomic_bit_test_and_setMODE'

  27610. 'atomic_bit_test_and_complementMODE'

  27611. 'atomic_bit_test_and_resetMODE'

  27612.      These patterns emit code for an atomic bitwise operation on memory

  27613.      with memory model semantics, and return the original value of the

  27614.      specified bit.  Operand 0 is an output operand which contains the

  27615.      value of the specified bit from the memory location before the

  27616.      operation was performed.  Operand 1 is the memory on which the

  27617.      atomic operation is performed.  Operand 2 is the bit within the

  27618.      operand, starting with least significant bit.  Operand 3 is the

  27619.      memory model to be used by the operation.  Operand 4 is a flag - it

  27620.      is 'const1_rtx' if operand 0 should contain the original value of

  27621.      the specified bit in the least significant bit of the operand, and

  27622.      'const0_rtx' if the bit should be in its original position in the

  27623.      operand.  'atomic_bit_test_and_setMODE' atomically sets the

  27624.      specified bit after remembering its original value,

  27625.      'atomic_bit_test_and_complementMODE' inverts the specified bit and

  27626.      'atomic_bit_test_and_resetMODE' clears the specified bit.

  27627. 

  27628.      If these patterns are not defined, attempts will be made to use

  27629.      'atomic_fetch_orMODE', 'atomic_fetch_xorMODE' or

  27630.      'atomic_fetch_andMODE' instruction patterns, or their 'sync'

  27631.      counterparts.  If none of these are available a compare-and-swap

  27632.      loop will be used.

  27633. 

  27634. 'mem_thread_fence'

  27635.      This pattern emits code required to implement a thread fence with

  27636.      memory model semantics.  Operand 0 is the memory model to be used.

  27637. 

  27638.      For the '__ATOMIC_RELAXED' model no instructions need to be issued

  27639.      and this expansion is not invoked.

  27640. 

  27641.      The compiler always emits a compiler memory barrier regardless of

  27642.      what expanding this pattern produced.

  27643. 

  27644.      If this pattern is not defined, the compiler falls back to

  27645.      expanding the 'memory_barrier' pattern, then to emitting

  27646.      '__sync_synchronize' library call, and finally to just placing a

  27647.      compiler memory barrier.

  27648. 

  27649. 'get_thread_pointerMODE'

  27650. 'set_thread_pointerMODE'

  27651.      These patterns emit code that reads/sets the TLS thread pointer.

  27652.      Currently, these are only needed if the target needs to support the

  27653.      '__builtin_thread_pointer' and '__builtin_set_thread_pointer'

  27654.      builtins.

  27655. 

  27656.      The get/set patterns have a single output/input operand

  27657.      respectively, with MODE intended to be 'Pmode'.

  27658. 

  27659. 'stack_protect_combined_set'

  27660.      This pattern, if defined, moves a 'ptr_mode' value from an address

  27661.      whose declaration RTX is given in operand 1 to the memory in

  27662.      operand 0 without leaving the value in a register afterward.  If

  27663.      several instructions are needed by the target to perform the

  27664.      operation (eg.  to load the address from a GOT entry then load the

  27665.      'ptr_mode' value and finally store it), it is the backend's

  27666.      responsibility to ensure no intermediate result gets spilled.  This

  27667.      is to avoid leaking the value some place that an attacker might use

  27668.      to rewrite the stack guard slot after having clobbered it.

  27669. 

  27670.      If this pattern is not defined, then the address declaration is

  27671.      expanded first in the standard way and a 'stack_protect_set'

  27672.      pattern is then generated to move the value from that address to

  27673.      the address in operand 0.

  27674. 

  27675. 'stack_protect_set'

  27676.      This pattern, if defined, moves a 'ptr_mode' value from the valid

  27677.      memory location in operand 1 to the memory in operand 0 without

  27678.      leaving the value in a register afterward.  This is to avoid

  27679.      leaking the value some place that an attacker might use to rewrite

  27680.      the stack guard slot after having clobbered it.

  27681. 

  27682.      Note: on targets where the addressing modes do not allow to load

  27683.      directly from stack guard address, the address is expanded in a

  27684.      standard way first which could cause some spills.

  27685. 

  27686.      If this pattern is not defined, then a plain move pattern is

  27687.      generated.

  27688. 

  27689. 'stack_protect_combined_test'

  27690.      This pattern, if defined, compares a 'ptr_mode' value from an

  27691.      address whose declaration RTX is given in operand 1 with the memory

  27692.      in operand 0 without leaving the value in a register afterward and

  27693.      branches to operand 2 if the values were equal.  If several

  27694.      instructions are needed by the target to perform the operation (eg.

  27695.      to load the address from a GOT entry then load the 'ptr_mode' value

  27696.      and finally store it), it is the backend's responsibility to ensure

  27697.      no intermediate result gets spilled.  This is to avoid leaking the

  27698.      value some place that an attacker might use to rewrite the stack

  27699.      guard slot after having clobbered it.

  27700. 

  27701.      If this pattern is not defined, then the address declaration is

  27702.      expanded first in the standard way and a 'stack_protect_test'

  27703.      pattern is then generated to compare the value from that address to

  27704.      the value at the memory in operand 0.

  27705. 

  27706. 'stack_protect_test'

  27707.      This pattern, if defined, compares a 'ptr_mode' value from the

  27708.      valid memory location in operand 1 with the memory in operand 0

  27709.      without leaving the value in a register afterward and branches to

  27710.      operand 2 if the values were equal.

  27711. 

  27712.      If this pattern is not defined, then a plain compare pattern and

  27713.      conditional branch pattern is used.

  27714. 

  27715. 'clear_cache'

  27716.      This pattern, if defined, flushes the instruction cache for a

  27717.      region of memory.  The region is bounded to by the Pmode pointers

  27718.      in operand 0 inclusive and operand 1 exclusive.

  27719. 

  27720.      If this pattern is not defined, a call to the library function

  27721.      '__clear_cache' is used.

  27722. 

  27723. 

  27724. File: gccint.info,  Node: Pattern Ordering,  Next: Dependent Patterns,  Prev: Standard Names,  Up: Machine Desc

  27725. 

  27726. 17.10 When the Order of Patterns Matters

  27727. ========================================

  27728. 

  27729. Sometimes an insn can match more than one instruction pattern.  Then the

  27730. pattern that appears first in the machine description is the one used.

  27731. Therefore, more specific patterns (patterns that will match fewer

  27732. things) and faster instructions (those that will produce better code

  27733. when they do match) should usually go first in the description.

  27734. 

  27735.  In some cases the effect of ordering the patterns can be used to hide a

  27736. pattern when it is not valid.  For example, the 68000 has an instruction

  27737. for converting a fullword to floating point and another for converting a

  27738. byte to floating point.  An instruction converting an integer to

  27739. floating point could match either one.  We put the pattern to convert

  27740. the fullword first to make sure that one will be used rather than the

  27741. other.  (Otherwise a large integer might be generated as a single-byte

  27742. immediate quantity, which would not work.)  Instead of using this

  27743. pattern ordering it would be possible to make the pattern for

  27744. convert-a-byte smart enough to deal properly with any constant value.

  27745. 

  27746. 

  27747. File: gccint.info,  Node: Dependent Patterns,  Next: Jump Patterns,  Prev: Pattern Ordering,  Up: Machine Desc

  27748. 

  27749. 17.11 Interdependence of Patterns

  27750. =================================

  27751. 

  27752. In some cases machines support instructions identical except for the

  27753. machine mode of one or more operands.  For example, there may be

  27754. "sign-extend halfword" and "sign-extend byte" instructions whose

  27755. patterns are

  27756. 

  27757.      (set (match_operand:SI 0 ...)

  27758.           (extend:SI (match_operand:HI 1 ...)))

  27759. 

  27760.      (set (match_operand:SI 0 ...)

  27761.           (extend:SI (match_operand:QI 1 ...)))

  27762. 

  27763. Constant integers do not specify a machine mode, so an instruction to

  27764. extend a constant value could match either pattern.  The pattern it

  27765. actually will match is the one that appears first in the file.  For

  27766. correct results, this must be the one for the widest possible mode

  27767. ('HImode', here).  If the pattern matches the 'QImode' instruction, the

  27768. results will be incorrect if the constant value does not actually fit

  27769. that mode.

  27770. 

  27771.  Such instructions to extend constants are rarely generated because they

  27772. are optimized away, but they do occasionally happen in nonoptimized

  27773. compilations.

  27774. 

  27775.  If a constraint in a pattern allows a constant, the reload pass may

  27776. replace a register with a constant permitted by the constraint in some

  27777. cases.  Similarly for memory references.  Because of this substitution,

  27778. you should not provide separate patterns for increment and decrement

  27779. instructions.  Instead, they should be generated from the same pattern

  27780. that supports register-register add insns by examining the operands and

  27781. generating the appropriate machine instruction.

  27782. 

  27783. 

  27784. File: gccint.info,  Node: Jump Patterns,  Next: Looping Patterns,  Prev: Dependent Patterns,  Up: Machine Desc

  27785. 

  27786. 17.12 Defining Jump Instruction Patterns

  27787. ========================================

  27788. 

  27789. GCC does not assume anything about how the machine realizes jumps.  The

  27790. machine description should define a single pattern, usually a

  27791. 'define_expand', which expands to all the required insns.

  27792. 

  27793.  Usually, this would be a comparison insn to set the condition code and

  27794. a separate branch insn testing the condition code and branching or not

  27795. according to its value.  For many machines, however, separating compares

  27796. and branches is limiting, which is why the more flexible approach with

  27797. one 'define_expand' is used in GCC. The machine description becomes

  27798. clearer for architectures that have compare-and-branch instructions but

  27799. no condition code.  It also works better when different sets of

  27800. comparison operators are supported by different kinds of conditional

  27801. branches (e.g. integer vs. floating-point), or by conditional branches

  27802. with respect to conditional stores.

  27803. 

  27804.  Two separate insns are always used if the machine description

  27805. represents a condition code register using the legacy RTL expression

  27806. '(cc0)', and on most machines that use a separate condition code

  27807. register (*note Condition Code::).  For machines that use '(cc0)', in

  27808. fact, the set and use of the condition code must be separate and

  27809. adjacent(1), thus allowing flags in 'cc_status' to be used (*note

  27810. Condition Code::) and so that the comparison and branch insns could be

  27811. located from each other by using the functions 'prev_cc0_setter' and

  27812. 'next_cc0_user'.

  27813. 

  27814.  Even in this case having a single entry point for conditional branches

  27815. is advantageous, because it handles equally well the case where a single

  27816. comparison instruction records the results of both signed and unsigned

  27817. comparison of the given operands (with the branch insns coming in

  27818. distinct signed and unsigned flavors) as in the x86 or SPARC, and the

  27819. case where there are distinct signed and unsigned compare instructions

  27820. and only one set of conditional branch instructions as in the PowerPC.

  27821. 

  27822.    ---------- Footnotes ----------

  27823. 

  27824.    (1) 'note' insns can separate them, though.

  27825. 

  27826. 

  27827. File: gccint.info,  Node: Looping Patterns,  Next: Insn Canonicalizations,  Prev: Jump Patterns,  Up: Machine Desc

  27828. 

  27829. 17.13 Defining Looping Instruction Patterns

  27830. ===========================================

  27831. 

  27832. Some machines have special jump instructions that can be utilized to

  27833. make loops more efficient.  A common example is the 68000 'dbra'

  27834. instruction which performs a decrement of a register and a branch if the

  27835. result was greater than zero.  Other machines, in particular digital

  27836. signal processors (DSPs), have special block repeat instructions to

  27837. provide low-overhead loop support.  For example, the TI TMS320C3x/C4x

  27838. DSPs have a block repeat instruction that loads special registers to

  27839. mark the top and end of a loop and to count the number of loop

  27840. iterations.  This avoids the need for fetching and executing a

  27841. 'dbra'-like instruction and avoids pipeline stalls associated with the

  27842. jump.

  27843. 

  27844.  GCC has two special named patterns to support low overhead looping.

  27845. They are 'doloop_begin' and 'doloop_end'.  These are emitted by the loop

  27846. optimizer for certain well-behaved loops with a finite number of loop

  27847. iterations using information collected during strength reduction.

  27848. 

  27849.  The 'doloop_end' pattern describes the actual looping instruction (or

  27850. the implicit looping operation) and the 'doloop_begin' pattern is an

  27851. optional companion pattern that can be used for initialization needed

  27852. for some low-overhead looping instructions.

  27853. 

  27854.  Note that some machines require the actual looping instruction to be

  27855. emitted at the top of the loop (e.g., the TMS320C3x/C4x DSPs).  Emitting

  27856. the true RTL for a looping instruction at the top of the loop can cause

  27857. problems with flow analysis.  So instead, a dummy 'doloop' insn is

  27858. emitted at the end of the loop.  The machine dependent reorg pass checks

  27859. for the presence of this 'doloop' insn and then searches back to the top

  27860. of the loop, where it inserts the true looping insn (provided there are

  27861. no instructions in the loop which would cause problems).  Any additional

  27862. labels can be emitted at this point.  In addition, if the desired

  27863. special iteration counter register was not allocated, this machine

  27864. dependent reorg pass could emit a traditional compare and jump

  27865. instruction pair.

  27866. 

  27867.  For the 'doloop_end' pattern, the loop optimizer allocates an

  27868. additional pseudo register as an iteration counter.  This pseudo

  27869. register cannot be used within the loop (i.e., general induction

  27870. variables cannot be derived from it), however, in many cases the loop

  27871. induction variable may become redundant and removed by the flow pass.

  27872. 

  27873.  The 'doloop_end' pattern must have a specific structure to be handled

  27874. correctly by GCC. The example below is taken (slightly simplified) from

  27875. the PDP-11 target:

  27876. 

  27877.      (define_expand "doloop_end"

  27878.        [(parallel [(set (pc)

  27879.                         (if_then_else

  27880.                          (ne (match_operand:HI 0 "nonimmediate_operand" "+r,!m")

  27881.                              (const_int 1))

  27882.                          (label_ref (match_operand 1 "" ""))

  27883.                          (pc)))

  27884.                    (set (match_dup 0)

  27885.                         (plus:HI (match_dup 0)

  27886.                               (const_int -1)))])]

  27887.        ""

  27888.        "{

  27889.          if (GET_MODE (operands[0]) != HImode)

  27890.            FAIL;

  27891.        }")

  27892. 

  27893.      (define_insn "doloop_end_insn"

  27894.        [(set (pc)

  27895.              (if_then_else

  27896.               (ne (match_operand:HI 0 "nonimmediate_operand" "+r,!m")

  27897.                   (const_int 1))

  27898.               (label_ref (match_operand 1 "" ""))

  27899.               (pc)))

  27900.         (set (match_dup 0)

  27901.              (plus:HI (match_dup 0)

  27902.                    (const_int -1)))]

  27903.        ""

  27904. 

  27905.        {

  27906.          if (which_alternative == 0)

  27907.            return "sob %0,%l1";

  27908. 

  27909.          /* emulate sob */

  27910.          output_asm_insn ("dec %0", operands);

  27911.          return "bne %l1";

  27912.        })

  27913. 

  27914.  The first part of the pattern describes the branch condition.  GCC

  27915. supports three cases for the way the target machine handles the loop

  27916. counter:

  27917.    * Loop terminates when the loop register decrements to zero.  This is

  27918.      represented by a 'ne' comparison of the register (its old value)

  27919.      with constant 1 (as in the example above).

  27920.    * Loop terminates when the loop register decrements to -1.  This is

  27921.      represented by a 'ne' comparison of the register with constant

  27922.      zero.

  27923.    * Loop terminates when the loop register decrements to a negative

  27924.      value.  This is represented by a 'ge' comparison of the register

  27925.      with constant zero.  For this case, GCC will attach a 'REG_NONNEG'

  27926.      note to the 'doloop_end' insn if it can determine that the register

  27927.      will be non-negative.

  27928. 

  27929.  Since the 'doloop_end' insn is a jump insn that also has an output, the

  27930. reload pass does not handle the output operand.  Therefore, the

  27931. constraint must allow for that operand to be in memory rather than a

  27932. register.  In the example shown above, that is handled (in the

  27933. 'doloop_end_insn' pattern) by using a loop instruction sequence that can

  27934. handle memory operands when the memory alternative appears.

  27935. 

  27936.  GCC does not check the mode of the loop register operand when

  27937. generating the 'doloop_end' pattern.  If the pattern is only valid for

  27938. some modes but not others, the pattern should be a 'define_expand'

  27939. pattern that checks the operand mode in the preparation code, and issues

  27940. 'FAIL' if an unsupported mode is found.  The example above does this,

  27941. since the machine instruction to be used only exists for 'HImode'.

  27942. 

  27943.  If the 'doloop_end' pattern is a 'define_expand', there must also be a

  27944. 'define_insn' or 'define_insn_and_split' matching the generated pattern.

  27945. Otherwise, the compiler will fail during loop optimization.

  27946. 

  27947. 

  27948. File: gccint.info,  Node: Insn Canonicalizations,  Next: Expander Definitions,  Prev: Looping Patterns,  Up: Machine Desc

  27949. 

  27950. 17.14 Canonicalization of Instructions

  27951. ======================================

  27952. 

  27953. There are often cases where multiple RTL expressions could represent an

  27954. operation performed by a single machine instruction.  This situation is

  27955. most commonly encountered with logical, branch, and multiply-accumulate

  27956. instructions.  In such cases, the compiler attempts to convert these

  27957. multiple RTL expressions into a single canonical form to reduce the

  27958. number of insn patterns required.

  27959. 

  27960.  In addition to algebraic simplifications, following canonicalizations

  27961. are performed:

  27962. 

  27963.    * For commutative and comparison operators, a constant is always made

  27964.      the second operand.  If a machine only supports a constant as the

  27965.      second operand, only patterns that match a constant in the second

  27966.      operand need be supplied.

  27967. 

  27968.    * For associative operators, a sequence of operators will always

  27969.      chain to the left; for instance, only the left operand of an

  27970.      integer 'plus' can itself be a 'plus'.  'and', 'ior', 'xor',

  27971.      'plus', 'mult', 'smin', 'smax', 'umin', and 'umax' are associative

  27972.      when applied to integers, and sometimes to floating-point.

  27973. 

  27974.    * For these operators, if only one operand is a 'neg', 'not', 'mult',

  27975.      'plus', or 'minus' expression, it will be the first operand.

  27976. 

  27977.    * In combinations of 'neg', 'mult', 'plus', and 'minus', the 'neg'

  27978.      operations (if any) will be moved inside the operations as far as

  27979.      possible.  For instance, '(neg (mult A B))' is canonicalized as

  27980.      '(mult (neg A) B)', but '(plus (mult (neg B) C) A)' is

  27981.      canonicalized as '(minus A (mult B C))'.

  27982. 

  27983.    * For the 'compare' operator, a constant is always the second operand

  27984.      if the first argument is a condition code register or '(cc0)'.

  27985. 

  27986.    * For instructions that inherently set a condition code register, the

  27987.      'compare' operator is always written as the first RTL expression of

  27988.      the 'parallel' instruction pattern.  For example,

  27989. 

  27990.           (define_insn ""

  27991.             [(set (reg:CCZ FLAGS_REG)

  27992.           	(compare:CCZ

  27993.           	  (plus:SI

  27994.           	    (match_operand:SI 1 "register_operand" "%r")

  27995.           	    (match_operand:SI 2 "register_operand" "r"))

  27996.           	  (const_int 0)))

  27997.              (set (match_operand:SI 0 "register_operand" "=r")

  27998.           	(plus:SI (match_dup 1) (match_dup 2)))]

  27999.             ""

  28000.             "addl %0, %1, %2")

  28001. 

  28002.    * An operand of 'neg', 'not', 'mult', 'plus', or 'minus' is made the

  28003.      first operand under the same conditions as above.

  28004. 

  28005.    * '(ltu (plus A B) B)' is converted to '(ltu (plus A B) A)'.

  28006.      Likewise with 'geu' instead of 'ltu'.

  28007. 

  28008.    * '(minus X (const_int N))' is converted to '(plus X (const_int

  28009.      -N))'.

  28010. 

  28011.    * Within address computations (i.e., inside 'mem'), a left shift is

  28012.      converted into the appropriate multiplication by a power of two.

  28013. 

  28014.    * De Morgan's Law is used to move bitwise negation inside a bitwise

  28015.      logical-and or logical-or operation.  If this results in only one

  28016.      operand being a 'not' expression, it will be the first one.

  28017. 

  28018.      A machine that has an instruction that performs a bitwise

  28019.      logical-and of one operand with the bitwise negation of the other

  28020.      should specify the pattern for that instruction as

  28021. 

  28022.           (define_insn ""

  28023.             [(set (match_operand:M 0 ...)

  28024.                   (and:M (not:M (match_operand:M 1 ...))

  28025.                                (match_operand:M 2 ...)))]

  28026.             "..."

  28027.             "...")

  28028. 

  28029.      Similarly, a pattern for a "NAND" instruction should be written

  28030. 

  28031.           (define_insn ""

  28032.             [(set (match_operand:M 0 ...)

  28033.                   (ior:M (not:M (match_operand:M 1 ...))

  28034.                                (not:M (match_operand:M 2 ...))))]

  28035.             "..."

  28036.             "...")

  28037. 

  28038.      In both cases, it is not necessary to include patterns for the many

  28039.      logically equivalent RTL expressions.

  28040. 

  28041.    * The only possible RTL expressions involving both bitwise

  28042.      exclusive-or and bitwise negation are '(xor:M X Y)' and '(not:M

  28043.      (xor:M X Y))'.

  28044. 

  28045.    * The sum of three items, one of which is a constant, will only

  28046.      appear in the form

  28047. 

  28048.           (plus:M (plus:M X Y) CONSTANT)

  28049. 

  28050.    * Equality comparisons of a group of bits (usually a single bit) with

  28051.      zero will be written using 'zero_extract' rather than the

  28052.      equivalent 'and' or 'sign_extract' operations.

  28053. 

  28054.    * '(sign_extend:M1 (mult:M2 (sign_extend:M2 X) (sign_extend:M2 Y)))'

  28055.      is converted to '(mult:M1 (sign_extend:M1 X) (sign_extend:M1 Y))',

  28056.      and likewise for 'zero_extend'.

  28057. 

  28058.    * '(sign_extend:M1 (mult:M2 (ashiftrt:M2 X S) (sign_extend:M2 Y)))'

  28059.      is converted to '(mult:M1 (sign_extend:M1 (ashiftrt:M2 X S))

  28060.      (sign_extend:M1 Y))', and likewise for patterns using 'zero_extend'

  28061.      and 'lshiftrt'.  If the second operand of 'mult' is also a shift,

  28062.      then that is extended also.  This transformation is only applied

  28063.      when it can be proven that the original operation had sufficient

  28064.      precision to prevent overflow.

  28065. 

  28066.  Further canonicalization rules are defined in the function

  28067. 'commutative_operand_precedence' in 'gcc/rtlanal.c'.

  28068. 

  28069. 

  28070. File: gccint.info,  Node: Expander Definitions,  Next: Insn Splitting,  Prev: Insn Canonicalizations,  Up: Machine Desc

  28071. 

  28072. 17.15 Defining RTL Sequences for Code Generation

  28073. ================================================

  28074. 

  28075. On some target machines, some standard pattern names for RTL generation

  28076. cannot be handled with single insn, but a sequence of RTL insns can

  28077. represent them.  For these target machines, you can write a

  28078. 'define_expand' to specify how to generate the sequence of RTL.

  28079. 

  28080.  A 'define_expand' is an RTL expression that looks almost like a

  28081. 'define_insn'; but, unlike the latter, a 'define_expand' is used only

  28082. for RTL generation and it can produce more than one RTL insn.

  28083. 

  28084.  A 'define_expand' RTX has four operands:

  28085. 

  28086.    * The name.  Each 'define_expand' must have a name, since the only

  28087.      use for it is to refer to it by name.

  28088. 

  28089.    * The RTL template.  This is a vector of RTL expressions representing

  28090.      a sequence of separate instructions.  Unlike 'define_insn', there

  28091.      is no implicit surrounding 'PARALLEL'.

  28092. 

  28093.    * The condition, a string containing a C expression.  This expression

  28094.      is used to express how the availability of this pattern depends on

  28095.      subclasses of target machine, selected by command-line options when

  28096.      GCC is run.  This is just like the condition of a 'define_insn'

  28097.      that has a standard name.  Therefore, the condition (if present)

  28098.      may not depend on the data in the insn being matched, but only the

  28099.      target-machine-type flags.  The compiler needs to test these

  28100.      conditions during initialization in order to learn exactly which

  28101.      named instructions are available in a particular run.

  28102. 

  28103.    * The preparation statements, a string containing zero or more C

  28104.      statements which are to be executed before RTL code is generated

  28105.      from the RTL template.

  28106. 

  28107.      Usually these statements prepare temporary registers for use as

  28108.      internal operands in the RTL template, but they can also generate

  28109.      RTL insns directly by calling routines such as 'emit_insn', etc.

  28110.      Any such insns precede the ones that come from the RTL template.

  28111. 

  28112.    * Optionally, a vector containing the values of attributes.  *Note

  28113.      Insn Attributes::.

  28114. 

  28115.  Every RTL insn emitted by a 'define_expand' must match some

  28116. 'define_insn' in the machine description.  Otherwise, the compiler will

  28117. crash when trying to generate code for the insn or trying to optimize

  28118. it.

  28119. 

  28120.  The RTL template, in addition to controlling generation of RTL insns,

  28121. also describes the operands that need to be specified when this pattern

  28122. is used.  In particular, it gives a predicate for each operand.

  28123. 

  28124.  A true operand, which needs to be specified in order to generate RTL

  28125. from the pattern, should be described with a 'match_operand' in its

  28126. first occurrence in the RTL template.  This enters information on the

  28127. operand's predicate into the tables that record such things.  GCC uses

  28128. the information to preload the operand into a register if that is

  28129. required for valid RTL code.  If the operand is referred to more than

  28130. once, subsequent references should use 'match_dup'.

  28131. 

  28132.  The RTL template may also refer to internal "operands" which are

  28133. temporary registers or labels used only within the sequence made by the

  28134. 'define_expand'.  Internal operands are substituted into the RTL

  28135. template with 'match_dup', never with 'match_operand'.  The values of

  28136. the internal operands are not passed in as arguments by the compiler

  28137. when it requests use of this pattern.  Instead, they are computed within

  28138. the pattern, in the preparation statements.  These statements compute

  28139. the values and store them into the appropriate elements of 'operands' so

  28140. that 'match_dup' can find them.

  28141. 

  28142.  There are two special macros defined for use in the preparation

  28143. statements: 'DONE' and 'FAIL'.  Use them with a following semicolon, as

  28144. a statement.

  28145. 

  28146. 'DONE'

  28147.      Use the 'DONE' macro to end RTL generation for the pattern.  The

  28148.      only RTL insns resulting from the pattern on this occasion will be

  28149.      those already emitted by explicit calls to 'emit_insn' within the

  28150.      preparation statements; the RTL template will not be generated.

  28151. 

  28152. 'FAIL'

  28153.      Make the pattern fail on this occasion.  When a pattern fails, it

  28154.      means that the pattern was not truly available.  The calling

  28155.      routines in the compiler will try other strategies for code

  28156.      generation using other patterns.

  28157. 

  28158.      Failure is currently supported only for binary (addition,

  28159.      multiplication, shifting, etc.)  and bit-field ('extv', 'extzv',

  28160.      and 'insv') operations.

  28161. 

  28162.  If the preparation falls through (invokes neither 'DONE' nor 'FAIL'),

  28163. then the 'define_expand' acts like a 'define_insn' in that the RTL

  28164. template is used to generate the insn.

  28165. 

  28166.  The RTL template is not used for matching, only for generating the

  28167. initial insn list.  If the preparation statement always invokes 'DONE'

  28168. or 'FAIL', the RTL template may be reduced to a simple list of operands,

  28169. such as this example:

  28170. 

  28171.      (define_expand "addsi3"

  28172.        [(match_operand:SI 0 "register_operand" "")

  28173.         (match_operand:SI 1 "register_operand" "")

  28174.         (match_operand:SI 2 "register_operand" "")]

  28175.        ""

  28176.        "

  28177.      {

  28178.        handle_add (operands[0], operands[1], operands[2]);

  28179.        DONE;

  28180.      }")

  28181. 

  28182.  Here is an example, the definition of left-shift for the SPUR chip:

  28183. 

  28184.      (define_expand "ashlsi3"

  28185.        [(set (match_operand:SI 0 "register_operand" "")

  28186.              (ashift:SI

  28187.                (match_operand:SI 1 "register_operand" "")

  28188.                (match_operand:SI 2 "nonmemory_operand" "")))]

  28189.        ""

  28190.        "

  28191. 

  28192.      {

  28193.        if (GET_CODE (operands[2]) != CONST_INT

  28194.            || (unsigned) INTVAL (operands[2]) > 3)

  28195.          FAIL;

  28196.      }")

  28197. 

  28198. This example uses 'define_expand' so that it can generate an RTL insn

  28199. for shifting when the shift-count is in the supported range of 0 to 3

  28200. but fail in other cases where machine insns aren't available.  When it

  28201. fails, the compiler tries another strategy using different patterns

  28202. (such as, a library call).

  28203. 

  28204.  If the compiler were able to handle nontrivial condition-strings in

  28205. patterns with names, then it would be possible to use a 'define_insn' in

  28206. that case.  Here is another case (zero-extension on the 68000) which

  28207. makes more use of the power of 'define_expand':

  28208. 

  28209.      (define_expand "zero_extendhisi2"

  28210.        [(set (match_operand:SI 0 "general_operand" "")

  28211.              (const_int 0))

  28212.         (set (strict_low_part

  28213.                (subreg:HI

  28214.                  (match_dup 0)

  28215.                  0))

  28216.              (match_operand:HI 1 "general_operand" ""))]

  28217.        ""

  28218.        "operands[1] = make_safe_from (operands[1], operands[0]);")

  28219. 

  28220. Here two RTL insns are generated, one to clear the entire output operand

  28221. and the other to copy the input operand into its low half.  This

  28222. sequence is incorrect if the input operand refers to [the old value of]

  28223. the output operand, so the preparation statement makes sure this isn't

  28224. so.  The function 'make_safe_from' copies the 'operands[1]' into a

  28225. temporary register if it refers to 'operands[0]'.  It does this by

  28226. emitting another RTL insn.

  28227. 

  28228.  Finally, a third example shows the use of an internal operand.

  28229. Zero-extension on the SPUR chip is done by 'and'-ing the result against

  28230. a halfword mask.  But this mask cannot be represented by a 'const_int'

  28231. because the constant value is too large to be legitimate on this

  28232. machine.  So it must be copied into a register with 'force_reg' and then

  28233. the register used in the 'and'.

  28234. 

  28235.      (define_expand "zero_extendhisi2"

  28236.        [(set (match_operand:SI 0 "register_operand" "")

  28237.              (and:SI (subreg:SI

  28238.                        (match_operand:HI 1 "register_operand" "")

  28239.                        0)

  28240.                      (match_dup 2)))]

  28241.        ""

  28242.        "operands[2]

  28243.           = force_reg (SImode, GEN_INT (65535)); ")

  28244. 

  28245.  _Note:_ If the 'define_expand' is used to serve a standard binary or

  28246. unary arithmetic operation or a bit-field operation, then the last insn

  28247. it generates must not be a 'code_label', 'barrier' or 'note'.  It must

  28248. be an 'insn', 'jump_insn' or 'call_insn'.  If you don't need a real insn

  28249. at the end, emit an insn to copy the result of the operation into

  28250. itself.  Such an insn will generate no code, but it can avoid problems

  28251. in the compiler.

  28252. 

  28253. 

  28254. File: gccint.info,  Node: Insn Splitting,  Next: Including Patterns,  Prev: Expander Definitions,  Up: Machine Desc

  28255. 

  28256. 17.16 Defining How to Split Instructions

  28257. ========================================

  28258. 

  28259. There are two cases where you should specify how to split a pattern into

  28260. multiple insns.  On machines that have instructions requiring delay

  28261. slots (*note Delay Slots::) or that have instructions whose output is

  28262. not available for multiple cycles (*note Processor pipeline

  28263. description::), the compiler phases that optimize these cases need to be

  28264. able to move insns into one-instruction delay slots.  However, some

  28265. insns may generate more than one machine instruction.  These insns

  28266. cannot be placed into a delay slot.

  28267. 

  28268.  Often you can rewrite the single insn as a list of individual insns,

  28269. each corresponding to one machine instruction.  The disadvantage of

  28270. doing so is that it will cause the compilation to be slower and require

  28271. more space.  If the resulting insns are too complex, it may also

  28272. suppress some optimizations.  The compiler splits the insn if there is a

  28273. reason to believe that it might improve instruction or delay slot

  28274. scheduling.

  28275. 

  28276.  The insn combiner phase also splits putative insns.  If three insns are

  28277. merged into one insn with a complex expression that cannot be matched by

  28278. some 'define_insn' pattern, the combiner phase attempts to split the

  28279. complex pattern into two insns that are recognized.  Usually it can

  28280. break the complex pattern into two patterns by splitting out some

  28281. subexpression.  However, in some other cases, such as performing an

  28282. addition of a large constant in two insns on a RISC machine, the way to

  28283. split the addition into two insns is machine-dependent.

  28284. 

  28285.  The 'define_split' definition tells the compiler how to split a complex

  28286. insn into several simpler insns.  It looks like this:

  28287. 

  28288.      (define_split

  28289.        [INSN-PATTERN]

  28290.        "CONDITION"

  28291.        [NEW-INSN-PATTERN-1

  28292.         NEW-INSN-PATTERN-2

  28293.         ...]

  28294.        "PREPARATION-STATEMENTS")

  28295. 

  28296.  INSN-PATTERN is a pattern that needs to be split and CONDITION is the

  28297. final condition to be tested, as in a 'define_insn'.  When an insn

  28298. matching INSN-PATTERN and satisfying CONDITION is found, it is replaced

  28299. in the insn list with the insns given by NEW-INSN-PATTERN-1,

  28300. NEW-INSN-PATTERN-2, etc.

  28301. 

  28302.  The PREPARATION-STATEMENTS are similar to those statements that are

  28303. specified for 'define_expand' (*note Expander Definitions::) and are

  28304. executed before the new RTL is generated to prepare for the generated

  28305. code or emit some insns whose pattern is not fixed.  Unlike those in

  28306. 'define_expand', however, these statements must not generate any new

  28307. pseudo-registers.  Once reload has completed, they also must not

  28308. allocate any space in the stack frame.

  28309. 

  28310.  There are two special macros defined for use in the preparation

  28311. statements: 'DONE' and 'FAIL'.  Use them with a following semicolon, as

  28312. a statement.

  28313. 

  28314. 'DONE'

  28315.      Use the 'DONE' macro to end RTL generation for the splitter.  The

  28316.      only RTL insns generated as replacement for the matched input insn

  28317.      will be those already emitted by explicit calls to 'emit_insn'

  28318.      within the preparation statements; the replacement pattern is not

  28319.      used.

  28320. 

  28321. 'FAIL'

  28322.      Make the 'define_split' fail on this occasion.  When a

  28323.      'define_split' fails, it means that the splitter was not truly

  28324.      available for the inputs it was given, and the input insn will not

  28325.      be split.

  28326. 

  28327.  If the preparation falls through (invokes neither 'DONE' nor 'FAIL'),

  28328. then the 'define_split' uses the replacement template.

  28329. 

  28330.  Patterns are matched against INSN-PATTERN in two different

  28331. circumstances.  If an insn needs to be split for delay slot scheduling

  28332. or insn scheduling, the insn is already known to be valid, which means

  28333. that it must have been matched by some 'define_insn' and, if

  28334. 'reload_completed' is nonzero, is known to satisfy the constraints of

  28335. that 'define_insn'.  In that case, the new insn patterns must also be

  28336. insns that are matched by some 'define_insn' and, if 'reload_completed'

  28337. is nonzero, must also satisfy the constraints of those definitions.

  28338. 

  28339.  As an example of this usage of 'define_split', consider the following

  28340. example from 'a29k.md', which splits a 'sign_extend' from 'HImode' to

  28341. 'SImode' into a pair of shift insns:

  28342. 

  28343.      (define_split

  28344.        [(set (match_operand:SI 0 "gen_reg_operand" "")

  28345.              (sign_extend:SI (match_operand:HI 1 "gen_reg_operand" "")))]

  28346.        ""

  28347.        [(set (match_dup 0)

  28348.              (ashift:SI (match_dup 1)

  28349.                         (const_int 16)))

  28350.         (set (match_dup 0)

  28351.              (ashiftrt:SI (match_dup 0)

  28352.                           (const_int 16)))]

  28353.        "

  28354.      { operands[1] = gen_lowpart (SImode, operands[1]); }")

  28355. 

  28356.  When the combiner phase tries to split an insn pattern, it is always

  28357. the case that the pattern is _not_ matched by any 'define_insn'.  The

  28358. combiner pass first tries to split a single 'set' expression and then

  28359. the same 'set' expression inside a 'parallel', but followed by a

  28360. 'clobber' of a pseudo-reg to use as a scratch register.  In these cases,

  28361. the combiner expects exactly one or two new insn patterns to be

  28362. generated.  It will verify that these patterns match some 'define_insn'

  28363. definitions, so you need not do this test in the 'define_split' (of

  28364. course, there is no point in writing a 'define_split' that will never

  28365. produce insns that match).

  28366. 

  28367.  Here is an example of this use of 'define_split', taken from

  28368. 'rs6000.md':

  28369. 

  28370.      (define_split

  28371.        [(set (match_operand:SI 0 "gen_reg_operand" "")

  28372.              (plus:SI (match_operand:SI 1 "gen_reg_operand" "")

  28373.                       (match_operand:SI 2 "non_add_cint_operand" "")))]

  28374.        ""

  28375.        [(set (match_dup 0) (plus:SI (match_dup 1) (match_dup 3)))

  28376.         (set (match_dup 0) (plus:SI (match_dup 0) (match_dup 4)))]

  28377.      "

  28378.      {

  28379.        int low = INTVAL (operands[2]) & 0xffff;

  28380.        int high = (unsigned) INTVAL (operands[2]) >> 16;

  28381. 

  28382.        if (low & 0x8000)

  28383.          high++, low |= 0xffff0000;

  28384. 

  28385.        operands[3] = GEN_INT (high << 16);

  28386.        operands[4] = GEN_INT (low);

  28387.      }")

  28388. 

  28389.  Here the predicate 'non_add_cint_operand' matches any 'const_int' that

  28390. is _not_ a valid operand of a single add insn.  The add with the smaller

  28391. displacement is written so that it can be substituted into the address

  28392. of a subsequent operation.

  28393. 

  28394.  An example that uses a scratch register, from the same file, generates

  28395. an equality comparison of a register and a large constant:

  28396. 

  28397.      (define_split

  28398.        [(set (match_operand:CC 0 "cc_reg_operand" "")

  28399.              (compare:CC (match_operand:SI 1 "gen_reg_operand" "")

  28400.                          (match_operand:SI 2 "non_short_cint_operand" "")))

  28401.         (clobber (match_operand:SI 3 "gen_reg_operand" ""))]

  28402.        "find_single_use (operands[0], insn, 0)

  28403.         && (GET_CODE (*find_single_use (operands[0], insn, 0)) == EQ

  28404.             || GET_CODE (*find_single_use (operands[0], insn, 0)) == NE)"

  28405.        [(set (match_dup 3) (xor:SI (match_dup 1) (match_dup 4)))

  28406.         (set (match_dup 0) (compare:CC (match_dup 3) (match_dup 5)))]

  28407.        "

  28408.      {

  28409.        /* Get the constant we are comparing against, C, and see what it

  28410.           looks like sign-extended to 16 bits.  Then see what constant

  28411.           could be XOR'ed with C to get the sign-extended value.  */

  28412. 

  28413.        int c = INTVAL (operands[2]);

  28414.        int sextc = (c << 16) >> 16;

  28415.        int xorv = c ^ sextc;

  28416. 

  28417.        operands[4] = GEN_INT (xorv);

  28418.        operands[5] = GEN_INT (sextc);

  28419.      }")

  28420. 

  28421.  To avoid confusion, don't write a single 'define_split' that accepts

  28422. some insns that match some 'define_insn' as well as some insns that

  28423. don't.  Instead, write two separate 'define_split' definitions, one for

  28424. the insns that are valid and one for the insns that are not valid.

  28425. 

  28426.  The splitter is allowed to split jump instructions into sequence of

  28427. jumps or create new jumps in while splitting non-jump instructions.  As

  28428. the control flow graph and branch prediction information needs to be

  28429. updated, several restriction apply.

  28430. 

  28431.  Splitting of jump instruction into sequence that over by another jump

  28432. instruction is always valid, as compiler expect identical behavior of

  28433. new jump.  When new sequence contains multiple jump instructions or new

  28434. labels, more assistance is needed.  Splitter is required to create only

  28435. unconditional jumps, or simple conditional jump instructions.

  28436. Additionally it must attach a 'REG_BR_PROB' note to each conditional

  28437. jump.  A global variable 'split_branch_probability' holds the

  28438. probability of the original branch in case it was a simple conditional

  28439. jump, -1 otherwise.  To simplify recomputing of edge frequencies, the

  28440. new sequence is required to have only forward jumps to the newly created

  28441. labels.

  28442. 

  28443.  For the common case where the pattern of a define_split exactly matches

  28444. the pattern of a define_insn, use 'define_insn_and_split'.  It looks

  28445. like this:

  28446. 

  28447.      (define_insn_and_split

  28448.        [INSN-PATTERN]

  28449.        "CONDITION"

  28450.        "OUTPUT-TEMPLATE"

  28451.        "SPLIT-CONDITION"

  28452.        [NEW-INSN-PATTERN-1

  28453.         NEW-INSN-PATTERN-2

  28454.         ...]

  28455.        "PREPARATION-STATEMENTS"

  28456.        [INSN-ATTRIBUTES])

  28457. 

  28458. 

  28459.  INSN-PATTERN, CONDITION, OUTPUT-TEMPLATE, and INSN-ATTRIBUTES are used

  28460. as in 'define_insn'.  The NEW-INSN-PATTERN vector and the

  28461. PREPARATION-STATEMENTS are used as in a 'define_split'.  The

  28462. SPLIT-CONDITION is also used as in 'define_split', with the additional

  28463. behavior that if the condition starts with '&&', the condition used for

  28464. the split will be the constructed as a logical "and" of the split

  28465. condition with the insn condition.  For example, from i386.md:

  28466. 

  28467.      (define_insn_and_split "zero_extendhisi2_and"

  28468.        [(set (match_operand:SI 0 "register_operand" "=r")

  28469.           (zero_extend:SI (match_operand:HI 1 "register_operand" "0")))

  28470.         (clobber (reg:CC 17))]

  28471.        "TARGET_ZERO_EXTEND_WITH_AND && !optimize_size"

  28472.        "#"

  28473.        "&& reload_completed"

  28474.        [(parallel [(set (match_dup 0)

  28475.                         (and:SI (match_dup 0) (const_int 65535)))

  28476.                    (clobber (reg:CC 17))])]

  28477.        ""

  28478.        [(set_attr "type" "alu1")])

  28479. 

  28480. 

  28481.  In this case, the actual split condition will be

  28482. 'TARGET_ZERO_EXTEND_WITH_AND && !optimize_size && reload_completed'.

  28483. 

  28484.  The 'define_insn_and_split' construction provides exactly the same

  28485. functionality as two separate 'define_insn' and 'define_split' patterns.

  28486. It exists for compactness, and as a maintenance tool to prevent having

  28487. to ensure the two patterns' templates match.

  28488. 

  28489.  It is sometimes useful to have a 'define_insn_and_split' that replaces

  28490. specific operands of an instruction but leaves the rest of the

  28491. instruction pattern unchanged.  You can do this directly with a

  28492. 'define_insn_and_split', but it requires a NEW-INSN-PATTERN-1 that

  28493. repeats most of the original INSN-PATTERN.  There is also the

  28494. complication that an implicit 'parallel' in INSN-PATTERN must become an

  28495. explicit 'parallel' in NEW-INSN-PATTERN-1, which is easy to overlook.  A

  28496. simpler alternative is to use 'define_insn_and_rewrite', which is a form

  28497. of 'define_insn_and_split' that automatically generates

  28498. NEW-INSN-PATTERN-1 by replacing each 'match_operand' in INSN-PATTERN

  28499. with a corresponding 'match_dup', and each 'match_operator' in the

  28500. pattern with a corresponding 'match_op_dup'.  The arguments are

  28501. otherwise identical to 'define_insn_and_split':

  28502. 

  28503.      (define_insn_and_rewrite

  28504.        [INSN-PATTERN]

  28505.        "CONDITION"

  28506.        "OUTPUT-TEMPLATE"

  28507.        "SPLIT-CONDITION"

  28508.        "PREPARATION-STATEMENTS"

  28509.        [INSN-ATTRIBUTES])

  28510. 

  28511.  The 'match_dup's and 'match_op_dup's in the new instruction pattern use

  28512. any new operand values that the PREPARATION-STATEMENTS store in the

  28513. 'operands' array, as for a normal 'define_insn_and_split'.

  28514. PREPARATION-STATEMENTS can also emit additional instructions before the

  28515. new instruction.  They can even emit an entirely different sequence of

  28516. instructions and use 'DONE' to avoid emitting a new form of the original

  28517. instruction.

  28518. 

  28519.  The split in a 'define_insn_and_rewrite' is only intended to apply to

  28520. existing instructions that match INSN-PATTERN.  SPLIT-CONDITION must

  28521. therefore start with '&&', so that the split condition applies on top of

  28522. CONDITION.

  28523. 

  28524.  Here is an example from the AArch64 SVE port, in which operand 1 is

  28525. known to be equivalent to an all-true constant and isn't used by the

  28526. output template:

  28527. 

  28528.      (define_insn_and_rewrite "*while_ult<GPI:mode><PRED_ALL:mode>_cc"

  28529.        [(set (reg:CC CC_REGNUM)

  28530.              (compare:CC

  28531.                (unspec:SI [(match_operand:PRED_ALL 1)

  28532.                            (unspec:PRED_ALL

  28533.                              [(match_operand:GPI 2 "aarch64_reg_or_zero" "rZ")

  28534.                               (match_operand:GPI 3 "aarch64_reg_or_zero" "rZ")]

  28535.                              UNSPEC_WHILE_LO)]

  28536.                           UNSPEC_PTEST_PTRUE)

  28537.                (const_int 0)))

  28538.         (set (match_operand:PRED_ALL 0 "register_operand" "=Upa")

  28539.              (unspec:PRED_ALL [(match_dup 2)

  28540.                                (match_dup 3)]

  28541.                               UNSPEC_WHILE_LO))]

  28542.        "TARGET_SVE"

  28543.        "whilelo\t%0.<PRED_ALL:Vetype>, %<w>2, %<w>3"

  28544.        ;; Force the compiler to drop the unused predicate operand, so that we

  28545.        ;; don't have an unnecessary PTRUE.

  28546.        "&& !CONSTANT_P (operands[1])"

  28547.        {

  28548.          operands[1] = CONSTM1_RTX (<MODE>mode);

  28549.        }

  28550.      )

  28551. 

  28552.  The splitter in this case simply replaces operand 1 with the constant

  28553. value that it is known to have.  The equivalent 'define_insn_and_split'

  28554. would be:

  28555. 

  28556.      (define_insn_and_split "*while_ult<GPI:mode><PRED_ALL:mode>_cc"

  28557.        [(set (reg:CC CC_REGNUM)

  28558.              (compare:CC

  28559.                (unspec:SI [(match_operand:PRED_ALL 1)

  28560.                            (unspec:PRED_ALL

  28561.                              [(match_operand:GPI 2 "aarch64_reg_or_zero" "rZ")

  28562.                               (match_operand:GPI 3 "aarch64_reg_or_zero" "rZ")]

  28563.                              UNSPEC_WHILE_LO)]

  28564.                           UNSPEC_PTEST_PTRUE)

  28565.                (const_int 0)))

  28566.         (set (match_operand:PRED_ALL 0 "register_operand" "=Upa")

  28567.              (unspec:PRED_ALL [(match_dup 2)

  28568.                                (match_dup 3)]

  28569.                               UNSPEC_WHILE_LO))]

  28570.        "TARGET_SVE"

  28571.        "whilelo\t%0.<PRED_ALL:Vetype>, %<w>2, %<w>3"

  28572.        ;; Force the compiler to drop the unused predicate operand, so that we

  28573.        ;; don't have an unnecessary PTRUE.

  28574.        "&& !CONSTANT_P (operands[1])"

  28575.        [(parallel

  28576.           [(set (reg:CC CC_REGNUM)

  28577.                 (compare:CC

  28578.                   (unspec:SI [(match_dup 1)

  28579.                               (unspec:PRED_ALL [(match_dup 2)

  28580.                                                 (match_dup 3)]

  28581.                                                UNSPEC_WHILE_LO)]

  28582.                              UNSPEC_PTEST_PTRUE)

  28583.                   (const_int 0)))

  28584.            (set (match_dup 0)

  28585.                 (unspec:PRED_ALL [(match_dup 2)

  28586.                                   (match_dup 3)]

  28587.                                  UNSPEC_WHILE_LO))])]

  28588.        {

  28589.          operands[1] = CONSTM1_RTX (<MODE>mode);

  28590.        }

  28591.      )

  28592. 

  28593. 

  28594. File: gccint.info,  Node: Including Patterns,  Next: Peephole Definitions,  Prev: Insn Splitting,  Up: Machine Desc

  28595. 

  28596. 17.17 Including Patterns in Machine Descriptions.

  28597. =================================================

  28598. 

  28599. The 'include' pattern tells the compiler tools where to look for

  28600. patterns that are in files other than in the file '.md'.  This is used

  28601. only at build time and there is no preprocessing allowed.

  28602. 

  28603.  It looks like:

  28604. 

  28605. 

  28606.      (include

  28607.        PATHNAME)

  28608. 

  28609.  For example:

  28610. 

  28611. 

  28612.      (include "filestuff")

  28613. 

  28614. 

  28615.  Where PATHNAME is a string that specifies the location of the file,

  28616. specifies the include file to be in 'gcc/config/target/filestuff'.  The

  28617. directory 'gcc/config/target' is regarded as the default directory.

  28618. 

  28619.  Machine descriptions may be split up into smaller more manageable

  28620. subsections and placed into subdirectories.

  28621. 

  28622.  By specifying:

  28623. 

  28624. 

  28625.      (include "BOGUS/filestuff")

  28626. 

  28627. 

  28628.  the include file is specified to be in

  28629. 'gcc/config/TARGET/BOGUS/filestuff'.

  28630. 

  28631.  Specifying an absolute path for the include file such as;

  28632. 

  28633.      (include "/u2/BOGUS/filestuff")

  28634. 

  28635.  is permitted but is not encouraged.

  28636. 

  28637. 17.17.1 RTL Generation Tool Options for Directory Search

  28638. --------------------------------------------------------

  28639. 

  28640. The '-IDIR' option specifies directories to search for machine

  28641. descriptions.  For example:

  28642. 

  28643. 

  28644.      genrecog -I/p1/abc/proc1 -I/p2/abcd/pro2 target.md

  28645. 

  28646. 

  28647.  Add the directory DIR to the head of the list of directories to be

  28648. searched for header files.  This can be used to override a system

  28649. machine definition file, substituting your own version, since these

  28650. directories are searched before the default machine description file

  28651. directories.  If you use more than one '-I' option, the directories are

  28652. scanned in left-to-right order; the standard default directory come

  28653. after.

  28654. 

  28655. 

  28656. File: gccint.info,  Node: Peephole Definitions,  Next: Insn Attributes,  Prev: Including Patterns,  Up: Machine Desc

  28657. 

  28658. 17.18 Machine-Specific Peephole Optimizers

  28659. ==========================================

  28660. 

  28661. In addition to instruction patterns the 'md' file may contain

  28662. definitions of machine-specific peephole optimizations.

  28663. 

  28664.  The combiner does not notice certain peephole optimizations when the

  28665. data flow in the program does not suggest that it should try them.  For

  28666. example, sometimes two consecutive insns related in purpose can be

  28667. combined even though the second one does not appear to use a register

  28668. computed in the first one.  A machine-specific peephole optimizer can

  28669. detect such opportunities.

  28670. 

  28671.  There are two forms of peephole definitions that may be used.  The

  28672. original 'define_peephole' is run at assembly output time to match insns

  28673. and substitute assembly text.  Use of 'define_peephole' is deprecated.

  28674. 

  28675.  A newer 'define_peephole2' matches insns and substitutes new insns.

  28676. The 'peephole2' pass is run after register allocation but before

  28677. scheduling, which may result in much better code for targets that do

  28678. scheduling.

  28679. 

  28680. * Menu:

  28681. 

  28682. * define_peephole::     RTL to Text Peephole Optimizers

  28683. * define_peephole2::    RTL to RTL Peephole Optimizers

  28684. 

  28685. 

  28686. File: gccint.info,  Node: define_peephole,  Next: define_peephole2,  Up: Peephole Definitions

  28687. 

  28688. 17.18.1 RTL to Text Peephole Optimizers

  28689. ---------------------------------------

  28690. 

  28691. A definition looks like this:

  28692. 

  28693.      (define_peephole

  28694.        [INSN-PATTERN-1

  28695.         INSN-PATTERN-2

  28696.         ...]

  28697.        "CONDITION"

  28698.        "TEMPLATE"

  28699.        "OPTIONAL-INSN-ATTRIBUTES")

  28700. 

  28701. The last string operand may be omitted if you are not using any

  28702. machine-specific information in this machine description.  If present,

  28703. it must obey the same rules as in a 'define_insn'.

  28704. 

  28705.  In this skeleton, INSN-PATTERN-1 and so on are patterns to match

  28706. consecutive insns.  The optimization applies to a sequence of insns when

  28707. INSN-PATTERN-1 matches the first one, INSN-PATTERN-2 matches the next,

  28708. and so on.

  28709. 

  28710.  Each of the insns matched by a peephole must also match a

  28711. 'define_insn'.  Peepholes are checked only at the last stage just before

  28712. code generation, and only optionally.  Therefore, any insn which would

  28713. match a peephole but no 'define_insn' will cause a crash in code

  28714. generation in an unoptimized compilation, or at various optimization

  28715. stages.

  28716. 

  28717.  The operands of the insns are matched with 'match_operands',

  28718. 'match_operator', and 'match_dup', as usual.  What is not usual is that

  28719. the operand numbers apply to all the insn patterns in the definition.

  28720. So, you can check for identical operands in two insns by using

  28721. 'match_operand' in one insn and 'match_dup' in the other.

  28722. 

  28723.  The operand constraints used in 'match_operand' patterns do not have

  28724. any direct effect on the applicability of the peephole, but they will be

  28725. validated afterward, so make sure your constraints are general enough to

  28726. apply whenever the peephole matches.  If the peephole matches but the

  28727. constraints are not satisfied, the compiler will crash.

  28728. 

  28729.  It is safe to omit constraints in all the operands of the peephole; or

  28730. you can write constraints which serve as a double-check on the criteria

  28731. previously tested.

  28732. 

  28733.  Once a sequence of insns matches the patterns, the CONDITION is

  28734. checked.  This is a C expression which makes the final decision whether

  28735. to perform the optimization (we do so if the expression is nonzero).  If

  28736. CONDITION is omitted (in other words, the string is empty) then the

  28737. optimization is applied to every sequence of insns that matches the

  28738. patterns.

  28739. 

  28740.  The defined peephole optimizations are applied after register

  28741. allocation is complete.  Therefore, the peephole definition can check

  28742. which operands have ended up in which kinds of registers, just by

  28743. looking at the operands.

  28744. 

  28745.  The way to refer to the operands in CONDITION is to write 'operands[I]'

  28746. for operand number I (as matched by '(match_operand I ...)').  Use the

  28747. variable 'insn' to refer to the last of the insns being matched; use

  28748. 'prev_active_insn' to find the preceding insns.

  28749. 

  28750.  When optimizing computations with intermediate results, you can use

  28751. CONDITION to match only when the intermediate results are not used

  28752. elsewhere.  Use the C expression 'dead_or_set_p (INSN, OP)', where INSN

  28753. is the insn in which you expect the value to be used for the last time

  28754. (from the value of 'insn', together with use of 'prev_nonnote_insn'),

  28755. and OP is the intermediate value (from 'operands[I]').

  28756. 

  28757.  Applying the optimization means replacing the sequence of insns with

  28758. one new insn.  The TEMPLATE controls ultimate output of assembler code

  28759. for this combined insn.  It works exactly like the template of a

  28760. 'define_insn'.  Operand numbers in this template are the same ones used

  28761. in matching the original sequence of insns.

  28762. 

  28763.  The result of a defined peephole optimizer does not need to match any

  28764. of the insn patterns in the machine description; it does not even have

  28765. an opportunity to match them.  The peephole optimizer definition itself

  28766. serves as the insn pattern to control how the insn is output.

  28767. 

  28768.  Defined peephole optimizers are run as assembler code is being output,

  28769. so the insns they produce are never combined or rearranged in any way.

  28770. 

  28771.  Here is an example, taken from the 68000 machine description:

  28772. 

  28773.      (define_peephole

  28774.        [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))

  28775.         (set (match_operand:DF 0 "register_operand" "=f")

  28776.              (match_operand:DF 1 "register_operand" "ad"))]

  28777.        "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"

  28778.      {

  28779.        rtx xoperands[2];

  28780.        xoperands[1] = gen_rtx_REG (SImode, REGNO (operands[1]) + 1);

  28781.      #ifdef MOTOROLA

  28782.        output_asm_insn ("move.l %1,(sp)", xoperands);

  28783.        output_asm_insn ("move.l %1,-(sp)", operands);

  28784.        return "fmove.d (sp)+,%0";

  28785.      #else

  28786.        output_asm_insn ("movel %1,sp@", xoperands);

  28787.        output_asm_insn ("movel %1,sp@-", operands);

  28788.        return "fmoved sp@+,%0";

  28789.      #endif

  28790.      })

  28791. 

  28792.  The effect of this optimization is to change

  28793. 

  28794.      jbsr _foobar

  28795.      addql #4,sp

  28796.      movel d1,sp@-

  28797.      movel d0,sp@-

  28798.      fmoved sp@+,fp0

  28799. 

  28800. into

  28801. 

  28802.      jbsr _foobar

  28803.      movel d1,sp@

  28804.      movel d0,sp@-

  28805.      fmoved sp@+,fp0

  28806. 

  28807.  INSN-PATTERN-1 and so on look _almost_ like the second operand of

  28808. 'define_insn'.  There is one important difference: the second operand of

  28809. 'define_insn' consists of one or more RTX's enclosed in square brackets.

  28810. Usually, there is only one: then the same action can be written as an

  28811. element of a 'define_peephole'.  But when there are multiple actions in

  28812. a 'define_insn', they are implicitly enclosed in a 'parallel'.  Then you

  28813. must explicitly write the 'parallel', and the square brackets within it,

  28814. in the 'define_peephole'.  Thus, if an insn pattern looks like this,

  28815. 

  28816.      (define_insn "divmodsi4"

  28817.        [(set (match_operand:SI 0 "general_operand" "=d")

  28818.              (div:SI (match_operand:SI 1 "general_operand" "0")

  28819.                      (match_operand:SI 2 "general_operand" "dmsK")))

  28820.         (set (match_operand:SI 3 "general_operand" "=d")

  28821.              (mod:SI (match_dup 1) (match_dup 2)))]

  28822.        "TARGET_68020"

  28823.        "divsl%.l %2,%3:%0")

  28824. 

  28825. then the way to mention this insn in a peephole is as follows:

  28826. 

  28827.      (define_peephole

  28828.        [...

  28829.         (parallel

  28830.          [(set (match_operand:SI 0 "general_operand" "=d")

  28831.                (div:SI (match_operand:SI 1 "general_operand" "0")

  28832.                        (match_operand:SI 2 "general_operand" "dmsK")))

  28833.           (set (match_operand:SI 3 "general_operand" "=d")

  28834.                (mod:SI (match_dup 1) (match_dup 2)))])

  28835.         ...]

  28836.        ...)

  28837. 

  28838. 

  28839. File: gccint.info,  Node: define_peephole2,  Prev: define_peephole,  Up: Peephole Definitions

  28840. 

  28841. 17.18.2 RTL to RTL Peephole Optimizers

  28842. --------------------------------------

  28843. 

  28844. The 'define_peephole2' definition tells the compiler how to substitute

  28845. one sequence of instructions for another sequence, what additional

  28846. scratch registers may be needed and what their lifetimes must be.

  28847. 

  28848.      (define_peephole2

  28849.        [INSN-PATTERN-1

  28850.         INSN-PATTERN-2

  28851.         ...]

  28852.        "CONDITION"

  28853.        [NEW-INSN-PATTERN-1

  28854.         NEW-INSN-PATTERN-2

  28855.         ...]

  28856.        "PREPARATION-STATEMENTS")

  28857. 

  28858.  The definition is almost identical to 'define_split' (*note Insn

  28859. Splitting::) except that the pattern to match is not a single

  28860. instruction, but a sequence of instructions.

  28861. 

  28862.  It is possible to request additional scratch registers for use in the

  28863. output template.  If appropriate registers are not free, the pattern

  28864. will simply not match.

  28865. 

  28866.  Scratch registers are requested with a 'match_scratch' pattern at the

  28867. top level of the input pattern.  The allocated register (initially) will

  28868. be dead at the point requested within the original sequence.  If the

  28869. scratch is used at more than a single point, a 'match_dup' pattern at

  28870. the top level of the input pattern marks the last position in the input

  28871. sequence at which the register must be available.

  28872. 

  28873.  Here is an example from the IA-32 machine description:

  28874. 

  28875.      (define_peephole2

  28876.        [(match_scratch:SI 2 "r")

  28877.         (parallel [(set (match_operand:SI 0 "register_operand" "")

  28878.                         (match_operator:SI 3 "arith_or_logical_operator"

  28879.                           [(match_dup 0)

  28880.                            (match_operand:SI 1 "memory_operand" "")]))

  28881.                    (clobber (reg:CC 17))])]

  28882.        "! optimize_size && ! TARGET_READ_MODIFY"

  28883.        [(set (match_dup 2) (match_dup 1))

  28884.         (parallel [(set (match_dup 0)

  28885.                         (match_op_dup 3 [(match_dup 0) (match_dup 2)]))

  28886.                    (clobber (reg:CC 17))])]

  28887.        "")

  28888. 

  28889. This pattern tries to split a load from its use in the hopes that we'll

  28890. be able to schedule around the memory load latency.  It allocates a

  28891. single 'SImode' register of class 'GENERAL_REGS' ('"r"') that needs to

  28892. be live only at the point just before the arithmetic.

  28893. 

  28894.  A real example requiring extended scratch lifetimes is harder to come

  28895. by, so here's a silly made-up example:

  28896. 

  28897.      (define_peephole2

  28898.        [(match_scratch:SI 4 "r")

  28899.         (set (match_operand:SI 0 "" "") (match_operand:SI 1 "" ""))

  28900.         (set (match_operand:SI 2 "" "") (match_dup 1))

  28901.         (match_dup 4)

  28902.         (set (match_operand:SI 3 "" "") (match_dup 1))]

  28903.        "/* determine 1 does not overlap 0 and 2 */"

  28904.        [(set (match_dup 4) (match_dup 1))

  28905.         (set (match_dup 0) (match_dup 4))

  28906.         (set (match_dup 2) (match_dup 4))

  28907.         (set (match_dup 3) (match_dup 4))]

  28908.        "")

  28909. 

  28910.  There are two special macros defined for use in the preparation

  28911. statements: 'DONE' and 'FAIL'.  Use them with a following semicolon, as

  28912. a statement.

  28913. 

  28914. 'DONE'

  28915.      Use the 'DONE' macro to end RTL generation for the peephole.  The

  28916.      only RTL insns generated as replacement for the matched input insn

  28917.      will be those already emitted by explicit calls to 'emit_insn'

  28918.      within the preparation statements; the replacement pattern is not

  28919.      used.

  28920. 

  28921. 'FAIL'

  28922.      Make the 'define_peephole2' fail on this occasion.  When a

  28923.      'define_peephole2' fails, it means that the replacement was not

  28924.      truly available for the particular inputs it was given.  In that

  28925.      case, GCC may still apply a later 'define_peephole2' that also

  28926.      matches the given insn pattern.  (Note that this is different from

  28927.      'define_split', where 'FAIL' prevents the input insn from being

  28928.      split at all.)

  28929. 

  28930.  If the preparation falls through (invokes neither 'DONE' nor 'FAIL'),

  28931. then the 'define_peephole2' uses the replacement template.

  28932. 

  28933. If we had not added the '(match_dup 4)' in the middle of the input

  28934. sequence, it might have been the case that the register we chose at the

  28935. beginning of the sequence is killed by the first or second 'set'.

  28936. 

  28937. 

  28938. File: gccint.info,  Node: Insn Attributes,  Next: Conditional Execution,  Prev: Peephole Definitions,  Up: Machine Desc

  28939. 

  28940. 17.19 Instruction Attributes

  28941. ============================

  28942. 

  28943. In addition to describing the instruction supported by the target

  28944. machine, the 'md' file also defines a group of "attributes" and a set of

  28945. values for each.  Every generated insn is assigned a value for each

  28946. attribute.  One possible attribute would be the effect that the insn has

  28947. on the machine's condition code.  This attribute can then be used by

  28948. 'NOTICE_UPDATE_CC' to track the condition codes.

  28949. 

  28950. * Menu:

  28951. 

  28952. * Defining Attributes:: Specifying attributes and their values.

  28953. * Expressions::         Valid expressions for attribute values.

  28954. * Tagging Insns::       Assigning attribute values to insns.

  28955. * Attr Example::        An example of assigning attributes.

  28956. * Insn Lengths::        Computing the length of insns.

  28957. * Constant Attributes:: Defining attributes that are constant.

  28958. * Mnemonic Attribute::  Obtain the instruction mnemonic as attribute value.

  28959. * Delay Slots::         Defining delay slots required for a machine.

  28960. * Processor pipeline description:: Specifying information for insn scheduling.

  28961. 

  28962. 

  28963. File: gccint.info,  Node: Defining Attributes,  Next: Expressions,  Up: Insn Attributes

  28964. 

  28965. 17.19.1 Defining Attributes and their Values

  28966. --------------------------------------------

  28967. 

  28968. The 'define_attr' expression is used to define each attribute required

  28969. by the target machine.  It looks like:

  28970. 

  28971.      (define_attr NAME LIST-OF-VALUES DEFAULT)

  28972. 

  28973.  NAME is a string specifying the name of the attribute being defined.

  28974. Some attributes are used in a special way by the rest of the compiler.

  28975. The 'enabled' attribute can be used to conditionally enable or disable

  28976. insn alternatives (*note Disable Insn Alternatives::).  The 'predicable'

  28977. attribute, together with a suitable 'define_cond_exec' (*note

  28978. Conditional Execution::), can be used to automatically generate

  28979. conditional variants of instruction patterns.  The 'mnemonic' attribute

  28980. can be used to check for the instruction mnemonic (*note Mnemonic

  28981. Attribute::).  The compiler internally uses the names 'ce_enabled' and

  28982. 'nonce_enabled', so they should not be used elsewhere as alternative

  28983. names.

  28984. 

  28985.  LIST-OF-VALUES is either a string that specifies a comma-separated list

  28986. of values that can be assigned to the attribute, or a null string to

  28987. indicate that the attribute takes numeric values.

  28988. 

  28989.  DEFAULT is an attribute expression that gives the value of this

  28990. attribute for insns that match patterns whose definition does not

  28991. include an explicit value for this attribute.  *Note Attr Example::, for

  28992. more information on the handling of defaults.  *Note Constant

  28993. Attributes::, for information on attributes that do not depend on any

  28994. particular insn.

  28995. 

  28996.  For each defined attribute, a number of definitions are written to the

  28997. 'insn-attr.h' file.  For cases where an explicit set of values is

  28998. specified for an attribute, the following are defined:

  28999. 

  29000.    * A '#define' is written for the symbol 'HAVE_ATTR_NAME'.

  29001. 

  29002.    * An enumerated class is defined for 'attr_NAME' with elements of the

  29003.      form 'UPPER-NAME_UPPER-VALUE' where the attribute name and value

  29004.      are first converted to uppercase.

  29005. 

  29006.    * A function 'get_attr_NAME' is defined that is passed an insn and

  29007.      returns the attribute value for that insn.

  29008. 

  29009.  For example, if the following is present in the 'md' file:

  29010. 

  29011.      (define_attr "type" "branch,fp,load,store,arith" ...)

  29012. 

  29013. the following lines will be written to the file 'insn-attr.h'.

  29014. 

  29015.      #define HAVE_ATTR_type 1

  29016.      enum attr_type {TYPE_BRANCH, TYPE_FP, TYPE_LOAD,

  29017.                       TYPE_STORE, TYPE_ARITH};

  29018.      extern enum attr_type get_attr_type ();

  29019. 

  29020.  If the attribute takes numeric values, no 'enum' type will be defined

  29021. and the function to obtain the attribute's value will return 'int'.

  29022. 

  29023.  There are attributes which are tied to a specific meaning.  These

  29024. attributes are not free to use for other purposes:

  29025. 

  29026. 'length'

  29027.      The 'length' attribute is used to calculate the length of emitted

  29028.      code chunks.  This is especially important when verifying branch

  29029.      distances.  *Note Insn Lengths::.

  29030. 

  29031. 'enabled'

  29032.      The 'enabled' attribute can be defined to prevent certain

  29033.      alternatives of an insn definition from being used during code

  29034.      generation.  *Note Disable Insn Alternatives::.

  29035. 

  29036. 'mnemonic'

  29037.      The 'mnemonic' attribute can be defined to implement instruction

  29038.      specific checks in e.g. the pipeline description.  *Note Mnemonic

  29039.      Attribute::.

  29040. 

  29041.  For each of these special attributes, the corresponding

  29042. 'HAVE_ATTR_NAME' '#define' is also written when the attribute is not

  29043. defined; in that case, it is defined as '0'.

  29044. 

  29045.  Another way of defining an attribute is to use:

  29046. 

  29047.      (define_enum_attr "ATTR" "ENUM" DEFAULT)

  29048. 

  29049.  This works in just the same way as 'define_attr', except that the list

  29050. of values is taken from a separate enumeration called ENUM (*note

  29051. define_enum::).  This form allows you to use the same list of values for

  29052. several attributes without having to repeat the list each time.  For

  29053. example:

  29054. 

  29055.      (define_enum "processor" [

  29056.        model_a

  29057.        model_b

  29058.        ...

  29059.      ])

  29060.      (define_enum_attr "arch" "processor"

  29061.        (const (symbol_ref "target_arch")))

  29062.      (define_enum_attr "tune" "processor"

  29063.        (const (symbol_ref "target_tune")))

  29064. 

  29065.  defines the same attributes as:

  29066. 

  29067.      (define_attr "arch" "model_a,model_b,..."

  29068.        (const (symbol_ref "target_arch")))

  29069.      (define_attr "tune" "model_a,model_b,..."

  29070.        (const (symbol_ref "target_tune")))

  29071. 

  29072.  but without duplicating the processor list.  The second example defines

  29073. two separate C enums ('attr_arch' and 'attr_tune') whereas the first

  29074. defines a single C enum ('processor').

  29075. 

  29076. 

  29077. File: gccint.info,  Node: Expressions,  Next: Tagging Insns,  Prev: Defining Attributes,  Up: Insn Attributes

  29078. 

  29079. 17.19.2 Attribute Expressions

  29080. -----------------------------

  29081. 

  29082. RTL expressions used to define attributes use the codes described above

  29083. plus a few specific to attribute definitions, to be discussed below.

  29084. Attribute value expressions must have one of the following forms:

  29085. 

  29086. '(const_int I)'

  29087.      The integer I specifies the value of a numeric attribute.  I must

  29088.      be non-negative.

  29089. 

  29090.      The value of a numeric attribute can be specified either with a

  29091.      'const_int', or as an integer represented as a string in

  29092.      'const_string', 'eq_attr' (see below), 'attr', 'symbol_ref', simple

  29093.      arithmetic expressions, and 'set_attr' overrides on specific

  29094.      instructions (*note Tagging Insns::).

  29095. 

  29096. '(const_string VALUE)'

  29097.      The string VALUE specifies a constant attribute value.  If VALUE is

  29098.      specified as '"*"', it means that the default value of the

  29099.      attribute is to be used for the insn containing this expression.

  29100.      '"*"' obviously cannot be used in the DEFAULT expression of a

  29101.      'define_attr'.

  29102. 

  29103.      If the attribute whose value is being specified is numeric, VALUE

  29104.      must be a string containing a non-negative integer (normally

  29105.      'const_int' would be used in this case).  Otherwise, it must

  29106.      contain one of the valid values for the attribute.

  29107. 

  29108. '(if_then_else TEST TRUE-VALUE FALSE-VALUE)'

  29109.      TEST specifies an attribute test, whose format is defined below.

  29110.      The value of this expression is TRUE-VALUE if TEST is true,

  29111.      otherwise it is FALSE-VALUE.

  29112. 

  29113. '(cond [TEST1 VALUE1 ...] DEFAULT)'

  29114.      The first operand of this expression is a vector containing an even

  29115.      number of expressions and consisting of pairs of TEST and VALUE

  29116.      expressions.  The value of the 'cond' expression is that of the

  29117.      VALUE corresponding to the first true TEST expression.  If none of

  29118.      the TEST expressions are true, the value of the 'cond' expression

  29119.      is that of the DEFAULT expression.

  29120. 

  29121.  TEST expressions can have one of the following forms:

  29122. 

  29123. '(const_int I)'

  29124.      This test is true if I is nonzero and false otherwise.

  29125. 

  29126. '(not TEST)'

  29127. '(ior TEST1 TEST2)'

  29128. '(and TEST1 TEST2)'

  29129.      These tests are true if the indicated logical function is true.

  29130. 

  29131. '(match_operand:M N PRED CONSTRAINTS)'

  29132.      This test is true if operand N of the insn whose attribute value is

  29133.      being determined has mode M (this part of the test is ignored if M

  29134.      is 'VOIDmode') and the function specified by the string PRED

  29135.      returns a nonzero value when passed operand N and mode M (this part

  29136.      of the test is ignored if PRED is the null string).

  29137. 

  29138.      The CONSTRAINTS operand is ignored and should be the null string.

  29139. 

  29140. '(match_test C-EXPR)'

  29141.      The test is true if C expression C-EXPR is true.  In non-constant

  29142.      attributes, C-EXPR has access to the following variables:

  29143. 

  29144.      INSN

  29145.           The rtl instruction under test.

  29146.      WHICH_ALTERNATIVE

  29147.           The 'define_insn' alternative that INSN matches.  *Note Output

  29148.           Statement::.

  29149.      OPERANDS

  29150.           An array of INSN's rtl operands.

  29151. 

  29152.      C-EXPR behaves like the condition in a C 'if' statement, so there

  29153.      is no need to explicitly convert the expression into a boolean 0 or

  29154.      1 value.  For example, the following two tests are equivalent:

  29155. 

  29156.           (match_test "x & 2")

  29157.           (match_test "(x & 2) != 0")

  29158. 

  29159. '(le ARITH1 ARITH2)'

  29160. '(leu ARITH1 ARITH2)'

  29161. '(lt ARITH1 ARITH2)'

  29162. '(ltu ARITH1 ARITH2)'

  29163. '(gt ARITH1 ARITH2)'

  29164. '(gtu ARITH1 ARITH2)'

  29165. '(ge ARITH1 ARITH2)'

  29166. '(geu ARITH1 ARITH2)'

  29167. '(ne ARITH1 ARITH2)'

  29168. '(eq ARITH1 ARITH2)'

  29169.      These tests are true if the indicated comparison of the two

  29170.      arithmetic expressions is true.  Arithmetic expressions are formed

  29171.      with 'plus', 'minus', 'mult', 'div', 'mod', 'abs', 'neg', 'and',

  29172.      'ior', 'xor', 'not', 'ashift', 'lshiftrt', and 'ashiftrt'

  29173.      expressions.

  29174. 

  29175.      'const_int' and 'symbol_ref' are always valid terms (*note Insn

  29176.      Lengths::,for additional forms).  'symbol_ref' is a string denoting

  29177.      a C expression that yields an 'int' when evaluated by the

  29178.      'get_attr_...' routine.  It should normally be a global variable.

  29179. 

  29180. '(eq_attr NAME VALUE)'

  29181.      NAME is a string specifying the name of an attribute.

  29182. 

  29183.      VALUE is a string that is either a valid value for attribute NAME,

  29184.      a comma-separated list of values, or '!' followed by a value or

  29185.      list.  If VALUE does not begin with a '!', this test is true if the

  29186.      value of the NAME attribute of the current insn is in the list

  29187.      specified by VALUE.  If VALUE begins with a '!', this test is true

  29188.      if the attribute's value is _not_ in the specified list.

  29189. 

  29190.      For example,

  29191. 

  29192.           (eq_attr "type" "load,store")

  29193. 

  29194.      is equivalent to

  29195. 

  29196.           (ior (eq_attr "type" "load") (eq_attr "type" "store"))

  29197. 

  29198.      If NAME specifies an attribute of 'alternative', it refers to the

  29199.      value of the compiler variable 'which_alternative' (*note Output

  29200.      Statement::) and the values must be small integers.  For example,

  29201. 

  29202.           (eq_attr "alternative" "2,3")

  29203. 

  29204.      is equivalent to

  29205. 

  29206.           (ior (eq (symbol_ref "which_alternative") (const_int 2))

  29207.                (eq (symbol_ref "which_alternative") (const_int 3)))

  29208. 

  29209.      Note that, for most attributes, an 'eq_attr' test is simplified in

  29210.      cases where the value of the attribute being tested is known for

  29211.      all insns matching a particular pattern.  This is by far the most

  29212.      common case.

  29213. 

  29214. '(attr_flag NAME)'

  29215.      The value of an 'attr_flag' expression is true if the flag

  29216.      specified by NAME is true for the 'insn' currently being scheduled.

  29217. 

  29218.      NAME is a string specifying one of a fixed set of flags to test.

  29219.      Test the flags 'forward' and 'backward' to determine the direction

  29220.      of a conditional branch.

  29221. 

  29222.      This example describes a conditional branch delay slot which can be

  29223.      nullified for forward branches that are taken (annul-true) or for

  29224.      backward branches which are not taken (annul-false).

  29225. 

  29226.           (define_delay (eq_attr "type" "cbranch")

  29227.             [(eq_attr "in_branch_delay" "true")

  29228.              (and (eq_attr "in_branch_delay" "true")

  29229.                   (attr_flag "forward"))

  29230.              (and (eq_attr "in_branch_delay" "true")

  29231.                   (attr_flag "backward"))])

  29232. 

  29233.      The 'forward' and 'backward' flags are false if the current 'insn'

  29234.      being scheduled is not a conditional branch.

  29235. 

  29236.      'attr_flag' is only used during delay slot scheduling and has no

  29237.      meaning to other passes of the compiler.

  29238. 

  29239. '(attr NAME)'

  29240.      The value of another attribute is returned.  This is most useful

  29241.      for numeric attributes, as 'eq_attr' and 'attr_flag' produce more

  29242.      efficient code for non-numeric attributes.

  29243. 

  29244. 

  29245. File: gccint.info,  Node: Tagging Insns,  Next: Attr Example,  Prev: Expressions,  Up: Insn Attributes

  29246. 

  29247. 17.19.3 Assigning Attribute Values to Insns

  29248. -------------------------------------------

  29249. 

  29250. The value assigned to an attribute of an insn is primarily determined by

  29251. which pattern is matched by that insn (or which 'define_peephole'

  29252. generated it).  Every 'define_insn' and 'define_peephole' can have an

  29253. optional last argument to specify the values of attributes for matching

  29254. insns.  The value of any attribute not specified in a particular insn is

  29255. set to the default value for that attribute, as specified in its

  29256. 'define_attr'.  Extensive use of default values for attributes permits

  29257. the specification of the values for only one or two attributes in the

  29258. definition of most insn patterns, as seen in the example in the next

  29259. section.

  29260. 

  29261.  The optional last argument of 'define_insn' and 'define_peephole' is a

  29262. vector of expressions, each of which defines the value for a single

  29263. attribute.  The most general way of assigning an attribute's value is to

  29264. use a 'set' expression whose first operand is an 'attr' expression

  29265. giving the name of the attribute being set.  The second operand of the

  29266. 'set' is an attribute expression (*note Expressions::) giving the value

  29267. of the attribute.

  29268. 

  29269.  When the attribute value depends on the 'alternative' attribute (i.e.,

  29270. which is the applicable alternative in the constraint of the insn), the

  29271. 'set_attr_alternative' expression can be used.  It allows the

  29272. specification of a vector of attribute expressions, one for each

  29273. alternative.

  29274. 

  29275.  When the generality of arbitrary attribute expressions is not required,

  29276. the simpler 'set_attr' expression can be used, which allows specifying a

  29277. string giving either a single attribute value or a list of attribute

  29278. values, one for each alternative.

  29279. 

  29280.  The form of each of the above specifications is shown below.  In each

  29281. case, NAME is a string specifying the attribute to be set.

  29282. 

  29283. '(set_attr NAME VALUE-STRING)'

  29284.      VALUE-STRING is either a string giving the desired attribute value,

  29285.      or a string containing a comma-separated list giving the values for

  29286.      succeeding alternatives.  The number of elements must match the

  29287.      number of alternatives in the constraint of the insn pattern.

  29288. 

  29289.      Note that it may be useful to specify '*' for some alternative, in

  29290.      which case the attribute will assume its default value for insns

  29291.      matching that alternative.

  29292. 

  29293. '(set_attr_alternative NAME [VALUE1 VALUE2 ...])'

  29294.      Depending on the alternative of the insn, the value will be one of

  29295.      the specified values.  This is a shorthand for using a 'cond' with

  29296.      tests on the 'alternative' attribute.

  29297. 

  29298. '(set (attr NAME) VALUE)'

  29299.      The first operand of this 'set' must be the special RTL expression

  29300.      'attr', whose sole operand is a string giving the name of the

  29301.      attribute being set.  VALUE is the value of the attribute.

  29302. 

  29303.  The following shows three different ways of representing the same

  29304. attribute value specification:

  29305. 

  29306.      (set_attr "type" "load,store,arith")

  29307. 

  29308.      (set_attr_alternative "type"

  29309.                            [(const_string "load") (const_string "store")

  29310.                             (const_string "arith")])

  29311. 

  29312.      (set (attr "type")

  29313.           (cond [(eq_attr "alternative" "1") (const_string "load")

  29314.                  (eq_attr "alternative" "2") (const_string "store")]

  29315.                 (const_string "arith")))

  29316. 

  29317.  The 'define_asm_attributes' expression provides a mechanism to specify

  29318. the attributes assigned to insns produced from an 'asm' statement.  It

  29319. has the form:

  29320. 

  29321.      (define_asm_attributes [ATTR-SETS])

  29322. 

  29323. where ATTR-SETS is specified the same as for both the 'define_insn' and

  29324. the 'define_peephole' expressions.

  29325. 

  29326.  These values will typically be the "worst case" attribute values.  For

  29327. example, they might indicate that the condition code will be clobbered.

  29328. 

  29329.  A specification for a 'length' attribute is handled specially.  The way

  29330. to compute the length of an 'asm' insn is to multiply the length

  29331. specified in the expression 'define_asm_attributes' by the number of

  29332. machine instructions specified in the 'asm' statement, determined by

  29333. counting the number of semicolons and newlines in the string.

  29334. Therefore, the value of the 'length' attribute specified in a

  29335. 'define_asm_attributes' should be the maximum possible length of a

  29336. single machine instruction.

  29337. 

  29338. 

  29339. File: gccint.info,  Node: Attr Example,  Next: Insn Lengths,  Prev: Tagging Insns,  Up: Insn Attributes

  29340. 

  29341. 17.19.4 Example of Attribute Specifications

  29342. -------------------------------------------

  29343. 

  29344. The judicious use of defaulting is important in the efficient use of

  29345. insn attributes.  Typically, insns are divided into "types" and an

  29346. attribute, customarily called 'type', is used to represent this value.

  29347. This attribute is normally used only to define the default value for

  29348. other attributes.  An example will clarify this usage.

  29349. 

  29350.  Assume we have a RISC machine with a condition code and in which only

  29351. full-word operations are performed in registers.  Let us assume that we

  29352. can divide all insns into loads, stores, (integer) arithmetic

  29353. operations, floating point operations, and branches.

  29354. 

  29355.  Here we will concern ourselves with determining the effect of an insn

  29356. on the condition code and will limit ourselves to the following possible

  29357. effects: The condition code can be set unpredictably (clobbered), not be

  29358. changed, be set to agree with the results of the operation, or only

  29359. changed if the item previously set into the condition code has been

  29360. modified.

  29361. 

  29362.  Here is part of a sample 'md' file for such a machine:

  29363. 

  29364.      (define_attr "type" "load,store,arith,fp,branch" (const_string "arith"))

  29365. 

  29366.      (define_attr "cc" "clobber,unchanged,set,change0"

  29367.                   (cond [(eq_attr "type" "load")

  29368.                              (const_string "change0")

  29369.                          (eq_attr "type" "store,branch")

  29370.                              (const_string "unchanged")

  29371.                          (eq_attr "type" "arith")

  29372.                              (if_then_else (match_operand:SI 0 "" "")

  29373.                                            (const_string "set")

  29374.                                            (const_string "clobber"))]

  29375.                         (const_string "clobber")))

  29376. 

  29377.      (define_insn ""

  29378.        [(set (match_operand:SI 0 "general_operand" "=r,r,m")

  29379.              (match_operand:SI 1 "general_operand" "r,m,r"))]

  29380.        ""

  29381.        "@

  29382.         move %0,%1

  29383.         load %0,%1

  29384.         store %0,%1"

  29385.        [(set_attr "type" "arith,load,store")])

  29386. 

  29387.  Note that we assume in the above example that arithmetic operations

  29388. performed on quantities smaller than a machine word clobber the

  29389. condition code since they will set the condition code to a value

  29390. corresponding to the full-word result.

  29391. 

  29392. 

  29393. File: gccint.info,  Node: Insn Lengths,  Next: Constant Attributes,  Prev: Attr Example,  Up: Insn Attributes

  29394. 

  29395. 17.19.5 Computing the Length of an Insn

  29396. ---------------------------------------

  29397. 

  29398. For many machines, multiple types of branch instructions are provided,

  29399. each for different length branch displacements.  In most cases, the

  29400. assembler will choose the correct instruction to use.  However, when the

  29401. assembler cannot do so, GCC can when a special attribute, the 'length'

  29402. attribute, is defined.  This attribute must be defined to have numeric

  29403. values by specifying a null string in its 'define_attr'.

  29404. 

  29405.  In the case of the 'length' attribute, two additional forms of

  29406. arithmetic terms are allowed in test expressions:

  29407. 

  29408. '(match_dup N)'

  29409.      This refers to the address of operand N of the current insn, which

  29410.      must be a 'label_ref'.

  29411. 

  29412. '(pc)'

  29413.      For non-branch instructions and backward branch instructions, this

  29414.      refers to the address of the current insn.  But for forward branch

  29415.      instructions, this refers to the address of the next insn, because

  29416.      the length of the current insn is to be computed.

  29417. 

  29418.  For normal insns, the length will be determined by value of the

  29419. 'length' attribute.  In the case of 'addr_vec' and 'addr_diff_vec' insn

  29420. patterns, the length is computed as the number of vectors multiplied by

  29421. the size of each vector.

  29422. 

  29423.  Lengths are measured in addressable storage units (bytes).

  29424. 

  29425.  Note that it is possible to call functions via the 'symbol_ref'

  29426. mechanism to compute the length of an insn.  However, if you use this

  29427. mechanism you must provide dummy clauses to express the maximum length

  29428. without using the function call.  You can an example of this in the 'pa'

  29429. machine description for the 'call_symref' pattern.

  29430. 

  29431.  The following macros can be used to refine the length computation:

  29432. 

  29433. 'ADJUST_INSN_LENGTH (INSN, LENGTH)'

  29434.      If defined, modifies the length assigned to instruction INSN as a

  29435.      function of the context in which it is used.  LENGTH is an lvalue

  29436.      that contains the initially computed length of the insn and should

  29437.      be updated with the correct length of the insn.

  29438. 

  29439.      This macro will normally not be required.  A case in which it is

  29440.      required is the ROMP.  On this machine, the size of an 'addr_vec'

  29441.      insn must be increased by two to compensate for the fact that

  29442.      alignment may be required.

  29443. 

  29444.  The routine that returns 'get_attr_length' (the value of the 'length'

  29445. attribute) can be used by the output routine to determine the form of

  29446. the branch instruction to be written, as the example below illustrates.

  29447. 

  29448.  As an example of the specification of variable-length branches,

  29449. consider the IBM 360.  If we adopt the convention that a register will

  29450. be set to the starting address of a function, we can jump to labels

  29451. within 4k of the start using a four-byte instruction.  Otherwise, we

  29452. need a six-byte sequence to load the address from memory and then branch

  29453. to it.

  29454. 

  29455.  On such a machine, a pattern for a branch instruction might be

  29456. specified as follows:

  29457. 

  29458.      (define_insn "jump"

  29459.        [(set (pc)

  29460.              (label_ref (match_operand 0 "" "")))]

  29461.        ""

  29462.      {

  29463.         return (get_attr_length (insn) == 4

  29464.                 ? "b %l0" : "l r15,=a(%l0); br r15");

  29465.      }

  29466.        [(set (attr "length")

  29467.              (if_then_else (lt (match_dup 0) (const_int 4096))

  29468.                            (const_int 4)

  29469.                            (const_int 6)))])

  29470. 

  29471. 

  29472. File: gccint.info,  Node: Constant Attributes,  Next: Mnemonic Attribute,  Prev: Insn Lengths,  Up: Insn Attributes

  29473. 

  29474. 17.19.6 Constant Attributes

  29475. ---------------------------

  29476. 

  29477. A special form of 'define_attr', where the expression for the default

  29478. value is a 'const' expression, indicates an attribute that is constant

  29479. for a given run of the compiler.  Constant attributes may be used to

  29480. specify which variety of processor is used.  For example,

  29481. 

  29482.      (define_attr "cpu" "m88100,m88110,m88000"

  29483.       (const

  29484.        (cond [(symbol_ref "TARGET_88100") (const_string "m88100")

  29485.               (symbol_ref "TARGET_88110") (const_string "m88110")]

  29486.              (const_string "m88000"))))

  29487. 

  29488.      (define_attr "memory" "fast,slow"

  29489.       (const

  29490.        (if_then_else (symbol_ref "TARGET_FAST_MEM")

  29491.                      (const_string "fast")

  29492.                      (const_string "slow"))))

  29493. 

  29494.  The routine generated for constant attributes has no parameters as it

  29495. does not depend on any particular insn.  RTL expressions used to define

  29496. the value of a constant attribute may use the 'symbol_ref' form, but may

  29497. not use either the 'match_operand' form or 'eq_attr' forms involving

  29498. insn attributes.

  29499. 

  29500. 

  29501. File: gccint.info,  Node: Mnemonic Attribute,  Next: Delay Slots,  Prev: Constant Attributes,  Up: Insn Attributes

  29502. 

  29503. 17.19.7 Mnemonic Attribute

  29504. --------------------------

  29505. 

  29506. The 'mnemonic' attribute is a string type attribute holding the

  29507. instruction mnemonic for an insn alternative.  The attribute values will

  29508. automatically be generated by the machine description parser if there is

  29509. an attribute definition in the md file:

  29510. 

  29511.      (define_attr "mnemonic" "unknown" (const_string "unknown"))

  29512. 

  29513.  The default value can be freely chosen as long as it does not collide

  29514. with any of the instruction mnemonics.  This value will be used whenever

  29515. the machine description parser is not able to determine the mnemonic

  29516. string.  This might be the case for output templates containing more

  29517. than a single instruction as in '"mvcle\t%0,%1,0\;jo\t.-4"'.

  29518. 

  29519.  The 'mnemonic' attribute set is not generated automatically if the

  29520. instruction string is generated via C code.

  29521. 

  29522.  An existing 'mnemonic' attribute set in an insn definition will not be

  29523. overriden by the md file parser.  That way it is possible to manually

  29524. set the instruction mnemonics for the cases where the md file parser

  29525. fails to determine it automatically.

  29526. 

  29527.  The 'mnemonic' attribute is useful for dealing with instruction

  29528. specific properties in the pipeline description without defining

  29529. additional insn attributes.

  29530. 

  29531.      (define_attr "ooo_expanded" ""

  29532.        (cond [(eq_attr "mnemonic" "dlr,dsgr,d,dsgf,stam,dsgfr,dlgr")

  29533.               (const_int 1)]

  29534.              (const_int 0)))

  29535. 

  29536. 

  29537. File: gccint.info,  Node: Delay Slots,  Next: Processor pipeline description,  Prev: Mnemonic Attribute,  Up: Insn Attributes

  29538. 

  29539. 17.19.8 Delay Slot Scheduling

  29540. -----------------------------

  29541. 

  29542. The insn attribute mechanism can be used to specify the requirements for

  29543. delay slots, if any, on a target machine.  An instruction is said to

  29544. require a "delay slot" if some instructions that are physically after

  29545. the instruction are executed as if they were located before it.  Classic

  29546. examples are branch and call instructions, which often execute the

  29547. following instruction before the branch or call is performed.

  29548. 

  29549.  On some machines, conditional branch instructions can optionally

  29550. "annul" instructions in the delay slot.  This means that the instruction

  29551. will not be executed for certain branch outcomes.  Both instructions

  29552. that annul if the branch is true and instructions that annul if the

  29553. branch is false are supported.

  29554. 

  29555.  Delay slot scheduling differs from instruction scheduling in that

  29556. determining whether an instruction needs a delay slot is dependent only

  29557. on the type of instruction being generated, not on data flow between the

  29558. instructions.  See the next section for a discussion of data-dependent

  29559. instruction scheduling.

  29560. 

  29561.  The requirement of an insn needing one or more delay slots is indicated

  29562. via the 'define_delay' expression.  It has the following form:

  29563. 

  29564.      (define_delay TEST

  29565.                    [DELAY-1 ANNUL-TRUE-1 ANNUL-FALSE-1

  29566.                     DELAY-2 ANNUL-TRUE-2 ANNUL-FALSE-2

  29567.                     ...])

  29568. 

  29569.  TEST is an attribute test that indicates whether this 'define_delay'

  29570. applies to a particular insn.  If so, the number of required delay slots

  29571. is determined by the length of the vector specified as the second

  29572. argument.  An insn placed in delay slot N must satisfy attribute test

  29573. DELAY-N.  ANNUL-TRUE-N is an attribute test that specifies which insns

  29574. may be annulled if the branch is true.  Similarly, ANNUL-FALSE-N

  29575. specifies which insns in the delay slot may be annulled if the branch is

  29576. false.  If annulling is not supported for that delay slot, '(nil)'

  29577. should be coded.

  29578. 

  29579.  For example, in the common case where branch and call insns require a

  29580. single delay slot, which may contain any insn other than a branch or

  29581. call, the following would be placed in the 'md' file:

  29582. 

  29583.      (define_delay (eq_attr "type" "branch,call")

  29584.                    [(eq_attr "type" "!branch,call") (nil) (nil)])

  29585. 

  29586.  Multiple 'define_delay' expressions may be specified.  In this case,

  29587. each such expression specifies different delay slot requirements and

  29588. there must be no insn for which tests in two 'define_delay' expressions

  29589. are both true.

  29590. 

  29591.  For example, if we have a machine that requires one delay slot for

  29592. branches but two for calls, no delay slot can contain a branch or call

  29593. insn, and any valid insn in the delay slot for the branch can be

  29594. annulled if the branch is true, we might represent this as follows:

  29595. 

  29596.      (define_delay (eq_attr "type" "branch")

  29597.         [(eq_attr "type" "!branch,call")

  29598.          (eq_attr "type" "!branch,call")

  29599.          (nil)])

  29600. 

  29601.      (define_delay (eq_attr "type" "call")

  29602.                    [(eq_attr "type" "!branch,call") (nil) (nil)

  29603.                     (eq_attr "type" "!branch,call") (nil) (nil)])

  29604. 

  29605. 

  29606. File: gccint.info,  Node: Processor pipeline description,  Prev: Delay Slots,  Up: Insn Attributes

  29607. 

  29608. 17.19.9 Specifying processor pipeline description

  29609. -------------------------------------------------

  29610. 

  29611. To achieve better performance, most modern processors (super-pipelined,

  29612. superscalar RISC, and VLIW processors) have many "functional units" on

  29613. which several instructions can be executed simultaneously.  An

  29614. instruction starts execution if its issue conditions are satisfied.  If

  29615. not, the instruction is stalled until its conditions are satisfied.

  29616. Such "interlock (pipeline) delay" causes interruption of the fetching of

  29617. successor instructions (or demands nop instructions, e.g. for some MIPS

  29618. processors).

  29619. 

  29620.  There are two major kinds of interlock delays in modern processors.

  29621. The first one is a data dependence delay determining "instruction

  29622. latency time".  The instruction execution is not started until all

  29623. source data have been evaluated by prior instructions (there are more

  29624. complex cases when the instruction execution starts even when the data

  29625. are not available but will be ready in given time after the instruction

  29626. execution start).  Taking the data dependence delays into account is

  29627. simple.  The data dependence (true, output, and anti-dependence) delay

  29628. between two instructions is given by a constant.  In most cases this

  29629. approach is adequate.  The second kind of interlock delays is a

  29630. reservation delay.  The reservation delay means that two instructions

  29631. under execution will be in need of shared processors resources, i.e.

  29632. buses, internal registers, and/or functional units, which are reserved

  29633. for some time.  Taking this kind of delay into account is complex

  29634. especially for modern RISC processors.

  29635. 

  29636.  The task of exploiting more processor parallelism is solved by an

  29637. instruction scheduler.  For a better solution to this problem, the

  29638. instruction scheduler has to have an adequate description of the

  29639. processor parallelism (or "pipeline description").  GCC machine

  29640. descriptions describe processor parallelism and functional unit

  29641. reservations for groups of instructions with the aid of "regular

  29642. expressions".

  29643. 

  29644.  The GCC instruction scheduler uses a "pipeline hazard recognizer" to

  29645. figure out the possibility of the instruction issue by the processor on

  29646. a given simulated processor cycle.  The pipeline hazard recognizer is

  29647. automatically generated from the processor pipeline description.  The

  29648. pipeline hazard recognizer generated from the machine description is

  29649. based on a deterministic finite state automaton (DFA): the instruction

  29650. issue is possible if there is a transition from one automaton state to

  29651. another one.  This algorithm is very fast, and furthermore, its speed is

  29652. not dependent on processor complexity(1).

  29653. 

  29654.  The rest of this section describes the directives that constitute an

  29655. automaton-based processor pipeline description.  The order of these

  29656. constructions within the machine description file is not important.

  29657. 

  29658.  The following optional construction describes names of automata

  29659. generated and used for the pipeline hazards recognition.  Sometimes the

  29660. generated finite state automaton used by the pipeline hazard recognizer

  29661. is large.  If we use more than one automaton and bind functional units

  29662. to the automata, the total size of the automata is usually less than the

  29663. size of the single automaton.  If there is no one such construction,

  29664. only one finite state automaton is generated.

  29665. 

  29666.      (define_automaton AUTOMATA-NAMES)

  29667. 

  29668.  AUTOMATA-NAMES is a string giving names of the automata.  The names are

  29669. separated by commas.  All the automata should have unique names.  The

  29670. automaton name is used in the constructions 'define_cpu_unit' and

  29671. 'define_query_cpu_unit'.

  29672. 

  29673.  Each processor functional unit used in the description of instruction

  29674. reservations should be described by the following construction.

  29675. 

  29676.      (define_cpu_unit UNIT-NAMES [AUTOMATON-NAME])

  29677. 

  29678.  UNIT-NAMES is a string giving the names of the functional units

  29679. separated by commas.  Don't use name 'nothing', it is reserved for other

  29680. goals.

  29681. 

  29682.  AUTOMATON-NAME is a string giving the name of the automaton with which

  29683. the unit is bound.  The automaton should be described in construction

  29684. 'define_automaton'.  You should give "automaton-name", if there is a

  29685. defined automaton.

  29686. 

  29687.  The assignment of units to automata are constrained by the uses of the

  29688. units in insn reservations.  The most important constraint is: if a unit

  29689. reservation is present on a particular cycle of an alternative for an

  29690. insn reservation, then some unit from the same automaton must be present

  29691. on the same cycle for the other alternatives of the insn reservation.

  29692. The rest of the constraints are mentioned in the description of the

  29693. subsequent constructions.

  29694. 

  29695.  The following construction describes CPU functional units analogously

  29696. to 'define_cpu_unit'.  The reservation of such units can be queried for

  29697. an automaton state.  The instruction scheduler never queries reservation

  29698. of functional units for given automaton state.  So as a rule, you don't

  29699. need this construction.  This construction could be used for future code

  29700. generation goals (e.g. to generate VLIW insn templates).

  29701. 

  29702.      (define_query_cpu_unit UNIT-NAMES [AUTOMATON-NAME])

  29703. 

  29704.  UNIT-NAMES is a string giving names of the functional units separated

  29705. by commas.

  29706. 

  29707.  AUTOMATON-NAME is a string giving the name of the automaton with which

  29708. the unit is bound.

  29709. 

  29710.  The following construction is the major one to describe pipeline

  29711. characteristics of an instruction.

  29712. 

  29713.      (define_insn_reservation INSN-NAME DEFAULT_LATENCY

  29714.                               CONDITION REGEXP)

  29715. 

  29716.  DEFAULT_LATENCY is a number giving latency time of the instruction.

  29717. There is an important difference between the old description and the

  29718. automaton based pipeline description.  The latency time is used for all

  29719. dependencies when we use the old description.  In the automaton based

  29720. pipeline description, the given latency time is only used for true

  29721. dependencies.  The cost of anti-dependencies is always zero and the cost

  29722. of output dependencies is the difference between latency times of the

  29723. producing and consuming insns (if the difference is negative, the cost

  29724. is considered to be zero).  You can always change the default costs for

  29725. any description by using the target hook 'TARGET_SCHED_ADJUST_COST'

  29726. (*note Scheduling::).

  29727. 

  29728.  INSN-NAME is a string giving the internal name of the insn.  The

  29729. internal names are used in constructions 'define_bypass' and in the

  29730. automaton description file generated for debugging.  The internal name

  29731. has nothing in common with the names in 'define_insn'.  It is a good

  29732. practice to use insn classes described in the processor manual.

  29733. 

  29734.  CONDITION defines what RTL insns are described by this construction.

  29735. You should remember that you will be in trouble if CONDITION for two or

  29736. more different 'define_insn_reservation' constructions is TRUE for an

  29737. insn.  In this case what reservation will be used for the insn is not

  29738. defined.  Such cases are not checked during generation of the pipeline

  29739. hazards recognizer because in general recognizing that two conditions

  29740. may have the same value is quite difficult (especially if the conditions

  29741. contain 'symbol_ref').  It is also not checked during the pipeline

  29742. hazard recognizer work because it would slow down the recognizer

  29743. considerably.

  29744. 

  29745.  REGEXP is a string describing the reservation of the cpu's functional

  29746. units by the instruction.  The reservations are described by a regular

  29747. expression according to the following syntax:

  29748. 

  29749.             regexp = regexp "," oneof

  29750.                    | oneof

  29751. 

  29752.             oneof = oneof "|" allof

  29753.                   | allof

  29754. 

  29755.             allof = allof "+" repeat

  29756.                   | repeat

  29757. 

  29758.             repeat = element "*" number

  29759.                    | element

  29760. 

  29761.             element = cpu_function_unit_name

  29762.                     | reservation_name

  29763.                     | result_name

  29764.                     | "nothing"

  29765.                     | "(" regexp ")"

  29766. 

  29767.    * ',' is used for describing the start of the next cycle in the

  29768.      reservation.

  29769. 

  29770.    * '|' is used for describing a reservation described by the first

  29771.      regular expression *or* a reservation described by the second

  29772.      regular expression *or* etc.

  29773. 

  29774.    * '+' is used for describing a reservation described by the first

  29775.      regular expression *and* a reservation described by the second

  29776.      regular expression *and* etc.

  29777. 

  29778.    * '*' is used for convenience and simply means a sequence in which

  29779.      the regular expression are repeated NUMBER times with cycle

  29780.      advancing (see ',').

  29781. 

  29782.    * 'cpu_function_unit_name' denotes reservation of the named

  29783.      functional unit.

  29784. 

  29785.    * 'reservation_name' -- see description of construction

  29786.      'define_reservation'.

  29787. 

  29788.    * 'nothing' denotes no unit reservations.

  29789. 

  29790.  Sometimes unit reservations for different insns contain common parts.

  29791. In such case, you can simplify the pipeline description by describing

  29792. the common part by the following construction

  29793. 

  29794.      (define_reservation RESERVATION-NAME REGEXP)

  29795. 

  29796.  RESERVATION-NAME is a string giving name of REGEXP.  Functional unit

  29797. names and reservation names are in the same name space.  So the

  29798. reservation names should be different from the functional unit names and

  29799. cannot be the reserved name 'nothing'.

  29800. 

  29801.  The following construction is used to describe exceptions in the

  29802. latency time for given instruction pair.  This is so called bypasses.

  29803. 

  29804.      (define_bypass NUMBER OUT_INSN_NAMES IN_INSN_NAMES

  29805.                     [GUARD])

  29806. 

  29807.  NUMBER defines when the result generated by the instructions given in

  29808. string OUT_INSN_NAMES will be ready for the instructions given in string

  29809. IN_INSN_NAMES.  Each of these strings is a comma-separated list of

  29810. filename-style globs and they refer to the names of

  29811. 'define_insn_reservation's.  For example:

  29812.      (define_bypass 1 "cpu1_load_*, cpu1_store_*" "cpu1_load_*")

  29813.  defines a bypass between instructions that start with 'cpu1_load_' or

  29814. 'cpu1_store_' and those that start with 'cpu1_load_'.

  29815. 

  29816.  GUARD is an optional string giving the name of a C function which

  29817. defines an additional guard for the bypass.  The function will get the

  29818. two insns as parameters.  If the function returns zero the bypass will

  29819. be ignored for this case.  The additional guard is necessary to

  29820. recognize complicated bypasses, e.g. when the consumer is only an

  29821. address of insn 'store' (not a stored value).

  29822. 

  29823.  If there are more one bypass with the same output and input insns, the

  29824. chosen bypass is the first bypass with a guard in description whose

  29825. guard function returns nonzero.  If there is no such bypass, then bypass

  29826. without the guard function is chosen.

  29827. 

  29828.  The following five constructions are usually used to describe VLIW

  29829. processors, or more precisely, to describe a placement of small

  29830. instructions into VLIW instruction slots.  They can be used for RISC

  29831. processors, too.

  29832. 

  29833.      (exclusion_set UNIT-NAMES UNIT-NAMES)

  29834.      (presence_set UNIT-NAMES PATTERNS)

  29835.      (final_presence_set UNIT-NAMES PATTERNS)

  29836.      (absence_set UNIT-NAMES PATTERNS)

  29837.      (final_absence_set UNIT-NAMES PATTERNS)

  29838. 

  29839.  UNIT-NAMES is a string giving names of functional units separated by

  29840. commas.

  29841. 

  29842.  PATTERNS is a string giving patterns of functional units separated by

  29843. comma.  Currently pattern is one unit or units separated by

  29844. white-spaces.

  29845. 

  29846.  The first construction ('exclusion_set') means that each functional

  29847. unit in the first string cannot be reserved simultaneously with a unit

  29848. whose name is in the second string and vice versa.  For example, the

  29849. construction is useful for describing processors (e.g. some SPARC

  29850. processors) with a fully pipelined floating point functional unit which

  29851. can execute simultaneously only single floating point insns or only

  29852. double floating point insns.

  29853. 

  29854.  The second construction ('presence_set') means that each functional

  29855. unit in the first string cannot be reserved unless at least one of

  29856. pattern of units whose names are in the second string is reserved.  This

  29857. is an asymmetric relation.  For example, it is useful for description

  29858. that VLIW 'slot1' is reserved after 'slot0' reservation.  We could

  29859. describe it by the following construction

  29860. 

  29861.      (presence_set "slot1" "slot0")

  29862. 

  29863.  Or 'slot1' is reserved only after 'slot0' and unit 'b0' reservation.

  29864. In this case we could write

  29865. 

  29866.      (presence_set "slot1" "slot0 b0")

  29867. 

  29868.  The third construction ('final_presence_set') is analogous to

  29869. 'presence_set'.  The difference between them is when checking is done.

  29870. When an instruction is issued in given automaton state reflecting all

  29871. current and planned unit reservations, the automaton state is changed.

  29872. The first state is a source state, the second one is a result state.

  29873. Checking for 'presence_set' is done on the source state reservation,

  29874. checking for 'final_presence_set' is done on the result reservation.

  29875. This construction is useful to describe a reservation which is actually

  29876. two subsequent reservations.  For example, if we use

  29877. 

  29878.      (presence_set "slot1" "slot0")

  29879. 

  29880.  the following insn will be never issued (because 'slot1' requires

  29881. 'slot0' which is absent in the source state).

  29882. 

  29883.      (define_reservation "insn_and_nop" "slot0 + slot1")

  29884. 

  29885.  but it can be issued if we use analogous 'final_presence_set'.

  29886. 

  29887.  The forth construction ('absence_set') means that each functional unit

  29888. in the first string can be reserved only if each pattern of units whose

  29889. names are in the second string is not reserved.  This is an asymmetric

  29890. relation (actually 'exclusion_set' is analogous to this one but it is

  29891. symmetric).  For example it might be useful in a VLIW description to say

  29892. that 'slot0' cannot be reserved after either 'slot1' or 'slot2' have

  29893. been reserved.  This can be described as:

  29894. 

  29895.      (absence_set "slot0" "slot1, slot2")

  29896. 

  29897.  Or 'slot2' cannot be reserved if 'slot0' and unit 'b0' are reserved or

  29898. 'slot1' and unit 'b1' are reserved.  In this case we could write

  29899. 

  29900.      (absence_set "slot2" "slot0 b0, slot1 b1")

  29901. 

  29902.  All functional units mentioned in a set should belong to the same

  29903. automaton.

  29904. 

  29905.  The last construction ('final_absence_set') is analogous to

  29906. 'absence_set' but checking is done on the result (state) reservation.

  29907. See comments for 'final_presence_set'.

  29908. 

  29909.  You can control the generator of the pipeline hazard recognizer with

  29910. the following construction.

  29911. 

  29912.      (automata_option OPTIONS)

  29913. 

  29914.  OPTIONS is a string giving options which affect the generated code.

  29915. Currently there are the following options:

  29916. 

  29917.    * "no-minimization" makes no minimization of the automaton.  This is

  29918.      only worth to do when we are debugging the description and need to

  29919.      look more accurately at reservations of states.

  29920. 

  29921.    * "time" means printing time statistics about the generation of

  29922.      automata.

  29923. 

  29924.    * "stats" means printing statistics about the generated automata such

  29925.      as the number of DFA states, NDFA states and arcs.

  29926. 

  29927.    * "v" means a generation of the file describing the result automata.

  29928.      The file has suffix '.dfa' and can be used for the description

  29929.      verification and debugging.

  29930. 

  29931.    * "w" means a generation of warning instead of error for non-critical

  29932.      errors.

  29933. 

  29934.    * "no-comb-vect" prevents the automaton generator from generating two

  29935.      data structures and comparing them for space efficiency.  Using a

  29936.      comb vector to represent transitions may be better, but it can be

  29937.      very expensive to construct.  This option is useful if the build

  29938.      process spends an unacceptably long time in genautomata.

  29939. 

  29940.    * "ndfa" makes nondeterministic finite state automata.  This affects

  29941.      the treatment of operator '|' in the regular expressions.  The

  29942.      usual treatment of the operator is to try the first alternative

  29943.      and, if the reservation is not possible, the second alternative.

  29944.      The nondeterministic treatment means trying all alternatives, some

  29945.      of them may be rejected by reservations in the subsequent insns.

  29946. 

  29947.    * "collapse-ndfa" modifies the behavior of the generator when

  29948.      producing an automaton.  An additional state transition to collapse

  29949.      a nondeterministic NDFA state to a deterministic DFA state is

  29950.      generated.  It can be triggered by passing 'const0_rtx' to

  29951.      state_transition.  In such an automaton, cycle advance transitions

  29952.      are available only for these collapsed states.  This option is

  29953.      useful for ports that want to use the 'ndfa' option, but also want

  29954.      to use 'define_query_cpu_unit' to assign units to insns issued in a

  29955.      cycle.

  29956. 

  29957.    * "progress" means output of a progress bar showing how many states

  29958.      were generated so far for automaton being processed.  This is

  29959.      useful during debugging a DFA description.  If you see too many

  29960.      generated states, you could interrupt the generator of the pipeline

  29961.      hazard recognizer and try to figure out a reason for generation of

  29962.      the huge automaton.

  29963. 

  29964.  As an example, consider a superscalar RISC machine which can issue

  29965. three insns (two integer insns and one floating point insn) on the cycle

  29966. but can finish only two insns.  To describe this, we define the

  29967. following functional units.

  29968. 

  29969.      (define_cpu_unit "i0_pipeline, i1_pipeline, f_pipeline")

  29970.      (define_cpu_unit "port0, port1")

  29971. 

  29972.  All simple integer insns can be executed in any integer pipeline and

  29973. their result is ready in two cycles.  The simple integer insns are

  29974. issued into the first pipeline unless it is reserved, otherwise they are

  29975. issued into the second pipeline.  Integer division and multiplication

  29976. insns can be executed only in the second integer pipeline and their

  29977. results are ready correspondingly in 9 and 4 cycles.  The integer

  29978. division is not pipelined, i.e. the subsequent integer division insn

  29979. cannot be issued until the current division insn finished.  Floating

  29980. point insns are fully pipelined and their results are ready in 3 cycles.

  29981. Where the result of a floating point insn is used by an integer insn, an

  29982. additional delay of one cycle is incurred.  To describe all of this we

  29983. could specify

  29984. 

  29985.      (define_cpu_unit "div")

  29986. 

  29987.      (define_insn_reservation "simple" 2 (eq_attr "type" "int")

  29988.                               "(i0_pipeline | i1_pipeline), (port0 | port1)")

  29989. 

  29990.      (define_insn_reservation "mult" 4 (eq_attr "type" "mult")

  29991.                               "i1_pipeline, nothing*2, (port0 | port1)")

  29992. 

  29993.      (define_insn_reservation "div" 9 (eq_attr "type" "div")

  29994.                               "i1_pipeline, div*7, div + (port0 | port1)")

  29995. 

  29996.      (define_insn_reservation "float" 3 (eq_attr "type" "float")

  29997.                               "f_pipeline, nothing, (port0 | port1))

  29998. 

  29999.      (define_bypass 4 "float" "simple,mult,div")

  30000. 

  30001.  To simplify the description we could describe the following reservation

  30002. 

  30003.      (define_reservation "finish" "port0|port1")

  30004. 

  30005.  and use it in all 'define_insn_reservation' as in the following

  30006. construction

  30007. 

  30008.      (define_insn_reservation "simple" 2 (eq_attr "type" "int")

  30009.                               "(i0_pipeline | i1_pipeline), finish")

  30010. 

  30011.    ---------- Footnotes ----------

  30012. 

  30013.    (1) However, the size of the automaton depends on processor

  30014. complexity.  To limit this effect, machine descriptions can split

  30015. orthogonal parts of the machine description among several automata: but

  30016. then, since each of these must be stepped independently, this does cause

  30017. a small decrease in the algorithm's performance.

  30018. 

  30019. 

  30020. File: gccint.info,  Node: Conditional Execution,  Next: Define Subst,  Prev: Insn Attributes,  Up: Machine Desc

  30021. 

  30022. 17.20 Conditional Execution

  30023. ===========================

  30024. 

  30025. A number of architectures provide for some form of conditional

  30026. execution, or predication.  The hallmark of this feature is the ability

  30027. to nullify most of the instructions in the instruction set.  When the

  30028. instruction set is large and not entirely symmetric, it can be quite

  30029. tedious to describe these forms directly in the '.md' file.  An

  30030. alternative is the 'define_cond_exec' template.

  30031. 

  30032.      (define_cond_exec

  30033.        [PREDICATE-PATTERN]

  30034.        "CONDITION"

  30035.        "OUTPUT-TEMPLATE"

  30036.        "OPTIONAL-INSN-ATTRIBUES")

  30037. 

  30038.  PREDICATE-PATTERN is the condition that must be true for the insn to be

  30039. executed at runtime and should match a relational operator.  One can use

  30040. 'match_operator' to match several relational operators at once.  Any

  30041. 'match_operand' operands must have no more than one alternative.

  30042. 

  30043.  CONDITION is a C expression that must be true for the generated pattern

  30044. to match.

  30045. 

  30046.  OUTPUT-TEMPLATE is a string similar to the 'define_insn' output

  30047. template (*note Output Template::), except that the '*' and '@' special

  30048. cases do not apply.  This is only useful if the assembly text for the

  30049. predicate is a simple prefix to the main insn.  In order to handle the

  30050. general case, there is a global variable 'current_insn_predicate' that

  30051. will contain the entire predicate if the current insn is predicated, and

  30052. will otherwise be 'NULL'.

  30053. 

  30054.  OPTIONAL-INSN-ATTRIBUTES is an optional vector of attributes that gets

  30055. appended to the insn attributes of the produced cond_exec rtx.  It can

  30056. be used to add some distinguishing attribute to cond_exec rtxs produced

  30057. that way.  An example usage would be to use this attribute in

  30058. conjunction with attributes on the main pattern to disable particular

  30059. alternatives under certain conditions.

  30060. 

  30061.  When 'define_cond_exec' is used, an implicit reference to the

  30062. 'predicable' instruction attribute is made.  *Note Insn Attributes::.

  30063. This attribute must be a boolean (i.e. have exactly two elements in its

  30064. LIST-OF-VALUES), with the possible values being 'no' and 'yes'.  The

  30065. default and all uses in the insns must be a simple constant, not a

  30066. complex expressions.  It may, however, depend on the alternative, by

  30067. using a comma-separated list of values.  If that is the case, the port

  30068. should also define an 'enabled' attribute (*note Disable Insn

  30069. Alternatives::), which should also allow only 'no' and 'yes' as its

  30070. values.

  30071. 

  30072.  For each 'define_insn' for which the 'predicable' attribute is true, a

  30073. new 'define_insn' pattern will be generated that matches a predicated

  30074. version of the instruction.  For example,

  30075. 

  30076.      (define_insn "addsi"

  30077.        [(set (match_operand:SI 0 "register_operand" "r")

  30078.              (plus:SI (match_operand:SI 1 "register_operand" "r")

  30079.                       (match_operand:SI 2 "register_operand" "r")))]

  30080.        "TEST1"

  30081.        "add %2,%1,%0")

  30082. 

  30083.      (define_cond_exec

  30084.        [(ne (match_operand:CC 0 "register_operand" "c")

  30085.             (const_int 0))]

  30086.        "TEST2"

  30087.        "(%0)")

  30088. 

  30089. generates a new pattern

  30090. 

  30091.      (define_insn ""

  30092.        [(cond_exec

  30093.           (ne (match_operand:CC 3 "register_operand" "c") (const_int 0))

  30094.           (set (match_operand:SI 0 "register_operand" "r")

  30095.                (plus:SI (match_operand:SI 1 "register_operand" "r")

  30096.                         (match_operand:SI 2 "register_operand" "r"))))]

  30097.        "(TEST2) && (TEST1)"

  30098.        "(%3) add %2,%1,%0")

  30099. 

  30100. 

  30101. File: gccint.info,  Node: Define Subst,  Next: Constant Definitions,  Prev: Conditional Execution,  Up: Machine Desc

  30102. 

  30103. 17.21 RTL Templates Transformations

  30104. ===================================

  30105. 

  30106. For some hardware architectures there are common cases when the RTL

  30107. templates for the instructions can be derived from the other RTL

  30108. templates using simple transformations.  E.g., 'i386.md' contains an RTL

  30109. template for the ordinary 'sub' instruction-- '*subsi_1', and for the

  30110. 'sub' instruction with subsequent zero-extension--'*subsi_1_zext'.  Such

  30111. cases can be easily implemented by a single meta-template capable of

  30112. generating a modified case based on the initial one:

  30113. 

  30114.      (define_subst "NAME"

  30115.        [INPUT-TEMPLATE]

  30116.        "CONDITION"

  30117.        [OUTPUT-TEMPLATE])

  30118.  INPUT-TEMPLATE is a pattern describing the source RTL template, which

  30119. will be transformed.

  30120. 

  30121.  CONDITION is a C expression that is conjunct with the condition from

  30122. the input-template to generate a condition to be used in the

  30123. output-template.

  30124. 

  30125.  OUTPUT-TEMPLATE is a pattern that will be used in the resulting

  30126. template.

  30127. 

  30128.  'define_subst' mechanism is tightly coupled with the notion of the

  30129. subst attribute (*note Subst Iterators::).  The use of 'define_subst' is

  30130. triggered by a reference to a subst attribute in the transforming RTL

  30131. template.  This reference initiates duplication of the source RTL

  30132. template and substitution of the attributes with their values.  The

  30133. source RTL template is left unchanged, while the copy is transformed by

  30134. 'define_subst'.  This transformation can fail in the case when the

  30135. source RTL template is not matched against the input-template of the

  30136. 'define_subst'.  In such case the copy is deleted.

  30137. 

  30138.  'define_subst' can be used only in 'define_insn' and 'define_expand',

  30139. it cannot be used in other expressions (e.g. in

  30140. 'define_insn_and_split').

  30141. 

  30142. * Menu:

  30143. 

  30144. * Define Subst Example::	    Example of 'define_subst' work.

  30145. * Define Subst Pattern Matching::   Process of template comparison.

  30146. * Define Subst Output Template::    Generation of output template.

  30147. 

  30148. 

  30149. File: gccint.info,  Node: Define Subst Example,  Next: Define Subst Pattern Matching,  Up: Define Subst

  30150. 

  30151. 17.21.1 'define_subst' Example

  30152. ------------------------------

  30153. 

  30154. To illustrate how 'define_subst' works, let us examine a simple template

  30155. transformation.

  30156. 

  30157.  Suppose there are two kinds of instructions: one that touches flags and

  30158. the other that does not.  The instructions of the second type could be

  30159. generated with the following 'define_subst':

  30160. 

  30161.      (define_subst "add_clobber_subst"

  30162.        [(set (match_operand:SI 0 "" "")

  30163.              (match_operand:SI 1 "" ""))]

  30164.        ""

  30165.        [(set (match_dup 0)

  30166.              (match_dup 1))

  30167.         (clobber (reg:CC FLAGS_REG))])

  30168. 

  30169.  This 'define_subst' can be applied to any RTL pattern containing 'set'

  30170. of mode SI and generates a copy with clobber when it is applied.

  30171. 

  30172.  Assume there is an RTL template for a 'max' instruction to be used in

  30173. 'define_subst' mentioned above:

  30174. 

  30175.      (define_insn "maxsi"

  30176.        [(set (match_operand:SI 0 "register_operand" "=r")

  30177.              (max:SI

  30178.                (match_operand:SI 1 "register_operand" "r")

  30179.                (match_operand:SI 2 "register_operand" "r")))]

  30180.        ""

  30181.        "max\t{%2, %1, %0|%0, %1, %2}"

  30182.       [...])

  30183. 

  30184.  To mark the RTL template for 'define_subst' application,

  30185. subst-attributes are used.  They should be declared in advance:

  30186. 

  30187.      (define_subst_attr "add_clobber_name" "add_clobber_subst" "_noclobber" "_clobber")

  30188. 

  30189.  Here 'add_clobber_name' is the attribute name, 'add_clobber_subst' is

  30190. the name of the corresponding 'define_subst', the third argument

  30191. ('_noclobber') is the attribute value that would be substituted into the

  30192. unchanged version of the source RTL template, and the last argument

  30193. ('_clobber') is the value that would be substituted into the second,

  30194. transformed, version of the RTL template.

  30195. 

  30196.  Once the subst-attribute has been defined, it should be used in RTL

  30197. templates which need to be processed by the 'define_subst'.  So, the

  30198. original RTL template should be changed:

  30199. 

  30200.      (define_insn "maxsi<add_clobber_name>"

  30201.        [(set (match_operand:SI 0 "register_operand" "=r")

  30202.              (max:SI

  30203.                (match_operand:SI 1 "register_operand" "r")

  30204.                (match_operand:SI 2 "register_operand" "r")))]

  30205.        ""

  30206.        "max\t{%2, %1, %0|%0, %1, %2}"

  30207.       [...])

  30208. 

  30209.  The result of the 'define_subst' usage would look like the following:

  30210. 

  30211.      (define_insn "maxsi_noclobber"

  30212.        [(set (match_operand:SI 0 "register_operand" "=r")

  30213.              (max:SI

  30214.                (match_operand:SI 1 "register_operand" "r")

  30215.                (match_operand:SI 2 "register_operand" "r")))]

  30216.        ""

  30217.        "max\t{%2, %1, %0|%0, %1, %2}"

  30218.       [...])

  30219.      (define_insn "maxsi_clobber"

  30220.        [(set (match_operand:SI 0 "register_operand" "=r")

  30221.              (max:SI

  30222.                (match_operand:SI 1 "register_operand" "r")

  30223.                (match_operand:SI 2 "register_operand" "r")))

  30224.         (clobber (reg:CC FLAGS_REG))]

  30225.        ""

  30226.        "max\t{%2, %1, %0|%0, %1, %2}"

  30227.       [...])

  30228. 

  30229. 

  30230. File: gccint.info,  Node: Define Subst Pattern Matching,  Next: Define Subst Output Template,  Prev: Define Subst Example,  Up: Define Subst

  30231. 

  30232. 17.21.2 Pattern Matching in 'define_subst'

  30233. ------------------------------------------

  30234. 

  30235. All expressions, allowed in 'define_insn' or 'define_expand', are

  30236. allowed in the input-template of 'define_subst', except 'match_par_dup',

  30237. 'match_scratch', 'match_parallel'.  The meanings of expressions in the

  30238. input-template were changed:

  30239. 

  30240.  'match_operand' matches any expression (possibly, a subtree in

  30241. RTL-template), if modes of the 'match_operand' and this expression are

  30242. the same, or mode of the 'match_operand' is 'VOIDmode', or this

  30243. expression is 'match_dup', 'match_op_dup'.  If the expression is

  30244. 'match_operand' too, and predicate of 'match_operand' from the input

  30245. pattern is not empty, then the predicates are compared.  That can be

  30246. used for more accurate filtering of accepted RTL-templates.

  30247. 

  30248.  'match_operator' matches common operators (like 'plus', 'minus'),

  30249. 'unspec', 'unspec_volatile' operators and 'match_operator's from the

  30250. original pattern if the modes match and 'match_operator' from the input

  30251. pattern has the same number of operands as the operator from the

  30252. original pattern.

  30253. 

  30254. 

  30255. File: gccint.info,  Node: Define Subst Output Template,  Prev: Define Subst Pattern Matching,  Up: Define Subst

  30256. 

  30257. 17.21.3 Generation of output template in 'define_subst'

  30258. -------------------------------------------------------

  30259. 

  30260. If all necessary checks for 'define_subst' application pass, a new

  30261. RTL-pattern, based on the output-template, is created to replace the old

  30262. template.  Like in input-patterns, meanings of some RTL expressions are

  30263. changed when they are used in output-patterns of a 'define_subst'.

  30264. Thus, 'match_dup' is used for copying the whole expression from the

  30265. original pattern, which matched corresponding 'match_operand' from the

  30266. input pattern.

  30267. 

  30268.  'match_dup N' is used in the output template to be replaced with the

  30269. expression from the original pattern, which matched 'match_operand N'

  30270. from the input pattern.  As a consequence, 'match_dup' cannot be used to

  30271. point to 'match_operand's from the output pattern, it should always

  30272. refer to a 'match_operand' from the input pattern.  If a 'match_dup N'

  30273. occurs more than once in the output template, its first occurrence is

  30274. replaced with the expression from the original pattern, and the

  30275. subsequent expressions are replaced with 'match_dup N', i.e., a

  30276. reference to the first expression.

  30277. 

  30278.  In the output template one can refer to the expressions from the

  30279. original pattern and create new ones.  For instance, some operands could

  30280. be added by means of standard 'match_operand'.

  30281. 

  30282.  After replacing 'match_dup' with some RTL-subtree from the original

  30283. pattern, it could happen that several 'match_operand's in the output

  30284. pattern have the same indexes.  It is unknown, how many and what indexes

  30285. would be used in the expression which would replace 'match_dup', so such

  30286. conflicts in indexes are inevitable.  To overcome this issue,

  30287. 'match_operands' and 'match_operators', which were introduced into the

  30288. output pattern, are renumerated when all 'match_dup's are replaced.

  30289. 

  30290.  Number of alternatives in 'match_operand's introduced into the output

  30291. template 'M' could differ from the number of alternatives in the

  30292. original pattern 'N', so in the resultant pattern there would be 'N*M'

  30293. alternatives.  Thus, constraints from the original pattern would be

  30294. duplicated 'N' times, constraints from the output pattern would be

  30295. duplicated 'M' times, producing all possible combinations.

  30296. 

  30297. 

  30298. File: gccint.info,  Node: Constant Definitions,  Next: Iterators,  Prev: Define Subst,  Up: Machine Desc

  30299. 

  30300. 17.22 Constant Definitions

  30301. ==========================

  30302. 

  30303. Using literal constants inside instruction patterns reduces legibility

  30304. and can be a maintenance problem.

  30305. 

  30306.  To overcome this problem, you may use the 'define_constants'

  30307. expression.  It contains a vector of name-value pairs.  From that point

  30308. on, wherever any of the names appears in the MD file, it is as if the

  30309. corresponding value had been written instead.  You may use

  30310. 'define_constants' multiple times; each appearance adds more constants

  30311. to the table.  It is an error to redefine a constant with a different

  30312. value.

  30313. 

  30314.  To come back to the a29k load multiple example, instead of

  30315. 

  30316.      (define_insn ""

  30317.        [(match_parallel 0 "load_multiple_operation"

  30318.           [(set (match_operand:SI 1 "gpc_reg_operand" "=r")

  30319.                 (match_operand:SI 2 "memory_operand" "m"))

  30320.            (use (reg:SI 179))

  30321.            (clobber (reg:SI 179))])]

  30322.        ""

  30323.        "loadm 0,0,%1,%2")

  30324. 

  30325.  You could write:

  30326. 

  30327.      (define_constants [

  30328.          (R_BP 177)

  30329.          (R_FC 178)

  30330.          (R_CR 179)

  30331.          (R_Q  180)

  30332.      ])

  30333. 

  30334.      (define_insn ""

  30335.        [(match_parallel 0 "load_multiple_operation"

  30336.           [(set (match_operand:SI 1 "gpc_reg_operand" "=r")

  30337.                 (match_operand:SI 2 "memory_operand" "m"))

  30338.            (use (reg:SI R_CR))

  30339.            (clobber (reg:SI R_CR))])]

  30340.        ""

  30341.        "loadm 0,0,%1,%2")

  30342. 

  30343.  The constants that are defined with a define_constant are also output

  30344. in the insn-codes.h header file as #defines.

  30345. 

  30346.  You can also use the machine description file to define enumerations.

  30347. Like the constants defined by 'define_constant', these enumerations are

  30348. visible to both the machine description file and the main C code.

  30349. 

  30350.  The syntax is as follows:

  30351. 

  30352.      (define_c_enum "NAME" [

  30353.        VALUE0

  30354.        VALUE1

  30355.        ...

  30356.        VALUEN

  30357.      ])

  30358. 

  30359.  This definition causes the equivalent of the following C code to appear

  30360. in 'insn-constants.h':

  30361. 

  30362.      enum NAME {

  30363.        VALUE0 = 0,

  30364.        VALUE1 = 1,

  30365.        ...

  30366.        VALUEN = N

  30367.      };

  30368.      #define NUM_CNAME_VALUES (N + 1)

  30369. 

  30370.  where CNAME is the capitalized form of NAME.  It also makes each VALUEI

  30371. available in the machine description file, just as if it had been

  30372. declared with:

  30373. 

  30374.      (define_constants [(VALUEI I)])

  30375. 

  30376.  Each VALUEI is usually an upper-case identifier and usually begins with

  30377. CNAME.

  30378. 

  30379.  You can split the enumeration definition into as many statements as you

  30380. like.  The above example is directly equivalent to:

  30381. 

  30382.      (define_c_enum "NAME" [VALUE0])

  30383.      (define_c_enum "NAME" [VALUE1])

  30384.      ...

  30385.      (define_c_enum "NAME" [VALUEN])

  30386. 

  30387.  Splitting the enumeration helps to improve the modularity of each

  30388. individual '.md' file.  For example, if a port defines its

  30389. synchronization instructions in a separate 'sync.md' file, it is

  30390. convenient to define all synchronization-specific enumeration values in

  30391. 'sync.md' rather than in the main '.md' file.

  30392. 

  30393.  Some enumeration names have special significance to GCC:

  30394. 

  30395. 'unspecv'

  30396.      If an enumeration called 'unspecv' is defined, GCC will use it when

  30397.      printing out 'unspec_volatile' expressions.  For example:

  30398. 

  30399.           (define_c_enum "unspecv" [

  30400.             UNSPECV_BLOCKAGE

  30401.           ])

  30402. 

  30403.      causes GCC to print '(unspec_volatile ... 0)' as:

  30404. 

  30405.           (unspec_volatile ... UNSPECV_BLOCKAGE)

  30406. 

  30407. 'unspec'

  30408.      If an enumeration called 'unspec' is defined, GCC will use it when

  30409.      printing out 'unspec' expressions.  GCC will also use it when

  30410.      printing out 'unspec_volatile' expressions unless an 'unspecv'

  30411.      enumeration is also defined.  You can therefore decide whether to

  30412.      keep separate enumerations for volatile and non-volatile

  30413.      expressions or whether to use the same enumeration for both.

  30414. 

  30415.  Another way of defining an enumeration is to use 'define_enum':

  30416. 

  30417.      (define_enum "NAME" [

  30418.        VALUE0

  30419.        VALUE1

  30420.        ...

  30421.        VALUEN

  30422.      ])

  30423. 

  30424.  This directive implies:

  30425. 

  30426.      (define_c_enum "NAME" [

  30427.        CNAME_CVALUE0

  30428.        CNAME_CVALUE1

  30429.        ...

  30430.        CNAME_CVALUEN

  30431.      ])

  30432. 

  30433.  where CVALUEI is the capitalized form of VALUEI.  However, unlike

  30434. 'define_c_enum', the enumerations defined by 'define_enum' can be used

  30435. in attribute specifications (*note define_enum_attr::).

  30436. 

  30437. 

  30438. File: gccint.info,  Node: Iterators,  Prev: Constant Definitions,  Up: Machine Desc

  30439. 

  30440. 17.23 Iterators

  30441. ===============

  30442. 

  30443. Ports often need to define similar patterns for more than one machine

  30444. mode or for more than one rtx code.  GCC provides some simple iterator

  30445. facilities to make this process easier.

  30446. 

  30447. * Menu:

  30448. 

  30449. * Mode Iterators::         Generating variations of patterns for different modes.

  30450. * Code Iterators::         Doing the same for codes.

  30451. * Int Iterators::          Doing the same for integers.

  30452. * Subst Iterators::	   Generating variations of patterns for define_subst.

  30453. * Parameterized Names::	   Specifying iterator values in C++ code.

  30454. 

  30455. 

  30456. File: gccint.info,  Node: Mode Iterators,  Next: Code Iterators,  Up: Iterators

  30457. 

  30458. 17.23.1 Mode Iterators

  30459. ----------------------

  30460. 

  30461. Ports often need to define similar patterns for two or more different

  30462. modes.  For example:

  30463. 

  30464.    * If a processor has hardware support for both single and double

  30465.      floating-point arithmetic, the 'SFmode' patterns tend to be very

  30466.      similar to the 'DFmode' ones.

  30467. 

  30468.    * If a port uses 'SImode' pointers in one configuration and 'DImode'

  30469.      pointers in another, it will usually have very similar 'SImode' and

  30470.      'DImode' patterns for manipulating pointers.

  30471. 

  30472.  Mode iterators allow several patterns to be instantiated from one '.md'

  30473. file template.  They can be used with any type of rtx-based construct,

  30474. such as a 'define_insn', 'define_split', or 'define_peephole2'.

  30475. 

  30476. * Menu:

  30477. 

  30478. * Defining Mode Iterators:: Defining a new mode iterator.

  30479. * Substitutions::           Combining mode iterators with substitutions

  30480. * Examples::                Examples

  30481. 

  30482. 

  30483. File: gccint.info,  Node: Defining Mode Iterators,  Next: Substitutions,  Up: Mode Iterators

  30484. 

  30485. 17.23.1.1 Defining Mode Iterators

  30486. .................................

  30487. 

  30488. The syntax for defining a mode iterator is:

  30489. 

  30490.      (define_mode_iterator NAME [(MODE1 "COND1") ... (MODEN "CONDN")])

  30491. 

  30492.  This allows subsequent '.md' file constructs to use the mode suffix

  30493. ':NAME'.  Every construct that does so will be expanded N times, once

  30494. with every use of ':NAME' replaced by ':MODE1', once with every use

  30495. replaced by ':MODE2', and so on.  In the expansion for a particular

  30496. MODEI, every C condition will also require that CONDI be true.

  30497. 

  30498.  For example:

  30499. 

  30500.      (define_mode_iterator P [(SI "Pmode == SImode") (DI "Pmode == DImode")])

  30501. 

  30502.  defines a new mode suffix ':P'.  Every construct that uses ':P' will be

  30503. expanded twice, once with every ':P' replaced by ':SI' and once with

  30504. every ':P' replaced by ':DI'.  The ':SI' version will only apply if

  30505. 'Pmode == SImode' and the ':DI' version will only apply if 'Pmode ==

  30506. DImode'.

  30507. 

  30508.  As with other '.md' conditions, an empty string is treated as "always

  30509. true".  '(MODE "")' can also be abbreviated to 'MODE'.  For example:

  30510. 

  30511.      (define_mode_iterator GPR [SI (DI "TARGET_64BIT")])

  30512. 

  30513.  means that the ':DI' expansion only applies if 'TARGET_64BIT' but that

  30514. the ':SI' expansion has no such constraint.

  30515. 

  30516.  Iterators are applied in the order they are defined.  This can be

  30517. significant if two iterators are used in a construct that requires

  30518. substitutions.  *Note Substitutions::.

  30519. 

  30520. 

  30521. File: gccint.info,  Node: Substitutions,  Next: Examples,  Prev: Defining Mode Iterators,  Up: Mode Iterators

  30522. 

  30523. 17.23.1.2 Substitution in Mode Iterators

  30524. ........................................

  30525. 

  30526. If an '.md' file construct uses mode iterators, each version of the

  30527. construct will often need slightly different strings or modes.  For

  30528. example:

  30529. 

  30530.    * When a 'define_expand' defines several 'addM3' patterns (*note

  30531.      Standard Names::), each expander will need to use the appropriate

  30532.      mode name for M.

  30533. 

  30534.    * When a 'define_insn' defines several instruction patterns, each

  30535.      instruction will often use a different assembler mnemonic.

  30536. 

  30537.    * When a 'define_insn' requires operands with different modes, using

  30538.      an iterator for one of the operand modes usually requires a

  30539.      specific mode for the other operand(s).

  30540. 

  30541.  GCC supports such variations through a system of "mode attributes".

  30542. There are two standard attributes: 'mode', which is the name of the mode

  30543. in lower case, and 'MODE', which is the same thing in upper case.  You

  30544. can define other attributes using:

  30545. 

  30546.      (define_mode_attr NAME [(MODE1 "VALUE1") ... (MODEN "VALUEN")])

  30547. 

  30548.  where NAME is the name of the attribute and VALUEI is the value

  30549. associated with MODEI.

  30550. 

  30551.  When GCC replaces some :ITERATOR with :MODE, it will scan each string

  30552. and mode in the pattern for sequences of the form '<ITERATOR:ATTR>',

  30553. where ATTR is the name of a mode attribute.  If the attribute is defined

  30554. for MODE, the whole '<...>' sequence will be replaced by the appropriate

  30555. attribute value.

  30556. 

  30557.  For example, suppose an '.md' file has:

  30558. 

  30559.      (define_mode_iterator P [(SI "Pmode == SImode") (DI "Pmode == DImode")])

  30560.      (define_mode_attr load [(SI "lw") (DI "ld")])

  30561. 

  30562.  If one of the patterns that uses ':P' contains the string

  30563. '"<P:load>\t%0,%1"', the 'SI' version of that pattern will use

  30564. '"lw\t%0,%1"' and the 'DI' version will use '"ld\t%0,%1"'.

  30565. 

  30566.  Here is an example of using an attribute for a mode:

  30567. 

  30568.      (define_mode_iterator LONG [SI DI])

  30569.      (define_mode_attr SHORT [(SI "HI") (DI "SI")])

  30570.      (define_insn ...

  30571.        (sign_extend:LONG (match_operand:<LONG:SHORT> ...)) ...)

  30572. 

  30573.  The 'ITERATOR:' prefix may be omitted, in which case the substitution

  30574. will be attempted for every iterator expansion.

  30575. 

  30576. 

  30577. File: gccint.info,  Node: Examples,  Prev: Substitutions,  Up: Mode Iterators

  30578. 

  30579. 17.23.1.3 Mode Iterator Examples

  30580. ................................

  30581. 

  30582. Here is an example from the MIPS port.  It defines the following modes

  30583. and attributes (among others):

  30584. 

  30585.      (define_mode_iterator GPR [SI (DI "TARGET_64BIT")])

  30586.      (define_mode_attr d [(SI "") (DI "d")])

  30587. 

  30588.  and uses the following template to define both 'subsi3' and 'subdi3':

  30589. 

  30590.      (define_insn "sub<mode>3"

  30591.        [(set (match_operand:GPR 0 "register_operand" "=d")

  30592.              (minus:GPR (match_operand:GPR 1 "register_operand" "d")

  30593.                         (match_operand:GPR 2 "register_operand" "d")))]

  30594.        ""

  30595.        "<d>subu\t%0,%1,%2"

  30596.        [(set_attr "type" "arith")

  30597.         (set_attr "mode" "<MODE>")])

  30598. 

  30599.  This is exactly equivalent to:

  30600. 

  30601.      (define_insn "subsi3"

  30602.        [(set (match_operand:SI 0 "register_operand" "=d")

  30603.              (minus:SI (match_operand:SI 1 "register_operand" "d")

  30604.                        (match_operand:SI 2 "register_operand" "d")))]

  30605.        ""

  30606.        "subu\t%0,%1,%2"

  30607.        [(set_attr "type" "arith")

  30608.         (set_attr "mode" "SI")])

  30609. 

  30610.      (define_insn "subdi3"

  30611.        [(set (match_operand:DI 0 "register_operand" "=d")

  30612.              (minus:DI (match_operand:DI 1 "register_operand" "d")

  30613.                        (match_operand:DI 2 "register_operand" "d")))]

  30614.        ""

  30615.        "dsubu\t%0,%1,%2"

  30616.        [(set_attr "type" "arith")

  30617.         (set_attr "mode" "DI")])

  30618. 

  30619. 

  30620. File: gccint.info,  Node: Code Iterators,  Next: Int Iterators,  Prev: Mode Iterators,  Up: Iterators

  30621. 

  30622. 17.23.2 Code Iterators

  30623. ----------------------

  30624. 

  30625. Code iterators operate in a similar way to mode iterators.  *Note Mode

  30626. Iterators::.

  30627. 

  30628.  The construct:

  30629. 

  30630.      (define_code_iterator NAME [(CODE1 "COND1") ... (CODEN "CONDN")])

  30631. 

  30632.  defines a pseudo rtx code NAME that can be instantiated as CODEI if

  30633. condition CONDI is true.  Each CODEI must have the same rtx format.

  30634. *Note RTL Classes::.

  30635. 

  30636.  As with mode iterators, each pattern that uses NAME will be expanded N

  30637. times, once with all uses of NAME replaced by CODE1, once with all uses

  30638. replaced by CODE2, and so on.  *Note Defining Mode Iterators::.

  30639. 

  30640.  It is possible to define attributes for codes as well as for modes.

  30641. There are two standard code attributes: 'code', the name of the code in

  30642. lower case, and 'CODE', the name of the code in upper case.  Other

  30643. attributes are defined using:

  30644. 

  30645.      (define_code_attr NAME [(CODE1 "VALUE1") ... (CODEN "VALUEN")])

  30646. 

  30647.  Instruction patterns can use code attributes as rtx codes, which can be

  30648. useful if two sets of codes act in tandem.  For example, the following

  30649. 'define_insn' defines two patterns, one calculating a signed absolute

  30650. difference and another calculating an unsigned absolute difference:

  30651. 

  30652.      (define_code_iterator any_max [smax umax])

  30653.      (define_code_attr paired_min [(smax "smin") (umax "umin")])

  30654.      (define_insn ...

  30655.        [(set (match_operand:SI 0 ...)

  30656.              (minus:SI (any_max:SI (match_operand:SI 1 ...)

  30657.                                    (match_operand:SI 2 ...))

  30658.                        (<paired_min>:SI (match_dup 1) (match_dup 2))))]

  30659.        ...)

  30660. 

  30661.  The signed version of the instruction uses 'smax' and 'smin' while the

  30662. unsigned version uses 'umax' and 'umin'.  There are no versions that

  30663. pair 'smax' with 'umin' or 'umax' with 'smin'.

  30664. 

  30665.  Here's an example of code iterators in action, taken from the MIPS

  30666. port:

  30667. 

  30668.      (define_code_iterator any_cond [unordered ordered unlt unge uneq ltgt unle ungt

  30669.                                      eq ne gt ge lt le gtu geu ltu leu])

  30670. 

  30671.      (define_expand "b<code>"

  30672.        [(set (pc)

  30673.              (if_then_else (any_cond:CC (cc0)

  30674.                                         (const_int 0))

  30675.                            (label_ref (match_operand 0 ""))

  30676.                            (pc)))]

  30677.        ""

  30678.      {

  30679.        gen_conditional_branch (operands, <CODE>);

  30680.        DONE;

  30681.      })

  30682. 

  30683.  This is equivalent to:

  30684. 

  30685.      (define_expand "bunordered"

  30686.        [(set (pc)

  30687.              (if_then_else (unordered:CC (cc0)

  30688.                                          (const_int 0))

  30689.                            (label_ref (match_operand 0 ""))

  30690.                            (pc)))]

  30691.        ""

  30692.      {

  30693.        gen_conditional_branch (operands, UNORDERED);

  30694.        DONE;

  30695.      })

  30696. 

  30697.      (define_expand "bordered"

  30698.        [(set (pc)

  30699.              (if_then_else (ordered:CC (cc0)

  30700.                                        (const_int 0))

  30701.                            (label_ref (match_operand 0 ""))

  30702.                            (pc)))]

  30703.        ""

  30704.      {

  30705.        gen_conditional_branch (operands, ORDERED);

  30706.        DONE;

  30707.      })

  30708. 

  30709.      ...

  30710. 

  30711. 

  30712. File: gccint.info,  Node: Int Iterators,  Next: Subst Iterators,  Prev: Code Iterators,  Up: Iterators

  30713. 

  30714. 17.23.3 Int Iterators

  30715. ---------------------

  30716. 

  30717. Int iterators operate in a similar way to code iterators.  *Note Code

  30718. Iterators::.

  30719. 

  30720.  The construct:

  30721. 

  30722.      (define_int_iterator NAME [(INT1 "COND1") ... (INTN "CONDN")])

  30723. 

  30724.  defines a pseudo integer constant NAME that can be instantiated as INTI

  30725. if condition CONDI is true.  Each INT must have the same rtx format.

  30726. *Note RTL Classes::.  Int iterators can appear in only those rtx fields

  30727. that have 'i' as the specifier.  This means that each INT has to be a

  30728. constant defined using define_constant or define_c_enum.

  30729. 

  30730.  As with mode and code iterators, each pattern that uses NAME will be

  30731. expanded N times, once with all uses of NAME replaced by INT1, once with

  30732. all uses replaced by INT2, and so on.  *Note Defining Mode Iterators::.

  30733. 

  30734.  It is possible to define attributes for ints as well as for codes and

  30735. modes.  Attributes are defined using:

  30736. 

  30737.      (define_int_attr NAME [(INT1 "VALUE1") ... (INTN "VALUEN")])

  30738. 

  30739.  Here's an example of int iterators in action, taken from the ARM port:

  30740. 

  30741.      (define_int_iterator QABSNEG [UNSPEC_VQABS UNSPEC_VQNEG])

  30742. 

  30743.      (define_int_attr absneg [(UNSPEC_VQABS "abs") (UNSPEC_VQNEG "neg")])

  30744. 

  30745.      (define_insn "neon_vq<absneg><mode>"

  30746.        [(set (match_operand:VDQIW 0 "s_register_operand" "=w")

  30747.      	(unspec:VDQIW [(match_operand:VDQIW 1 "s_register_operand" "w")

  30748.      		       (match_operand:SI 2 "immediate_operand" "i")]

  30749.      		      QABSNEG))]

  30750.        "TARGET_NEON"

  30751.        "vq<absneg>.<V_s_elem>\t%<V_reg>0, %<V_reg>1"

  30752.        [(set_attr "type" "neon_vqneg_vqabs")]

  30753.      )

  30754. 

  30755. 

  30756.  This is equivalent to:

  30757. 

  30758.      (define_insn "neon_vqabs<mode>"

  30759.        [(set (match_operand:VDQIW 0 "s_register_operand" "=w")

  30760.      	(unspec:VDQIW [(match_operand:VDQIW 1 "s_register_operand" "w")

  30761.      		       (match_operand:SI 2 "immediate_operand" "i")]

  30762.      		      UNSPEC_VQABS))]

  30763.        "TARGET_NEON"

  30764.        "vqabs.<V_s_elem>\t%<V_reg>0, %<V_reg>1"

  30765.        [(set_attr "type" "neon_vqneg_vqabs")]

  30766.      )

  30767. 

  30768.      (define_insn "neon_vqneg<mode>"

  30769.        [(set (match_operand:VDQIW 0 "s_register_operand" "=w")

  30770.      	(unspec:VDQIW [(match_operand:VDQIW 1 "s_register_operand" "w")

  30771.      		       (match_operand:SI 2 "immediate_operand" "i")]

  30772.      		      UNSPEC_VQNEG))]

  30773.        "TARGET_NEON"

  30774.        "vqneg.<V_s_elem>\t%<V_reg>0, %<V_reg>1"

  30775.        [(set_attr "type" "neon_vqneg_vqabs")]

  30776.      )

  30777. 

  30778. 

  30779. 

  30780. File: gccint.info,  Node: Subst Iterators,  Next: Parameterized Names,  Prev: Int Iterators,  Up: Iterators

  30781. 

  30782. 17.23.4 Subst Iterators

  30783. -----------------------

  30784. 

  30785. Subst iterators are special type of iterators with the following

  30786. restrictions: they could not be declared explicitly, they always have

  30787. only two values, and they do not have explicit dedicated name.

  30788. Subst-iterators are triggered only when corresponding subst-attribute is

  30789. used in RTL-pattern.

  30790. 

  30791.  Subst iterators transform templates in the following way: the templates

  30792. are duplicated, the subst-attributes in these templates are replaced

  30793. with the corresponding values, and a new attribute is implicitly added

  30794. to the given 'define_insn'/'define_expand'.  The name of the added

  30795. attribute matches the name of 'define_subst'.  Such attributes are

  30796. declared implicitly, and it is not allowed to have a 'define_attr' named

  30797. as a 'define_subst'.

  30798. 

  30799.  Each subst iterator is linked to a 'define_subst'.  It is declared

  30800. implicitly by the first appearance of the corresponding

  30801. 'define_subst_attr', and it is not allowed to define it explicitly.

  30802. 

  30803.  Declarations of subst-attributes have the following syntax:

  30804. 

  30805.      (define_subst_attr "NAME"

  30806.        "SUBST-NAME"

  30807.        "NO-SUBST-VALUE"

  30808.        "SUBST-APPLIED-VALUE")

  30809. 

  30810.  NAME is a string with which the given subst-attribute could be referred

  30811. to.

  30812. 

  30813.  SUBST-NAME shows which 'define_subst' should be applied to an

  30814. RTL-template if the given subst-attribute is present in the

  30815. RTL-template.

  30816. 

  30817.  NO-SUBST-VALUE is a value with which subst-attribute would be replaced

  30818. in the first copy of the original RTL-template.

  30819. 

  30820.  SUBST-APPLIED-VALUE is a value with which subst-attribute would be

  30821. replaced in the second copy of the original RTL-template.

  30822. 

  30823. 

  30824. File: gccint.info,  Node: Parameterized Names,  Prev: Subst Iterators,  Up: Iterators

  30825. 

  30826. 17.23.5 Parameterized Names

  30827. ---------------------------

  30828. 

  30829. Ports sometimes need to apply iterators using C++ code, in order to get

  30830. the code or RTL pattern for a specific instruction.  For example,

  30831. suppose we have the 'neon_vq<absneg><mode>' pattern given above:

  30832. 

  30833.      (define_int_iterator QABSNEG [UNSPEC_VQABS UNSPEC_VQNEG])

  30834. 

  30835.      (define_int_attr absneg [(UNSPEC_VQABS "abs") (UNSPEC_VQNEG "neg")])

  30836. 

  30837.      (define_insn "neon_vq<absneg><mode>"

  30838.        [(set (match_operand:VDQIW 0 "s_register_operand" "=w")

  30839.      	(unspec:VDQIW [(match_operand:VDQIW 1 "s_register_operand" "w")

  30840.      		       (match_operand:SI 2 "immediate_operand" "i")]

  30841.      		      QABSNEG))]

  30842.        ...

  30843.      )

  30844. 

  30845.  A port might need to generate this pattern for a variable 'QABSNEG'

  30846. value and a variable 'VDQIW' mode.  There are two ways of doing this.

  30847. The first is to build the rtx for the pattern directly from C++ code;

  30848. this is a valid technique and avoids any risk of combinatorial

  30849. explosion.  The second is to prefix the instruction name with the

  30850. special character '@', which tells GCC to generate the four additional

  30851. functions below.  In each case, NAME is the name of the instruction

  30852. without the leading '@' character, without the '<...>' placeholders, and

  30853. with any underscore before a '<...>' placeholder removed if keeping it

  30854. would lead to a double or trailing underscore.

  30855. 

  30856. 'insn_code maybe_code_for_NAME (I1, I2, ...)'

  30857.      See whether replacing the first '<...>' placeholder with iterator

  30858.      value I1, the second with iterator value I2, and so on, gives a

  30859.      valid instruction.  Return its code if so, otherwise return

  30860.      'CODE_FOR_nothing'.

  30861. 

  30862. 'insn_code code_for_NAME (I1, I2, ...)'

  30863.      Same, but abort the compiler if the requested instruction does not

  30864.      exist.

  30865. 

  30866. 'rtx maybe_gen_NAME (I1, I2, ..., OP0, OP1, ...)'

  30867.      Check for a valid instruction in the same way as

  30868.      'maybe_code_for_NAME'.  If the instruction exists, generate an

  30869.      instance of it using the operand values given by OP0, OP1, and so

  30870.      on, otherwise return null.

  30871. 

  30872. 'rtx gen_NAME (I1, I2, ..., OP0, OP1, ...)'

  30873.      Same, but abort the compiler if the requested instruction does not

  30874.      exist, or if the instruction generator invoked the 'FAIL' macro.

  30875. 

  30876.  For example, changing the pattern above to:

  30877. 

  30878.      (define_insn "@neon_vq<absneg><mode>"

  30879.        [(set (match_operand:VDQIW 0 "s_register_operand" "=w")

  30880.      	(unspec:VDQIW [(match_operand:VDQIW 1 "s_register_operand" "w")

  30881.      		       (match_operand:SI 2 "immediate_operand" "i")]

  30882.      		      QABSNEG))]

  30883.        ...

  30884.      )

  30885. 

  30886.  would define the same patterns as before, but in addition would

  30887. generate the four functions below:

  30888. 

  30889.      insn_code maybe_code_for_neon_vq (int, machine_mode);

  30890.      insn_code code_for_neon_vq (int, machine_mode);

  30891.      rtx maybe_gen_neon_vq (int, machine_mode, rtx, rtx, rtx);

  30892.      rtx gen_neon_vq (int, machine_mode, rtx, rtx, rtx);

  30893. 

  30894.  Calling 'code_for_neon_vq (UNSPEC_VQABS, V8QImode)' would then give

  30895. 'CODE_FOR_neon_vqabsv8qi'.

  30896. 

  30897.  It is possible to have multiple '@' patterns with the same name and

  30898. same types of iterator.  For example:

  30899. 

  30900.      (define_insn "@some_arithmetic_op<mode>"

  30901.        [(set (match_operand:INTEGER_MODES 0 "register_operand") ...)]

  30902.        ...

  30903.      )

  30904. 

  30905.      (define_insn "@some_arithmetic_op<mode>"

  30906.        [(set (match_operand:FLOAT_MODES 0 "register_operand") ...)]

  30907.        ...

  30908.      )

  30909. 

  30910.  would produce a single set of functions that handles both

  30911. 'INTEGER_MODES' and 'FLOAT_MODES'.

  30912. 

  30913.  It is also possible for these '@' patterns to have different numbers of

  30914. operands from each other.  For example, patterns with a binary rtl code

  30915. might take three operands (one output and two inputs) while patterns

  30916. with a ternary rtl code might take four operands (one output and three

  30917. inputs).  This combination would produce separate 'maybe_gen_NAME' and

  30918. 'gen_NAME' functions for each operand count, but it would still produce

  30919. a single 'maybe_code_for_NAME' and a single 'code_for_NAME'.

  30920. 

  30921. 

  30922. File: gccint.info,  Node: Target Macros,  Next: Host Config,  Prev: Machine Desc,  Up: Top

  30923. 

  30924. 18 Target Description Macros and Functions

  30925. ******************************************

  30926. 

  30927. In addition to the file 'MACHINE.md', a machine description includes a C

  30928. header file conventionally given the name 'MACHINE.h' and a C source

  30929. file named 'MACHINE.c'.  The header file defines numerous macros that

  30930. convey the information about the target machine that does not fit into

  30931. the scheme of the '.md' file.  The file 'tm.h' should be a link to

  30932. 'MACHINE.h'.  The header file 'config.h' includes 'tm.h' and most

  30933. compiler source files include 'config.h'.  The source file defines a

  30934. variable 'targetm', which is a structure containing pointers to

  30935. functions and data relating to the target machine.  'MACHINE.c' should

  30936. also contain their definitions, if they are not defined elsewhere in

  30937. GCC, and other functions called through the macros defined in the '.h'

  30938. file.

  30939. 

  30940. * Menu:

  30941. 

  30942. * Target Structure::    The 'targetm' variable.

  30943. * Driver::              Controlling how the driver runs the compilation passes.

  30944. * Run-time Target::     Defining '-m' options like '-m68000' and '-m68020'.

  30945. * Per-Function Data::   Defining data structures for per-function information.

  30946. * Storage Layout::      Defining sizes and alignments of data.

  30947. * Type Layout::         Defining sizes and properties of basic user data types.

  30948. * Registers::           Naming and describing the hardware registers.

  30949. * Register Classes::    Defining the classes of hardware registers.

  30950. * Stack and Calling::   Defining which way the stack grows and by how much.

  30951. * Varargs::             Defining the varargs macros.

  30952. * Trampolines::         Code set up at run time to enter a nested function.

  30953. * Library Calls::       Controlling how library routines are implicitly called.

  30954. * Addressing Modes::    Defining addressing modes valid for memory operands.

  30955. * Anchored Addresses::  Defining how '-fsection-anchors' should work.

  30956. * Condition Code::      Defining how insns update the condition code.

  30957. * Costs::               Defining relative costs of different operations.

  30958. * Scheduling::          Adjusting the behavior of the instruction scheduler.

  30959. * Sections::            Dividing storage into text, data, and other sections.

  30960. * PIC::                 Macros for position independent code.

  30961. * Assembler Format::    Defining how to write insns and pseudo-ops to output.

  30962. * Debugging Info::      Defining the format of debugging output.

  30963. * Floating Point::      Handling floating point for cross-compilers.

  30964. * Mode Switching::      Insertion of mode-switching instructions.

  30965. * Target Attributes::   Defining target-specific uses of '__attribute__'.

  30966. * Emulated TLS::        Emulated TLS support.

  30967. * MIPS Coprocessors::   MIPS coprocessor support and how to customize it.

  30968. * PCH Target::          Validity checking for precompiled headers.

  30969. * C++ ABI::             Controlling C++ ABI changes.

  30970. * D Language and ABI::  Controlling D ABI changes.

  30971. * Named Address Spaces:: Adding support for named address spaces

  30972. * Misc::                Everything else.

  30973. 

  30974. 

  30975. File: gccint.info,  Node: Target Structure,  Next: Driver,  Up: Target Macros

  30976. 

  30977. 18.1 The Global 'targetm' Variable

  30978. ==================================

  30979. 

  30980.  -- Variable: struct gcc_target targetm

  30981.      The target '.c' file must define the global 'targetm' variable

  30982.      which contains pointers to functions and data relating to the

  30983.      target machine.  The variable is declared in 'target.h';

  30984.      'target-def.h' defines the macro 'TARGET_INITIALIZER' which is used

  30985.      to initialize the variable, and macros for the default initializers

  30986.      for elements of the structure.  The '.c' file should override those

  30987.      macros for which the default definition is inappropriate.  For

  30988.      example:

  30989.           #include "target.h"

  30990.           #include "target-def.h"

  30991. 

  30992.           /* Initialize the GCC target structure.  */

  30993. 

  30994.           #undef TARGET_COMP_TYPE_ATTRIBUTES

  30995.           #define TARGET_COMP_TYPE_ATTRIBUTES MACHINE_comp_type_attributes

  30996. 

  30997.           struct gcc_target targetm = TARGET_INITIALIZER;

  30998. 

  30999.  Where a macro should be defined in the '.c' file in this manner to form

  31000. part of the 'targetm' structure, it is documented below as a "Target

  31001. Hook" with a prototype.  Many macros will change in future from being

  31002. defined in the '.h' file to being part of the 'targetm' structure.

  31003. 

  31004.  Similarly, there is a 'targetcm' variable for hooks that are specific

  31005. to front ends for C-family languages, documented as "C Target Hook".

  31006. This is declared in 'c-family/c-target.h', the initializer

  31007. 'TARGETCM_INITIALIZER' in 'c-family/c-target-def.h'.  If targets

  31008. initialize 'targetcm' themselves, they should set

  31009. 'target_has_targetcm=yes' in 'config.gcc'; otherwise a default

  31010. definition is used.

  31011. 

  31012.  Similarly, there is a 'targetm_common' variable for hooks that are

  31013. shared between the compiler driver and the compilers proper, documented

  31014. as "Common Target Hook".  This is declared in 'common/common-target.h',

  31015. the initializer 'TARGETM_COMMON_INITIALIZER' in

  31016. 'common/common-target-def.h'.  If targets initialize 'targetm_common'

  31017. themselves, they should set 'target_has_targetm_common=yes' in

  31018. 'config.gcc'; otherwise a default definition is used.

  31019. 

  31020.  Similarly, there is a 'targetdm' variable for hooks that are specific

  31021. to the D language front end, documented as "D Target Hook".  This is

  31022. declared in 'd/d-target.h', the initializer 'TARGETDM_INITIALIZER' in

  31023. 'd/d-target-def.h'.  If targets initialize 'targetdm' themselves, they

  31024. should set 'target_has_targetdm=yes' in 'config.gcc'; otherwise a

  31025. default definition is used.

  31026. 

  31027. 

  31028. File: gccint.info,  Node: Driver,  Next: Run-time Target,  Prev: Target Structure,  Up: Target Macros

  31029. 

  31030. 18.2 Controlling the Compilation Driver, 'gcc'

  31031. ==============================================

  31032. 

  31033. You can control the compilation driver.

  31034. 

  31035.  -- Macro: DRIVER_SELF_SPECS

  31036.      A list of specs for the driver itself.  It should be a suitable

  31037.      initializer for an array of strings, with no surrounding braces.

  31038. 

  31039.      The driver applies these specs to its own command line between

  31040.      loading default 'specs' files (but not command-line specified ones)

  31041.      and choosing the multilib directory or running any subcommands.  It

  31042.      applies them in the order given, so each spec can depend on the

  31043.      options added by earlier ones.  It is also possible to remove

  31044.      options using '%<OPTION' in the usual way.

  31045. 

  31046.      This macro can be useful when a port has several interdependent

  31047.      target options.  It provides a way of standardizing the command

  31048.      line so that the other specs are easier to write.

  31049. 

  31050.      Do not define this macro if it does not need to do anything.

  31051. 

  31052.  -- Macro: OPTION_DEFAULT_SPECS

  31053.      A list of specs used to support configure-time default options

  31054.      (i.e. '--with' options) in the driver.  It should be a suitable

  31055.      initializer for an array of structures, each containing two

  31056.      strings, without the outermost pair of surrounding braces.

  31057. 

  31058.      The first item in the pair is the name of the default.  This must

  31059.      match the code in 'config.gcc' for the target.  The second item is

  31060.      a spec to apply if a default with this name was specified.  The

  31061.      string '%(VALUE)' in the spec will be replaced by the value of the

  31062.      default everywhere it occurs.

  31063. 

  31064.      The driver will apply these specs to its own command line between

  31065.      loading default 'specs' files and processing 'DRIVER_SELF_SPECS',

  31066.      using the same mechanism as 'DRIVER_SELF_SPECS'.

  31067. 

  31068.      Do not define this macro if it does not need to do anything.

  31069. 

  31070.  -- Macro: CPP_SPEC

  31071.      A C string constant that tells the GCC driver program options to

  31072.      pass to CPP.  It can also specify how to translate options you give

  31073.      to GCC into options for GCC to pass to the CPP.

  31074. 

  31075.      Do not define this macro if it does not need to do anything.

  31076. 

  31077.  -- Macro: CPLUSPLUS_CPP_SPEC

  31078.      This macro is just like 'CPP_SPEC', but is used for C++, rather

  31079.      than C.  If you do not define this macro, then the value of

  31080.      'CPP_SPEC' (if any) will be used instead.

  31081. 

  31082.  -- Macro: CC1_SPEC

  31083.      A C string constant that tells the GCC driver program options to

  31084.      pass to 'cc1', 'cc1plus', 'f771', and the other language front

  31085.      ends.  It can also specify how to translate options you give to GCC

  31086.      into options for GCC to pass to front ends.

  31087. 

  31088.      Do not define this macro if it does not need to do anything.

  31089. 

  31090.  -- Macro: CC1PLUS_SPEC

  31091.      A C string constant that tells the GCC driver program options to

  31092.      pass to 'cc1plus'.  It can also specify how to translate options

  31093.      you give to GCC into options for GCC to pass to the 'cc1plus'.

  31094. 

  31095.      Do not define this macro if it does not need to do anything.  Note

  31096.      that everything defined in CC1_SPEC is already passed to 'cc1plus'

  31097.      so there is no need to duplicate the contents of CC1_SPEC in

  31098.      CC1PLUS_SPEC.

  31099. 

  31100.  -- Macro: ASM_SPEC

  31101.      A C string constant that tells the GCC driver program options to

  31102.      pass to the assembler.  It can also specify how to translate

  31103.      options you give to GCC into options for GCC to pass to the

  31104.      assembler.  See the file 'sun3.h' for an example of this.

  31105. 

  31106.      Do not define this macro if it does not need to do anything.

  31107. 

  31108.  -- Macro: ASM_FINAL_SPEC

  31109.      A C string constant that tells the GCC driver program how to run

  31110.      any programs which cleanup after the normal assembler.  Normally,

  31111.      this is not needed.  See the file 'mips.h' for an example of this.

  31112. 

  31113.      Do not define this macro if it does not need to do anything.

  31114. 

  31115.  -- Macro: AS_NEEDS_DASH_FOR_PIPED_INPUT

  31116.      Define this macro, with no value, if the driver should give the

  31117.      assembler an argument consisting of a single dash, '-', to instruct

  31118.      it to read from its standard input (which will be a pipe connected

  31119.      to the output of the compiler proper).  This argument is given

  31120.      after any '-o' option specifying the name of the output file.

  31121. 

  31122.      If you do not define this macro, the assembler is assumed to read

  31123.      its standard input if given no non-option arguments.  If your

  31124.      assembler cannot read standard input at all, use a '%{pipe:%e}'

  31125.      construct; see 'mips.h' for instance.

  31126. 

  31127.  -- Macro: LINK_SPEC

  31128.      A C string constant that tells the GCC driver program options to

  31129.      pass to the linker.  It can also specify how to translate options

  31130.      you give to GCC into options for GCC to pass to the linker.

  31131. 

  31132.      Do not define this macro if it does not need to do anything.

  31133. 

  31134.  -- Macro: LIB_SPEC

  31135.      Another C string constant used much like 'LINK_SPEC'.  The

  31136.      difference between the two is that 'LIB_SPEC' is used at the end of

  31137.      the command given to the linker.

  31138. 

  31139.      If this macro is not defined, a default is provided that loads the

  31140.      standard C library from the usual place.  See 'gcc.c'.

  31141. 

  31142.  -- Macro: LIBGCC_SPEC

  31143.      Another C string constant that tells the GCC driver program how and

  31144.      when to place a reference to 'libgcc.a' into the linker command

  31145.      line.  This constant is placed both before and after the value of

  31146.      'LIB_SPEC'.

  31147. 

  31148.      If this macro is not defined, the GCC driver provides a default

  31149.      that passes the string '-lgcc' to the linker.

  31150. 

  31151.  -- Macro: REAL_LIBGCC_SPEC

  31152.      By default, if 'ENABLE_SHARED_LIBGCC' is defined, the 'LIBGCC_SPEC'

  31153.      is not directly used by the driver program but is instead modified

  31154.      to refer to different versions of 'libgcc.a' depending on the

  31155.      values of the command line flags '-static', '-shared',

  31156.      '-static-libgcc', and '-shared-libgcc'.  On targets where these

  31157.      modifications are inappropriate, define 'REAL_LIBGCC_SPEC' instead.

  31158.      'REAL_LIBGCC_SPEC' tells the driver how to place a reference to

  31159.      'libgcc' on the link command line, but, unlike 'LIBGCC_SPEC', it is

  31160.      used unmodified.

  31161. 

  31162.  -- Macro: USE_LD_AS_NEEDED

  31163.      A macro that controls the modifications to 'LIBGCC_SPEC' mentioned

  31164.      in 'REAL_LIBGCC_SPEC'.  If nonzero, a spec will be generated that

  31165.      uses '--as-needed' or equivalent options and the shared 'libgcc' in

  31166.      place of the static exception handler library, when linking without

  31167.      any of '-static', '-static-libgcc', or '-shared-libgcc'.

  31168. 

  31169.  -- Macro: LINK_EH_SPEC

  31170.      If defined, this C string constant is added to 'LINK_SPEC'.  When

  31171.      'USE_LD_AS_NEEDED' is zero or undefined, it also affects the

  31172.      modifications to 'LIBGCC_SPEC' mentioned in 'REAL_LIBGCC_SPEC'.

  31173. 

  31174.  -- Macro: STARTFILE_SPEC

  31175.      Another C string constant used much like 'LINK_SPEC'.  The

  31176.      difference between the two is that 'STARTFILE_SPEC' is used at the

  31177.      very beginning of the command given to the linker.

  31178. 

  31179.      If this macro is not defined, a default is provided that loads the

  31180.      standard C startup file from the usual place.  See 'gcc.c'.

  31181. 

  31182.  -- Macro: ENDFILE_SPEC

  31183.      Another C string constant used much like 'LINK_SPEC'.  The

  31184.      difference between the two is that 'ENDFILE_SPEC' is used at the

  31185.      very end of the command given to the linker.

  31186. 

  31187.      Do not define this macro if it does not need to do anything.

  31188. 

  31189.  -- Macro: THREAD_MODEL_SPEC

  31190.      GCC '-v' will print the thread model GCC was configured to use.

  31191.      However, this doesn't work on platforms that are multilibbed on

  31192.      thread models, such as AIX 4.3.  On such platforms, define

  31193.      'THREAD_MODEL_SPEC' such that it evaluates to a string without

  31194.      blanks that names one of the recognized thread models.  '%*', the

  31195.      default value of this macro, will expand to the value of

  31196.      'thread_file' set in 'config.gcc'.

  31197. 

  31198.  -- Macro: SYSROOT_SUFFIX_SPEC

  31199.      Define this macro to add a suffix to the target sysroot when GCC is

  31200.      configured with a sysroot.  This will cause GCC to search for

  31201.      usr/lib, et al, within sysroot+suffix.

  31202. 

  31203.  -- Macro: SYSROOT_HEADERS_SUFFIX_SPEC

  31204.      Define this macro to add a headers_suffix to the target sysroot

  31205.      when GCC is configured with a sysroot.  This will cause GCC to pass

  31206.      the updated sysroot+headers_suffix to CPP, causing it to search for

  31207.      usr/include, et al, within sysroot+headers_suffix.

  31208. 

  31209.  -- Macro: EXTRA_SPECS

  31210.      Define this macro to provide additional specifications to put in

  31211.      the 'specs' file that can be used in various specifications like

  31212.      'CC1_SPEC'.

  31213. 

  31214.      The definition should be an initializer for an array of structures,

  31215.      containing a string constant, that defines the specification name,

  31216.      and a string constant that provides the specification.

  31217. 

  31218.      Do not define this macro if it does not need to do anything.

  31219. 

  31220.      'EXTRA_SPECS' is useful when an architecture contains several

  31221.      related targets, which have various '..._SPECS' which are similar

  31222.      to each other, and the maintainer would like one central place to

  31223.      keep these definitions.

  31224. 

  31225.      For example, the PowerPC System V.4 targets use 'EXTRA_SPECS' to

  31226.      define either '_CALL_SYSV' when the System V calling sequence is

  31227.      used or '_CALL_AIX' when the older AIX-based calling sequence is

  31228.      used.

  31229. 

  31230.      The 'config/rs6000/rs6000.h' target file defines:

  31231. 

  31232.           #define EXTRA_SPECS \

  31233.             { "cpp_sysv_default", CPP_SYSV_DEFAULT },

  31234. 

  31235.           #define CPP_SYS_DEFAULT ""

  31236. 

  31237.      The 'config/rs6000/sysv.h' target file defines:

  31238.           #undef CPP_SPEC

  31239.           #define CPP_SPEC \

  31240.           "%{posix: -D_POSIX_SOURCE } \

  31241.           %{mcall-sysv: -D_CALL_SYSV } \

  31242.           %{!mcall-sysv: %(cpp_sysv_default) } \

  31243.           %{msoft-float: -D_SOFT_FLOAT} %{mcpu=403: -D_SOFT_FLOAT}"

  31244. 

  31245.           #undef CPP_SYSV_DEFAULT

  31246.           #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"

  31247. 

  31248.      while the 'config/rs6000/eabiaix.h' target file defines

  31249.      'CPP_SYSV_DEFAULT' as:

  31250. 

  31251.           #undef CPP_SYSV_DEFAULT

  31252.           #define CPP_SYSV_DEFAULT "-D_CALL_AIX"

  31253. 

  31254.  -- Macro: LINK_LIBGCC_SPECIAL_1

  31255.      Define this macro if the driver program should find the library

  31256.      'libgcc.a'.  If you do not define this macro, the driver program

  31257.      will pass the argument '-lgcc' to tell the linker to do the search.

  31258. 

  31259.  -- Macro: LINK_GCC_C_SEQUENCE_SPEC

  31260.      The sequence in which libgcc and libc are specified to the linker.

  31261.      By default this is '%G %L %G'.

  31262. 

  31263.  -- Macro: POST_LINK_SPEC

  31264.      Define this macro to add additional steps to be executed after

  31265.      linker.  The default value of this macro is empty string.

  31266. 

  31267.  -- Macro: LINK_COMMAND_SPEC

  31268.      A C string constant giving the complete command line need to

  31269.      execute the linker.  When you do this, you will need to update your

  31270.      port each time a change is made to the link command line within

  31271.      'gcc.c'.  Therefore, define this macro only if you need to

  31272.      completely redefine the command line for invoking the linker and

  31273.      there is no other way to accomplish the effect you need.

  31274.      Overriding this macro may be avoidable by overriding

  31275.      'LINK_GCC_C_SEQUENCE_SPEC' instead.

  31276. 

  31277.  -- Common Target Hook: bool TARGET_ALWAYS_STRIP_DOTDOT

  31278.      True if '..' components should always be removed from directory

  31279.      names computed relative to GCC's internal directories, false

  31280.      (default) if such components should be preserved and directory

  31281.      names containing them passed to other tools such as the linker.

  31282. 

  31283.  -- Macro: MULTILIB_DEFAULTS

  31284.      Define this macro as a C expression for the initializer of an array

  31285.      of string to tell the driver program which options are defaults for

  31286.      this target and thus do not need to be handled specially when using

  31287.      'MULTILIB_OPTIONS'.

  31288. 

  31289.      Do not define this macro if 'MULTILIB_OPTIONS' is not defined in

  31290.      the target makefile fragment or if none of the options listed in

  31291.      'MULTILIB_OPTIONS' are set by default.  *Note Target Fragment::.

  31292. 

  31293.  -- Macro: RELATIVE_PREFIX_NOT_LINKDIR

  31294.      Define this macro to tell 'gcc' that it should only translate a

  31295.      '-B' prefix into a '-L' linker option if the prefix indicates an

  31296.      absolute file name.

  31297. 

  31298.  -- Macro: MD_EXEC_PREFIX

  31299.      If defined, this macro is an additional prefix to try after

  31300.      'STANDARD_EXEC_PREFIX'.  'MD_EXEC_PREFIX' is not searched when the

  31301.      compiler is built as a cross compiler.  If you define

  31302.      'MD_EXEC_PREFIX', then be sure to add it to the list of directories

  31303.      used to find the assembler in 'configure.ac'.

  31304. 

  31305.  -- Macro: STANDARD_STARTFILE_PREFIX

  31306.      Define this macro as a C string constant if you wish to override

  31307.      the standard choice of 'libdir' as the default prefix to try when

  31308.      searching for startup files such as 'crt0.o'.

  31309.      'STANDARD_STARTFILE_PREFIX' is not searched when the compiler is

  31310.      built as a cross compiler.

  31311. 

  31312.  -- Macro: STANDARD_STARTFILE_PREFIX_1

  31313.      Define this macro as a C string constant if you wish to override

  31314.      the standard choice of '/lib' as a prefix to try after the default

  31315.      prefix when searching for startup files such as 'crt0.o'.

  31316.      'STANDARD_STARTFILE_PREFIX_1' is not searched when the compiler is

  31317.      built as a cross compiler.

  31318. 

  31319.  -- Macro: STANDARD_STARTFILE_PREFIX_2

  31320.      Define this macro as a C string constant if you wish to override

  31321.      the standard choice of '/lib' as yet another prefix to try after

  31322.      the default prefix when searching for startup files such as

  31323.      'crt0.o'.  'STANDARD_STARTFILE_PREFIX_2' is not searched when the

  31324.      compiler is built as a cross compiler.

  31325. 

  31326.  -- Macro: MD_STARTFILE_PREFIX

  31327.      If defined, this macro supplies an additional prefix to try after

  31328.      the standard prefixes.  'MD_EXEC_PREFIX' is not searched when the

  31329.      compiler is built as a cross compiler.

  31330. 

  31331.  -- Macro: MD_STARTFILE_PREFIX_1

  31332.      If defined, this macro supplies yet another prefix to try after the

  31333.      standard prefixes.  It is not searched when the compiler is built

  31334.      as a cross compiler.

  31335. 

  31336.  -- Macro: INIT_ENVIRONMENT

  31337.      Define this macro as a C string constant if you wish to set

  31338.      environment variables for programs called by the driver, such as

  31339.      the assembler and loader.  The driver passes the value of this

  31340.      macro to 'putenv' to initialize the necessary environment

  31341.      variables.

  31342. 

  31343.  -- Macro: LOCAL_INCLUDE_DIR

  31344.      Define this macro as a C string constant if you wish to override

  31345.      the standard choice of '/usr/local/include' as the default prefix

  31346.      to try when searching for local header files.  'LOCAL_INCLUDE_DIR'

  31347.      comes before 'NATIVE_SYSTEM_HEADER_DIR' (set in 'config.gcc',

  31348.      normally '/usr/include') in the search order.

  31349. 

  31350.      Cross compilers do not search either '/usr/local/include' or its

  31351.      replacement.

  31352. 

  31353.  -- Macro: NATIVE_SYSTEM_HEADER_COMPONENT

  31354.      The "component" corresponding to 'NATIVE_SYSTEM_HEADER_DIR'.  See

  31355.      'INCLUDE_DEFAULTS', below, for the description of components.  If

  31356.      you do not define this macro, no component is used.

  31357. 

  31358.  -- Macro: INCLUDE_DEFAULTS

  31359.      Define this macro if you wish to override the entire default search

  31360.      path for include files.  For a native compiler, the default search

  31361.      path usually consists of 'GCC_INCLUDE_DIR', 'LOCAL_INCLUDE_DIR',

  31362.      'GPLUSPLUS_INCLUDE_DIR', and 'NATIVE_SYSTEM_HEADER_DIR'.  In

  31363.      addition, 'GPLUSPLUS_INCLUDE_DIR' and 'GCC_INCLUDE_DIR' are defined

  31364.      automatically by 'Makefile', and specify private search areas for

  31365.      GCC.  The directory 'GPLUSPLUS_INCLUDE_DIR' is used only for C++

  31366.      programs.

  31367. 

  31368.      The definition should be an initializer for an array of structures.

  31369.      Each array element should have four elements: the directory name (a

  31370.      string constant), the component name (also a string constant), a

  31371.      flag for C++-only directories, and a flag showing that the includes

  31372.      in the directory don't need to be wrapped in 'extern 'C'' when

  31373.      compiling C++.  Mark the end of the array with a null element.

  31374. 

  31375.      The component name denotes what GNU package the include file is

  31376.      part of, if any, in all uppercase letters.  For example, it might

  31377.      be 'GCC' or 'BINUTILS'.  If the package is part of a

  31378.      vendor-supplied operating system, code the component name as '0'.

  31379. 

  31380.      For example, here is the definition used for VAX/VMS:

  31381. 

  31382.           #define INCLUDE_DEFAULTS \

  31383.           {                                       \

  31384.             { "GNU_GXX_INCLUDE:", "G++", 1, 1},   \

  31385.             { "GNU_CC_INCLUDE:", "GCC", 0, 0},    \

  31386.             { "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0},  \

  31387.             { ".", 0, 0, 0},                      \

  31388.             { 0, 0, 0, 0}                         \

  31389.           }

  31390. 

  31391.  Here is the order of prefixes tried for exec files:

  31392. 

  31393.   1. Any prefixes specified by the user with '-B'.

  31394. 

  31395.   2. The environment variable 'GCC_EXEC_PREFIX' or, if 'GCC_EXEC_PREFIX'

  31396.      is not set and the compiler has not been installed in the

  31397.      configure-time PREFIX, the location in which the compiler has

  31398.      actually been installed.

  31399. 

  31400.   3. The directories specified by the environment variable

  31401.      'COMPILER_PATH'.

  31402. 

  31403.   4. The macro 'STANDARD_EXEC_PREFIX', if the compiler has been

  31404.      installed in the configured-time PREFIX.

  31405. 

  31406.   5. The location '/usr/libexec/gcc/', but only if this is a native

  31407.      compiler.

  31408. 

  31409.   6. The location '/usr/lib/gcc/', but only if this is a native

  31410.      compiler.

  31411. 

  31412.   7. The macro 'MD_EXEC_PREFIX', if defined, but only if this is a

  31413.      native compiler.

  31414. 

  31415.  Here is the order of prefixes tried for startfiles:

  31416. 

  31417.   1. Any prefixes specified by the user with '-B'.

  31418. 

  31419.   2. The environment variable 'GCC_EXEC_PREFIX' or its automatically

  31420.      determined value based on the installed toolchain location.

  31421. 

  31422.   3. The directories specified by the environment variable

  31423.      'LIBRARY_PATH' (or port-specific name; native only, cross compilers

  31424.      do not use this).

  31425. 

  31426.   4. The macro 'STANDARD_EXEC_PREFIX', but only if the toolchain is

  31427.      installed in the configured PREFIX or this is a native compiler.

  31428. 

  31429.   5. The location '/usr/lib/gcc/', but only if this is a native

  31430.      compiler.

  31431. 

  31432.   6. The macro 'MD_EXEC_PREFIX', if defined, but only if this is a

  31433.      native compiler.

  31434. 

  31435.   7. The macro 'MD_STARTFILE_PREFIX', if defined, but only if this is a

  31436.      native compiler, or we have a target system root.

  31437. 

  31438.   8. The macro 'MD_STARTFILE_PREFIX_1', if defined, but only if this is

  31439.      a native compiler, or we have a target system root.

  31440. 

  31441.   9. The macro 'STANDARD_STARTFILE_PREFIX', with any sysroot

  31442.      modifications.  If this path is relative it will be prefixed by

  31443.      'GCC_EXEC_PREFIX' and the machine suffix or 'STANDARD_EXEC_PREFIX'

  31444.      and the machine suffix.

  31445. 

  31446.   10. The macro 'STANDARD_STARTFILE_PREFIX_1', but only if this is a

  31447.      native compiler, or we have a target system root.  The default for

  31448.      this macro is '/lib/'.

  31449. 

  31450.   11. The macro 'STANDARD_STARTFILE_PREFIX_2', but only if this is a

  31451.      native compiler, or we have a target system root.  The default for

  31452.      this macro is '/usr/lib/'.

  31453. 

  31454. 

  31455. File: gccint.info,  Node: Run-time Target,  Next: Per-Function Data,  Prev: Driver,  Up: Target Macros

  31456. 

  31457. 18.3 Run-time Target Specification

  31458. ==================================

  31459. 

  31460. Here are run-time target specifications.

  31461. 

  31462.  -- Macro: TARGET_CPU_CPP_BUILTINS ()

  31463.      This function-like macro expands to a block of code that defines

  31464.      built-in preprocessor macros and assertions for the target CPU,

  31465.      using the functions 'builtin_define', 'builtin_define_std' and

  31466.      'builtin_assert'.  When the front end calls this macro it provides

  31467.      a trailing semicolon, and since it has finished command line option

  31468.      processing your code can use those results freely.

  31469. 

  31470.      'builtin_assert' takes a string in the form you pass to the

  31471.      command-line option '-A', such as 'cpu=mips', and creates the

  31472.      assertion.  'builtin_define' takes a string in the form accepted by

  31473.      option '-D' and unconditionally defines the macro.

  31474. 

  31475.      'builtin_define_std' takes a string representing the name of an

  31476.      object-like macro.  If it doesn't lie in the user's namespace,

  31477.      'builtin_define_std' defines it unconditionally.  Otherwise, it

  31478.      defines a version with two leading underscores, and another version

  31479.      with two leading and trailing underscores, and defines the original

  31480.      only if an ISO standard was not requested on the command line.  For

  31481.      example, passing 'unix' defines '__unix', '__unix__' and possibly

  31482.      'unix'; passing '_mips' defines '__mips', '__mips__' and possibly

  31483.      '_mips', and passing '_ABI64' defines only '_ABI64'.

  31484. 

  31485.      You can also test for the C dialect being compiled.  The variable

  31486.      'c_language' is set to one of 'clk_c', 'clk_cplusplus' or

  31487.      'clk_objective_c'.  Note that if we are preprocessing assembler,

  31488.      this variable will be 'clk_c' but the function-like macro

  31489.      'preprocessing_asm_p()' will return true, so you might want to

  31490.      check for that first.  If you need to check for strict ANSI, the

  31491.      variable 'flag_iso' can be used.  The function-like macro

  31492.      'preprocessing_trad_p()' can be used to check for traditional

  31493.      preprocessing.

  31494. 

  31495.  -- Macro: TARGET_OS_CPP_BUILTINS ()

  31496.      Similarly to 'TARGET_CPU_CPP_BUILTINS' but this macro is optional

  31497.      and is used for the target operating system instead.

  31498. 

  31499.  -- Macro: TARGET_OBJFMT_CPP_BUILTINS ()

  31500.      Similarly to 'TARGET_CPU_CPP_BUILTINS' but this macro is optional

  31501.      and is used for the target object format.  'elfos.h' uses this

  31502.      macro to define '__ELF__', so you probably do not need to define it

  31503.      yourself.

  31504. 

  31505.  -- Variable: extern int target_flags

  31506.      This variable is declared in 'options.h', which is included before

  31507.      any target-specific headers.

  31508. 

  31509.  -- Common Target Hook: int TARGET_DEFAULT_TARGET_FLAGS

  31510.      This variable specifies the initial value of 'target_flags'.  Its

  31511.      default setting is 0.

  31512. 

  31513.  -- Common Target Hook: bool TARGET_HANDLE_OPTION (struct gcc_options

  31514.           *OPTS, struct gcc_options *OPTS_SET, const struct

  31515.           cl_decoded_option *DECODED, location_t LOC)

  31516.      This hook is called whenever the user specifies one of the

  31517.      target-specific options described by the '.opt' definition files

  31518.      (*note Options::).  It has the opportunity to do some

  31519.      option-specific processing and should return true if the option is

  31520.      valid.  The default definition does nothing but return true.

  31521. 

  31522.      DECODED specifies the option and its arguments.  OPTS and OPTS_SET

  31523.      are the 'gcc_options' structures to be used for storing option

  31524.      state, and LOC is the location at which the option was passed

  31525.      ('UNKNOWN_LOCATION' except for options passed via attributes).

  31526. 

  31527.  -- C Target Hook: bool TARGET_HANDLE_C_OPTION (size_t CODE, const char

  31528.           *ARG, int VALUE)

  31529.      This target hook is called whenever the user specifies one of the

  31530.      target-specific C language family options described by the '.opt'

  31531.      definition files(*note Options::).  It has the opportunity to do

  31532.      some option-specific processing and should return true if the

  31533.      option is valid.  The arguments are like for

  31534.      'TARGET_HANDLE_OPTION'.  The default definition does nothing but

  31535.      return false.

  31536. 

  31537.      In general, you should use 'TARGET_HANDLE_OPTION' to handle

  31538.      options.  However, if processing an option requires routines that

  31539.      are only available in the C (and related language) front ends, then

  31540.      you should use 'TARGET_HANDLE_C_OPTION' instead.

  31541. 

  31542.  -- C Target Hook: tree TARGET_OBJC_CONSTRUCT_STRING_OBJECT (tree

  31543.           STRING)

  31544.      Targets may provide a string object type that can be used within

  31545.      and between C, C++ and their respective Objective-C dialects.  A

  31546.      string object might, for example, embed encoding and length

  31547.      information.  These objects are considered opaque to the compiler

  31548.      and handled as references.  An ideal implementation makes the

  31549.      composition of the string object match that of the Objective-C

  31550.      'NSString' ('NXString' for GNUStep), allowing efficient

  31551.      interworking between C-only and Objective-C code.  If a target

  31552.      implements string objects then this hook should return a reference

  31553.      to such an object constructed from the normal 'C' string

  31554.      representation provided in STRING.  At present, the hook is used by

  31555.      Objective-C only, to obtain a common-format string object when the

  31556.      target provides one.

  31557. 

  31558.  -- C Target Hook: void TARGET_OBJC_DECLARE_UNRESOLVED_CLASS_REFERENCE

  31559.           (const char *CLASSNAME)

  31560.      Declare that Objective C class CLASSNAME is referenced by the

  31561.      current TU.

  31562. 

  31563.  -- C Target Hook: void TARGET_OBJC_DECLARE_CLASS_DEFINITION (const char

  31564.           *CLASSNAME)

  31565.      Declare that Objective C class CLASSNAME is defined by the current

  31566.      TU.

  31567. 

  31568.  -- C Target Hook: bool TARGET_STRING_OBJECT_REF_TYPE_P (const_tree

  31569.           STRINGREF)

  31570.      If a target implements string objects then this hook should return

  31571.      'true' if STRINGREF is a valid reference to such an object.

  31572. 

  31573.  -- C Target Hook: void TARGET_CHECK_STRING_OBJECT_FORMAT_ARG (tree

  31574.           FORMAT_ARG, tree ARGS_LIST)

  31575.      If a target implements string objects then this hook should should

  31576.      provide a facility to check the function arguments in ARGS_LIST

  31577.      against the format specifiers in FORMAT_ARG where the type of

  31578.      FORMAT_ARG is one recognized as a valid string reference type.

  31579. 

  31580.  -- Target Hook: void TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE (void)

  31581.      This target function is similar to the hook

  31582.      'TARGET_OPTION_OVERRIDE' but is called when the optimize level is

  31583.      changed via an attribute or pragma or when it is reset at the end

  31584.      of the code affected by the attribute or pragma.  It is not called

  31585.      at the beginning of compilation when 'TARGET_OPTION_OVERRIDE' is

  31586.      called so if you want to perform these actions then, you should

  31587.      have 'TARGET_OPTION_OVERRIDE' call

  31588.      'TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE'.

  31589. 

  31590.  -- Macro: C_COMMON_OVERRIDE_OPTIONS

  31591.      This is similar to the 'TARGET_OPTION_OVERRIDE' hook but is only

  31592.      used in the C language frontends (C, Objective-C, C++,

  31593.      Objective-C++) and so can be used to alter option flag variables

  31594.      which only exist in those frontends.

  31595. 

  31596.  -- Common Target Hook: const struct default_options *

  31597.           TARGET_OPTION_OPTIMIZATION_TABLE

  31598.      Some machines may desire to change what optimizations are performed

  31599.      for various optimization levels.  This variable, if defined,

  31600.      describes options to enable at particular sets of optimization

  31601.      levels.  These options are processed once just after the

  31602.      optimization level is determined and before the remainder of the

  31603.      command options have been parsed, so may be overridden by other

  31604.      options passed explicitly.

  31605. 

  31606.      This processing is run once at program startup and when the

  31607.      optimization options are changed via '#pragma GCC optimize' or by

  31608.      using the 'optimize' attribute.

  31609. 

  31610.  -- Common Target Hook: void TARGET_OPTION_INIT_STRUCT (struct

  31611.           gcc_options *OPTS)

  31612.      Set target-dependent initial values of fields in OPTS.

  31613. 

  31614.  -- Macro: SWITCHABLE_TARGET

  31615.      Some targets need to switch between substantially different

  31616.      subtargets during compilation.  For example, the MIPS target has

  31617.      one subtarget for the traditional MIPS architecture and another for

  31618.      MIPS16.  Source code can switch between these two subarchitectures

  31619.      using the 'mips16' and 'nomips16' attributes.

  31620. 

  31621.      Such subtargets can differ in things like the set of available

  31622.      registers, the set of available instructions, the costs of various

  31623.      operations, and so on.  GCC caches a lot of this type of

  31624.      information in global variables, and recomputing them for each

  31625.      subtarget takes a significant amount of time.  The compiler

  31626.      therefore provides a facility for maintaining several versions of

  31627.      the global variables and quickly switching between them; see

  31628.      'target-globals.h' for details.

  31629. 

  31630.      Define this macro to 1 if your target needs this facility.  The

  31631.      default is 0.

  31632. 

  31633.  -- Target Hook: bool TARGET_FLOAT_EXCEPTIONS_ROUNDING_SUPPORTED_P

  31634.           (void)

  31635.      Returns true if the target supports IEEE 754 floating-point

  31636.      exceptions and rounding modes, false otherwise.  This is intended

  31637.      to relate to the 'float' and 'double' types, but not necessarily

  31638.      'long double'.  By default, returns true if the 'adddf3'

  31639.      instruction pattern is available and false otherwise, on the

  31640.      assumption that hardware floating point supports exceptions and

  31641.      rounding modes but software floating point does not.

  31642. 

  31643. 

  31644. File: gccint.info,  Node: Per-Function Data,  Next: Storage Layout,  Prev: Run-time Target,  Up: Target Macros

  31645. 

  31646. 18.4 Defining data structures for per-function information.

  31647. ===========================================================

  31648. 

  31649. If the target needs to store information on a per-function basis, GCC

  31650. provides a macro and a couple of variables to allow this.  Note, just

  31651. using statics to store the information is a bad idea, since GCC supports

  31652. nested functions, so you can be halfway through encoding one function

  31653. when another one comes along.

  31654. 

  31655.  GCC defines a data structure called 'struct function' which contains

  31656. all of the data specific to an individual function.  This structure

  31657. contains a field called 'machine' whose type is 'struct machine_function

  31658. *', which can be used by targets to point to their own specific data.

  31659. 

  31660.  If a target needs per-function specific data it should define the type

  31661. 'struct machine_function' and also the macro 'INIT_EXPANDERS'.  This

  31662. macro should be used to initialize the function pointer

  31663. 'init_machine_status'.  This pointer is explained below.

  31664. 

  31665.  One typical use of per-function, target specific data is to create an

  31666. RTX to hold the register containing the function's return address.  This

  31667. RTX can then be used to implement the '__builtin_return_address'

  31668. function, for level 0.

  31669. 

  31670.  Note--earlier implementations of GCC used a single data area to hold

  31671. all of the per-function information.  Thus when processing of a nested

  31672. function began the old per-function data had to be pushed onto a stack,

  31673. and when the processing was finished, it had to be popped off the stack.

  31674. GCC used to provide function pointers called 'save_machine_status' and

  31675. 'restore_machine_status' to handle the saving and restoring of the

  31676. target specific information.  Since the single data area approach is no

  31677. longer used, these pointers are no longer supported.

  31678. 

  31679.  -- Macro: INIT_EXPANDERS

  31680.      Macro called to initialize any target specific information.  This

  31681.      macro is called once per function, before generation of any RTL has

  31682.      begun.  The intention of this macro is to allow the initialization

  31683.      of the function pointer 'init_machine_status'.

  31684. 

  31685.  -- Variable: void (*)(struct function *) init_machine_status

  31686.      If this function pointer is non-'NULL' it will be called once per

  31687.      function, before function compilation starts, in order to allow the

  31688.      target to perform any target specific initialization of the 'struct

  31689.      function' structure.  It is intended that this would be used to

  31690.      initialize the 'machine' of that structure.

  31691. 

  31692.      'struct machine_function' structures are expected to be freed by

  31693.      GC.  Generally, any memory that they reference must be allocated by

  31694.      using GC allocation, including the structure itself.

  31695. 

  31696. 

  31697. File: gccint.info,  Node: Storage Layout,  Next: Type Layout,  Prev: Per-Function Data,  Up: Target Macros

  31698. 

  31699. 18.5 Storage Layout

  31700. ===================

  31701. 

  31702. Note that the definitions of the macros in this table which are sizes or

  31703. alignments measured in bits do not need to be constant.  They can be C

  31704. expressions that refer to static variables, such as the 'target_flags'.

  31705. *Note Run-time Target::.

  31706. 

  31707.  -- Macro: BITS_BIG_ENDIAN

  31708.      Define this macro to have the value 1 if the most significant bit

  31709.      in a byte has the lowest number; otherwise define it to have the

  31710.      value zero.  This means that bit-field instructions count from the

  31711.      most significant bit.  If the machine has no bit-field

  31712.      instructions, then this must still be defined, but it doesn't

  31713.      matter which value it is defined to.  This macro need not be a

  31714.      constant.

  31715. 

  31716.      This macro does not affect the way structure fields are packed into

  31717.      bytes or words; that is controlled by 'BYTES_BIG_ENDIAN'.

  31718. 

  31719.  -- Macro: BYTES_BIG_ENDIAN

  31720.      Define this macro to have the value 1 if the most significant byte

  31721.      in a word has the lowest number.  This macro need not be a

  31722.      constant.

  31723. 

  31724.  -- Macro: WORDS_BIG_ENDIAN

  31725.      Define this macro to have the value 1 if, in a multiword object,

  31726.      the most significant word has the lowest number.  This applies to

  31727.      both memory locations and registers; see 'REG_WORDS_BIG_ENDIAN' if

  31728.      the order of words in memory is not the same as the order in

  31729.      registers.  This macro need not be a constant.

  31730. 

  31731.  -- Macro: REG_WORDS_BIG_ENDIAN

  31732.      On some machines, the order of words in a multiword object differs

  31733.      between registers in memory.  In such a situation, define this

  31734.      macro to describe the order of words in a register.  The macro

  31735.      'WORDS_BIG_ENDIAN' controls the order of words in memory.

  31736. 

  31737.  -- Macro: FLOAT_WORDS_BIG_ENDIAN

  31738.      Define this macro to have the value 1 if 'DFmode', 'XFmode' or

  31739.      'TFmode' floating point numbers are stored in memory with the word

  31740.      containing the sign bit at the lowest address; otherwise define it

  31741.      to have the value 0.  This macro need not be a constant.

  31742. 

  31743.      You need not define this macro if the ordering is the same as for

  31744.      multi-word integers.

  31745. 

  31746.  -- Macro: BITS_PER_WORD

  31747.      Number of bits in a word.  If you do not define this macro, the

  31748.      default is 'BITS_PER_UNIT * UNITS_PER_WORD'.

  31749. 

  31750.  -- Macro: MAX_BITS_PER_WORD

  31751.      Maximum number of bits in a word.  If this is undefined, the

  31752.      default is 'BITS_PER_WORD'.  Otherwise, it is the constant value

  31753.      that is the largest value that 'BITS_PER_WORD' can have at

  31754.      run-time.

  31755. 

  31756.  -- Macro: UNITS_PER_WORD

  31757.      Number of storage units in a word; normally the size of a

  31758.      general-purpose register, a power of two from 1 or 8.

  31759. 

  31760.  -- Macro: MIN_UNITS_PER_WORD

  31761.      Minimum number of units in a word.  If this is undefined, the

  31762.      default is 'UNITS_PER_WORD'.  Otherwise, it is the constant value

  31763.      that is the smallest value that 'UNITS_PER_WORD' can have at

  31764.      run-time.

  31765. 

  31766.  -- Macro: POINTER_SIZE

  31767.      Width of a pointer, in bits.  You must specify a value no wider

  31768.      than the width of 'Pmode'.  If it is not equal to the width of

  31769.      'Pmode', you must define 'POINTERS_EXTEND_UNSIGNED'.  If you do not

  31770.      specify a value the default is 'BITS_PER_WORD'.

  31771. 

  31772.  -- Macro: POINTERS_EXTEND_UNSIGNED

  31773.      A C expression that determines how pointers should be extended from

  31774.      'ptr_mode' to either 'Pmode' or 'word_mode'.  It is greater than

  31775.      zero if pointers should be zero-extended, zero if they should be

  31776.      sign-extended, and negative if some other sort of conversion is

  31777.      needed.  In the last case, the extension is done by the target's

  31778.      'ptr_extend' instruction.

  31779. 

  31780.      You need not define this macro if the 'ptr_mode', 'Pmode' and

  31781.      'word_mode' are all the same width.

  31782. 

  31783.  -- Macro: PROMOTE_MODE (M, UNSIGNEDP, TYPE)

  31784.      A macro to update M and UNSIGNEDP when an object whose type is TYPE

  31785.      and which has the specified mode and signedness is to be stored in

  31786.      a register.  This macro is only called when TYPE is a scalar type.

  31787. 

  31788.      On most RISC machines, which only have operations that operate on a

  31789.      full register, define this macro to set M to 'word_mode' if M is an

  31790.      integer mode narrower than 'BITS_PER_WORD'.  In most cases, only

  31791.      integer modes should be widened because wider-precision

  31792.      floating-point operations are usually more expensive than their

  31793.      narrower counterparts.

  31794. 

  31795.      For most machines, the macro definition does not change UNSIGNEDP.

  31796.      However, some machines, have instructions that preferentially

  31797.      handle either signed or unsigned quantities of certain modes.  For

  31798.      example, on the DEC Alpha, 32-bit loads from memory and 32-bit add

  31799.      instructions sign-extend the result to 64 bits.  On such machines,

  31800.      set UNSIGNEDP according to which kind of extension is more

  31801.      efficient.

  31802. 

  31803.      Do not define this macro if it would never modify M.

  31804. 

  31805.  -- Target Hook: enum flt_eval_method TARGET_C_EXCESS_PRECISION (enum

  31806.           excess_precision_type TYPE)

  31807.      Return a value, with the same meaning as the C99 macro

  31808.      'FLT_EVAL_METHOD' that describes which excess precision should be

  31809.      applied.  TYPE is either 'EXCESS_PRECISION_TYPE_IMPLICIT',

  31810.      'EXCESS_PRECISION_TYPE_FAST', or 'EXCESS_PRECISION_TYPE_STANDARD'.

  31811.      For 'EXCESS_PRECISION_TYPE_IMPLICIT', the target should return

  31812.      which precision and range operations will be implictly evaluated in

  31813.      regardless of the excess precision explicitly added.  For

  31814.      'EXCESS_PRECISION_TYPE_STANDARD' and 'EXCESS_PRECISION_TYPE_FAST',

  31815.      the target should return the explicit excess precision that should

  31816.      be added depending on the value set for

  31817.      '-fexcess-precision=[standard|fast]'.  Note that unpredictable

  31818.      explicit excess precision does not make sense, so a target should

  31819.      never return 'FLT_EVAL_METHOD_UNPREDICTABLE' when TYPE is

  31820.      'EXCESS_PRECISION_TYPE_STANDARD' or 'EXCESS_PRECISION_TYPE_FAST'.

  31821. 

  31822.  -- Target Hook: machine_mode TARGET_PROMOTE_FUNCTION_MODE (const_tree

  31823.           TYPE, machine_mode MODE, int *PUNSIGNEDP, const_tree FUNTYPE,

  31824.           int FOR_RETURN)

  31825.      Like 'PROMOTE_MODE', but it is applied to outgoing function

  31826.      arguments or function return values.  The target hook should return

  31827.      the new mode and possibly change '*PUNSIGNEDP' if the promotion

  31828.      should change signedness.  This function is called only for scalar

  31829.      _or pointer_ types.

  31830. 

  31831.      FOR_RETURN allows to distinguish the promotion of arguments and

  31832.      return values.  If it is '1', a return value is being promoted and

  31833.      'TARGET_FUNCTION_VALUE' must perform the same promotions done here.

  31834.      If it is '2', the returned mode should be that of the register in

  31835.      which an incoming parameter is copied, or the outgoing result is

  31836.      computed; then the hook should return the same mode as

  31837.      'promote_mode', though the signedness may be different.

  31838. 

  31839.      TYPE can be NULL when promoting function arguments of libcalls.

  31840. 

  31841.      The default is to not promote arguments and return values.  You can

  31842.      also define the hook to

  31843.      'default_promote_function_mode_always_promote' if you would like to

  31844.      apply the same rules given by 'PROMOTE_MODE'.

  31845. 

  31846.  -- Macro: PARM_BOUNDARY

  31847.      Normal alignment required for function parameters on the stack, in

  31848.      bits.  All stack parameters receive at least this much alignment

  31849.      regardless of data type.  On most machines, this is the same as the

  31850.      size of an integer.

  31851. 

  31852.  -- Macro: STACK_BOUNDARY

  31853.      Define this macro to the minimum alignment enforced by hardware for

  31854.      the stack pointer on this machine.  The definition is a C

  31855.      expression for the desired alignment (measured in bits).  This

  31856.      value is used as a default if 'PREFERRED_STACK_BOUNDARY' is not

  31857.      defined.  On most machines, this should be the same as

  31858.      'PARM_BOUNDARY'.

  31859. 

  31860.  -- Macro: PREFERRED_STACK_BOUNDARY

  31861.      Define this macro if you wish to preserve a certain alignment for

  31862.      the stack pointer, greater than what the hardware enforces.  The

  31863.      definition is a C expression for the desired alignment (measured in

  31864.      bits).  This macro must evaluate to a value equal to or larger than

  31865.      'STACK_BOUNDARY'.

  31866. 

  31867.  -- Macro: INCOMING_STACK_BOUNDARY

  31868.      Define this macro if the incoming stack boundary may be different

  31869.      from 'PREFERRED_STACK_BOUNDARY'.  This macro must evaluate to a

  31870.      value equal to or larger than 'STACK_BOUNDARY'.

  31871. 

  31872.  -- Macro: FUNCTION_BOUNDARY

  31873.      Alignment required for a function entry point, in bits.

  31874. 

  31875.  -- Macro: BIGGEST_ALIGNMENT

  31876.      Biggest alignment that any data type can require on this machine,

  31877.      in bits.  Note that this is not the biggest alignment that is

  31878.      supported, just the biggest alignment that, when violated, may

  31879.      cause a fault.

  31880. 

  31881.  -- Target Hook: HOST_WIDE_INT TARGET_ABSOLUTE_BIGGEST_ALIGNMENT

  31882.      If defined, this target hook specifies the absolute biggest

  31883.      alignment that a type or variable can have on this machine,

  31884.      otherwise, 'BIGGEST_ALIGNMENT' is used.

  31885. 

  31886.  -- Macro: MALLOC_ABI_ALIGNMENT

  31887.      Alignment, in bits, a C conformant malloc implementation has to

  31888.      provide.  If not defined, the default value is 'BITS_PER_WORD'.

  31889. 

  31890.  -- Macro: ATTRIBUTE_ALIGNED_VALUE

  31891.      Alignment used by the '__attribute__ ((aligned))' construct.  If

  31892.      not defined, the default value is 'BIGGEST_ALIGNMENT'.

  31893. 

  31894.  -- Macro: MINIMUM_ATOMIC_ALIGNMENT

  31895.      If defined, the smallest alignment, in bits, that can be given to

  31896.      an object that can be referenced in one operation, without

  31897.      disturbing any nearby object.  Normally, this is 'BITS_PER_UNIT',

  31898.      but may be larger on machines that don't have byte or half-word

  31899.      store operations.

  31900. 

  31901.  -- Macro: BIGGEST_FIELD_ALIGNMENT

  31902.      Biggest alignment that any structure or union field can require on

  31903.      this machine, in bits.  If defined, this overrides

  31904.      'BIGGEST_ALIGNMENT' for structure and union fields only, unless the

  31905.      field alignment has been set by the '__attribute__ ((aligned (N)))'

  31906.      construct.

  31907. 

  31908.  -- Macro: ADJUST_FIELD_ALIGN (FIELD, TYPE, COMPUTED)

  31909.      An expression for the alignment of a structure field FIELD of type

  31910.      TYPE if the alignment computed in the usual way (including applying

  31911.      of 'BIGGEST_ALIGNMENT' and 'BIGGEST_FIELD_ALIGNMENT' to the

  31912.      alignment) is COMPUTED.  It overrides alignment only if the field

  31913.      alignment has not been set by the '__attribute__ ((aligned (N)))'

  31914.      construct.  Note that FIELD may be 'NULL_TREE' in case we just

  31915.      query for the minimum alignment of a field of type TYPE in

  31916.      structure context.

  31917. 

  31918.  -- Macro: MAX_STACK_ALIGNMENT

  31919.      Biggest stack alignment guaranteed by the backend.  Use this macro

  31920.      to specify the maximum alignment of a variable on stack.

  31921. 

  31922.      If not defined, the default value is 'STACK_BOUNDARY'.

  31923. 

  31924.  -- Macro: MAX_OFILE_ALIGNMENT

  31925.      Biggest alignment supported by the object file format of this

  31926.      machine.  Use this macro to limit the alignment which can be

  31927.      specified using the '__attribute__ ((aligned (N)))' construct for

  31928.      functions and objects with static storage duration.  The alignment

  31929.      of automatic objects may exceed the object file format maximum up

  31930.      to the maximum supported by GCC. If not defined, the default value

  31931.      is 'BIGGEST_ALIGNMENT'.

  31932. 

  31933.      On systems that use ELF, the default (in 'config/elfos.h') is the

  31934.      largest supported 32-bit ELF section alignment representable on a

  31935.      32-bit host e.g. '(((uint64_t) 1 << 28) * 8)'.  On 32-bit ELF the

  31936.      largest supported section alignment in bits is '(0x80000000 * 8)',

  31937.      but this is not representable on 32-bit hosts.

  31938. 

  31939.  -- Target Hook: HOST_WIDE_INT TARGET_STATIC_RTX_ALIGNMENT (machine_mode

  31940.           MODE)

  31941.      This hook returns the preferred alignment in bits for a

  31942.      statically-allocated rtx, such as a constant pool entry.  MODE is

  31943.      the mode of the rtx.  The default implementation returns

  31944.      'GET_MODE_ALIGNMENT (MODE)'.

  31945. 

  31946.  -- Macro: DATA_ALIGNMENT (TYPE, BASIC-ALIGN)

  31947.      If defined, a C expression to compute the alignment for a variable

  31948.      in the static store.  TYPE is the data type, and BASIC-ALIGN is the

  31949.      alignment that the object would ordinarily have.  The value of this

  31950.      macro is used instead of that alignment to align the object.

  31951. 

  31952.      If this macro is not defined, then BASIC-ALIGN is used.

  31953. 

  31954.      One use of this macro is to increase alignment of medium-size data

  31955.      to make it all fit in fewer cache lines.  Another is to cause

  31956.      character arrays to be word-aligned so that 'strcpy' calls that

  31957.      copy constants to character arrays can be done inline.

  31958. 

  31959.  -- Macro: DATA_ABI_ALIGNMENT (TYPE, BASIC-ALIGN)

  31960.      Similar to 'DATA_ALIGNMENT', but for the cases where the ABI

  31961.      mandates some alignment increase, instead of optimization only

  31962.      purposes.  E.g. AMD x86-64 psABI says that variables with array

  31963.      type larger than 15 bytes must be aligned to 16 byte boundaries.

  31964. 

  31965.      If this macro is not defined, then BASIC-ALIGN is used.

  31966. 

  31967.  -- Target Hook: HOST_WIDE_INT TARGET_CONSTANT_ALIGNMENT (const_tree

  31968.           CONSTANT, HOST_WIDE_INT BASIC_ALIGN)

  31969.      This hook returns the alignment in bits of a constant that is being

  31970.      placed in memory.  CONSTANT is the constant and BASIC_ALIGN is the

  31971.      alignment that the object would ordinarily have.

  31972. 

  31973.      The default definition just returns BASIC_ALIGN.

  31974. 

  31975.      The typical use of this hook is to increase alignment for string

  31976.      constants to be word aligned so that 'strcpy' calls that copy

  31977.      constants can be done inline.  The function

  31978.      'constant_alignment_word_strings' provides such a definition.

  31979. 

  31980.  -- Macro: LOCAL_ALIGNMENT (TYPE, BASIC-ALIGN)

  31981.      If defined, a C expression to compute the alignment for a variable

  31982.      in the local store.  TYPE is the data type, and BASIC-ALIGN is the

  31983.      alignment that the object would ordinarily have.  The value of this

  31984.      macro is used instead of that alignment to align the object.

  31985. 

  31986.      If this macro is not defined, then BASIC-ALIGN is used.

  31987. 

  31988.      One use of this macro is to increase alignment of medium-size data

  31989.      to make it all fit in fewer cache lines.

  31990. 

  31991.      If the value of this macro has a type, it should be an unsigned

  31992.      type.

  31993. 

  31994.  -- Target Hook: HOST_WIDE_INT TARGET_VECTOR_ALIGNMENT (const_tree TYPE)

  31995.      This hook can be used to define the alignment for a vector of type

  31996.      TYPE, in order to comply with a platform ABI. The default is to

  31997.      require natural alignment for vector types.  The alignment returned

  31998.      by this hook must be a power-of-two multiple of the default

  31999.      alignment of the vector element type.

  32000. 

  32001.  -- Macro: STACK_SLOT_ALIGNMENT (TYPE, MODE, BASIC-ALIGN)

  32002.      If defined, a C expression to compute the alignment for stack slot.

  32003.      TYPE is the data type, MODE is the widest mode available, and

  32004.      BASIC-ALIGN is the alignment that the slot would ordinarily have.

  32005.      The value of this macro is used instead of that alignment to align

  32006.      the slot.

  32007. 

  32008.      If this macro is not defined, then BASIC-ALIGN is used when TYPE is

  32009.      'NULL'.  Otherwise, 'LOCAL_ALIGNMENT' will be used.

  32010. 

  32011.      This macro is to set alignment of stack slot to the maximum

  32012.      alignment of all possible modes which the slot may have.

  32013. 

  32014.      If the value of this macro has a type, it should be an unsigned

  32015.      type.

  32016. 

  32017.  -- Macro: LOCAL_DECL_ALIGNMENT (DECL)

  32018.      If defined, a C expression to compute the alignment for a local

  32019.      variable DECL.

  32020. 

  32021.      If this macro is not defined, then 'LOCAL_ALIGNMENT (TREE_TYPE

  32022.      (DECL), DECL_ALIGN (DECL))' is used.

  32023. 

  32024.      One use of this macro is to increase alignment of medium-size data

  32025.      to make it all fit in fewer cache lines.

  32026. 

  32027.      If the value of this macro has a type, it should be an unsigned

  32028.      type.

  32029. 

  32030.  -- Macro: MINIMUM_ALIGNMENT (EXP, MODE, ALIGN)

  32031.      If defined, a C expression to compute the minimum required

  32032.      alignment for dynamic stack realignment purposes for EXP (a type or

  32033.      decl), MODE, assuming normal alignment ALIGN.

  32034. 

  32035.      If this macro is not defined, then ALIGN will be used.

  32036. 

  32037.  -- Macro: EMPTY_FIELD_BOUNDARY

  32038.      Alignment in bits to be given to a structure bit-field that follows

  32039.      an empty field such as 'int : 0;'.

  32040. 

  32041.      If 'PCC_BITFIELD_TYPE_MATTERS' is true, it overrides this macro.

  32042. 

  32043.  -- Macro: STRUCTURE_SIZE_BOUNDARY

  32044.      Number of bits which any structure or union's size must be a

  32045.      multiple of.  Each structure or union's size is rounded up to a

  32046.      multiple of this.

  32047. 

  32048.      If you do not define this macro, the default is the same as

  32049.      'BITS_PER_UNIT'.

  32050. 

  32051.  -- Macro: STRICT_ALIGNMENT

  32052.      Define this macro to be the value 1 if instructions will fail to

  32053.      work if given data not on the nominal alignment.  If instructions

  32054.      will merely go slower in that case, define this macro as 0.

  32055. 

  32056.  -- Macro: PCC_BITFIELD_TYPE_MATTERS

  32057.      Define this if you wish to imitate the way many other C compilers

  32058.      handle alignment of bit-fields and the structures that contain

  32059.      them.

  32060. 

  32061.      The behavior is that the type written for a named bit-field ('int',

  32062.      'short', or other integer type) imposes an alignment for the entire

  32063.      structure, as if the structure really did contain an ordinary field

  32064.      of that type.  In addition, the bit-field is placed within the

  32065.      structure so that it would fit within such a field, not crossing a

  32066.      boundary for it.

  32067. 

  32068.      Thus, on most machines, a named bit-field whose type is written as

  32069.      'int' would not cross a four-byte boundary, and would force

  32070.      four-byte alignment for the whole structure.  (The alignment used

  32071.      may not be four bytes; it is controlled by the other alignment

  32072.      parameters.)

  32073. 

  32074.      An unnamed bit-field will not affect the alignment of the

  32075.      containing structure.

  32076. 

  32077.      If the macro is defined, its definition should be a C expression; a

  32078.      nonzero value for the expression enables this behavior.

  32079. 

  32080.      Note that if this macro is not defined, or its value is zero, some

  32081.      bit-fields may cross more than one alignment boundary.  The

  32082.      compiler can support such references if there are 'insv', 'extv',

  32083.      and 'extzv' insns that can directly reference memory.

  32084. 

  32085.      The other known way of making bit-fields work is to define

  32086.      'STRUCTURE_SIZE_BOUNDARY' as large as 'BIGGEST_ALIGNMENT'.  Then

  32087.      every structure can be accessed with fullwords.

  32088. 

  32089.      Unless the machine has bit-field instructions or you define

  32090.      'STRUCTURE_SIZE_BOUNDARY' that way, you must define

  32091.      'PCC_BITFIELD_TYPE_MATTERS' to have a nonzero value.

  32092. 

  32093.      If your aim is to make GCC use the same conventions for laying out

  32094.      bit-fields as are used by another compiler, here is how to

  32095.      investigate what the other compiler does.  Compile and run this

  32096.      program:

  32097. 

  32098.           struct foo1

  32099.           {

  32100.             char x;

  32101.             char :0;

  32102.             char y;

  32103.           };

  32104. 

  32105.           struct foo2

  32106.           {

  32107.             char x;

  32108.             int :0;

  32109.             char y;

  32110.           };

  32111. 

  32112.           main ()

  32113.           {

  32114.             printf ("Size of foo1 is %d\n",

  32115.                     sizeof (struct foo1));

  32116.             printf ("Size of foo2 is %d\n",

  32117.                     sizeof (struct foo2));

  32118.             exit (0);

  32119.           }

  32120. 

  32121.      If this prints 2 and 5, then the compiler's behavior is what you

  32122.      would get from 'PCC_BITFIELD_TYPE_MATTERS'.

  32123. 

  32124.  -- Macro: BITFIELD_NBYTES_LIMITED

  32125.      Like 'PCC_BITFIELD_TYPE_MATTERS' except that its effect is limited

  32126.      to aligning a bit-field within the structure.

  32127. 

  32128.  -- Target Hook: bool TARGET_ALIGN_ANON_BITFIELD (void)

  32129.      When 'PCC_BITFIELD_TYPE_MATTERS' is true this hook will determine

  32130.      whether unnamed bitfields affect the alignment of the containing

  32131.      structure.  The hook should return true if the structure should

  32132.      inherit the alignment requirements of an unnamed bitfield's type.

  32133. 

  32134.  -- Target Hook: bool TARGET_NARROW_VOLATILE_BITFIELD (void)

  32135.      This target hook should return 'true' if accesses to volatile

  32136.      bitfields should use the narrowest mode possible.  It should return

  32137.      'false' if these accesses should use the bitfield container type.

  32138. 

  32139.      The default is 'false'.

  32140. 

  32141.  -- Target Hook: bool TARGET_MEMBER_TYPE_FORCES_BLK (const_tree FIELD,

  32142.           machine_mode MODE)

  32143.      Return true if a structure, union or array containing FIELD should

  32144.      be accessed using 'BLKMODE'.

  32145. 

  32146.      If FIELD is the only field in the structure, MODE is its mode,

  32147.      otherwise MODE is VOIDmode.  MODE is provided in the case where

  32148.      structures of one field would require the structure's mode to

  32149.      retain the field's mode.

  32150. 

  32151.      Normally, this is not needed.

  32152. 

  32153.  -- Macro: ROUND_TYPE_ALIGN (TYPE, COMPUTED, SPECIFIED)

  32154.      Define this macro as an expression for the alignment of a type

  32155.      (given by TYPE as a tree node) if the alignment computed in the

  32156.      usual way is COMPUTED and the alignment explicitly specified was

  32157.      SPECIFIED.

  32158. 

  32159.      The default is to use SPECIFIED if it is larger; otherwise, use the

  32160.      smaller of COMPUTED and 'BIGGEST_ALIGNMENT'

  32161. 

  32162.  -- Macro: MAX_FIXED_MODE_SIZE

  32163.      An integer expression for the size in bits of the largest integer

  32164.      machine mode that should actually be used.  All integer machine

  32165.      modes of this size or smaller can be used for structures and unions

  32166.      with the appropriate sizes.  If this macro is undefined,

  32167.      'GET_MODE_BITSIZE (DImode)' is assumed.

  32168. 

  32169.  -- Macro: STACK_SAVEAREA_MODE (SAVE_LEVEL)

  32170.      If defined, an expression of type 'machine_mode' that specifies the

  32171.      mode of the save area operand of a 'save_stack_LEVEL' named pattern

  32172.      (*note Standard Names::).  SAVE_LEVEL is one of 'SAVE_BLOCK',

  32173.      'SAVE_FUNCTION', or 'SAVE_NONLOCAL' and selects which of the three

  32174.      named patterns is having its mode specified.

  32175. 

  32176.      You need not define this macro if it always returns 'Pmode'.  You

  32177.      would most commonly define this macro if the 'save_stack_LEVEL'

  32178.      patterns need to support both a 32- and a 64-bit mode.

  32179. 

  32180.  -- Macro: STACK_SIZE_MODE

  32181.      If defined, an expression of type 'machine_mode' that specifies the

  32182.      mode of the size increment operand of an 'allocate_stack' named

  32183.      pattern (*note Standard Names::).

  32184. 

  32185.      You need not define this macro if it always returns 'word_mode'.

  32186.      You would most commonly define this macro if the 'allocate_stack'

  32187.      pattern needs to support both a 32- and a 64-bit mode.

  32188. 

  32189.  -- Target Hook: scalar_int_mode TARGET_LIBGCC_CMP_RETURN_MODE (void)

  32190.      This target hook should return the mode to be used for the return

  32191.      value of compare instructions expanded to libgcc calls.  If not

  32192.      defined 'word_mode' is returned which is the right choice for a

  32193.      majority of targets.

  32194. 

  32195.  -- Target Hook: scalar_int_mode TARGET_LIBGCC_SHIFT_COUNT_MODE (void)

  32196.      This target hook should return the mode to be used for the shift

  32197.      count operand of shift instructions expanded to libgcc calls.  If

  32198.      not defined 'word_mode' is returned which is the right choice for a

  32199.      majority of targets.

  32200. 

  32201.  -- Target Hook: scalar_int_mode TARGET_UNWIND_WORD_MODE (void)

  32202.      Return machine mode to be used for '_Unwind_Word' type.  The

  32203.      default is to use 'word_mode'.

  32204. 

  32205.  -- Target Hook: bool TARGET_MS_BITFIELD_LAYOUT_P (const_tree

  32206.           RECORD_TYPE)

  32207.      This target hook returns 'true' if bit-fields in the given

  32208.      RECORD_TYPE are to be laid out following the rules of Microsoft

  32209.      Visual C/C++, namely: (i) a bit-field won't share the same storage

  32210.      unit with the previous bit-field if their underlying types have

  32211.      different sizes, and the bit-field will be aligned to the highest

  32212.      alignment of the underlying types of itself and of the previous

  32213.      bit-field; (ii) a zero-sized bit-field will affect the alignment of

  32214.      the whole enclosing structure, even if it is unnamed; except that

  32215.      (iii) a zero-sized bit-field will be disregarded unless it follows

  32216.      another bit-field of nonzero size.  If this hook returns 'true',

  32217.      other macros that control bit-field layout are ignored.

  32218. 

  32219.      When a bit-field is inserted into a packed record, the whole size

  32220.      of the underlying type is used by one or more same-size adjacent

  32221.      bit-fields (that is, if its long:3, 32 bits is used in the record,

  32222.      and any additional adjacent long bit-fields are packed into the

  32223.      same chunk of 32 bits.  However, if the size changes, a new field

  32224.      of that size is allocated).  In an unpacked record, this is the

  32225.      same as using alignment, but not equivalent when packing.

  32226. 

  32227.      If both MS bit-fields and '__attribute__((packed))' are used, the

  32228.      latter will take precedence.  If '__attribute__((packed))' is used

  32229.      on a single field when MS bit-fields are in use, it will take

  32230.      precedence for that field, but the alignment of the rest of the

  32231.      structure may affect its placement.

  32232. 

  32233.  -- Target Hook: bool TARGET_DECIMAL_FLOAT_SUPPORTED_P (void)

  32234.      Returns true if the target supports decimal floating point.

  32235. 

  32236.  -- Target Hook: bool TARGET_FIXED_POINT_SUPPORTED_P (void)

  32237.      Returns true if the target supports fixed-point arithmetic.

  32238. 

  32239.  -- Target Hook: void TARGET_EXPAND_TO_RTL_HOOK (void)

  32240.      This hook is called just before expansion into rtl, allowing the

  32241.      target to perform additional initializations or analysis before the

  32242.      expansion.  For example, the rs6000 port uses it to allocate a

  32243.      scratch stack slot for use in copying SDmode values between memory

  32244.      and floating point registers whenever the function being expanded

  32245.      has any SDmode usage.

  32246. 

  32247.  -- Target Hook: void TARGET_INSTANTIATE_DECLS (void)

  32248.      This hook allows the backend to perform additional instantiations

  32249.      on rtl that are not actually in any insns yet, but will be later.

  32250. 

  32251.  -- Target Hook: const char * TARGET_MANGLE_TYPE (const_tree TYPE)

  32252.      If your target defines any fundamental types, or any types your

  32253.      target uses should be mangled differently from the default, define

  32254.      this hook to return the appropriate encoding for these types as

  32255.      part of a C++ mangled name.  The TYPE argument is the tree

  32256.      structure representing the type to be mangled.  The hook may be

  32257.      applied to trees which are not target-specific fundamental types;

  32258.      it should return 'NULL' for all such types, as well as arguments it

  32259.      does not recognize.  If the return value is not 'NULL', it must

  32260.      point to a statically-allocated string constant.

  32261. 

  32262.      Target-specific fundamental types might be new fundamental types or

  32263.      qualified versions of ordinary fundamental types.  Encode new

  32264.      fundamental types as 'u N NAME', where NAME is the name used for

  32265.      the type in source code, and N is the length of NAME in decimal.

  32266.      Encode qualified versions of ordinary types as 'U N NAME CODE',

  32267.      where NAME is the name used for the type qualifier in source code,

  32268.      N is the length of NAME as above, and CODE is the code used to

  32269.      represent the unqualified version of this type.  (See

  32270.      'write_builtin_type' in 'cp/mangle.c' for the list of codes.)  In

  32271.      both cases the spaces are for clarity; do not include any spaces in

  32272.      your string.

  32273. 

  32274.      This hook is applied to types prior to typedef resolution.  If the

  32275.      mangled name for a particular type depends only on that type's main

  32276.      variant, you can perform typedef resolution yourself using

  32277.      'TYPE_MAIN_VARIANT' before mangling.

  32278. 

  32279.      The default version of this hook always returns 'NULL', which is

  32280.      appropriate for a target that does not define any new fundamental

  32281.      types.

  32282. 

  32283. 

  32284. File: gccint.info,  Node: Type Layout,  Next: Registers,  Prev: Storage Layout,  Up: Target Macros

  32285. 

  32286. 18.6 Layout of Source Language Data Types

  32287. =========================================

  32288. 

  32289. These macros define the sizes and other characteristics of the standard

  32290. basic data types used in programs being compiled.  Unlike the macros in

  32291. the previous section, these apply to specific features of C and related

  32292. languages, rather than to fundamental aspects of storage layout.

  32293. 

  32294.  -- Macro: INT_TYPE_SIZE

  32295.      A C expression for the size in bits of the type 'int' on the target

  32296.      machine.  If you don't define this, the default is one word.

  32297. 

  32298.  -- Macro: SHORT_TYPE_SIZE

  32299.      A C expression for the size in bits of the type 'short' on the

  32300.      target machine.  If you don't define this, the default is half a

  32301.      word.  (If this would be less than one storage unit, it is rounded

  32302.      up to one unit.)

  32303. 

  32304.  -- Macro: LONG_TYPE_SIZE

  32305.      A C expression for the size in bits of the type 'long' on the

  32306.      target machine.  If you don't define this, the default is one word.

  32307. 

  32308.  -- Macro: ADA_LONG_TYPE_SIZE

  32309.      On some machines, the size used for the Ada equivalent of the type

  32310.      'long' by a native Ada compiler differs from that used by C.  In

  32311.      that situation, define this macro to be a C expression to be used

  32312.      for the size of that type.  If you don't define this, the default

  32313.      is the value of 'LONG_TYPE_SIZE'.

  32314. 

  32315.  -- Macro: LONG_LONG_TYPE_SIZE

  32316.      A C expression for the size in bits of the type 'long long' on the

  32317.      target machine.  If you don't define this, the default is two

  32318.      words.  If you want to support GNU Ada on your machine, the value

  32319.      of this macro must be at least 64.

  32320. 

  32321.  -- Macro: CHAR_TYPE_SIZE

  32322.      A C expression for the size in bits of the type 'char' on the

  32323.      target machine.  If you don't define this, the default is

  32324.      'BITS_PER_UNIT'.

  32325. 

  32326.  -- Macro: BOOL_TYPE_SIZE

  32327.      A C expression for the size in bits of the C++ type 'bool' and C99

  32328.      type '_Bool' on the target machine.  If you don't define this, and

  32329.      you probably shouldn't, the default is 'CHAR_TYPE_SIZE'.

  32330. 

  32331.  -- Macro: FLOAT_TYPE_SIZE

  32332.      A C expression for the size in bits of the type 'float' on the

  32333.      target machine.  If you don't define this, the default is one word.

  32334. 

  32335.  -- Macro: DOUBLE_TYPE_SIZE

  32336.      A C expression for the size in bits of the type 'double' on the

  32337.      target machine.  If you don't define this, the default is two

  32338.      words.

  32339. 

  32340.  -- Macro: LONG_DOUBLE_TYPE_SIZE

  32341.      A C expression for the size in bits of the type 'long double' on

  32342.      the target machine.  If you don't define this, the default is two

  32343.      words.

  32344. 

  32345.  -- Macro: SHORT_FRACT_TYPE_SIZE

  32346.      A C expression for the size in bits of the type 'short _Fract' on

  32347.      the target machine.  If you don't define this, the default is

  32348.      'BITS_PER_UNIT'.

  32349. 

  32350.  -- Macro: FRACT_TYPE_SIZE

  32351.      A C expression for the size in bits of the type '_Fract' on the

  32352.      target machine.  If you don't define this, the default is

  32353.      'BITS_PER_UNIT * 2'.

  32354. 

  32355.  -- Macro: LONG_FRACT_TYPE_SIZE

  32356.      A C expression for the size in bits of the type 'long _Fract' on

  32357.      the target machine.  If you don't define this, the default is

  32358.      'BITS_PER_UNIT * 4'.

  32359. 

  32360.  -- Macro: LONG_LONG_FRACT_TYPE_SIZE

  32361.      A C expression for the size in bits of the type 'long long _Fract'

  32362.      on the target machine.  If you don't define this, the default is

  32363.      'BITS_PER_UNIT * 8'.

  32364. 

  32365.  -- Macro: SHORT_ACCUM_TYPE_SIZE

  32366.      A C expression for the size in bits of the type 'short _Accum' on

  32367.      the target machine.  If you don't define this, the default is

  32368.      'BITS_PER_UNIT * 2'.

  32369. 

  32370.  -- Macro: ACCUM_TYPE_SIZE

  32371.      A C expression for the size in bits of the type '_Accum' on the

  32372.      target machine.  If you don't define this, the default is

  32373.      'BITS_PER_UNIT * 4'.

  32374. 

  32375.  -- Macro: LONG_ACCUM_TYPE_SIZE

  32376.      A C expression for the size in bits of the type 'long _Accum' on

  32377.      the target machine.  If you don't define this, the default is

  32378.      'BITS_PER_UNIT * 8'.

  32379. 

  32380.  -- Macro: LONG_LONG_ACCUM_TYPE_SIZE

  32381.      A C expression for the size in bits of the type 'long long _Accum'

  32382.      on the target machine.  If you don't define this, the default is

  32383.      'BITS_PER_UNIT * 16'.

  32384. 

  32385.  -- Macro: LIBGCC2_GNU_PREFIX

  32386.      This macro corresponds to the 'TARGET_LIBFUNC_GNU_PREFIX' target

  32387.      hook and should be defined if that hook is overriden to be true.

  32388.      It causes function names in libgcc to be changed to use a '__gnu_'

  32389.      prefix for their name rather than the default '__'.  A port which

  32390.      uses this macro should also arrange to use 't-gnu-prefix' in the

  32391.      libgcc 'config.host'.

  32392. 

  32393.  -- Macro: WIDEST_HARDWARE_FP_SIZE

  32394.      A C expression for the size in bits of the widest floating-point

  32395.      format supported by the hardware.  If you define this macro, you

  32396.      must specify a value less than or equal to the value of

  32397.      'LONG_DOUBLE_TYPE_SIZE'.  If you do not define this macro, the

  32398.      value of 'LONG_DOUBLE_TYPE_SIZE' is the default.

  32399. 

  32400.  -- Macro: DEFAULT_SIGNED_CHAR

  32401.      An expression whose value is 1 or 0, according to whether the type

  32402.      'char' should be signed or unsigned by default.  The user can

  32403.      always override this default with the options '-fsigned-char' and

  32404.      '-funsigned-char'.

  32405. 

  32406.  -- Target Hook: bool TARGET_DEFAULT_SHORT_ENUMS (void)

  32407.      This target hook should return true if the compiler should give an

  32408.      'enum' type only as many bytes as it takes to represent the range

  32409.      of possible values of that type.  It should return false if all

  32410.      'enum' types should be allocated like 'int'.

  32411. 

  32412.      The default is to return false.

  32413. 

  32414.  -- Macro: SIZE_TYPE

  32415.      A C expression for a string describing the name of the data type to

  32416.      use for size values.  The typedef name 'size_t' is defined using

  32417.      the contents of the string.

  32418. 

  32419.      The string can contain more than one keyword.  If so, separate them

  32420.      with spaces, and write first any length keyword, then 'unsigned' if

  32421.      appropriate, and finally 'int'.  The string must exactly match one

  32422.      of the data type names defined in the function

  32423.      'c_common_nodes_and_builtins' in the file 'c-family/c-common.c'.

  32424.      You may not omit 'int' or change the order--that would cause the

  32425.      compiler to crash on startup.

  32426. 

  32427.      If you don't define this macro, the default is '"long unsigned

  32428.      int"'.

  32429. 

  32430.  -- Macro: SIZETYPE

  32431.      GCC defines internal types ('sizetype', 'ssizetype', 'bitsizetype'

  32432.      and 'sbitsizetype') for expressions dealing with size.  This macro

  32433.      is a C expression for a string describing the name of the data type

  32434.      from which the precision of 'sizetype' is extracted.

  32435. 

  32436.      The string has the same restrictions as 'SIZE_TYPE' string.

  32437. 

  32438.      If you don't define this macro, the default is 'SIZE_TYPE'.

  32439. 

  32440.  -- Macro: PTRDIFF_TYPE

  32441.      A C expression for a string describing the name of the data type to

  32442.      use for the result of subtracting two pointers.  The typedef name

  32443.      'ptrdiff_t' is defined using the contents of the string.  See

  32444.      'SIZE_TYPE' above for more information.

  32445. 

  32446.      If you don't define this macro, the default is '"long int"'.

  32447. 

  32448.  -- Macro: WCHAR_TYPE

  32449.      A C expression for a string describing the name of the data type to

  32450.      use for wide characters.  The typedef name 'wchar_t' is defined

  32451.      using the contents of the string.  See 'SIZE_TYPE' above for more

  32452.      information.

  32453. 

  32454.      If you don't define this macro, the default is '"int"'.

  32455. 

  32456.  -- Macro: WCHAR_TYPE_SIZE

  32457.      A C expression for the size in bits of the data type for wide

  32458.      characters.  This is used in 'cpp', which cannot make use of

  32459.      'WCHAR_TYPE'.

  32460. 

  32461.  -- Macro: WINT_TYPE

  32462.      A C expression for a string describing the name of the data type to

  32463.      use for wide characters passed to 'printf' and returned from

  32464.      'getwc'.  The typedef name 'wint_t' is defined using the contents

  32465.      of the string.  See 'SIZE_TYPE' above for more information.

  32466. 

  32467.      If you don't define this macro, the default is '"unsigned int"'.

  32468. 

  32469.  -- Macro: INTMAX_TYPE

  32470.      A C expression for a string describing the name of the data type

  32471.      that can represent any value of any standard or extended signed

  32472.      integer type.  The typedef name 'intmax_t' is defined using the

  32473.      contents of the string.  See 'SIZE_TYPE' above for more

  32474.      information.

  32475. 

  32476.      If you don't define this macro, the default is the first of

  32477.      '"int"', '"long int"', or '"long long int"' that has as much

  32478.      precision as 'long long int'.

  32479. 

  32480.  -- Macro: UINTMAX_TYPE

  32481.      A C expression for a string describing the name of the data type

  32482.      that can represent any value of any standard or extended unsigned

  32483.      integer type.  The typedef name 'uintmax_t' is defined using the

  32484.      contents of the string.  See 'SIZE_TYPE' above for more

  32485.      information.

  32486. 

  32487.      If you don't define this macro, the default is the first of

  32488.      '"unsigned int"', '"long unsigned int"', or '"long long unsigned

  32489.      int"' that has as much precision as 'long long unsigned int'.

  32490. 

  32491.  -- Macro: SIG_ATOMIC_TYPE

  32492.  -- Macro: INT8_TYPE

  32493.  -- Macro: INT16_TYPE

  32494.  -- Macro: INT32_TYPE

  32495.  -- Macro: INT64_TYPE

  32496.  -- Macro: UINT8_TYPE

  32497.  -- Macro: UINT16_TYPE

  32498.  -- Macro: UINT32_TYPE

  32499.  -- Macro: UINT64_TYPE

  32500.  -- Macro: INT_LEAST8_TYPE

  32501.  -- Macro: INT_LEAST16_TYPE

  32502.  -- Macro: INT_LEAST32_TYPE

  32503.  -- Macro: INT_LEAST64_TYPE

  32504.  -- Macro: UINT_LEAST8_TYPE

  32505.  -- Macro: UINT_LEAST16_TYPE

  32506.  -- Macro: UINT_LEAST32_TYPE

  32507.  -- Macro: UINT_LEAST64_TYPE

  32508.  -- Macro: INT_FAST8_TYPE

  32509.  -- Macro: INT_FAST16_TYPE

  32510.  -- Macro: INT_FAST32_TYPE

  32511.  -- Macro: INT_FAST64_TYPE

  32512.  -- Macro: UINT_FAST8_TYPE

  32513.  -- Macro: UINT_FAST16_TYPE

  32514.  -- Macro: UINT_FAST32_TYPE

  32515.  -- Macro: UINT_FAST64_TYPE

  32516.  -- Macro: INTPTR_TYPE

  32517.  -- Macro: UINTPTR_TYPE

  32518.      C expressions for the standard types 'sig_atomic_t', 'int8_t',

  32519.      'int16_t', 'int32_t', 'int64_t', 'uint8_t', 'uint16_t', 'uint32_t',

  32520.      'uint64_t', 'int_least8_t', 'int_least16_t', 'int_least32_t',

  32521.      'int_least64_t', 'uint_least8_t', 'uint_least16_t',

  32522.      'uint_least32_t', 'uint_least64_t', 'int_fast8_t', 'int_fast16_t',

  32523.      'int_fast32_t', 'int_fast64_t', 'uint_fast8_t', 'uint_fast16_t',

  32524.      'uint_fast32_t', 'uint_fast64_t', 'intptr_t', and 'uintptr_t'.  See

  32525.      'SIZE_TYPE' above for more information.

  32526. 

  32527.      If any of these macros evaluates to a null pointer, the

  32528.      corresponding type is not supported; if GCC is configured to

  32529.      provide '<stdint.h>' in such a case, the header provided may not

  32530.      conform to C99, depending on the type in question.  The defaults

  32531.      for all of these macros are null pointers.

  32532. 

  32533.  -- Macro: TARGET_PTRMEMFUNC_VBIT_LOCATION

  32534.      The C++ compiler represents a pointer-to-member-function with a

  32535.      struct that looks like:

  32536. 

  32537.             struct {

  32538.               union {

  32539.                 void (*fn)();

  32540.                 ptrdiff_t vtable_index;

  32541.               };

  32542.               ptrdiff_t delta;

  32543.             };

  32544. 

  32545.      The C++ compiler must use one bit to indicate whether the function

  32546.      that will be called through a pointer-to-member-function is

  32547.      virtual.  Normally, we assume that the low-order bit of a function

  32548.      pointer must always be zero.  Then, by ensuring that the

  32549.      vtable_index is odd, we can distinguish which variant of the union

  32550.      is in use.  But, on some platforms function pointers can be odd,

  32551.      and so this doesn't work.  In that case, we use the low-order bit

  32552.      of the 'delta' field, and shift the remainder of the 'delta' field

  32553.      to the left.

  32554. 

  32555.      GCC will automatically make the right selection about where to

  32556.      store this bit using the 'FUNCTION_BOUNDARY' setting for your

  32557.      platform.  However, some platforms such as ARM/Thumb have

  32558.      'FUNCTION_BOUNDARY' set such that functions always start at even

  32559.      addresses, but the lowest bit of pointers to functions indicate

  32560.      whether the function at that address is in ARM or Thumb mode.  If

  32561.      this is the case of your architecture, you should define this macro

  32562.      to 'ptrmemfunc_vbit_in_delta'.

  32563. 

  32564.      In general, you should not have to define this macro.  On

  32565.      architectures in which function addresses are always even,

  32566.      according to 'FUNCTION_BOUNDARY', GCC will automatically define

  32567.      this macro to 'ptrmemfunc_vbit_in_pfn'.

  32568. 

  32569.  -- Macro: TARGET_VTABLE_USES_DESCRIPTORS

  32570.      Normally, the C++ compiler uses function pointers in vtables.  This

  32571.      macro allows the target to change to use "function descriptors"

  32572.      instead.  Function descriptors are found on targets for whom a

  32573.      function pointer is actually a small data structure.  Normally the

  32574.      data structure consists of the actual code address plus a data

  32575.      pointer to which the function's data is relative.

  32576. 

  32577.      If vtables are used, the value of this macro should be the number

  32578.      of words that the function descriptor occupies.

  32579. 

  32580.  -- Macro: TARGET_VTABLE_ENTRY_ALIGN

  32581.      By default, the vtable entries are void pointers, the so the

  32582.      alignment is the same as pointer alignment.  The value of this

  32583.      macro specifies the alignment of the vtable entry in bits.  It

  32584.      should be defined only when special alignment is necessary.  */

  32585. 

  32586.  -- Macro: TARGET_VTABLE_DATA_ENTRY_DISTANCE

  32587.      There are a few non-descriptor entries in the vtable at offsets

  32588.      below zero.  If these entries must be padded (say, to preserve the

  32589.      alignment specified by 'TARGET_VTABLE_ENTRY_ALIGN'), set this to

  32590.      the number of words in each data entry.

  32591. 

  32592. 

  32593. File: gccint.info,  Node: Registers,  Next: Register Classes,  Prev: Type Layout,  Up: Target Macros

  32594. 

  32595. 18.7 Register Usage

  32596. ===================

  32597. 

  32598. This section explains how to describe what registers the target machine

  32599. has, and how (in general) they can be used.

  32600. 

  32601.  The description of which registers a specific instruction can use is

  32602. done with register classes; see *note Register Classes::.  For

  32603. information on using registers to access a stack frame, see *note Frame

  32604. Registers::.  For passing values in registers, see *note Register

  32605. Arguments::.  For returning values in registers, see *note Scalar

  32606. Return::.

  32607. 

  32608. * Menu:

  32609. 

  32610. * Register Basics::             Number and kinds of registers.

  32611. * Allocation Order::            Order in which registers are allocated.

  32612. * Values in Registers::         What kinds of values each reg can hold.

  32613. * Leaf Functions::              Renumbering registers for leaf functions.

  32614. * Stack Registers::             Handling a register stack such as 80387.

  32615. 

  32616. 

  32617. File: gccint.info,  Node: Register Basics,  Next: Allocation Order,  Up: Registers

  32618. 

  32619. 18.7.1 Basic Characteristics of Registers

  32620. -----------------------------------------

  32621. 

  32622. Registers have various characteristics.

  32623. 

  32624.  -- Macro: FIRST_PSEUDO_REGISTER

  32625.      Number of hardware registers known to the compiler.  They receive

  32626.      numbers 0 through 'FIRST_PSEUDO_REGISTER-1'; thus, the first pseudo

  32627.      register's number really is assigned the number

  32628.      'FIRST_PSEUDO_REGISTER'.

  32629. 

  32630.  -- Macro: FIXED_REGISTERS

  32631.      An initializer that says which registers are used for fixed

  32632.      purposes all throughout the compiled code and are therefore not

  32633.      available for general allocation.  These would include the stack

  32634.      pointer, the frame pointer (except on machines where that can be

  32635.      used as a general register when no frame pointer is needed), the

  32636.      program counter on machines where that is considered one of the

  32637.      addressable registers, and any other numbered register with a

  32638.      standard use.

  32639. 

  32640.      This information is expressed as a sequence of numbers, separated

  32641.      by commas and surrounded by braces.  The Nth number is 1 if

  32642.      register N is fixed, 0 otherwise.

  32643. 

  32644.      The table initialized from this macro, and the table initialized by

  32645.      the following one, may be overridden at run time either

  32646.      automatically, by the actions of the macro

  32647.      'CONDITIONAL_REGISTER_USAGE', or by the user with the command

  32648.      options '-ffixed-REG', '-fcall-used-REG' and '-fcall-saved-REG'.

  32649. 

  32650.  -- Macro: CALL_USED_REGISTERS

  32651.      Like 'FIXED_REGISTERS' but has 1 for each register that is

  32652.      clobbered (in general) by function calls as well as for fixed

  32653.      registers.  This macro therefore identifies the registers that are

  32654.      not available for general allocation of values that must live

  32655.      across function calls.

  32656. 

  32657.      If a register has 0 in 'CALL_USED_REGISTERS', the compiler

  32658.      automatically saves it on function entry and restores it on

  32659.      function exit, if the register is used within the function.

  32660. 

  32661.      Exactly one of 'CALL_USED_REGISTERS' and

  32662.      'CALL_REALLY_USED_REGISTERS' must be defined.  Modern ports should

  32663.      define 'CALL_REALLY_USED_REGISTERS'.

  32664. 

  32665.  -- Macro: CALL_REALLY_USED_REGISTERS

  32666.      Like 'CALL_USED_REGISTERS' except this macro doesn't require that

  32667.      the entire set of 'FIXED_REGISTERS' be included.

  32668.      ('CALL_USED_REGISTERS' must be a superset of 'FIXED_REGISTERS').

  32669. 

  32670.      Exactly one of 'CALL_USED_REGISTERS' and

  32671.      'CALL_REALLY_USED_REGISTERS' must be defined.  Modern ports should

  32672.      define 'CALL_REALLY_USED_REGISTERS'.

  32673. 

  32674.  -- Target Hook: const predefined_function_abi & TARGET_FNTYPE_ABI

  32675.           (const_tree TYPE)

  32676.      Return the ABI used by a function with type TYPE; see the

  32677.      definition of 'predefined_function_abi' for details of the ABI

  32678.      descriptor.  Targets only need to define this hook if they support

  32679.      interoperability between several ABIs in the same translation unit.

  32680. 

  32681.  -- Target Hook: const predefined_function_abi & TARGET_INSN_CALLEE_ABI

  32682.           (const rtx_insn *INSN)

  32683.      This hook returns a description of the ABI used by the target of

  32684.      call instruction INSN; see the definition of

  32685.      'predefined_function_abi' for details of the ABI descriptor.  Only

  32686.      the global function 'insn_callee_abi' should call this hook

  32687.      directly.

  32688. 

  32689.      Targets only need to define this hook if they support

  32690.      interoperability between several ABIs in the same translation unit.

  32691. 

  32692.  -- Target Hook: bool TARGET_HARD_REGNO_CALL_PART_CLOBBERED (unsigned

  32693.           int ABI_ID, unsigned int REGNO, machine_mode MODE)

  32694.      ABIs usually specify that calls must preserve the full contents of

  32695.      a particular register, or that calls can alter any part of a

  32696.      particular register.  This information is captured by the target

  32697.      macro 'CALL_REALLY_USED_REGISTERS'.  However, some ABIs specify

  32698.      that calls must preserve certain bits of a particular register but

  32699.      can alter others.  This hook should return true if this applies to

  32700.      at least one of the registers in '(reg:MODE REGNO)', and if as a

  32701.      result the call would alter part of the MODE value.  For example,

  32702.      if a call preserves the low 32 bits of a 64-bit hard register REGNO

  32703.      but can clobber the upper 32 bits, this hook should return true for

  32704.      a 64-bit mode but false for a 32-bit mode.

  32705. 

  32706.      The value of ABI_ID comes from the 'predefined_function_abi'

  32707.      structure that describes the ABI of the call; see the definition of

  32708.      the structure for more details.  If (as is usual) the target uses

  32709.      the same ABI for all functions in a translation unit, ABI_ID is

  32710.      always 0.

  32711. 

  32712.      The default implementation returns false, which is correct for

  32713.      targets that don't have partly call-clobbered registers.

  32714. 

  32715.  -- Target Hook: const char * TARGET_GET_MULTILIB_ABI_NAME (void)

  32716.      This hook returns name of multilib ABI name.

  32717. 

  32718.  -- Target Hook: void TARGET_CONDITIONAL_REGISTER_USAGE (void)

  32719.      This hook may conditionally modify five variables 'fixed_regs',

  32720.      'call_used_regs', 'global_regs', 'reg_names', and

  32721.      'reg_class_contents', to take into account any dependence of these

  32722.      register sets on target flags.  The first three of these are of

  32723.      type 'char []' (interpreted as boolean vectors).  'global_regs' is

  32724.      a 'const char *[]', and 'reg_class_contents' is a 'HARD_REG_SET'.

  32725.      Before the macro is called, 'fixed_regs', 'call_used_regs',

  32726.      'reg_class_contents', and 'reg_names' have been initialized from

  32727.      'FIXED_REGISTERS', 'CALL_USED_REGISTERS', 'REG_CLASS_CONTENTS', and

  32728.      'REGISTER_NAMES', respectively.  'global_regs' has been cleared,

  32729.      and any '-ffixed-REG', '-fcall-used-REG' and '-fcall-saved-REG'

  32730.      command options have been applied.

  32731. 

  32732.      If the usage of an entire class of registers depends on the target

  32733.      flags, you may indicate this to GCC by using this macro to modify

  32734.      'fixed_regs' and 'call_used_regs' to 1 for each of the registers in

  32735.      the classes which should not be used by GCC.  Also make

  32736.      'define_register_constraint's return 'NO_REGS' for constraints that

  32737.      shouldn't be used.

  32738. 

  32739.      (However, if this class is not included in 'GENERAL_REGS' and all

  32740.      of the insn patterns whose constraints permit this class are

  32741.      controlled by target switches, then GCC will automatically avoid

  32742.      using these registers when the target switches are opposed to

  32743.      them.)

  32744. 

  32745.  -- Macro: INCOMING_REGNO (OUT)

  32746.      Define this macro if the target machine has register windows.  This

  32747.      C expression returns the register number as seen by the called

  32748.      function corresponding to the register number OUT as seen by the

  32749.      calling function.  Return OUT if register number OUT is not an

  32750.      outbound register.

  32751. 

  32752.  -- Macro: OUTGOING_REGNO (IN)

  32753.      Define this macro if the target machine has register windows.  This

  32754.      C expression returns the register number as seen by the calling

  32755.      function corresponding to the register number IN as seen by the

  32756.      called function.  Return IN if register number IN is not an inbound

  32757.      register.

  32758. 

  32759.  -- Macro: LOCAL_REGNO (REGNO)

  32760.      Define this macro if the target machine has register windows.  This

  32761.      C expression returns true if the register is call-saved but is in

  32762.      the register window.  Unlike most call-saved registers, such

  32763.      registers need not be explicitly restored on function exit or

  32764.      during non-local gotos.

  32765. 

  32766.  -- Macro: PC_REGNUM

  32767.      If the program counter has a register number, define this as that

  32768.      register number.  Otherwise, do not define it.

  32769. 

  32770. 

  32771. File: gccint.info,  Node: Allocation Order,  Next: Values in Registers,  Prev: Register Basics,  Up: Registers

  32772. 

  32773. 18.7.2 Order of Allocation of Registers

  32774. ---------------------------------------

  32775. 

  32776. Registers are allocated in order.

  32777. 

  32778.  -- Macro: REG_ALLOC_ORDER

  32779.      If defined, an initializer for a vector of integers, containing the

  32780.      numbers of hard registers in the order in which GCC should prefer

  32781.      to use them (from most preferred to least).

  32782. 

  32783.      If this macro is not defined, registers are used lowest numbered

  32784.      first (all else being equal).

  32785. 

  32786.      One use of this macro is on machines where the highest numbered

  32787.      registers must always be saved and the save-multiple-registers

  32788.      instruction supports only sequences of consecutive registers.  On

  32789.      such machines, define 'REG_ALLOC_ORDER' to be an initializer that

  32790.      lists the highest numbered allocable register first.

  32791. 

  32792.  -- Macro: ADJUST_REG_ALLOC_ORDER

  32793.      A C statement (sans semicolon) to choose the order in which to

  32794.      allocate hard registers for pseudo-registers local to a basic

  32795.      block.

  32796. 

  32797.      Store the desired register order in the array 'reg_alloc_order'.

  32798.      Element 0 should be the register to allocate first; element 1, the

  32799.      next register; and so on.

  32800. 

  32801.      The macro body should not assume anything about the contents of

  32802.      'reg_alloc_order' before execution of the macro.

  32803. 

  32804.      On most machines, it is not necessary to define this macro.

  32805. 

  32806.  -- Macro: HONOR_REG_ALLOC_ORDER

  32807.      Normally, IRA tries to estimate the costs for saving a register in

  32808.      the prologue and restoring it in the epilogue.  This discourages it

  32809.      from using call-saved registers.  If a machine wants to ensure that

  32810.      IRA allocates registers in the order given by REG_ALLOC_ORDER even

  32811.      if some call-saved registers appear earlier than call-used ones,

  32812.      then define this macro as a C expression to nonzero.  Default is 0.

  32813. 

  32814.  -- Macro: IRA_HARD_REGNO_ADD_COST_MULTIPLIER (REGNO)

  32815.      In some case register allocation order is not enough for the

  32816.      Integrated Register Allocator (IRA) to generate a good code.  If

  32817.      this macro is defined, it should return a floating point value

  32818.      based on REGNO.  The cost of using REGNO for a pseudo will be

  32819.      increased by approximately the pseudo's usage frequency times the

  32820.      value returned by this macro.  Not defining this macro is

  32821.      equivalent to having it always return '0.0'.

  32822. 

  32823.      On most machines, it is not necessary to define this macro.

  32824. 

  32825. 

  32826. File: gccint.info,  Node: Values in Registers,  Next: Leaf Functions,  Prev: Allocation Order,  Up: Registers

  32827. 

  32828. 18.7.3 How Values Fit in Registers

  32829. ----------------------------------

  32830. 

  32831. This section discusses the macros that describe which kinds of values

  32832. (specifically, which machine modes) each register can hold, and how many

  32833. consecutive registers are needed for a given mode.

  32834. 

  32835.  -- Target Hook: unsigned int TARGET_HARD_REGNO_NREGS (unsigned int

  32836.           REGNO, machine_mode MODE)

  32837.      This hook returns the number of consecutive hard registers,

  32838.      starting at register number REGNO, required to hold a value of mode

  32839.      MODE.  This hook must never return zero, even if a register cannot

  32840.      hold the requested mode - indicate that with

  32841.      'TARGET_HARD_REGNO_MODE_OK' and/or 'TARGET_CAN_CHANGE_MODE_CLASS'

  32842.      instead.

  32843. 

  32844.      The default definition returns the number of words in MODE.

  32845. 

  32846.  -- Macro: HARD_REGNO_NREGS_HAS_PADDING (REGNO, MODE)

  32847.      A C expression that is nonzero if a value of mode MODE, stored in

  32848.      memory, ends with padding that causes it to take up more space than

  32849.      in registers starting at register number REGNO (as determined by

  32850.      multiplying GCC's notion of the size of the register when

  32851.      containing this mode by the number of registers returned by

  32852.      'TARGET_HARD_REGNO_NREGS').  By default this is zero.

  32853. 

  32854.      For example, if a floating-point value is stored in three 32-bit

  32855.      registers but takes up 128 bits in memory, then this would be

  32856.      nonzero.

  32857. 

  32858.      This macros only needs to be defined if there are cases where

  32859.      'subreg_get_info' would otherwise wrongly determine that a 'subreg'

  32860.      can be represented by an offset to the register number, when in

  32861.      fact such a 'subreg' would contain some of the padding not stored

  32862.      in registers and so not be representable.

  32863. 

  32864.  -- Macro: HARD_REGNO_NREGS_WITH_PADDING (REGNO, MODE)

  32865.      For values of REGNO and MODE for which

  32866.      'HARD_REGNO_NREGS_HAS_PADDING' returns nonzero, a C expression

  32867.      returning the greater number of registers required to hold the

  32868.      value including any padding.  In the example above, the value would

  32869.      be four.

  32870. 

  32871.  -- Macro: REGMODE_NATURAL_SIZE (MODE)

  32872.      Define this macro if the natural size of registers that hold values

  32873.      of mode MODE is not the word size.  It is a C expression that

  32874.      should give the natural size in bytes for the specified mode.  It

  32875.      is used by the register allocator to try to optimize its results.

  32876.      This happens for example on SPARC 64-bit where the natural size of

  32877.      floating-point registers is still 32-bit.

  32878. 

  32879.  -- Target Hook: bool TARGET_HARD_REGNO_MODE_OK (unsigned int REGNO,

  32880.           machine_mode MODE)

  32881.      This hook returns true if it is permissible to store a value of

  32882.      mode MODE in hard register number REGNO (or in several registers

  32883.      starting with that one).  The default definition returns true

  32884.      unconditionally.

  32885. 

  32886.      You need not include code to check for the numbers of fixed

  32887.      registers, because the allocation mechanism considers them to be

  32888.      always occupied.

  32889. 

  32890.      On some machines, double-precision values must be kept in even/odd

  32891.      register pairs.  You can implement that by defining this hook to

  32892.      reject odd register numbers for such modes.

  32893. 

  32894.      The minimum requirement for a mode to be OK in a register is that

  32895.      the 'movMODE' instruction pattern support moves between the

  32896.      register and other hard register in the same class and that moving

  32897.      a value into the register and back out not alter it.

  32898. 

  32899.      Since the same instruction used to move 'word_mode' will work for

  32900.      all narrower integer modes, it is not necessary on any machine for

  32901.      this hook to distinguish between these modes, provided you define

  32902.      patterns 'movhi', etc., to take advantage of this.  This is useful

  32903.      because of the interaction between 'TARGET_HARD_REGNO_MODE_OK' and

  32904.      'TARGET_MODES_TIEABLE_P'; it is very desirable for all integer

  32905.      modes to be tieable.

  32906. 

  32907.      Many machines have special registers for floating point arithmetic.

  32908.      Often people assume that floating point machine modes are allowed

  32909.      only in floating point registers.  This is not true.  Any registers

  32910.      that can hold integers can safely _hold_ a floating point machine

  32911.      mode, whether or not floating arithmetic can be done on it in those

  32912.      registers.  Integer move instructions can be used to move the

  32913.      values.

  32914. 

  32915.      On some machines, though, the converse is true: fixed-point machine

  32916.      modes may not go in floating registers.  This is true if the

  32917.      floating registers normalize any value stored in them, because

  32918.      storing a non-floating value there would garble it.  In this case,

  32919.      'TARGET_HARD_REGNO_MODE_OK' should reject fixed-point machine modes

  32920.      in floating registers.  But if the floating registers do not

  32921.      automatically normalize, if you can store any bit pattern in one

  32922.      and retrieve it unchanged without a trap, then any machine mode may

  32923.      go in a floating register, so you can define this hook to say so.

  32924. 

  32925.      The primary significance of special floating registers is rather

  32926.      that they are the registers acceptable in floating point arithmetic

  32927.      instructions.  However, this is of no concern to

  32928.      'TARGET_HARD_REGNO_MODE_OK'.  You handle it by writing the proper

  32929.      constraints for those instructions.

  32930. 

  32931.      On some machines, the floating registers are especially slow to

  32932.      access, so that it is better to store a value in a stack frame than

  32933.      in such a register if floating point arithmetic is not being done.

  32934.      As long as the floating registers are not in class 'GENERAL_REGS',

  32935.      they will not be used unless some pattern's constraint asks for

  32936.      one.

  32937. 

  32938.  -- Macro: HARD_REGNO_RENAME_OK (FROM, TO)

  32939.      A C expression that is nonzero if it is OK to rename a hard

  32940.      register FROM to another hard register TO.

  32941. 

  32942.      One common use of this macro is to prevent renaming of a register

  32943.      to another register that is not saved by a prologue in an interrupt

  32944.      handler.

  32945. 

  32946.      The default is always nonzero.

  32947. 

  32948.  -- Target Hook: bool TARGET_MODES_TIEABLE_P (machine_mode MODE1,

  32949.           machine_mode MODE2)

  32950.      This hook returns true if a value of mode MODE1 is accessible in

  32951.      mode MODE2 without copying.

  32952. 

  32953.      If 'TARGET_HARD_REGNO_MODE_OK (R, MODE1)' and

  32954.      'TARGET_HARD_REGNO_MODE_OK (R, MODE2)' are always the same for any

  32955.      R, then 'TARGET_MODES_TIEABLE_P (MODE1, MODE2)' should be true.  If

  32956.      they differ for any R, you should define this hook to return false

  32957.      unless some other mechanism ensures the accessibility of the value

  32958.      in a narrower mode.

  32959. 

  32960.      You should define this hook to return true in as many cases as

  32961.      possible since doing so will allow GCC to perform better register

  32962.      allocation.  The default definition returns true unconditionally.

  32963. 

  32964.  -- Target Hook: bool TARGET_HARD_REGNO_SCRATCH_OK (unsigned int REGNO)

  32965.      This target hook should return 'true' if it is OK to use a hard

  32966.      register REGNO as scratch reg in peephole2.

  32967. 

  32968.      One common use of this macro is to prevent using of a register that

  32969.      is not saved by a prologue in an interrupt handler.

  32970. 

  32971.      The default version of this hook always returns 'true'.

  32972. 

  32973.  -- Macro: AVOID_CCMODE_COPIES

  32974.      Define this macro if the compiler should avoid copies to/from

  32975.      'CCmode' registers.  You should only define this macro if support

  32976.      for copying to/from 'CCmode' is incomplete.

  32977. 

  32978. 

  32979. File: gccint.info,  Node: Leaf Functions,  Next: Stack Registers,  Prev: Values in Registers,  Up: Registers

  32980. 

  32981. 18.7.4 Handling Leaf Functions

  32982. ------------------------------

  32983. 

  32984. On some machines, a leaf function (i.e., one which makes no calls) can

  32985. run more efficiently if it does not make its own register window.  Often

  32986. this means it is required to receive its arguments in the registers

  32987. where they are passed by the caller, instead of the registers where they

  32988. would normally arrive.

  32989. 

  32990.  The special treatment for leaf functions generally applies only when

  32991. other conditions are met; for example, often they may use only those

  32992. registers for its own variables and temporaries.  We use the term "leaf

  32993. function" to mean a function that is suitable for this special handling,

  32994. so that functions with no calls are not necessarily "leaf functions".

  32995. 

  32996.  GCC assigns register numbers before it knows whether the function is

  32997. suitable for leaf function treatment.  So it needs to renumber the

  32998. registers in order to output a leaf function.  The following macros

  32999. accomplish this.

  33000. 

  33001.  -- Macro: LEAF_REGISTERS

  33002.      Name of a char vector, indexed by hard register number, which

  33003.      contains 1 for a register that is allowable in a candidate for leaf

  33004.      function treatment.

  33005. 

  33006.      If leaf function treatment involves renumbering the registers, then

  33007.      the registers marked here should be the ones before

  33008.      renumbering--those that GCC would ordinarily allocate.  The

  33009.      registers which will actually be used in the assembler code, after

  33010.      renumbering, should not be marked with 1 in this vector.

  33011. 

  33012.      Define this macro only if the target machine offers a way to

  33013.      optimize the treatment of leaf functions.

  33014. 

  33015.  -- Macro: LEAF_REG_REMAP (REGNO)

  33016.      A C expression whose value is the register number to which REGNO

  33017.      should be renumbered, when a function is treated as a leaf

  33018.      function.

  33019. 

  33020.      If REGNO is a register number which should not appear in a leaf

  33021.      function before renumbering, then the expression should yield -1,

  33022.      which will cause the compiler to abort.

  33023. 

  33024.      Define this macro only if the target machine offers a way to

  33025.      optimize the treatment of leaf functions, and registers need to be

  33026.      renumbered to do this.

  33027. 

  33028.  'TARGET_ASM_FUNCTION_PROLOGUE' and 'TARGET_ASM_FUNCTION_EPILOGUE' must

  33029. usually treat leaf functions specially.  They can test the C variable

  33030. 'current_function_is_leaf' which is nonzero for leaf functions.

  33031. 'current_function_is_leaf' is set prior to local register allocation and

  33032. is valid for the remaining compiler passes.  They can also test the C

  33033. variable 'current_function_uses_only_leaf_regs' which is nonzero for

  33034. leaf functions which only use leaf registers.

  33035. 'current_function_uses_only_leaf_regs' is valid after all passes that

  33036. modify the instructions have been run and is only useful if

  33037. 'LEAF_REGISTERS' is defined.

  33038. 

  33039. 

  33040. File: gccint.info,  Node: Stack Registers,  Prev: Leaf Functions,  Up: Registers

  33041. 

  33042. 18.7.5 Registers That Form a Stack

  33043. ----------------------------------

  33044. 

  33045. There are special features to handle computers where some of the

  33046. "registers" form a stack.  Stack registers are normally written by

  33047. pushing onto the stack, and are numbered relative to the top of the

  33048. stack.

  33049. 

  33050.  Currently, GCC can only handle one group of stack-like registers, and

  33051. they must be consecutively numbered.  Furthermore, the existing support

  33052. for stack-like registers is specific to the 80387 floating point

  33053. coprocessor.  If you have a new architecture that uses stack-like

  33054. registers, you will need to do substantial work on 'reg-stack.c' and

  33055. write your machine description to cooperate with it, as well as defining

  33056. these macros.

  33057. 

  33058.  -- Macro: STACK_REGS

  33059.      Define this if the machine has any stack-like registers.

  33060. 

  33061.  -- Macro: STACK_REG_COVER_CLASS

  33062.      This is a cover class containing the stack registers.  Define this

  33063.      if the machine has any stack-like registers.

  33064. 

  33065.  -- Macro: FIRST_STACK_REG

  33066.      The number of the first stack-like register.  This one is the top

  33067.      of the stack.

  33068. 

  33069.  -- Macro: LAST_STACK_REG

  33070.      The number of the last stack-like register.  This one is the bottom

  33071.      of the stack.

  33072. 

  33073. 

  33074. File: gccint.info,  Node: Register Classes,  Next: Stack and Calling,  Prev: Registers,  Up: Target Macros

  33075. 

  33076. 18.8 Register Classes

  33077. =====================

  33078. 

  33079. On many machines, the numbered registers are not all equivalent.  For

  33080. example, certain registers may not be allowed for indexed addressing;

  33081. certain registers may not be allowed in some instructions.  These

  33082. machine restrictions are described to the compiler using "register

  33083. classes".

  33084. 

  33085.  You define a number of register classes, giving each one a name and

  33086. saying which of the registers belong to it.  Then you can specify

  33087. register classes that are allowed as operands to particular instruction

  33088. patterns.

  33089. 

  33090.  In general, each register will belong to several classes.  In fact, one

  33091. class must be named 'ALL_REGS' and contain all the registers.  Another

  33092. class must be named 'NO_REGS' and contain no registers.  Often the union

  33093. of two classes will be another class; however, this is not required.

  33094. 

  33095.  One of the classes must be named 'GENERAL_REGS'.  There is nothing

  33096. terribly special about the name, but the operand constraint letters 'r'

  33097. and 'g' specify this class.  If 'GENERAL_REGS' is the same as

  33098. 'ALL_REGS', just define it as a macro which expands to 'ALL_REGS'.

  33099. 

  33100.  Order the classes so that if class X is contained in class Y then X has

  33101. a lower class number than Y.

  33102. 

  33103.  The way classes other than 'GENERAL_REGS' are specified in operand

  33104. constraints is through machine-dependent operand constraint letters.

  33105. You can define such letters to correspond to various classes, then use

  33106. them in operand constraints.

  33107. 

  33108.  You must define the narrowest register classes for allocatable

  33109. registers, so that each class either has no subclasses, or that for some

  33110. mode, the move cost between registers within the class is cheaper than

  33111. moving a register in the class to or from memory (*note Costs::).

  33112. 

  33113.  You should define a class for the union of two classes whenever some

  33114. instruction allows both classes.  For example, if an instruction allows

  33115. either a floating point (coprocessor) register or a general register for

  33116. a certain operand, you should define a class 'FLOAT_OR_GENERAL_REGS'

  33117. which includes both of them.  Otherwise you will get suboptimal code, or

  33118. even internal compiler errors when reload cannot find a register in the

  33119. class computed via 'reg_class_subunion'.

  33120. 

  33121.  You must also specify certain redundant information about the register

  33122. classes: for each class, which classes contain it and which ones are

  33123. contained in it; for each pair of classes, the largest class contained

  33124. in their union.

  33125. 

  33126.  When a value occupying several consecutive registers is expected in a

  33127. certain class, all the registers used must belong to that class.

  33128. Therefore, register classes cannot be used to enforce a requirement for

  33129. a register pair to start with an even-numbered register.  The way to

  33130. specify this requirement is with 'TARGET_HARD_REGNO_MODE_OK'.

  33131. 

  33132.  Register classes used for input-operands of bitwise-and or shift

  33133. instructions have a special requirement: each such class must have, for

  33134. each fixed-point machine mode, a subclass whose registers can transfer

  33135. that mode to or from memory.  For example, on some machines, the

  33136. operations for single-byte values ('QImode') are limited to certain

  33137. registers.  When this is so, each register class that is used in a

  33138. bitwise-and or shift instruction must have a subclass consisting of

  33139. registers from which single-byte values can be loaded or stored.  This

  33140. is so that 'PREFERRED_RELOAD_CLASS' can always have a possible value to

  33141. return.

  33142. 

  33143.  -- Data type: enum reg_class

  33144.      An enumerated type that must be defined with all the register class

  33145.      names as enumerated values.  'NO_REGS' must be first.  'ALL_REGS'

  33146.      must be the last register class, followed by one more enumerated

  33147.      value, 'LIM_REG_CLASSES', which is not a register class but rather

  33148.      tells how many classes there are.

  33149. 

  33150.      Each register class has a number, which is the value of casting the

  33151.      class name to type 'int'.  The number serves as an index in many of

  33152.      the tables described below.

  33153. 

  33154.  -- Macro: N_REG_CLASSES

  33155.      The number of distinct register classes, defined as follows:

  33156. 

  33157.           #define N_REG_CLASSES (int) LIM_REG_CLASSES

  33158. 

  33159.  -- Macro: REG_CLASS_NAMES

  33160.      An initializer containing the names of the register classes as C

  33161.      string constants.  These names are used in writing some of the

  33162.      debugging dumps.

  33163. 

  33164.  -- Macro: REG_CLASS_CONTENTS

  33165.      An initializer containing the contents of the register classes, as

  33166.      integers which are bit masks.  The Nth integer specifies the

  33167.      contents of class N.  The way the integer MASK is interpreted is

  33168.      that register R is in the class if 'MASK & (1 << R)' is 1.

  33169. 

  33170.      When the machine has more than 32 registers, an integer does not

  33171.      suffice.  Then the integers are replaced by sub-initializers,

  33172.      braced groupings containing several integers.  Each sub-initializer

  33173.      must be suitable as an initializer for the type 'HARD_REG_SET'

  33174.      which is defined in 'hard-reg-set.h'.  In this situation, the first

  33175.      integer in each sub-initializer corresponds to registers 0 through

  33176.      31, the second integer to registers 32 through 63, and so on.

  33177. 

  33178.  -- Macro: REGNO_REG_CLASS (REGNO)

  33179.      A C expression whose value is a register class containing hard

  33180.      register REGNO.  In general there is more than one such class;

  33181.      choose a class which is "minimal", meaning that no smaller class

  33182.      also contains the register.

  33183. 

  33184.  -- Macro: BASE_REG_CLASS

  33185.      A macro whose definition is the name of the class to which a valid

  33186.      base register must belong.  A base register is one used in an

  33187.      address which is the register value plus a displacement.

  33188. 

  33189.  -- Macro: MODE_BASE_REG_CLASS (MODE)

  33190.      This is a variation of the 'BASE_REG_CLASS' macro which allows the

  33191.      selection of a base register in a mode dependent manner.  If MODE

  33192.      is VOIDmode then it should return the same value as

  33193.      'BASE_REG_CLASS'.

  33194. 

  33195.  -- Macro: MODE_BASE_REG_REG_CLASS (MODE)

  33196.      A C expression whose value is the register class to which a valid

  33197.      base register must belong in order to be used in a base plus index

  33198.      register address.  You should define this macro if base plus index

  33199.      addresses have different requirements than other base register

  33200.      uses.

  33201. 

  33202.  -- Macro: MODE_CODE_BASE_REG_CLASS (MODE, ADDRESS_SPACE, OUTER_CODE,

  33203.           INDEX_CODE)

  33204.      A C expression whose value is the register class to which a valid

  33205.      base register for a memory reference in mode MODE to address space

  33206.      ADDRESS_SPACE must belong.  OUTER_CODE and INDEX_CODE define the

  33207.      context in which the base register occurs.  OUTER_CODE is the code

  33208.      of the immediately enclosing expression ('MEM' for the top level of

  33209.      an address, 'ADDRESS' for something that occurs in an

  33210.      'address_operand').  INDEX_CODE is the code of the corresponding

  33211.      index expression if OUTER_CODE is 'PLUS'; 'SCRATCH' otherwise.

  33212. 

  33213.  -- Macro: INDEX_REG_CLASS

  33214.      A macro whose definition is the name of the class to which a valid

  33215.      index register must belong.  An index register is one used in an

  33216.      address where its value is either multiplied by a scale factor or

  33217.      added to another register (as well as added to a displacement).

  33218. 

  33219.  -- Macro: REGNO_OK_FOR_BASE_P (NUM)

  33220.      A C expression which is nonzero if register number NUM is suitable

  33221.      for use as a base register in operand addresses.

  33222. 

  33223.  -- Macro: REGNO_MODE_OK_FOR_BASE_P (NUM, MODE)

  33224.      A C expression that is just like 'REGNO_OK_FOR_BASE_P', except that

  33225.      that expression may examine the mode of the memory reference in

  33226.      MODE.  You should define this macro if the mode of the memory

  33227.      reference affects whether a register may be used as a base

  33228.      register.  If you define this macro, the compiler will use it

  33229.      instead of 'REGNO_OK_FOR_BASE_P'.  The mode may be 'VOIDmode' for

  33230.      addresses that appear outside a 'MEM', i.e., as an

  33231.      'address_operand'.

  33232. 

  33233.  -- Macro: REGNO_MODE_OK_FOR_REG_BASE_P (NUM, MODE)

  33234.      A C expression which is nonzero if register number NUM is suitable

  33235.      for use as a base register in base plus index operand addresses,

  33236.      accessing memory in mode MODE.  It may be either a suitable hard

  33237.      register or a pseudo register that has been allocated such a hard

  33238.      register.  You should define this macro if base plus index

  33239.      addresses have different requirements than other base register

  33240.      uses.

  33241. 

  33242.      Use of this macro is deprecated; please use the more general

  33243.      'REGNO_MODE_CODE_OK_FOR_BASE_P'.

  33244. 

  33245.  -- Macro: REGNO_MODE_CODE_OK_FOR_BASE_P (NUM, MODE, ADDRESS_SPACE,

  33246.           OUTER_CODE, INDEX_CODE)

  33247.      A C expression which is nonzero if register number NUM is suitable

  33248.      for use as a base register in operand addresses, accessing memory

  33249.      in mode MODE in address space ADDRESS_SPACE.  This is similar to

  33250.      'REGNO_MODE_OK_FOR_BASE_P', except that that expression may examine

  33251.      the context in which the register appears in the memory reference.

  33252.      OUTER_CODE is the code of the immediately enclosing expression

  33253.      ('MEM' if at the top level of the address, 'ADDRESS' for something

  33254.      that occurs in an 'address_operand').  INDEX_CODE is the code of

  33255.      the corresponding index expression if OUTER_CODE is 'PLUS';

  33256.      'SCRATCH' otherwise.  The mode may be 'VOIDmode' for addresses that

  33257.      appear outside a 'MEM', i.e., as an 'address_operand'.

  33258. 

  33259.  -- Macro: REGNO_OK_FOR_INDEX_P (NUM)

  33260.      A C expression which is nonzero if register number NUM is suitable

  33261.      for use as an index register in operand addresses.  It may be

  33262.      either a suitable hard register or a pseudo register that has been

  33263.      allocated such a hard register.

  33264. 

  33265.      The difference between an index register and a base register is

  33266.      that the index register may be scaled.  If an address involves the

  33267.      sum of two registers, neither one of them scaled, then either one

  33268.      may be labeled the "base" and the other the "index"; but whichever

  33269.      labeling is used must fit the machine's constraints of which

  33270.      registers may serve in each capacity.  The compiler will try both

  33271.      labelings, looking for one that is valid, and will reload one or

  33272.      both registers only if neither labeling works.

  33273. 

  33274.  -- Target Hook: reg_class_t TARGET_PREFERRED_RENAME_CLASS (reg_class_t

  33275.           RCLASS)

  33276.      A target hook that places additional preference on the register

  33277.      class to use when it is necessary to rename a register in class

  33278.      RCLASS to another class, or perhaps NO_REGS, if no preferred

  33279.      register class is found or hook 'preferred_rename_class' is not

  33280.      implemented.  Sometimes returning a more restrictive class makes

  33281.      better code.  For example, on ARM, thumb-2 instructions using

  33282.      'LO_REGS' may be smaller than instructions using 'GENERIC_REGS'.

  33283.      By returning 'LO_REGS' from 'preferred_rename_class', code size can

  33284.      be reduced.

  33285. 

  33286.  -- Target Hook: reg_class_t TARGET_PREFERRED_RELOAD_CLASS (rtx X,

  33287.           reg_class_t RCLASS)

  33288.      A target hook that places additional restrictions on the register

  33289.      class to use when it is necessary to copy value X into a register

  33290.      in class RCLASS.  The value is a register class; perhaps RCLASS, or

  33291.      perhaps another, smaller class.

  33292. 

  33293.      The default version of this hook always returns value of 'rclass'

  33294.      argument.

  33295. 

  33296.      Sometimes returning a more restrictive class makes better code.

  33297.      For example, on the 68000, when X is an integer constant that is in

  33298.      range for a 'moveq' instruction, the value of this macro is always

  33299.      'DATA_REGS' as long as RCLASS includes the data registers.

  33300.      Requiring a data register guarantees that a 'moveq' will be used.

  33301. 

  33302.      One case where 'TARGET_PREFERRED_RELOAD_CLASS' must not return

  33303.      RCLASS is if X is a legitimate constant which cannot be loaded into

  33304.      some register class.  By returning 'NO_REGS' you can force X into a

  33305.      memory location.  For example, rs6000 can load immediate values

  33306.      into general-purpose registers, but does not have an instruction

  33307.      for loading an immediate value into a floating-point register, so

  33308.      'TARGET_PREFERRED_RELOAD_CLASS' returns 'NO_REGS' when X is a

  33309.      floating-point constant.  If the constant can't be loaded into any

  33310.      kind of register, code generation will be better if

  33311.      'TARGET_LEGITIMATE_CONSTANT_P' makes the constant illegitimate

  33312.      instead of using 'TARGET_PREFERRED_RELOAD_CLASS'.

  33313. 

  33314.      If an insn has pseudos in it after register allocation, reload will

  33315.      go through the alternatives and call repeatedly

  33316.      'TARGET_PREFERRED_RELOAD_CLASS' to find the best one.  Returning

  33317.      'NO_REGS', in this case, makes reload add a '!' in front of the

  33318.      constraint: the x86 back-end uses this feature to discourage usage

  33319.      of 387 registers when math is done in the SSE registers (and vice

  33320.      versa).

  33321. 

  33322.  -- Macro: PREFERRED_RELOAD_CLASS (X, CLASS)

  33323.      A C expression that places additional restrictions on the register

  33324.      class to use when it is necessary to copy value X into a register

  33325.      in class CLASS.  The value is a register class; perhaps CLASS, or

  33326.      perhaps another, smaller class.  On many machines, the following

  33327.      definition is safe:

  33328. 

  33329.           #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS

  33330. 

  33331.      Sometimes returning a more restrictive class makes better code.

  33332.      For example, on the 68000, when X is an integer constant that is in

  33333.      range for a 'moveq' instruction, the value of this macro is always

  33334.      'DATA_REGS' as long as CLASS includes the data registers.

  33335.      Requiring a data register guarantees that a 'moveq' will be used.

  33336. 

  33337.      One case where 'PREFERRED_RELOAD_CLASS' must not return CLASS is if

  33338.      X is a legitimate constant which cannot be loaded into some

  33339.      register class.  By returning 'NO_REGS' you can force X into a

  33340.      memory location.  For example, rs6000 can load immediate values

  33341.      into general-purpose registers, but does not have an instruction

  33342.      for loading an immediate value into a floating-point register, so

  33343.      'PREFERRED_RELOAD_CLASS' returns 'NO_REGS' when X is a

  33344.      floating-point constant.  If the constant cannot be loaded into any

  33345.      kind of register, code generation will be better if

  33346.      'TARGET_LEGITIMATE_CONSTANT_P' makes the constant illegitimate

  33347.      instead of using 'TARGET_PREFERRED_RELOAD_CLASS'.

  33348. 

  33349.      If an insn has pseudos in it after register allocation, reload will

  33350.      go through the alternatives and call repeatedly

  33351.      'PREFERRED_RELOAD_CLASS' to find the best one.  Returning

  33352.      'NO_REGS', in this case, makes reload add a '!' in front of the

  33353.      constraint: the x86 back-end uses this feature to discourage usage

  33354.      of 387 registers when math is done in the SSE registers (and vice

  33355.      versa).

  33356. 

  33357.  -- Target Hook: reg_class_t TARGET_PREFERRED_OUTPUT_RELOAD_CLASS (rtx

  33358.           X, reg_class_t RCLASS)

  33359.      Like 'TARGET_PREFERRED_RELOAD_CLASS', but for output reloads

  33360.      instead of input reloads.

  33361. 

  33362.      The default version of this hook always returns value of 'rclass'

  33363.      argument.

  33364. 

  33365.      You can also use 'TARGET_PREFERRED_OUTPUT_RELOAD_CLASS' to

  33366.      discourage reload from using some alternatives, like

  33367.      'TARGET_PREFERRED_RELOAD_CLASS'.

  33368. 

  33369.  -- Macro: LIMIT_RELOAD_CLASS (MODE, CLASS)

  33370.      A C expression that places additional restrictions on the register

  33371.      class to use when it is necessary to be able to hold a value of

  33372.      mode MODE in a reload register for which class CLASS would

  33373.      ordinarily be used.

  33374. 

  33375.      Unlike 'PREFERRED_RELOAD_CLASS', this macro should be used when

  33376.      there are certain modes that simply cannot go in certain reload

  33377.      classes.

  33378. 

  33379.      The value is a register class; perhaps CLASS, or perhaps another,

  33380.      smaller class.

  33381. 

  33382.      Don't define this macro unless the target machine has limitations

  33383.      which require the macro to do something nontrivial.

  33384. 

  33385.  -- Target Hook: reg_class_t TARGET_SECONDARY_RELOAD (bool IN_P, rtx X,

  33386.           reg_class_t RELOAD_CLASS, machine_mode RELOAD_MODE,

  33387.           secondary_reload_info *SRI)

  33388.      Many machines have some registers that cannot be copied directly to

  33389.      or from memory or even from other types of registers.  An example

  33390.      is the 'MQ' register, which on most machines, can only be copied to

  33391.      or from general registers, but not memory.  Below, we shall be

  33392.      using the term 'intermediate register' when a move operation cannot

  33393.      be performed directly, but has to be done by copying the source

  33394.      into the intermediate register first, and then copying the

  33395.      intermediate register to the destination.  An intermediate register

  33396.      always has the same mode as source and destination.  Since it holds

  33397.      the actual value being copied, reload might apply optimizations to

  33398.      re-use an intermediate register and eliding the copy from the

  33399.      source when it can determine that the intermediate register still

  33400.      holds the required value.

  33401. 

  33402.      Another kind of secondary reload is required on some machines which

  33403.      allow copying all registers to and from memory, but require a

  33404.      scratch register for stores to some memory locations (e.g., those

  33405.      with symbolic address on the RT, and those with certain symbolic

  33406.      address on the SPARC when compiling PIC).  Scratch registers need

  33407.      not have the same mode as the value being copied, and usually hold

  33408.      a different value than that being copied.  Special patterns in the

  33409.      md file are needed to describe how the copy is performed with the

  33410.      help of the scratch register; these patterns also describe the

  33411.      number, register class(es) and mode(s) of the scratch register(s).

  33412. 

  33413.      In some cases, both an intermediate and a scratch register are

  33414.      required.

  33415. 

  33416.      For input reloads, this target hook is called with nonzero IN_P,

  33417.      and X is an rtx that needs to be copied to a register of class

  33418.      RELOAD_CLASS in RELOAD_MODE.  For output reloads, this target hook

  33419.      is called with zero IN_P, and a register of class RELOAD_CLASS

  33420.      needs to be copied to rtx X in RELOAD_MODE.

  33421. 

  33422.      If copying a register of RELOAD_CLASS from/to X requires an

  33423.      intermediate register, the hook 'secondary_reload' should return

  33424.      the register class required for this intermediate register.  If no

  33425.      intermediate register is required, it should return NO_REGS. If

  33426.      more than one intermediate register is required, describe the one

  33427.      that is closest in the copy chain to the reload register.

  33428. 

  33429.      If scratch registers are needed, you also have to describe how to

  33430.      perform the copy from/to the reload register to/from this closest

  33431.      intermediate register.  Or if no intermediate register is required,

  33432.      but still a scratch register is needed, describe the copy from/to

  33433.      the reload register to/from the reload operand X.

  33434. 

  33435.      You do this by setting 'sri->icode' to the instruction code of a

  33436.      pattern in the md file which performs the move.  Operands 0 and 1

  33437.      are the output and input of this copy, respectively.  Operands from

  33438.      operand 2 onward are for scratch operands.  These scratch operands

  33439.      must have a mode, and a single-register-class output constraint.

  33440. 

  33441.      When an intermediate register is used, the 'secondary_reload' hook

  33442.      will be called again to determine how to copy the intermediate

  33443.      register to/from the reload operand X, so your hook must also have

  33444.      code to handle the register class of the intermediate operand.

  33445. 

  33446.      X might be a pseudo-register or a 'subreg' of a pseudo-register,

  33447.      which could either be in a hard register or in memory.  Use

  33448.      'true_regnum' to find out; it will return -1 if the pseudo is in

  33449.      memory and the hard register number if it is in a register.

  33450. 

  33451.      Scratch operands in memory (constraint '"=m"' / '"=&m"') are

  33452.      currently not supported.  For the time being, you will have to

  33453.      continue to use 'TARGET_SECONDARY_MEMORY_NEEDED' for that purpose.

  33454. 

  33455.      'copy_cost' also uses this target hook to find out how values are

  33456.      copied.  If you want it to include some extra cost for the need to

  33457.      allocate (a) scratch register(s), set 'sri->extra_cost' to the

  33458.      additional cost.  Or if two dependent moves are supposed to have a

  33459.      lower cost than the sum of the individual moves due to expected

  33460.      fortuitous scheduling and/or special forwarding logic, you can set

  33461.      'sri->extra_cost' to a negative amount.

  33462. 

  33463.  -- Macro: SECONDARY_RELOAD_CLASS (CLASS, MODE, X)

  33464.  -- Macro: SECONDARY_INPUT_RELOAD_CLASS (CLASS, MODE, X)

  33465.  -- Macro: SECONDARY_OUTPUT_RELOAD_CLASS (CLASS, MODE, X)

  33466.      These macros are obsolete, new ports should use the target hook

  33467.      'TARGET_SECONDARY_RELOAD' instead.

  33468. 

  33469.      These are obsolete macros, replaced by the

  33470.      'TARGET_SECONDARY_RELOAD' target hook.  Older ports still define

  33471.      these macros to indicate to the reload phase that it may need to

  33472.      allocate at least one register for a reload in addition to the

  33473.      register to contain the data.  Specifically, if copying X to a

  33474.      register CLASS in MODE requires an intermediate register, you were

  33475.      supposed to define 'SECONDARY_INPUT_RELOAD_CLASS' to return the

  33476.      largest register class all of whose registers can be used as

  33477.      intermediate registers or scratch registers.

  33478. 

  33479.      If copying a register CLASS in MODE to X requires an intermediate

  33480.      or scratch register, 'SECONDARY_OUTPUT_RELOAD_CLASS' was supposed

  33481.      to be defined be defined to return the largest register class

  33482.      required.  If the requirements for input and output reloads were

  33483.      the same, the macro 'SECONDARY_RELOAD_CLASS' should have been used

  33484.      instead of defining both macros identically.

  33485. 

  33486.      The values returned by these macros are often 'GENERAL_REGS'.

  33487.      Return 'NO_REGS' if no spare register is needed; i.e., if X can be

  33488.      directly copied to or from a register of CLASS in MODE without

  33489.      requiring a scratch register.  Do not define this macro if it would

  33490.      always return 'NO_REGS'.

  33491. 

  33492.      If a scratch register is required (either with or without an

  33493.      intermediate register), you were supposed to define patterns for

  33494.      'reload_inM' or 'reload_outM', as required (*note Standard Names::.

  33495.      These patterns, which were normally implemented with a

  33496.      'define_expand', should be similar to the 'movM' patterns, except

  33497.      that operand 2 is the scratch register.

  33498. 

  33499.      These patterns need constraints for the reload register and scratch

  33500.      register that contain a single register class.  If the original

  33501.      reload register (whose class is CLASS) can meet the constraint

  33502.      given in the pattern, the value returned by these macros is used

  33503.      for the class of the scratch register.  Otherwise, two additional

  33504.      reload registers are required.  Their classes are obtained from the

  33505.      constraints in the insn pattern.

  33506. 

  33507.      X might be a pseudo-register or a 'subreg' of a pseudo-register,

  33508.      which could either be in a hard register or in memory.  Use

  33509.      'true_regnum' to find out; it will return -1 if the pseudo is in

  33510.      memory and the hard register number if it is in a register.

  33511. 

  33512.      These macros should not be used in the case where a particular

  33513.      class of registers can only be copied to memory and not to another

  33514.      class of registers.  In that case, secondary reload registers are

  33515.      not needed and would not be helpful.  Instead, a stack location

  33516.      must be used to perform the copy and the 'movM' pattern should use

  33517.      memory as an intermediate storage.  This case often occurs between

  33518.      floating-point and general registers.

  33519. 

  33520.  -- Target Hook: bool TARGET_SECONDARY_MEMORY_NEEDED (machine_mode MODE,

  33521.           reg_class_t CLASS1, reg_class_t CLASS2)

  33522.      Certain machines have the property that some registers cannot be

  33523.      copied to some other registers without using memory.  Define this

  33524.      hook on those machines to return true if objects of mode M in

  33525.      registers of CLASS1 can only be copied to registers of class CLASS2

  33526.      by storing a register of CLASS1 into memory and loading that memory

  33527.      location into a register of CLASS2.  The default definition returns

  33528.      false for all inputs.

  33529. 

  33530.  -- Macro: SECONDARY_MEMORY_NEEDED_RTX (MODE)

  33531.      Normally when 'TARGET_SECONDARY_MEMORY_NEEDED' is defined, the

  33532.      compiler allocates a stack slot for a memory location needed for

  33533.      register copies.  If this macro is defined, the compiler instead

  33534.      uses the memory location defined by this macro.

  33535. 

  33536.      Do not define this macro if you do not define

  33537.      'TARGET_SECONDARY_MEMORY_NEEDED'.

  33538. 

  33539.  -- Target Hook: machine_mode TARGET_SECONDARY_MEMORY_NEEDED_MODE

  33540.           (machine_mode MODE)

  33541.      If 'TARGET_SECONDARY_MEMORY_NEEDED' tells the compiler to use

  33542.      memory when moving between two particular registers of mode MODE,

  33543.      this hook specifies the mode that the memory should have.

  33544. 

  33545.      The default depends on 'TARGET_LRA_P'.  Without LRA, the default is

  33546.      to use a word-sized mode for integral modes that are smaller than a

  33547.      a word.  This is right thing to do on most machines because it

  33548.      ensures that all bits of the register are copied and prevents

  33549.      accesses to the registers in a narrower mode, which some machines

  33550.      prohibit for floating-point registers.

  33551. 

  33552.      However, this default behavior is not correct on some machines,

  33553.      such as the DEC Alpha, that store short integers in floating-point

  33554.      registers differently than in integer registers.  On those

  33555.      machines, the default widening will not work correctly and you must

  33556.      define this hook to suppress that widening in some cases.  See the

  33557.      file 'alpha.c' for details.

  33558. 

  33559.      With LRA, the default is to use MODE unmodified.

  33560. 

  33561.  -- Target Hook: void TARGET_SELECT_EARLY_REMAT_MODES (sbitmap MODES)

  33562.      On some targets, certain modes cannot be held in registers around a

  33563.      standard ABI call and are relatively expensive to spill to the

  33564.      stack.  The early rematerialization pass can help in such cases by

  33565.      aggressively recomputing values after calls, so that they don't

  33566.      need to be spilled.

  33567. 

  33568.      This hook returns the set of such modes by setting the associated

  33569.      bits in MODES.  The default implementation selects no modes, which

  33570.      has the effect of disabling the early rematerialization pass.

  33571. 

  33572.  -- Target Hook: bool TARGET_CLASS_LIKELY_SPILLED_P (reg_class_t RCLASS)

  33573.      A target hook which returns 'true' if pseudos that have been

  33574.      assigned to registers of class RCLASS would likely be spilled

  33575.      because registers of RCLASS are needed for spill registers.

  33576. 

  33577.      The default version of this target hook returns 'true' if RCLASS

  33578.      has exactly one register and 'false' otherwise.  On most machines,

  33579.      this default should be used.  For generally register-starved

  33580.      machines, such as i386, or machines with right register

  33581.      constraints, such as SH, this hook can be used to avoid excessive

  33582.      spilling.

  33583. 

  33584.      This hook is also used by some of the global intra-procedural code

  33585.      transformations to throtle code motion, to avoid increasing

  33586.      register pressure.

  33587. 

  33588.  -- Target Hook: unsigned char TARGET_CLASS_MAX_NREGS (reg_class_t

  33589.           RCLASS, machine_mode MODE)

  33590.      A target hook returns the maximum number of consecutive registers

  33591.      of class RCLASS needed to hold a value of mode MODE.

  33592. 

  33593.      This is closely related to the macro 'TARGET_HARD_REGNO_NREGS'.  In

  33594.      fact, the value returned by 'TARGET_CLASS_MAX_NREGS (RCLASS, MODE)'

  33595.      target hook should be the maximum value of 'TARGET_HARD_REGNO_NREGS

  33596.      (REGNO, MODE)' for all REGNO values in the class RCLASS.

  33597. 

  33598.      This target hook helps control the handling of multiple-word values

  33599.      in the reload pass.

  33600. 

  33601.      The default version of this target hook returns the size of MODE in

  33602.      words.

  33603. 

  33604.  -- Macro: CLASS_MAX_NREGS (CLASS, MODE)

  33605.      A C expression for the maximum number of consecutive registers of

  33606.      class CLASS needed to hold a value of mode MODE.

  33607. 

  33608.      This is closely related to the macro 'TARGET_HARD_REGNO_NREGS'.  In

  33609.      fact, the value of the macro 'CLASS_MAX_NREGS (CLASS, MODE)' should

  33610.      be the maximum value of 'TARGET_HARD_REGNO_NREGS (REGNO, MODE)' for

  33611.      all REGNO values in the class CLASS.

  33612. 

  33613.      This macro helps control the handling of multiple-word values in

  33614.      the reload pass.

  33615. 

  33616.  -- Target Hook: bool TARGET_CAN_CHANGE_MODE_CLASS (machine_mode FROM,

  33617.           machine_mode TO, reg_class_t RCLASS)

  33618.      This hook returns true if it is possible to bitcast values held in

  33619.      registers of class RCLASS from mode FROM to mode TO and if doing so

  33620.      preserves the low-order bits that are common to both modes.  The

  33621.      result is only meaningful if RCLASS has registers that can hold

  33622.      both 'from' and 'to'.  The default implementation returns true.

  33623. 

  33624.      As an example of when such bitcasting is invalid, loading 32-bit

  33625.      integer or floating-point objects into floating-point registers on

  33626.      Alpha extends them to 64 bits.  Therefore loading a 64-bit object

  33627.      and then storing it as a 32-bit object does not store the low-order

  33628.      32 bits, as would be the case for a normal register.  Therefore,

  33629.      'alpha.h' defines 'TARGET_CAN_CHANGE_MODE_CLASS' to return:

  33630. 

  33631.           (GET_MODE_SIZE (from) == GET_MODE_SIZE (to)

  33632.            || !reg_classes_intersect_p (FLOAT_REGS, rclass))

  33633. 

  33634.      Even if storing from a register in mode TO would be valid, if both

  33635.      FROM and 'raw_reg_mode' for RCLASS are wider than 'word_mode', then

  33636.      we must prevent TO narrowing the mode.  This happens when the

  33637.      middle-end assumes that it can load or store pieces of an N-word

  33638.      pseudo, and that the pseudo will eventually be allocated to N

  33639.      'word_mode' hard registers.  Failure to prevent this kind of mode

  33640.      change will result in the entire 'raw_reg_mode' being modified

  33641.      instead of the partial value that the middle-end intended.

  33642. 

  33643.  -- Target Hook: reg_class_t TARGET_IRA_CHANGE_PSEUDO_ALLOCNO_CLASS

  33644.           (int, REG_CLASS_T, REG_CLASS_T)

  33645.      A target hook which can change allocno class for given pseudo from

  33646.      allocno and best class calculated by IRA.

  33647. 

  33648.      The default version of this target hook always returns given class.

  33649. 

  33650.  -- Target Hook: bool TARGET_LRA_P (void)

  33651.      A target hook which returns true if we use LRA instead of reload

  33652.      pass.  The default version of this target hook returns true.  New

  33653.      ports should use LRA, and existing ports are encouraged to convert.

  33654. 

  33655.  -- Target Hook: int TARGET_REGISTER_PRIORITY (int)

  33656.      A target hook which returns the register priority number to which

  33657.      the register HARD_REGNO belongs to.  The bigger the number, the

  33658.      more preferable the hard register usage (when all other conditions

  33659.      are the same).  This hook can be used to prefer some hard register

  33660.      over others in LRA. For example, some x86-64 register usage needs

  33661.      additional prefix which makes instructions longer.  The hook can

  33662.      return lower priority number for such registers make them less

  33663.      favorable and as result making the generated code smaller.  The

  33664.      default version of this target hook returns always zero.

  33665. 

  33666.  -- Target Hook: bool TARGET_REGISTER_USAGE_LEVELING_P (void)

  33667.      A target hook which returns true if we need register usage

  33668.      leveling.  That means if a few hard registers are equally good for

  33669.      the assignment, we choose the least used hard register.  The

  33670.      register usage leveling may be profitable for some targets.  Don't

  33671.      use the usage leveling for targets with conditional execution or

  33672.      targets with big register files as it hurts if-conversion and

  33673.      cross-jumping optimizations.  The default version of this target

  33674.      hook returns always false.

  33675. 

  33676.  -- Target Hook: bool TARGET_DIFFERENT_ADDR_DISPLACEMENT_P (void)

  33677.      A target hook which returns true if an address with the same

  33678.      structure can have different maximal legitimate displacement.  For

  33679.      example, the displacement can depend on memory mode or on operand

  33680.      combinations in the insn.  The default version of this target hook

  33681.      returns always false.

  33682. 

  33683.  -- Target Hook: bool TARGET_CANNOT_SUBSTITUTE_MEM_EQUIV_P (rtx SUBST)

  33684.      A target hook which returns 'true' if SUBST can't substitute safely

  33685.      pseudos with equivalent memory values during register allocation.

  33686.      The default version of this target hook returns 'false'.  On most

  33687.      machines, this default should be used.  For generally machines with

  33688.      non orthogonal register usage for addressing, such as SH, this hook

  33689.      can be used to avoid excessive spilling.

  33690. 

  33691.  -- Target Hook: bool TARGET_LEGITIMIZE_ADDRESS_DISPLACEMENT (rtx

  33692.           *OFFSET1, rtx *OFFSET2, poly_int64 ORIG_OFFSET, machine_mode

  33693.           MODE)

  33694.      This hook tries to split address offset ORIG_OFFSET into two parts:

  33695.      one that should be added to the base address to create a local

  33696.      anchor point, and an additional offset that can be applied to the

  33697.      anchor to address a value of mode MODE.  The idea is that the local

  33698.      anchor could be shared by other accesses to nearby locations.

  33699. 

  33700.      The hook returns true if it succeeds, storing the offset of the

  33701.      anchor from the base in OFFSET1 and the offset of the final address

  33702.      from the anchor in OFFSET2.  The default implementation returns

  33703.      false.

  33704. 

  33705.  -- Target Hook: reg_class_t TARGET_SPILL_CLASS (reg_class_t,

  33706.           MACHINE_MODE)

  33707.      This hook defines a class of registers which could be used for

  33708.      spilling pseudos of the given mode and class, or 'NO_REGS' if only

  33709.      memory should be used.  Not defining this hook is equivalent to

  33710.      returning 'NO_REGS' for all inputs.

  33711. 

  33712.  -- Target Hook: bool TARGET_ADDITIONAL_ALLOCNO_CLASS_P (reg_class_t)

  33713.      This hook should return 'true' if given class of registers should

  33714.      be an allocno class in any way.  Usually RA uses only one register

  33715.      class from all classes containing the same register set.  In some

  33716.      complicated cases, you need to have two or more such classes as

  33717.      allocno ones for RA correct work.  Not defining this hook is

  33718.      equivalent to returning 'false' for all inputs.

  33719. 

  33720.  -- Target Hook: scalar_int_mode TARGET_CSTORE_MODE (enum insn_code

  33721.           ICODE)

  33722.      This hook defines the machine mode to use for the boolean result of

  33723.      conditional store patterns.  The ICODE argument is the instruction

  33724.      code for the cstore being performed.  Not definiting this hook is

  33725.      the same as accepting the mode encoded into operand 0 of the cstore

  33726.      expander patterns.

  33727. 

  33728.  -- Target Hook: int TARGET_COMPUTE_PRESSURE_CLASSES (enum reg_class

  33729.           *PRESSURE_CLASSES)

  33730.      A target hook which lets a backend compute the set of pressure

  33731.      classes to be used by those optimization passes which take register

  33732.      pressure into account, as opposed to letting IRA compute them.  It

  33733.      returns the number of register classes stored in the array

  33734.      PRESSURE_CLASSES.

  33735. 

  33736. 

  33737. File: gccint.info,  Node: Stack and Calling,  Next: Varargs,  Prev: Register Classes,  Up: Target Macros

  33738. 

  33739. 18.9 Stack Layout and Calling Conventions

  33740. =========================================

  33741. 

  33742. This describes the stack layout and calling conventions.

  33743. 

  33744. * Menu:

  33745. 

  33746. * Frame Layout::

  33747. * Exception Handling::

  33748. * Stack Checking::

  33749. * Frame Registers::

  33750. * Elimination::

  33751. * Stack Arguments::

  33752. * Register Arguments::

  33753. * Scalar Return::

  33754. * Aggregate Return::

  33755. * Caller Saves::

  33756. * Function Entry::

  33757. * Profiling::

  33758. * Tail Calls::

  33759. * Shrink-wrapping separate components::

  33760. * Stack Smashing Protection::

  33761. * Miscellaneous Register Hooks::

  33762. 

  33763. 

  33764. File: gccint.info,  Node: Frame Layout,  Next: Exception Handling,  Up: Stack and Calling

  33765. 

  33766. 18.9.1 Basic Stack Layout

  33767. -------------------------

  33768. 

  33769. Here is the basic stack layout.

  33770. 

  33771.  -- Macro: STACK_GROWS_DOWNWARD

  33772.      Define this macro to be true if pushing a word onto the stack moves

  33773.      the stack pointer to a smaller address, and false otherwise.

  33774. 

  33775.  -- Macro: STACK_PUSH_CODE

  33776.      This macro defines the operation used when something is pushed on

  33777.      the stack.  In RTL, a push operation will be '(set (mem

  33778.      (STACK_PUSH_CODE (reg sp))) ...)'

  33779. 

  33780.      The choices are 'PRE_DEC', 'POST_DEC', 'PRE_INC', and 'POST_INC'.

  33781.      Which of these is correct depends on the stack direction and on

  33782.      whether the stack pointer points to the last item on the stack or

  33783.      whether it points to the space for the next item on the stack.

  33784. 

  33785.      The default is 'PRE_DEC' when 'STACK_GROWS_DOWNWARD' is true, which

  33786.      is almost always right, and 'PRE_INC' otherwise, which is often

  33787.      wrong.

  33788. 

  33789.  -- Macro: FRAME_GROWS_DOWNWARD

  33790.      Define this macro to nonzero value if the addresses of local

  33791.      variable slots are at negative offsets from the frame pointer.

  33792. 

  33793.  -- Macro: ARGS_GROW_DOWNWARD

  33794.      Define this macro if successive arguments to a function occupy

  33795.      decreasing addresses on the stack.

  33796. 

  33797.  -- Target Hook: HOST_WIDE_INT TARGET_STARTING_FRAME_OFFSET (void)

  33798.      This hook returns the offset from the frame pointer to the first

  33799.      local variable slot to be allocated.  If 'FRAME_GROWS_DOWNWARD', it

  33800.      is the offset to _end_ of the first slot allocated, otherwise it is

  33801.      the offset to _beginning_ of the first slot allocated.  The default

  33802.      implementation returns 0.

  33803. 

  33804.  -- Macro: STACK_ALIGNMENT_NEEDED

  33805.      Define to zero to disable final alignment of the stack during

  33806.      reload.  The nonzero default for this macro is suitable for most

  33807.      ports.

  33808. 

  33809.      On ports where 'TARGET_STARTING_FRAME_OFFSET' is nonzero or where

  33810.      there is a register save block following the local block that

  33811.      doesn't require alignment to 'STACK_BOUNDARY', it may be beneficial

  33812.      to disable stack alignment and do it in the backend.

  33813. 

  33814.  -- Macro: STACK_POINTER_OFFSET

  33815.      Offset from the stack pointer register to the first location at

  33816.      which outgoing arguments are placed.  If not specified, the default

  33817.      value of zero is used.  This is the proper value for most machines.

  33818. 

  33819.      If 'ARGS_GROW_DOWNWARD', this is the offset to the location above

  33820.      the first location at which outgoing arguments are placed.

  33821. 

  33822.  -- Macro: FIRST_PARM_OFFSET (FUNDECL)

  33823.      Offset from the argument pointer register to the first argument's

  33824.      address.  On some machines it may depend on the data type of the

  33825.      function.

  33826. 

  33827.      If 'ARGS_GROW_DOWNWARD', this is the offset to the location above

  33828.      the first argument's address.

  33829. 

  33830.  -- Macro: STACK_DYNAMIC_OFFSET (FUNDECL)

  33831.      Offset from the stack pointer register to an item dynamically

  33832.      allocated on the stack, e.g., by 'alloca'.

  33833. 

  33834.      The default value for this macro is 'STACK_POINTER_OFFSET' plus the

  33835.      length of the outgoing arguments.  The default is correct for most

  33836.      machines.  See 'function.c' for details.

  33837. 

  33838.  -- Macro: INITIAL_FRAME_ADDRESS_RTX

  33839.      A C expression whose value is RTL representing the address of the

  33840.      initial stack frame.  This address is passed to 'RETURN_ADDR_RTX'

  33841.      and 'DYNAMIC_CHAIN_ADDRESS'.  If you don't define this macro, a

  33842.      reasonable default value will be used.  Define this macro in order

  33843.      to make frame pointer elimination work in the presence of

  33844.      '__builtin_frame_address (count)' and '__builtin_return_address

  33845.      (count)' for 'count' not equal to zero.

  33846. 

  33847.  -- Macro: DYNAMIC_CHAIN_ADDRESS (FRAMEADDR)

  33848.      A C expression whose value is RTL representing the address in a

  33849.      stack frame where the pointer to the caller's frame is stored.

  33850.      Assume that FRAMEADDR is an RTL expression for the address of the

  33851.      stack frame itself.

  33852. 

  33853.      If you don't define this macro, the default is to return the value

  33854.      of FRAMEADDR--that is, the stack frame address is also the address

  33855.      of the stack word that points to the previous frame.

  33856. 

  33857.  -- Macro: SETUP_FRAME_ADDRESSES

  33858.      A C expression that produces the machine-specific code to setup the

  33859.      stack so that arbitrary frames can be accessed.  For example, on

  33860.      the SPARC, we must flush all of the register windows to the stack

  33861.      before we can access arbitrary stack frames.  You will seldom need

  33862.      to define this macro.  The default is to do nothing.

  33863. 

  33864.  -- Target Hook: rtx TARGET_BUILTIN_SETJMP_FRAME_VALUE (void)

  33865.      This target hook should return an rtx that is used to store the

  33866.      address of the current frame into the built in 'setjmp' buffer.

  33867.      The default value, 'virtual_stack_vars_rtx', is correct for most

  33868.      machines.  One reason you may need to define this target hook is if

  33869.      'hard_frame_pointer_rtx' is the appropriate value on your machine.

  33870. 

  33871.  -- Macro: FRAME_ADDR_RTX (FRAMEADDR)

  33872.      A C expression whose value is RTL representing the value of the

  33873.      frame address for the current frame.  FRAMEADDR is the frame

  33874.      pointer of the current frame.  This is used for

  33875.      __builtin_frame_address.  You need only define this macro if the

  33876.      frame address is not the same as the frame pointer.  Most machines

  33877.      do not need to define it.

  33878. 

  33879.  -- Macro: RETURN_ADDR_RTX (COUNT, FRAMEADDR)

  33880.      A C expression whose value is RTL representing the value of the

  33881.      return address for the frame COUNT steps up from the current frame,

  33882.      after the prologue.  FRAMEADDR is the frame pointer of the COUNT

  33883.      frame, or the frame pointer of the COUNT - 1 frame if

  33884.      'RETURN_ADDR_IN_PREVIOUS_FRAME' is nonzero.

  33885. 

  33886.      The value of the expression must always be the correct address when

  33887.      COUNT is zero, but may be 'NULL_RTX' if there is no way to

  33888.      determine the return address of other frames.

  33889. 

  33890.  -- Macro: RETURN_ADDR_IN_PREVIOUS_FRAME

  33891.      Define this macro to nonzero value if the return address of a

  33892.      particular stack frame is accessed from the frame pointer of the

  33893.      previous stack frame.  The zero default for this macro is suitable

  33894.      for most ports.

  33895. 

  33896.  -- Macro: INCOMING_RETURN_ADDR_RTX

  33897.      A C expression whose value is RTL representing the location of the

  33898.      incoming return address at the beginning of any function, before

  33899.      the prologue.  This RTL is either a 'REG', indicating that the

  33900.      return value is saved in 'REG', or a 'MEM' representing a location

  33901.      in the stack.

  33902. 

  33903.      You only need to define this macro if you want to support call

  33904.      frame debugging information like that provided by DWARF 2.

  33905. 

  33906.      If this RTL is a 'REG', you should also define

  33907.      'DWARF_FRAME_RETURN_COLUMN' to 'DWARF_FRAME_REGNUM (REGNO)'.

  33908. 

  33909.  -- Macro: DWARF_ALT_FRAME_RETURN_COLUMN

  33910.      A C expression whose value is an integer giving a DWARF 2 column

  33911.      number that may be used as an alternative return column.  The

  33912.      column must not correspond to any gcc hard register (that is, it

  33913.      must not be in the range of 'DWARF_FRAME_REGNUM').

  33914. 

  33915.      This macro can be useful if 'DWARF_FRAME_RETURN_COLUMN' is set to a

  33916.      general register, but an alternative column needs to be used for

  33917.      signal frames.  Some targets have also used different frame return

  33918.      columns over time.

  33919. 

  33920.  -- Macro: DWARF_ZERO_REG

  33921.      A C expression whose value is an integer giving a DWARF 2 register

  33922.      number that is considered to always have the value zero.  This

  33923.      should only be defined if the target has an architected zero

  33924.      register, and someone decided it was a good idea to use that

  33925.      register number to terminate the stack backtrace.  New ports should

  33926.      avoid this.

  33927. 

  33928.  -- Target Hook: void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char

  33929.           *LABEL, rtx PATTERN, int INDEX)

  33930.      This target hook allows the backend to emit frame-related insns

  33931.      that contain UNSPECs or UNSPEC_VOLATILEs.  The DWARF 2 call frame

  33932.      debugging info engine will invoke it on insns of the form

  33933.           (set (reg) (unspec [...] UNSPEC_INDEX))

  33934.      and

  33935.           (set (reg) (unspec_volatile [...] UNSPECV_INDEX)).

  33936.      to let the backend emit the call frame instructions.  LABEL is the

  33937.      CFI label attached to the insn, PATTERN is the pattern of the insn

  33938.      and INDEX is 'UNSPEC_INDEX' or 'UNSPECV_INDEX'.

  33939. 

  33940.  -- Target Hook: unsigned int TARGET_DWARF_POLY_INDETERMINATE_VALUE

  33941.           (unsigned int I, unsigned int *FACTOR, int *OFFSET)

  33942.      Express the value of 'poly_int' indeterminate I as a DWARF

  33943.      expression, with I counting from 1.  Return the number of a DWARF

  33944.      register R and set '*FACTOR' and '*OFFSET' such that the value of

  33945.      the indeterminate is:

  33946.           value_of(R) / FACTOR - OFFSET

  33947. 

  33948.      A target only needs to define this hook if it sets

  33949.      'NUM_POLY_INT_COEFFS' to a value greater than 1.

  33950. 

  33951.  -- Macro: INCOMING_FRAME_SP_OFFSET

  33952.      A C expression whose value is an integer giving the offset, in

  33953.      bytes, from the value of the stack pointer register to the top of

  33954.      the stack frame at the beginning of any function, before the

  33955.      prologue.  The top of the frame is defined to be the value of the

  33956.      stack pointer in the previous frame, just before the call

  33957.      instruction.

  33958. 

  33959.      You only need to define this macro if you want to support call

  33960.      frame debugging information like that provided by DWARF 2.

  33961. 

  33962.  -- Macro: DEFAULT_INCOMING_FRAME_SP_OFFSET

  33963.      Like 'INCOMING_FRAME_SP_OFFSET', but must be the same for all

  33964.      functions of the same ABI, and when using GAS '.cfi_*' directives

  33965.      must also agree with the default CFI GAS emits.  Define this macro

  33966.      only if 'INCOMING_FRAME_SP_OFFSET' can have different values

  33967.      between different functions of the same ABI or when

  33968.      'INCOMING_FRAME_SP_OFFSET' does not agree with GAS default CFI.

  33969. 

  33970.  -- Macro: ARG_POINTER_CFA_OFFSET (FUNDECL)

  33971.      A C expression whose value is an integer giving the offset, in

  33972.      bytes, from the argument pointer to the canonical frame address

  33973.      (cfa).  The final value should coincide with that calculated by

  33974.      'INCOMING_FRAME_SP_OFFSET'.  Which is unfortunately not usable

  33975.      during virtual register instantiation.

  33976. 

  33977.      The default value for this macro is 'FIRST_PARM_OFFSET (fundecl) +

  33978.      crtl->args.pretend_args_size', which is correct for most machines;

  33979.      in general, the arguments are found immediately before the stack

  33980.      frame.  Note that this is not the case on some targets that save

  33981.      registers into the caller's frame, such as SPARC and rs6000, and so

  33982.      such targets need to define this macro.

  33983. 

  33984.      You only need to define this macro if the default is incorrect, and

  33985.      you want to support call frame debugging information like that

  33986.      provided by DWARF 2.

  33987. 

  33988.  -- Macro: FRAME_POINTER_CFA_OFFSET (FUNDECL)

  33989.      If defined, a C expression whose value is an integer giving the

  33990.      offset in bytes from the frame pointer to the canonical frame

  33991.      address (cfa).  The final value should coincide with that

  33992.      calculated by 'INCOMING_FRAME_SP_OFFSET'.

  33993. 

  33994.      Normally the CFA is calculated as an offset from the argument

  33995.      pointer, via 'ARG_POINTER_CFA_OFFSET', but if the argument pointer

  33996.      is variable due to the ABI, this may not be possible.  If this

  33997.      macro is defined, it implies that the virtual register

  33998.      instantiation should be based on the frame pointer instead of the

  33999.      argument pointer.  Only one of 'FRAME_POINTER_CFA_OFFSET' and

  34000.      'ARG_POINTER_CFA_OFFSET' should be defined.

  34001. 

  34002.  -- Macro: CFA_FRAME_BASE_OFFSET (FUNDECL)

  34003.      If defined, a C expression whose value is an integer giving the

  34004.      offset in bytes from the canonical frame address (cfa) to the frame

  34005.      base used in DWARF 2 debug information.  The default is zero.  A

  34006.      different value may reduce the size of debug information on some

  34007.      ports.

  34008. 

  34009. 

  34010. File: gccint.info,  Node: Exception Handling,  Next: Stack Checking,  Prev: Frame Layout,  Up: Stack and Calling

  34011. 

  34012. 18.9.2 Exception Handling Support

  34013. ---------------------------------

  34014. 

  34015.  -- Macro: EH_RETURN_DATA_REGNO (N)

  34016.      A C expression whose value is the Nth register number used for data

  34017.      by exception handlers, or 'INVALID_REGNUM' if fewer than N

  34018.      registers are usable.

  34019. 

  34020.      The exception handling library routines communicate with the

  34021.      exception handlers via a set of agreed upon registers.  Ideally

  34022.      these registers should be call-clobbered; it is possible to use

  34023.      call-saved registers, but may negatively impact code size.  The

  34024.      target must support at least 2 data registers, but should define 4

  34025.      if there are enough free registers.

  34026. 

  34027.      You must define this macro if you want to support call frame

  34028.      exception handling like that provided by DWARF 2.

  34029. 

  34030.  -- Macro: EH_RETURN_STACKADJ_RTX

  34031.      A C expression whose value is RTL representing a location in which

  34032.      to store a stack adjustment to be applied before function return.

  34033.      This is used to unwind the stack to an exception handler's call

  34034.      frame.  It will be assigned zero on code paths that return

  34035.      normally.

  34036. 

  34037.      Typically this is a call-clobbered hard register that is otherwise

  34038.      untouched by the epilogue, but could also be a stack slot.

  34039. 

  34040.      Do not define this macro if the stack pointer is saved and restored

  34041.      by the regular prolog and epilog code in the call frame itself; in

  34042.      this case, the exception handling library routines will update the

  34043.      stack location to be restored in place.  Otherwise, you must define

  34044.      this macro if you want to support call frame exception handling

  34045.      like that provided by DWARF 2.

  34046. 

  34047.  -- Macro: EH_RETURN_HANDLER_RTX

  34048.      A C expression whose value is RTL representing a location in which

  34049.      to store the address of an exception handler to which we should

  34050.      return.  It will not be assigned on code paths that return

  34051.      normally.

  34052. 

  34053.      Typically this is the location in the call frame at which the

  34054.      normal return address is stored.  For targets that return by

  34055.      popping an address off the stack, this might be a memory address

  34056.      just below the _target_ call frame rather than inside the current

  34057.      call frame.  If defined, 'EH_RETURN_STACKADJ_RTX' will have already

  34058.      been assigned, so it may be used to calculate the location of the

  34059.      target call frame.

  34060. 

  34061.      Some targets have more complex requirements than storing to an

  34062.      address calculable during initial code generation.  In that case

  34063.      the 'eh_return' instruction pattern should be used instead.

  34064. 

  34065.      If you want to support call frame exception handling, you must

  34066.      define either this macro or the 'eh_return' instruction pattern.

  34067. 

  34068.  -- Macro: RETURN_ADDR_OFFSET

  34069.      If defined, an integer-valued C expression for which rtl will be

  34070.      generated to add it to the exception handler address before it is

  34071.      searched in the exception handling tables, and to subtract it again

  34072.      from the address before using it to return to the exception

  34073.      handler.

  34074. 

  34075.  -- Macro: ASM_PREFERRED_EH_DATA_FORMAT (CODE, GLOBAL)

  34076.      This macro chooses the encoding of pointers embedded in the

  34077.      exception handling sections.  If at all possible, this should be

  34078.      defined such that the exception handling section will not require

  34079.      dynamic relocations, and so may be read-only.

  34080. 

  34081.      CODE is 0 for data, 1 for code labels, 2 for function pointers.

  34082.      GLOBAL is true if the symbol may be affected by dynamic

  34083.      relocations.  The macro should return a combination of the

  34084.      'DW_EH_PE_*' defines as found in 'dwarf2.h'.

  34085. 

  34086.      If this macro is not defined, pointers will not be encoded but

  34087.      represented directly.

  34088. 

  34089.  -- Macro: ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (FILE, ENCODING, SIZE,

  34090.           ADDR, DONE)

  34091.      This macro allows the target to emit whatever special magic is

  34092.      required to represent the encoding chosen by

  34093.      'ASM_PREFERRED_EH_DATA_FORMAT'.  Generic code takes care of

  34094.      pc-relative and indirect encodings; this must be defined if the

  34095.      target uses text-relative or data-relative encodings.

  34096. 

  34097.      This is a C statement that branches to DONE if the format was

  34098.      handled.  ENCODING is the format chosen, SIZE is the number of

  34099.      bytes that the format occupies, ADDR is the 'SYMBOL_REF' to be

  34100.      emitted.

  34101. 

  34102.  -- Macro: MD_FALLBACK_FRAME_STATE_FOR (CONTEXT, FS)

  34103.      This macro allows the target to add CPU and operating system

  34104.      specific code to the call-frame unwinder for use when there is no

  34105.      unwind data available.  The most common reason to implement this

  34106.      macro is to unwind through signal frames.

  34107. 

  34108.      This macro is called from 'uw_frame_state_for' in 'unwind-dw2.c',

  34109.      'unwind-dw2-xtensa.c' and 'unwind-ia64.c'.  CONTEXT is an

  34110.      '_Unwind_Context'; FS is an '_Unwind_FrameState'.  Examine

  34111.      'context->ra' for the address of the code being executed and

  34112.      'context->cfa' for the stack pointer value.  If the frame can be

  34113.      decoded, the register save addresses should be updated in FS and

  34114.      the macro should evaluate to '_URC_NO_REASON'.  If the frame cannot

  34115.      be decoded, the macro should evaluate to '_URC_END_OF_STACK'.

  34116. 

  34117.      For proper signal handling in Java this macro is accompanied by

  34118.      'MAKE_THROW_FRAME', defined in 'libjava/include/*-signal.h'

  34119.      headers.

  34120. 

  34121.  -- Macro: MD_HANDLE_UNWABI (CONTEXT, FS)

  34122.      This macro allows the target to add operating system specific code

  34123.      to the call-frame unwinder to handle the IA-64 '.unwabi' unwinding

  34124.      directive, usually used for signal or interrupt frames.

  34125. 

  34126.      This macro is called from 'uw_update_context' in libgcc's

  34127.      'unwind-ia64.c'.  CONTEXT is an '_Unwind_Context'; FS is an

  34128.      '_Unwind_FrameState'.  Examine 'fs->unwabi' for the abi and context

  34129.      in the '.unwabi' directive.  If the '.unwabi' directive can be

  34130.      handled, the register save addresses should be updated in FS.

  34131. 

  34132.  -- Macro: TARGET_USES_WEAK_UNWIND_INFO

  34133.      A C expression that evaluates to true if the target requires unwind

  34134.      info to be given comdat linkage.  Define it to be '1' if comdat

  34135.      linkage is necessary.  The default is '0'.

  34136. 

  34137. 

  34138. File: gccint.info,  Node: Stack Checking,  Next: Frame Registers,  Prev: Exception Handling,  Up: Stack and Calling

  34139. 

  34140. 18.9.3 Specifying How Stack Checking is Done

  34141. --------------------------------------------

  34142. 

  34143. GCC will check that stack references are within the boundaries of the

  34144. stack, if the option '-fstack-check' is specified, in one of three ways:

  34145. 

  34146.   1. If the value of the 'STACK_CHECK_BUILTIN' macro is nonzero, GCC

  34147.      will assume that you have arranged for full stack checking to be

  34148.      done at appropriate places in the configuration files.  GCC will

  34149.      not do other special processing.

  34150. 

  34151.   2. If 'STACK_CHECK_BUILTIN' is zero and the value of the

  34152.      'STACK_CHECK_STATIC_BUILTIN' macro is nonzero, GCC will assume that

  34153.      you have arranged for static stack checking (checking of the static

  34154.      stack frame of functions) to be done at appropriate places in the

  34155.      configuration files.  GCC will only emit code to do dynamic stack

  34156.      checking (checking on dynamic stack allocations) using the third

  34157.      approach below.

  34158. 

  34159.   3. If neither of the above are true, GCC will generate code to

  34160.      periodically "probe" the stack pointer using the values of the

  34161.      macros defined below.

  34162. 

  34163.  If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is

  34164. defined, GCC will change its allocation strategy for large objects if

  34165. the option '-fstack-check' is specified: they will always be allocated

  34166. dynamically if their size exceeds 'STACK_CHECK_MAX_VAR_SIZE' bytes.

  34167. 

  34168.  -- Macro: STACK_CHECK_BUILTIN

  34169.      A nonzero value if stack checking is done by the configuration

  34170.      files in a machine-dependent manner.  You should define this macro

  34171.      if stack checking is required by the ABI of your machine or if you

  34172.      would like to do stack checking in some more efficient way than the

  34173.      generic approach.  The default value of this macro is zero.

  34174. 

  34175.  -- Macro: STACK_CHECK_STATIC_BUILTIN

  34176.      A nonzero value if static stack checking is done by the

  34177.      configuration files in a machine-dependent manner.  You should

  34178.      define this macro if you would like to do static stack checking in

  34179.      some more efficient way than the generic approach.  The default

  34180.      value of this macro is zero.

  34181. 

  34182.  -- Macro: STACK_CHECK_PROBE_INTERVAL_EXP

  34183.      An integer specifying the interval at which GCC must generate stack

  34184.      probe instructions, defined as 2 raised to this integer.  You will

  34185.      normally define this macro so that the interval be no larger than

  34186.      the size of the "guard pages" at the end of a stack area.  The

  34187.      default value of 12 (4096-byte interval) is suitable for most

  34188.      systems.

  34189. 

  34190.  -- Macro: STACK_CHECK_MOVING_SP

  34191.      An integer which is nonzero if GCC should move the stack pointer

  34192.      page by page when doing probes.  This can be necessary on systems

  34193.      where the stack pointer contains the bottom address of the memory

  34194.      area accessible to the executing thread at any point in time.  In

  34195.      this situation an alternate signal stack is required in order to be

  34196.      able to recover from a stack overflow.  The default value of this

  34197.      macro is zero.

  34198. 

  34199.  -- Macro: STACK_CHECK_PROTECT

  34200.      The number of bytes of stack needed to recover from a stack

  34201.      overflow, for languages where such a recovery is supported.  The

  34202.      default value of 4KB/8KB with the 'setjmp'/'longjmp'-based

  34203.      exception handling mechanism and 8KB/12KB with other exception

  34204.      handling mechanisms should be adequate for most architectures and

  34205.      operating systems.

  34206. 

  34207.  The following macros are relevant only if neither STACK_CHECK_BUILTIN

  34208. nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether

  34209. in the opposite case.

  34210. 

  34211.  -- Macro: STACK_CHECK_MAX_FRAME_SIZE

  34212.      The maximum size of a stack frame, in bytes.  GCC will generate

  34213.      probe instructions in non-leaf functions to ensure at least this

  34214.      many bytes of stack are available.  If a stack frame is larger than

  34215.      this size, stack checking will not be reliable and GCC will issue a

  34216.      warning.  The default is chosen so that GCC only generates one

  34217.      instruction on most systems.  You should normally not change the

  34218.      default value of this macro.

  34219. 

  34220.  -- Macro: STACK_CHECK_FIXED_FRAME_SIZE

  34221.      GCC uses this value to generate the above warning message.  It

  34222.      represents the amount of fixed frame used by a function, not

  34223.      including space for any callee-saved registers, temporaries and

  34224.      user variables.  You need only specify an upper bound for this

  34225.      amount and will normally use the default of four words.

  34226. 

  34227.  -- Macro: STACK_CHECK_MAX_VAR_SIZE

  34228.      The maximum size, in bytes, of an object that GCC will place in the

  34229.      fixed area of the stack frame when the user specifies

  34230.      '-fstack-check'.  GCC computed the default from the values of the

  34231.      above macros and you will normally not need to override that

  34232.      default.

  34233. 

  34234.  -- Target Hook: HOST_WIDE_INT

  34235.           TARGET_STACK_CLASH_PROTECTION_ALLOCA_PROBE_RANGE (void)

  34236.      Some targets have an ABI defined interval for which no probing

  34237.      needs to be done.  When a probe does need to be done this same

  34238.      interval is used as the probe distance up when doing stack clash

  34239.      protection for alloca.  On such targets this value can be set to

  34240.      override the default probing up interval.  Define this variable to

  34241.      return nonzero if such a probe range is required or zero otherwise.

  34242.      Defining this hook also requires your functions which make use of

  34243.      alloca to have at least 8 byesof outgoing arguments.  If this is

  34244.      not the case the stack will be corrupted.  You need not define this

  34245.      macro if it would always have the value zero.

  34246. 

  34247. 

  34248. File: gccint.info,  Node: Frame Registers,  Next: Elimination,  Prev: Stack Checking,  Up: Stack and Calling

  34249. 

  34250. 18.9.4 Registers That Address the Stack Frame

  34251. ---------------------------------------------

  34252. 

  34253. This discusses registers that address the stack frame.

  34254. 

  34255.  -- Macro: STACK_POINTER_REGNUM

  34256.      The register number of the stack pointer register, which must also

  34257.      be a fixed register according to 'FIXED_REGISTERS'.  On most

  34258.      machines, the hardware determines which register this is.

  34259. 

  34260.  -- Macro: FRAME_POINTER_REGNUM

  34261.      The register number of the frame pointer register, which is used to

  34262.      access automatic variables in the stack frame.  On some machines,

  34263.      the hardware determines which register this is.  On other machines,

  34264.      you can choose any register you wish for this purpose.

  34265. 

  34266.  -- Macro: HARD_FRAME_POINTER_REGNUM

  34267.      On some machines the offset between the frame pointer and starting

  34268.      offset of the automatic variables is not known until after register

  34269.      allocation has been done (for example, because the saved registers

  34270.      are between these two locations).  On those machines, define

  34271.      'FRAME_POINTER_REGNUM' the number of a special, fixed register to

  34272.      be used internally until the offset is known, and define

  34273.      'HARD_FRAME_POINTER_REGNUM' to be the actual hard register number

  34274.      used for the frame pointer.

  34275. 

  34276.      You should define this macro only in the very rare circumstances

  34277.      when it is not possible to calculate the offset between the frame

  34278.      pointer and the automatic variables until after register allocation

  34279.      has been completed.  When this macro is defined, you must also

  34280.      indicate in your definition of 'ELIMINABLE_REGS' how to eliminate

  34281.      'FRAME_POINTER_REGNUM' into either 'HARD_FRAME_POINTER_REGNUM' or

  34282.      'STACK_POINTER_REGNUM'.

  34283. 

  34284.      Do not define this macro if it would be the same as

  34285.      'FRAME_POINTER_REGNUM'.

  34286. 

  34287.  -- Macro: ARG_POINTER_REGNUM

  34288.      The register number of the arg pointer register, which is used to

  34289.      access the function's argument list.  On some machines, this is the

  34290.      same as the frame pointer register.  On some machines, the hardware

  34291.      determines which register this is.  On other machines, you can

  34292.      choose any register you wish for this purpose.  If this is not the

  34293.      same register as the frame pointer register, then you must mark it

  34294.      as a fixed register according to 'FIXED_REGISTERS', or arrange to

  34295.      be able to eliminate it (*note Elimination::).

  34296. 

  34297.  -- Macro: HARD_FRAME_POINTER_IS_FRAME_POINTER

  34298.      Define this to a preprocessor constant that is nonzero if

  34299.      'hard_frame_pointer_rtx' and 'frame_pointer_rtx' should be the

  34300.      same.  The default definition is '(HARD_FRAME_POINTER_REGNUM ==

  34301.      FRAME_POINTER_REGNUM)'; you only need to define this macro if that

  34302.      definition is not suitable for use in preprocessor conditionals.

  34303. 

  34304.  -- Macro: HARD_FRAME_POINTER_IS_ARG_POINTER

  34305.      Define this to a preprocessor constant that is nonzero if

  34306.      'hard_frame_pointer_rtx' and 'arg_pointer_rtx' should be the same.

  34307.      The default definition is '(HARD_FRAME_POINTER_REGNUM ==

  34308.      ARG_POINTER_REGNUM)'; you only need to define this macro if that

  34309.      definition is not suitable for use in preprocessor conditionals.

  34310. 

  34311.  -- Macro: RETURN_ADDRESS_POINTER_REGNUM

  34312.      The register number of the return address pointer register, which

  34313.      is used to access the current function's return address from the

  34314.      stack.  On some machines, the return address is not at a fixed

  34315.      offset from the frame pointer or stack pointer or argument pointer.

  34316.      This register can be defined to point to the return address on the

  34317.      stack, and then be converted by 'ELIMINABLE_REGS' into either the

  34318.      frame pointer or stack pointer.

  34319. 

  34320.      Do not define this macro unless there is no other way to get the

  34321.      return address from the stack.

  34322. 

  34323.  -- Macro: STATIC_CHAIN_REGNUM

  34324.  -- Macro: STATIC_CHAIN_INCOMING_REGNUM

  34325.      Register numbers used for passing a function's static chain

  34326.      pointer.  If register windows are used, the register number as seen

  34327.      by the called function is 'STATIC_CHAIN_INCOMING_REGNUM', while the

  34328.      register number as seen by the calling function is

  34329.      'STATIC_CHAIN_REGNUM'.  If these registers are the same,

  34330.      'STATIC_CHAIN_INCOMING_REGNUM' need not be defined.

  34331. 

  34332.      The static chain register need not be a fixed register.

  34333. 

  34334.      If the static chain is passed in memory, these macros should not be

  34335.      defined; instead, the 'TARGET_STATIC_CHAIN' hook should be used.

  34336. 

  34337.  -- Target Hook: rtx TARGET_STATIC_CHAIN (const_tree FNDECL_OR_TYPE,

  34338.           bool INCOMING_P)

  34339.      This hook replaces the use of 'STATIC_CHAIN_REGNUM' et al for

  34340.      targets that may use different static chain locations for different

  34341.      nested functions.  This may be required if the target has function

  34342.      attributes that affect the calling conventions of the function and

  34343.      those calling conventions use different static chain locations.

  34344. 

  34345.      The default version of this hook uses 'STATIC_CHAIN_REGNUM' et al.

  34346. 

  34347.      If the static chain is passed in memory, this hook should be used

  34348.      to provide rtx giving 'mem' expressions that denote where they are

  34349.      stored.  Often the 'mem' expression as seen by the caller will be

  34350.      at an offset from the stack pointer and the 'mem' expression as

  34351.      seen by the callee will be at an offset from the frame pointer.

  34352.      The variables 'stack_pointer_rtx', 'frame_pointer_rtx', and

  34353.      'arg_pointer_rtx' will have been initialized and should be used to

  34354.      refer to those items.

  34355. 

  34356.  -- Macro: DWARF_FRAME_REGISTERS

  34357.      This macro specifies the maximum number of hard registers that can

  34358.      be saved in a call frame.  This is used to size data structures

  34359.      used in DWARF2 exception handling.

  34360. 

  34361.      Prior to GCC 3.0, this macro was needed in order to establish a

  34362.      stable exception handling ABI in the face of adding new hard

  34363.      registers for ISA extensions.  In GCC 3.0 and later, the EH ABI is

  34364.      insulated from changes in the number of hard registers.

  34365.      Nevertheless, this macro can still be used to reduce the runtime

  34366.      memory requirements of the exception handling routines, which can

  34367.      be substantial if the ISA contains a lot of registers that are not

  34368.      call-saved.

  34369. 

  34370.      If this macro is not defined, it defaults to

  34371.      'FIRST_PSEUDO_REGISTER'.

  34372. 

  34373.  -- Macro: PRE_GCC3_DWARF_FRAME_REGISTERS

  34374. 

  34375.      This macro is similar to 'DWARF_FRAME_REGISTERS', but is provided

  34376.      for backward compatibility in pre GCC 3.0 compiled code.

  34377. 

  34378.      If this macro is not defined, it defaults to

  34379.      'DWARF_FRAME_REGISTERS'.

  34380. 

  34381.  -- Macro: DWARF_REG_TO_UNWIND_COLUMN (REGNO)

  34382. 

  34383.      Define this macro if the target's representation for dwarf

  34384.      registers is different than the internal representation for unwind

  34385.      column.  Given a dwarf register, this macro should return the

  34386.      internal unwind column number to use instead.

  34387. 

  34388.  -- Macro: DWARF_FRAME_REGNUM (REGNO)

  34389. 

  34390.      Define this macro if the target's representation for dwarf

  34391.      registers used in .eh_frame or .debug_frame is different from that

  34392.      used in other debug info sections.  Given a GCC hard register

  34393.      number, this macro should return the .eh_frame register number.

  34394.      The default is 'DBX_REGISTER_NUMBER (REGNO)'.

  34395. 

  34396.  -- Macro: DWARF2_FRAME_REG_OUT (REGNO, FOR_EH)

  34397. 

  34398.      Define this macro to map register numbers held in the call frame

  34399.      info that GCC has collected using 'DWARF_FRAME_REGNUM' to those

  34400.      that should be output in .debug_frame ('FOR_EH' is zero) and

  34401.      .eh_frame ('FOR_EH' is nonzero).  The default is to return 'REGNO'.

  34402. 

  34403.  -- Macro: REG_VALUE_IN_UNWIND_CONTEXT

  34404. 

  34405.      Define this macro if the target stores register values as

  34406.      '_Unwind_Word' type in unwind context.  It should be defined if

  34407.      target register size is larger than the size of 'void *'.  The

  34408.      default is to store register values as 'void *' type.

  34409. 

  34410.  -- Macro: ASSUME_EXTENDED_UNWIND_CONTEXT

  34411. 

  34412.      Define this macro to be 1 if the target always uses extended unwind

  34413.      context with version, args_size and by_value fields.  If it is

  34414.      undefined, it will be defined to 1 when

  34415.      'REG_VALUE_IN_UNWIND_CONTEXT' is defined and 0 otherwise.

  34416. 

  34417.  -- Macro: DWARF_LAZY_REGISTER_VALUE (REGNO, VALUE)

  34418.      Define this macro if the target has pseudo DWARF registers whose

  34419.      values need to be computed lazily on demand by the unwinder (such

  34420.      as when referenced in a CFA expression).  The macro returns true if

  34421.      REGNO is such a register and stores its value in '*VALUE' if so.

  34422. 

  34423. 

  34424. File: gccint.info,  Node: Elimination,  Next: Stack Arguments,  Prev: Frame Registers,  Up: Stack and Calling

  34425. 

  34426. 18.9.5 Eliminating Frame Pointer and Arg Pointer

  34427. ------------------------------------------------

  34428. 

  34429. This is about eliminating the frame pointer and arg pointer.

  34430. 

  34431.  -- Target Hook: bool TARGET_FRAME_POINTER_REQUIRED (void)

  34432.      This target hook should return 'true' if a function must have and

  34433.      use a frame pointer.  This target hook is called in the reload

  34434.      pass.  If its return value is 'true' the function will have a frame

  34435.      pointer.

  34436. 

  34437.      This target hook can in principle examine the current function and

  34438.      decide according to the facts, but on most machines the constant

  34439.      'false' or the constant 'true' suffices.  Use 'false' when the

  34440.      machine allows code to be generated with no frame pointer, and

  34441.      doing so saves some time or space.  Use 'true' when there is no

  34442.      possible advantage to avoiding a frame pointer.

  34443. 

  34444.      In certain cases, the compiler does not know how to produce valid

  34445.      code without a frame pointer.  The compiler recognizes those cases

  34446.      and automatically gives the function a frame pointer regardless of

  34447.      what 'targetm.frame_pointer_required' returns.  You don't need to

  34448.      worry about them.

  34449. 

  34450.      In a function that does not require a frame pointer, the frame

  34451.      pointer register can be allocated for ordinary usage, unless you

  34452.      mark it as a fixed register.  See 'FIXED_REGISTERS' for more

  34453.      information.

  34454. 

  34455.      Default return value is 'false'.

  34456. 

  34457.  -- Macro: ELIMINABLE_REGS

  34458.      This macro specifies a table of register pairs used to eliminate

  34459.      unneeded registers that point into the stack frame.

  34460. 

  34461.      The definition of this macro is a list of structure

  34462.      initializations, each of which specifies an original and

  34463.      replacement register.

  34464. 

  34465.      On some machines, the position of the argument pointer is not known

  34466.      until the compilation is completed.  In such a case, a separate

  34467.      hard register must be used for the argument pointer.  This register

  34468.      can be eliminated by replacing it with either the frame pointer or

  34469.      the argument pointer, depending on whether or not the frame pointer

  34470.      has been eliminated.

  34471. 

  34472.      In this case, you might specify:

  34473.           #define ELIMINABLE_REGS  \

  34474.           {{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM}, \

  34475.            {ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM}, \

  34476.            {FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM}}

  34477. 

  34478.      Note that the elimination of the argument pointer with the stack

  34479.      pointer is specified first since that is the preferred elimination.

  34480. 

  34481.  -- Target Hook: bool TARGET_CAN_ELIMINATE (const int FROM_REG, const

  34482.           int TO_REG)

  34483.      This target hook should return 'true' if the compiler is allowed to

  34484.      try to replace register number FROM_REG with register number

  34485.      TO_REG.  This target hook will usually be 'true', since most of the

  34486.      cases preventing register elimination are things that the compiler

  34487.      already knows about.

  34488. 

  34489.      Default return value is 'true'.

  34490. 

  34491.  -- Macro: INITIAL_ELIMINATION_OFFSET (FROM-REG, TO-REG, OFFSET-VAR)

  34492.      This macro returns the initial difference between the specified

  34493.      pair of registers.  The value would be computed from information

  34494.      such as the result of 'get_frame_size ()' and the tables of

  34495.      registers 'df_regs_ever_live_p' and 'call_used_regs'.

  34496. 

  34497.  -- Target Hook: void TARGET_COMPUTE_FRAME_LAYOUT (void)

  34498.      This target hook is called once each time the frame layout needs to

  34499.      be recalculated.  The calculations can be cached by the target and

  34500.      can then be used by 'INITIAL_ELIMINATION_OFFSET' instead of

  34501.      re-computing the layout on every invocation of that hook.  This is

  34502.      particularly useful for targets that have an expensive frame layout

  34503.      function.  Implementing this callback is optional.

  34504. 

  34505. 

  34506. File: gccint.info,  Node: Stack Arguments,  Next: Register Arguments,  Prev: Elimination,  Up: Stack and Calling

  34507. 

  34508. 18.9.6 Passing Function Arguments on the Stack

  34509. ----------------------------------------------

  34510. 

  34511. The macros in this section control how arguments are passed on the

  34512. stack.  See the following section for other macros that control passing

  34513. certain arguments in registers.

  34514. 

  34515.  -- Target Hook: bool TARGET_PROMOTE_PROTOTYPES (const_tree FNTYPE)

  34516.      This target hook returns 'true' if an argument declared in a

  34517.      prototype as an integral type smaller than 'int' should actually be

  34518.      passed as an 'int'.  In addition to avoiding errors in certain

  34519.      cases of mismatch, it also makes for better code on certain

  34520.      machines.  The default is to not promote prototypes.

  34521. 

  34522.  -- Macro: PUSH_ARGS

  34523.      A C expression.  If nonzero, push insns will be used to pass

  34524.      outgoing arguments.  If the target machine does not have a push

  34525.      instruction, set it to zero.  That directs GCC to use an alternate

  34526.      strategy: to allocate the entire argument block and then store the

  34527.      arguments into it.  When 'PUSH_ARGS' is nonzero, 'PUSH_ROUNDING'

  34528.      must be defined too.

  34529. 

  34530.  -- Macro: PUSH_ARGS_REVERSED

  34531.      A C expression.  If nonzero, function arguments will be evaluated

  34532.      from last to first, rather than from first to last.  If this macro

  34533.      is not defined, it defaults to 'PUSH_ARGS' on targets where the

  34534.      stack and args grow in opposite directions, and 0 otherwise.

  34535. 

  34536.  -- Macro: PUSH_ROUNDING (NPUSHED)

  34537.      A C expression that is the number of bytes actually pushed onto the

  34538.      stack when an instruction attempts to push NPUSHED bytes.

  34539. 

  34540.      On some machines, the definition

  34541. 

  34542.           #define PUSH_ROUNDING(BYTES) (BYTES)

  34543. 

  34544.      will suffice.  But on other machines, instructions that appear to

  34545.      push one byte actually push two bytes in an attempt to maintain

  34546.      alignment.  Then the definition should be

  34547. 

  34548.           #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)

  34549. 

  34550.      If the value of this macro has a type, it should be an unsigned

  34551.      type.

  34552. 

  34553.  -- Macro: ACCUMULATE_OUTGOING_ARGS

  34554.      A C expression.  If nonzero, the maximum amount of space required

  34555.      for outgoing arguments will be computed and placed into

  34556.      'crtl->outgoing_args_size'.  No space will be pushed onto the stack

  34557.      for each call; instead, the function prologue should increase the

  34558.      stack frame size by this amount.

  34559. 

  34560.      Setting both 'PUSH_ARGS' and 'ACCUMULATE_OUTGOING_ARGS' is not

  34561.      proper.

  34562. 

  34563.  -- Macro: REG_PARM_STACK_SPACE (FNDECL)

  34564.      Define this macro if functions should assume that stack space has

  34565.      been allocated for arguments even when their values are passed in

  34566.      registers.

  34567. 

  34568.      The value of this macro is the size, in bytes, of the area reserved

  34569.      for arguments passed in registers for the function represented by

  34570.      FNDECL, which can be zero if GCC is calling a library function.

  34571.      The argument FNDECL can be the FUNCTION_DECL, or the type itself of

  34572.      the function.

  34573. 

  34574.      This space can be allocated by the caller, or be a part of the

  34575.      machine-dependent stack frame: 'OUTGOING_REG_PARM_STACK_SPACE' says

  34576.      which.

  34577. 

  34578.  -- Macro: INCOMING_REG_PARM_STACK_SPACE (FNDECL)

  34579.      Like 'REG_PARM_STACK_SPACE', but for incoming register arguments.

  34580.      Define this macro if space guaranteed when compiling a function

  34581.      body is different to space required when making a call, a situation

  34582.      that can arise with K&R style function definitions.

  34583. 

  34584.  -- Macro: OUTGOING_REG_PARM_STACK_SPACE (FNTYPE)

  34585.      Define this to a nonzero value if it is the responsibility of the

  34586.      caller to allocate the area reserved for arguments passed in

  34587.      registers when calling a function of FNTYPE.  FNTYPE may be NULL if

  34588.      the function called is a library function.

  34589. 

  34590.      If 'ACCUMULATE_OUTGOING_ARGS' is defined, this macro controls

  34591.      whether the space for these arguments counts in the value of

  34592.      'crtl->outgoing_args_size'.

  34593. 

  34594.  -- Macro: STACK_PARMS_IN_REG_PARM_AREA

  34595.      Define this macro if 'REG_PARM_STACK_SPACE' is defined, but the

  34596.      stack parameters don't skip the area specified by it.

  34597. 

  34598.      Normally, when a parameter is not passed in registers, it is placed

  34599.      on the stack beyond the 'REG_PARM_STACK_SPACE' area.  Defining this

  34600.      macro suppresses this behavior and causes the parameter to be

  34601.      passed on the stack in its natural location.

  34602. 

  34603.  -- Target Hook: poly_int64 TARGET_RETURN_POPS_ARGS (tree FUNDECL, tree

  34604.           FUNTYPE, poly_int64 SIZE)

  34605.      This target hook returns the number of bytes of its own arguments

  34606.      that a function pops on returning, or 0 if the function pops no

  34607.      arguments and the caller must therefore pop them all after the

  34608.      function returns.

  34609. 

  34610.      FUNDECL is a C variable whose value is a tree node that describes

  34611.      the function in question.  Normally it is a node of type

  34612.      'FUNCTION_DECL' that describes the declaration of the function.

  34613.      From this you can obtain the 'DECL_ATTRIBUTES' of the function.

  34614. 

  34615.      FUNTYPE is a C variable whose value is a tree node that describes

  34616.      the function in question.  Normally it is a node of type

  34617.      'FUNCTION_TYPE' that describes the data type of the function.  From

  34618.      this it is possible to obtain the data types of the value and

  34619.      arguments (if known).

  34620. 

  34621.      When a call to a library function is being considered, FUNDECL will

  34622.      contain an identifier node for the library function.  Thus, if you

  34623.      need to distinguish among various library functions, you can do so

  34624.      by their names.  Note that "library function" in this context means

  34625.      a function used to perform arithmetic, whose name is known

  34626.      specially in the compiler and was not mentioned in the C code being

  34627.      compiled.

  34628. 

  34629.      SIZE is the number of bytes of arguments passed on the stack.  If a

  34630.      variable number of bytes is passed, it is zero, and argument

  34631.      popping will always be the responsibility of the calling function.

  34632. 

  34633.      On the VAX, all functions always pop their arguments, so the

  34634.      definition of this macro is SIZE.  On the 68000, using the standard

  34635.      calling convention, no functions pop their arguments, so the value

  34636.      of the macro is always 0 in this case.  But an alternative calling

  34637.      convention is available in which functions that take a fixed number

  34638.      of arguments pop them but other functions (such as 'printf') pop

  34639.      nothing (the caller pops all).  When this convention is in use,

  34640.      FUNTYPE is examined to determine whether a function takes a fixed

  34641.      number of arguments.

  34642. 

  34643.  -- Macro: CALL_POPS_ARGS (CUM)

  34644.      A C expression that should indicate the number of bytes a call

  34645.      sequence pops off the stack.  It is added to the value of

  34646.      'RETURN_POPS_ARGS' when compiling a function call.

  34647. 

  34648.      CUM is the variable in which all arguments to the called function

  34649.      have been accumulated.

  34650. 

  34651.      On certain architectures, such as the SH5, a call trampoline is

  34652.      used that pops certain registers off the stack, depending on the

  34653.      arguments that have been passed to the function.  Since this is a

  34654.      property of the call site, not of the called function,

  34655.      'RETURN_POPS_ARGS' is not appropriate.

  34656. 

  34657. 

  34658. File: gccint.info,  Node: Register Arguments,  Next: Scalar Return,  Prev: Stack Arguments,  Up: Stack and Calling

  34659. 

  34660. 18.9.7 Passing Arguments in Registers

  34661. -------------------------------------

  34662. 

  34663. This section describes the macros which let you control how various

  34664. types of arguments are passed in registers or how they are arranged in

  34665. the stack.

  34666. 

  34667.  -- Target Hook: rtx TARGET_FUNCTION_ARG (cumulative_args_t CA, const

  34668.           function_arg_info &ARG)

  34669.      Return an RTX indicating whether function argument ARG is passed in

  34670.      a register and if so, which register.  Argument CA summarizes all

  34671.      the previous arguments.

  34672. 

  34673.      The return value is usually either a 'reg' RTX for the hard

  34674.      register in which to pass the argument, or zero to pass the

  34675.      argument on the stack.

  34676. 

  34677.      The return value can be a 'const_int' which means argument is

  34678.      passed in a target specific slot with specified number.  Target

  34679.      hooks should be used to store or load argument in such case.  See

  34680.      'TARGET_STORE_BOUNDS_FOR_ARG' and 'TARGET_LOAD_BOUNDS_FOR_ARG' for

  34681.      more information.

  34682. 

  34683.      The value of the expression can also be a 'parallel' RTX.  This is

  34684.      used when an argument is passed in multiple locations.  The mode of

  34685.      the 'parallel' should be the mode of the entire argument.  The

  34686.      'parallel' holds any number of 'expr_list' pairs; each one

  34687.      describes where part of the argument is passed.  In each

  34688.      'expr_list' the first operand must be a 'reg' RTX for the hard

  34689.      register in which to pass this part of the argument, and the mode

  34690.      of the register RTX indicates how large this part of the argument

  34691.      is.  The second operand of the 'expr_list' is a 'const_int' which

  34692.      gives the offset in bytes into the entire argument of where this

  34693.      part starts.  As a special exception the first 'expr_list' in the

  34694.      'parallel' RTX may have a first operand of zero.  This indicates

  34695.      that the entire argument is also stored on the stack.

  34696. 

  34697.      The last time this hook is called, it is called with 'MODE ==

  34698.      VOIDmode', and its result is passed to the 'call' or 'call_value'

  34699.      pattern as operands 2 and 3 respectively.

  34700. 

  34701.      The usual way to make the ISO library 'stdarg.h' work on a machine

  34702.      where some arguments are usually passed in registers, is to cause

  34703.      nameless arguments to be passed on the stack instead.  This is done

  34704.      by making 'TARGET_FUNCTION_ARG' return 0 whenever NAMED is 'false'.

  34705. 

  34706.      You may use the hook 'targetm.calls.must_pass_in_stack' in the

  34707.      definition of this macro to determine if this argument is of a type

  34708.      that must be passed in the stack.  If 'REG_PARM_STACK_SPACE' is not

  34709.      defined and 'TARGET_FUNCTION_ARG' returns nonzero for such an

  34710.      argument, the compiler will abort.  If 'REG_PARM_STACK_SPACE' is

  34711.      defined, the argument will be computed in the stack and then loaded

  34712.      into a register.

  34713. 

  34714.  -- Target Hook: bool TARGET_MUST_PASS_IN_STACK (const function_arg_info

  34715.           &ARG)

  34716.      This target hook should return 'true' if we should not pass ARG

  34717.      solely in registers.  The file 'expr.h' defines a definition that

  34718.      is usually appropriate, refer to 'expr.h' for additional

  34719.      documentation.

  34720. 

  34721.  -- Target Hook: rtx TARGET_FUNCTION_INCOMING_ARG (cumulative_args_t CA,

  34722.           const function_arg_info &ARG)

  34723.      Define this hook if the caller and callee on the target have

  34724.      different views of where arguments are passed.  Also define this

  34725.      hook if there are functions that are never directly called, but are

  34726.      invoked by the hardware and which have nonstandard calling

  34727.      conventions.

  34728. 

  34729.      In this case 'TARGET_FUNCTION_ARG' computes the register in which

  34730.      the caller passes the value, and 'TARGET_FUNCTION_INCOMING_ARG'

  34731.      should be defined in a similar fashion to tell the function being

  34732.      called where the arguments will arrive.

  34733. 

  34734.      'TARGET_FUNCTION_INCOMING_ARG' can also return arbitrary address

  34735.      computation using hard register, which can be forced into a

  34736.      register, so that it can be used to pass special arguments.

  34737. 

  34738.      If 'TARGET_FUNCTION_INCOMING_ARG' is not defined,

  34739.      'TARGET_FUNCTION_ARG' serves both purposes.

  34740. 

  34741.  -- Target Hook: bool TARGET_USE_PSEUDO_PIC_REG (void)

  34742.      This hook should return 1 in case pseudo register should be created

  34743.      for pic_offset_table_rtx during function expand.

  34744. 

  34745.  -- Target Hook: void TARGET_INIT_PIC_REG (void)

  34746.      Perform a target dependent initialization of pic_offset_table_rtx.

  34747.      This hook is called at the start of register allocation.

  34748. 

  34749.  -- Target Hook: int TARGET_ARG_PARTIAL_BYTES (cumulative_args_t CUM,

  34750.           const function_arg_info &ARG)

  34751.      This target hook returns the number of bytes at the beginning of an

  34752.      argument that must be put in registers.  The value must be zero for

  34753.      arguments that are passed entirely in registers or that are

  34754.      entirely pushed on the stack.

  34755. 

  34756.      On some machines, certain arguments must be passed partially in

  34757.      registers and partially in memory.  On these machines, typically

  34758.      the first few words of arguments are passed in registers, and the

  34759.      rest on the stack.  If a multi-word argument (a 'double' or a

  34760.      structure) crosses that boundary, its first few words must be

  34761.      passed in registers and the rest must be pushed.  This macro tells

  34762.      the compiler when this occurs, and how many bytes should go in

  34763.      registers.

  34764. 

  34765.      'TARGET_FUNCTION_ARG' for these arguments should return the first

  34766.      register to be used by the caller for this argument; likewise

  34767.      'TARGET_FUNCTION_INCOMING_ARG', for the called function.

  34768. 

  34769.  -- Target Hook: bool TARGET_PASS_BY_REFERENCE (cumulative_args_t CUM,

  34770.           const function_arg_info &ARG)

  34771.      This target hook should return 'true' if argument ARG at the

  34772.      position indicated by CUM should be passed by reference.  This

  34773.      predicate is queried after target independent reasons for being

  34774.      passed by reference, such as 'TREE_ADDRESSABLE (ARG.type)'.

  34775. 

  34776.      If the hook returns true, a copy of that argument is made in memory

  34777.      and a pointer to the argument is passed instead of the argument

  34778.      itself.  The pointer is passed in whatever way is appropriate for

  34779.      passing a pointer to that type.

  34780. 

  34781.  -- Target Hook: bool TARGET_CALLEE_COPIES (cumulative_args_t CUM, const

  34782.           function_arg_info &ARG)

  34783.      The function argument described by the parameters to this hook is

  34784.      known to be passed by reference.  The hook should return true if

  34785.      the function argument should be copied by the callee instead of

  34786.      copied by the caller.

  34787. 

  34788.      For any argument for which the hook returns true, if it can be

  34789.      determined that the argument is not modified, then a copy need not

  34790.      be generated.

  34791. 

  34792.      The default version of this hook always returns false.

  34793. 

  34794.  -- Macro: CUMULATIVE_ARGS

  34795.      A C type for declaring a variable that is used as the first

  34796.      argument of 'TARGET_FUNCTION_ARG' and other related values.  For

  34797.      some target machines, the type 'int' suffices and can hold the

  34798.      number of bytes of argument so far.

  34799. 

  34800.      There is no need to record in 'CUMULATIVE_ARGS' anything about the

  34801.      arguments that have been passed on the stack.  The compiler has

  34802.      other variables to keep track of that.  For target machines on

  34803.      which all arguments are passed on the stack, there is no need to

  34804.      store anything in 'CUMULATIVE_ARGS'; however, the data structure

  34805.      must exist and should not be empty, so use 'int'.

  34806. 

  34807.  -- Macro: OVERRIDE_ABI_FORMAT (FNDECL)

  34808.      If defined, this macro is called before generating any code for a

  34809.      function, but after the CFUN descriptor for the function has been

  34810.      created.  The back end may use this macro to update CFUN to reflect

  34811.      an ABI other than that which would normally be used by default.  If

  34812.      the compiler is generating code for a compiler-generated function,

  34813.      FNDECL may be 'NULL'.

  34814. 

  34815.  -- Macro: INIT_CUMULATIVE_ARGS (CUM, FNTYPE, LIBNAME, FNDECL,

  34816.           N_NAMED_ARGS)

  34817.      A C statement (sans semicolon) for initializing the variable CUM

  34818.      for the state at the beginning of the argument list.  The variable

  34819.      has type 'CUMULATIVE_ARGS'.  The value of FNTYPE is the tree node

  34820.      for the data type of the function which will receive the args, or 0

  34821.      if the args are to a compiler support library function.  For direct

  34822.      calls that are not libcalls, FNDECL contain the declaration node of

  34823.      the function.  FNDECL is also set when 'INIT_CUMULATIVE_ARGS' is

  34824.      used to find arguments for the function being compiled.

  34825.      N_NAMED_ARGS is set to the number of named arguments, including a

  34826.      structure return address if it is passed as a parameter, when

  34827.      making a call.  When processing incoming arguments, N_NAMED_ARGS is

  34828.      set to -1.

  34829. 

  34830.      When processing a call to a compiler support library function,

  34831.      LIBNAME identifies which one.  It is a 'symbol_ref' rtx which

  34832.      contains the name of the function, as a string.  LIBNAME is 0 when

  34833.      an ordinary C function call is being processed.  Thus, each time

  34834.      this macro is called, either LIBNAME or FNTYPE is nonzero, but

  34835.      never both of them at once.

  34836. 

  34837.  -- Macro: INIT_CUMULATIVE_LIBCALL_ARGS (CUM, MODE, LIBNAME)

  34838.      Like 'INIT_CUMULATIVE_ARGS' but only used for outgoing libcalls, it

  34839.      gets a 'MODE' argument instead of FNTYPE, that would be 'NULL'.

  34840.      INDIRECT would always be zero, too.  If this macro is not defined,

  34841.      'INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname, 0)' is used instead.

  34842. 

  34843.  -- Macro: INIT_CUMULATIVE_INCOMING_ARGS (CUM, FNTYPE, LIBNAME)

  34844.      Like 'INIT_CUMULATIVE_ARGS' but overrides it for the purposes of

  34845.      finding the arguments for the function being compiled.  If this

  34846.      macro is undefined, 'INIT_CUMULATIVE_ARGS' is used instead.

  34847. 

  34848.      The value passed for LIBNAME is always 0, since library routines

  34849.      with special calling conventions are never compiled with GCC.  The

  34850.      argument LIBNAME exists for symmetry with 'INIT_CUMULATIVE_ARGS'.

  34851. 

  34852.  -- Target Hook: void TARGET_FUNCTION_ARG_ADVANCE (cumulative_args_t CA,

  34853.           const function_arg_info &ARG)

  34854.      This hook updates the summarizer variable pointed to by CA to

  34855.      advance past argument ARG in the argument list.  Once this is done,

  34856.      the variable CUM is suitable for analyzing the _following_ argument

  34857.      with 'TARGET_FUNCTION_ARG', etc.

  34858. 

  34859.      This hook need not do anything if the argument in question was

  34860.      passed on the stack.  The compiler knows how to track the amount of

  34861.      stack space used for arguments without any special help.

  34862. 

  34863.  -- Target Hook: HOST_WIDE_INT TARGET_FUNCTION_ARG_OFFSET (machine_mode

  34864.           MODE, const_tree TYPE)

  34865.      This hook returns the number of bytes to add to the offset of an

  34866.      argument of type TYPE and mode MODE when passed in memory.  This is

  34867.      needed for the SPU, which passes 'char' and 'short' arguments in

  34868.      the preferred slot that is in the middle of the quad word instead

  34869.      of starting at the top.  The default implementation returns 0.

  34870. 

  34871.  -- Target Hook: pad_direction TARGET_FUNCTION_ARG_PADDING (machine_mode

  34872.           MODE, const_tree TYPE)

  34873.      This hook determines whether, and in which direction, to pad out an

  34874.      argument of mode MODE and type TYPE.  It returns 'PAD_UPWARD' to

  34875.      insert padding above the argument, 'PAD_DOWNWARD' to insert padding

  34876.      below the argument, or 'PAD_NONE' to inhibit padding.

  34877. 

  34878.      The _amount_ of padding is not controlled by this hook, but by

  34879.      'TARGET_FUNCTION_ARG_ROUND_BOUNDARY'.  It is always just enough to

  34880.      reach the next multiple of that boundary.

  34881. 

  34882.      This hook has a default definition that is right for most systems.

  34883.      For little-endian machines, the default is to pad upward.  For

  34884.      big-endian machines, the default is to pad downward for an argument

  34885.      of constant size shorter than an 'int', and upward otherwise.

  34886. 

  34887.  -- Macro: PAD_VARARGS_DOWN

  34888.      If defined, a C expression which determines whether the default

  34889.      implementation of va_arg will attempt to pad down before reading

  34890.      the next argument, if that argument is smaller than its aligned

  34891.      space as controlled by 'PARM_BOUNDARY'.  If this macro is not

  34892.      defined, all such arguments are padded down if 'BYTES_BIG_ENDIAN'

  34893.      is true.

  34894. 

  34895.  -- Macro: BLOCK_REG_PADDING (MODE, TYPE, FIRST)

  34896.      Specify padding for the last element of a block move between

  34897.      registers and memory.  FIRST is nonzero if this is the only

  34898.      element.  Defining this macro allows better control of register

  34899.      function parameters on big-endian machines, without using

  34900.      'PARALLEL' rtl.  In particular, 'MUST_PASS_IN_STACK' need not test

  34901.      padding and mode of types in registers, as there is no longer a

  34902.      "wrong" part of a register; For example, a three byte aggregate may

  34903.      be passed in the high part of a register if so required.

  34904. 

  34905.  -- Target Hook: unsigned int TARGET_FUNCTION_ARG_BOUNDARY (machine_mode

  34906.           MODE, const_tree TYPE)

  34907.      This hook returns the alignment boundary, in bits, of an argument

  34908.      with the specified mode and type.  The default hook returns

  34909.      'PARM_BOUNDARY' for all arguments.

  34910. 

  34911.  -- Target Hook: unsigned int TARGET_FUNCTION_ARG_ROUND_BOUNDARY

  34912.           (machine_mode MODE, const_tree TYPE)

  34913.      Normally, the size of an argument is rounded up to 'PARM_BOUNDARY',

  34914.      which is the default value for this hook.  You can define this hook

  34915.      to return a different value if an argument size must be rounded to

  34916.      a larger value.

  34917. 

  34918.  -- Macro: FUNCTION_ARG_REGNO_P (REGNO)

  34919.      A C expression that is nonzero if REGNO is the number of a hard

  34920.      register in which function arguments are sometimes passed.  This

  34921.      does _not_ include implicit arguments such as the static chain and

  34922.      the structure-value address.  On many machines, no registers can be

  34923.      used for this purpose since all function arguments are pushed on

  34924.      the stack.

  34925. 

  34926.  -- Target Hook: bool TARGET_SPLIT_COMPLEX_ARG (const_tree TYPE)

  34927.      This hook should return true if parameter of type TYPE are passed

  34928.      as two scalar parameters.  By default, GCC will attempt to pack

  34929.      complex arguments into the target's word size.  Some ABIs require

  34930.      complex arguments to be split and treated as their individual

  34931.      components.  For example, on AIX64, complex floats should be passed

  34932.      in a pair of floating point registers, even though a complex float

  34933.      would fit in one 64-bit floating point register.

  34934. 

  34935.      The default value of this hook is 'NULL', which is treated as

  34936.      always false.

  34937. 

  34938.  -- Target Hook: tree TARGET_BUILD_BUILTIN_VA_LIST (void)

  34939.      This hook returns a type node for 'va_list' for the target.  The

  34940.      default version of the hook returns 'void*'.

  34941. 

  34942.  -- Target Hook: int TARGET_ENUM_VA_LIST_P (int IDX, const char **PNAME,

  34943.           tree *PTREE)

  34944.      This target hook is used in function 'c_common_nodes_and_builtins'

  34945.      to iterate through the target specific builtin types for va_list.

  34946.      The variable IDX is used as iterator.  PNAME has to be a pointer to

  34947.      a 'const char *' and PTREE a pointer to a 'tree' typed variable.

  34948.      The arguments PNAME and PTREE are used to store the result of this

  34949.      macro and are set to the name of the va_list builtin type and its

  34950.      internal type.  If the return value of this macro is zero, then

  34951.      there is no more element.  Otherwise the IDX should be increased

  34952.      for the next call of this macro to iterate through all types.

  34953. 

  34954.  -- Target Hook: tree TARGET_FN_ABI_VA_LIST (tree FNDECL)

  34955.      This hook returns the va_list type of the calling convention

  34956.      specified by FNDECL.  The default version of this hook returns

  34957.      'va_list_type_node'.

  34958. 

  34959.  -- Target Hook: tree TARGET_CANONICAL_VA_LIST_TYPE (tree TYPE)

  34960.      This hook returns the va_list type of the calling convention

  34961.      specified by the type of TYPE.  If TYPE is not a valid va_list

  34962.      type, it returns 'NULL_TREE'.

  34963. 

  34964.  -- Target Hook: tree TARGET_GIMPLIFY_VA_ARG_EXPR (tree VALIST, tree

  34965.           TYPE, gimple_seq *PRE_P, gimple_seq *POST_P)

  34966.      This hook performs target-specific gimplification of 'VA_ARG_EXPR'.

  34967.      The first two parameters correspond to the arguments to 'va_arg';

  34968.      the latter two are as in 'gimplify.c:gimplify_expr'.

  34969. 

  34970.  -- Target Hook: bool TARGET_VALID_POINTER_MODE (scalar_int_mode MODE)

  34971.      Define this to return nonzero if the port can handle pointers with

  34972.      machine mode MODE.  The default version of this hook returns true

  34973.      for both 'ptr_mode' and 'Pmode'.

  34974. 

  34975.  -- Target Hook: bool TARGET_REF_MAY_ALIAS_ERRNO (ao_ref *REF)

  34976.      Define this to return nonzero if the memory reference REF may alias

  34977.      with the system C library errno location.  The default version of

  34978.      this hook assumes the system C library errno location is either a

  34979.      declaration of type int or accessed by dereferencing a pointer to

  34980.      int.

  34981. 

  34982.  -- Target Hook: machine_mode TARGET_TRANSLATE_MODE_ATTRIBUTE

  34983.           (machine_mode MODE)

  34984.      Define this hook if during mode attribute processing, the port

  34985.      should translate machine_mode MODE to another mode.  For example,

  34986.      rs6000's 'KFmode', when it is the same as 'TFmode'.

  34987. 

  34988.      The default version of the hook returns that mode that was passed

  34989.      in.

  34990. 

  34991.  -- Target Hook: bool TARGET_SCALAR_MODE_SUPPORTED_P (scalar_mode MODE)

  34992.      Define this to return nonzero if the port is prepared to handle

  34993.      insns involving scalar mode MODE.  For a scalar mode to be

  34994.      considered supported, all the basic arithmetic and comparisons must

  34995.      work.

  34996. 

  34997.      The default version of this hook returns true for any mode required

  34998.      to handle the basic C types (as defined by the port).  Included

  34999.      here are the double-word arithmetic supported by the code in

  35000.      'optabs.c'.

  35001. 

  35002.  -- Target Hook: bool TARGET_VECTOR_MODE_SUPPORTED_P (machine_mode MODE)

  35003.      Define this to return nonzero if the port is prepared to handle

  35004.      insns involving vector mode MODE.  At the very least, it must have

  35005.      move patterns for this mode.

  35006. 

  35007.  -- Target Hook: bool TARGET_COMPATIBLE_VECTOR_TYPES_P (const_tree

  35008.           TYPE1, const_tree TYPE2)

  35009.      Return true if there is no target-specific reason for treating

  35010.      vector types TYPE1 and TYPE2 as distinct types.  The caller has

  35011.      already checked for target-independent reasons, meaning that the

  35012.      types are known to have the same mode, to have the same number of

  35013.      elements, and to have what the caller considers to be compatible

  35014.      element types.

  35015. 

  35016.      The main reason for defining this hook is to reject pairs of types

  35017.      that are handled differently by the target's calling convention.

  35018.      For example, when a new N-bit vector architecture is added to a

  35019.      target, the target may want to handle normal N-bit 'VECTOR_TYPE'

  35020.      arguments and return values in the same way as before, to maintain

  35021.      backwards compatibility.  However, it may also provide new,

  35022.      architecture-specific 'VECTOR_TYPE's that are passed and returned

  35023.      in a more efficient way.  It is then important to maintain a

  35024.      distinction between the "normal" 'VECTOR_TYPE's and the new

  35025.      architecture-specific ones.

  35026. 

  35027.      The default implementation returns true, which is correct for most

  35028.      targets.

  35029. 

  35030.  -- Target Hook: opt_machine_mode TARGET_ARRAY_MODE (machine_mode MODE,

  35031.           unsigned HOST_WIDE_INT NELEMS)

  35032.      Return the mode that GCC should use for an array that has NELEMS

  35033.      elements, with each element having mode MODE.  Return no mode if

  35034.      the target has no special requirements.  In the latter case, GCC

  35035.      looks for an integer mode of the appropriate size if available and

  35036.      uses BLKmode otherwise.  Usually the search for the integer mode is

  35037.      limited to 'MAX_FIXED_MODE_SIZE', but the

  35038.      'TARGET_ARRAY_MODE_SUPPORTED_P' hook allows a larger mode to be

  35039.      used in specific cases.

  35040. 

  35041.      The main use of this hook is to specify that an array of vectors

  35042.      should also have a vector mode.  The default implementation returns

  35043.      no mode.

  35044. 

  35045.  -- Target Hook: bool TARGET_ARRAY_MODE_SUPPORTED_P (machine_mode MODE,

  35046.           unsigned HOST_WIDE_INT NELEMS)

  35047.      Return true if GCC should try to use a scalar mode to store an

  35048.      array of NELEMS elements, given that each element has mode MODE.

  35049.      Returning true here overrides the usual 'MAX_FIXED_MODE' limit and

  35050.      allows GCC to use any defined integer mode.

  35051. 

  35052.      One use of this hook is to support vector load and store operations

  35053.      that operate on several homogeneous vectors.  For example, ARM NEON

  35054.      has operations like:

  35055. 

  35056.           int8x8x3_t vld3_s8 (const int8_t *)

  35057. 

  35058.      where the return type is defined as:

  35059. 

  35060.           typedef struct int8x8x3_t

  35061.           {

  35062.             int8x8_t val[3];

  35063.           } int8x8x3_t;

  35064. 

  35065.      If this hook allows 'val' to have a scalar mode, then 'int8x8x3_t'

  35066.      can have the same mode.  GCC can then store 'int8x8x3_t's in

  35067.      registers rather than forcing them onto the stack.

  35068. 

  35069.  -- Target Hook: bool TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P

  35070.           (scalar_float_mode MODE)

  35071.      Define this to return nonzero if libgcc provides support for the

  35072.      floating-point mode MODE, which is known to pass

  35073.      'TARGET_SCALAR_MODE_SUPPORTED_P'.  The default version of this hook

  35074.      returns true for all of 'SFmode', 'DFmode', 'XFmode' and 'TFmode',

  35075.      if such modes exist.

  35076. 

  35077.  -- Target Hook: opt_scalar_float_mode TARGET_FLOATN_MODE (int N, bool

  35078.           EXTENDED)

  35079.      Define this to return the machine mode to use for the type

  35080.      '_FloatN', if EXTENDED is false, or the type '_FloatNx', if

  35081.      EXTENDED is true.  If such a type is not supported, return

  35082.      'opt_scalar_float_mode ()'.  The default version of this hook

  35083.      returns 'SFmode' for '_Float32', 'DFmode' for '_Float64' and

  35084.      '_Float32x' and 'TFmode' for '_Float128', if those modes exist and

  35085.      satisfy the requirements for those types and pass

  35086.      'TARGET_SCALAR_MODE_SUPPORTED_P' and

  35087.      'TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P'; for '_Float64x', it

  35088.      returns the first of 'XFmode' and 'TFmode' that exists and

  35089.      satisfies the same requirements; for other types, it returns

  35090.      'opt_scalar_float_mode ()'.  The hook is only called for values of

  35091.      N and EXTENDED that are valid according to ISO/IEC TS 18661-3:2015;

  35092.      that is, N is one of 32, 64, 128, or, if EXTENDED is false, 16 or

  35093.      greater than 128 and a multiple of 32.

  35094. 

  35095.  -- Target Hook: bool TARGET_FLOATN_BUILTIN_P (int FUNC)

  35096.      Define this to return true if the '_FloatN' and '_FloatNx' built-in

  35097.      functions should implicitly enable the built-in function without

  35098.      the '__builtin_' prefix in addition to the normal built-in function

  35099.      with the '__builtin_' prefix.  The default is to only enable

  35100.      built-in functions without the '__builtin_' prefix for the GNU C

  35101.      langauge.  In strict ANSI/ISO mode, the built-in function without

  35102.      the '__builtin_' prefix is not enabled.  The argument 'FUNC' is the

  35103.      'enum built_in_function' id of the function to be enabled.

  35104. 

  35105.  -- Target Hook: bool TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P

  35106.           (machine_mode MODE)

  35107.      Define this to return nonzero for machine modes for which the port

  35108.      has small register classes.  If this target hook returns nonzero

  35109.      for a given MODE, the compiler will try to minimize the lifetime of

  35110.      registers in MODE.  The hook may be called with 'VOIDmode' as

  35111.      argument.  In this case, the hook is expected to return nonzero if

  35112.      it returns nonzero for any mode.

  35113. 

  35114.      On some machines, it is risky to let hard registers live across

  35115.      arbitrary insns.  Typically, these machines have instructions that

  35116.      require values to be in specific registers (like an accumulator),

  35117.      and reload will fail if the required hard register is used for

  35118.      another purpose across such an insn.

  35119. 

  35120.      Passes before reload do not know which hard registers will be used

  35121.      in an instruction, but the machine modes of the registers set or

  35122.      used in the instruction are already known.  And for some machines,

  35123.      register classes are small for, say, integer registers but not for

  35124.      floating point registers.  For example, the AMD x86-64 architecture

  35125.      requires specific registers for the legacy x86 integer

  35126.      instructions, but there are many SSE registers for floating point

  35127.      operations.  On such targets, a good strategy may be to return

  35128.      nonzero from this hook for 'INTEGRAL_MODE_P' machine modes but zero

  35129.      for the SSE register classes.

  35130. 

  35131.      The default version of this hook returns false for any mode.  It is

  35132.      always safe to redefine this hook to return with a nonzero value.

  35133.      But if you unnecessarily define it, you will reduce the amount of

  35134.      optimizations that can be performed in some cases.  If you do not

  35135.      define this hook to return a nonzero value when it is required, the

  35136.      compiler will run out of spill registers and print a fatal error

  35137.      message.

  35138. 

  35139. 

  35140. File: gccint.info,  Node: Scalar Return,  Next: Aggregate Return,  Prev: Register Arguments,  Up: Stack and Calling

  35141. 

  35142. 18.9.8 How Scalar Function Values Are Returned

  35143. ----------------------------------------------

  35144. 

  35145. This section discusses the macros that control returning scalars as

  35146. values--values that can fit in registers.

  35147. 

  35148.  -- Target Hook: rtx TARGET_FUNCTION_VALUE (const_tree RET_TYPE,

  35149.           const_tree FN_DECL_OR_TYPE, bool OUTGOING)

  35150. 

  35151.      Define this to return an RTX representing the place where a

  35152.      function returns or receives a value of data type RET_TYPE, a tree

  35153.      node representing a data type.  FN_DECL_OR_TYPE is a tree node

  35154.      representing 'FUNCTION_DECL' or 'FUNCTION_TYPE' of a function being

  35155.      called.  If OUTGOING is false, the hook should compute the register

  35156.      in which the caller will see the return value.  Otherwise, the hook

  35157.      should return an RTX representing the place where a function

  35158.      returns a value.

  35159. 

  35160.      On many machines, only 'TYPE_MODE (RET_TYPE)' is relevant.

  35161.      (Actually, on most machines, scalar values are returned in the same

  35162.      place regardless of mode.)  The value of the expression is usually

  35163.      a 'reg' RTX for the hard register where the return value is stored.

  35164.      The value can also be a 'parallel' RTX, if the return value is in

  35165.      multiple places.  See 'TARGET_FUNCTION_ARG' for an explanation of

  35166.      the 'parallel' form.  Note that the callee will populate every

  35167.      location specified in the 'parallel', but if the first element of

  35168.      the 'parallel' contains the whole return value, callers will use

  35169.      that element as the canonical location and ignore the others.  The

  35170.      m68k port uses this type of 'parallel' to return pointers in both

  35171.      '%a0' (the canonical location) and '%d0'.

  35172. 

  35173.      If 'TARGET_PROMOTE_FUNCTION_RETURN' returns true, you must apply

  35174.      the same promotion rules specified in 'PROMOTE_MODE' if VALTYPE is

  35175.      a scalar type.

  35176. 

  35177.      If the precise function being called is known, FUNC is a tree node

  35178.      ('FUNCTION_DECL') for it; otherwise, FUNC is a null pointer.  This

  35179.      makes it possible to use a different value-returning convention for

  35180.      specific functions when all their calls are known.

  35181. 

  35182.      Some target machines have "register windows" so that the register

  35183.      in which a function returns its value is not the same as the one in

  35184.      which the caller sees the value.  For such machines, you should

  35185.      return different RTX depending on OUTGOING.

  35186. 

  35187.      'TARGET_FUNCTION_VALUE' is not used for return values with

  35188.      aggregate data types, because these are returned in another way.

  35189.      See 'TARGET_STRUCT_VALUE_RTX' and related macros, below.

  35190. 

  35191.  -- Macro: FUNCTION_VALUE (VALTYPE, FUNC)

  35192.      This macro has been deprecated.  Use 'TARGET_FUNCTION_VALUE' for a

  35193.      new target instead.

  35194. 

  35195.  -- Macro: LIBCALL_VALUE (MODE)

  35196.      A C expression to create an RTX representing the place where a

  35197.      library function returns a value of mode MODE.

  35198. 

  35199.      Note that "library function" in this context means a compiler

  35200.      support routine, used to perform arithmetic, whose name is known

  35201.      specially by the compiler and was not mentioned in the C code being

  35202.      compiled.

  35203. 

  35204.  -- Target Hook: rtx TARGET_LIBCALL_VALUE (machine_mode MODE, const_rtx

  35205.           FUN)

  35206.      Define this hook if the back-end needs to know the name of the

  35207.      libcall function in order to determine where the result should be

  35208.      returned.

  35209. 

  35210.      The mode of the result is given by MODE and the name of the called

  35211.      library function is given by FUN.  The hook should return an RTX

  35212.      representing the place where the library function result will be

  35213.      returned.

  35214. 

  35215.      If this hook is not defined, then LIBCALL_VALUE will be used.

  35216. 

  35217.  -- Macro: FUNCTION_VALUE_REGNO_P (REGNO)

  35218.      A C expression that is nonzero if REGNO is the number of a hard

  35219.      register in which the values of called function may come back.

  35220. 

  35221.      A register whose use for returning values is limited to serving as

  35222.      the second of a pair (for a value of type 'double', say) need not

  35223.      be recognized by this macro.  So for most machines, this definition

  35224.      suffices:

  35225. 

  35226.           #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)

  35227. 

  35228.      If the machine has register windows, so that the caller and the

  35229.      called function use different registers for the return value, this

  35230.      macro should recognize only the caller's register numbers.

  35231. 

  35232.      This macro has been deprecated.  Use

  35233.      'TARGET_FUNCTION_VALUE_REGNO_P' for a new target instead.

  35234. 

  35235.  -- Target Hook: bool TARGET_FUNCTION_VALUE_REGNO_P (const unsigned int

  35236.           REGNO)

  35237.      A target hook that return 'true' if REGNO is the number of a hard

  35238.      register in which the values of called function may come back.

  35239. 

  35240.      A register whose use for returning values is limited to serving as

  35241.      the second of a pair (for a value of type 'double', say) need not

  35242.      be recognized by this target hook.

  35243. 

  35244.      If the machine has register windows, so that the caller and the

  35245.      called function use different registers for the return value, this

  35246.      target hook should recognize only the caller's register numbers.

  35247. 

  35248.      If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be

  35249.      used.

  35250. 

  35251.  -- Macro: APPLY_RESULT_SIZE

  35252.      Define this macro if 'untyped_call' and 'untyped_return' need more

  35253.      space than is implied by 'FUNCTION_VALUE_REGNO_P' for saving and

  35254.      restoring an arbitrary return value.

  35255. 

  35256.  -- Target Hook: bool TARGET_OMIT_STRUCT_RETURN_REG

  35257.      Normally, when a function returns a structure by memory, the

  35258.      address is passed as an invisible pointer argument, but the

  35259.      compiler also arranges to return the address from the function like

  35260.      it would a normal pointer return value.  Define this to true if

  35261.      that behavior is undesirable on your target.

  35262. 

  35263.  -- Target Hook: bool TARGET_RETURN_IN_MSB (const_tree TYPE)

  35264.      This hook should return true if values of type TYPE are returned at

  35265.      the most significant end of a register (in other words, if they are

  35266.      padded at the least significant end).  You can assume that TYPE is

  35267.      returned in a register; the caller is required to check this.

  35268. 

  35269.      Note that the register provided by 'TARGET_FUNCTION_VALUE' must be

  35270.      able to hold the complete return value.  For example, if a 1-, 2-

  35271.      or 3-byte structure is returned at the most significant end of a

  35272.      4-byte register, 'TARGET_FUNCTION_VALUE' should provide an 'SImode'

  35273.      rtx.

  35274. 

  35275. 

  35276. File: gccint.info,  Node: Aggregate Return,  Next: Caller Saves,  Prev: Scalar Return,  Up: Stack and Calling

  35277. 

  35278. 18.9.9 How Large Values Are Returned

  35279. ------------------------------------

  35280. 

  35281. When a function value's mode is 'BLKmode' (and in some other cases), the

  35282. value is not returned according to 'TARGET_FUNCTION_VALUE' (*note Scalar

  35283. Return::).  Instead, the caller passes the address of a block of memory

  35284. in which the value should be stored.  This address is called the

  35285. "structure value address".

  35286. 

  35287.  This section describes how to control returning structure values in

  35288. memory.

  35289. 

  35290.  -- Target Hook: bool TARGET_RETURN_IN_MEMORY (const_tree TYPE,

  35291.           const_tree FNTYPE)

  35292.      This target hook should return a nonzero value to say to return the

  35293.      function value in memory, just as large structures are always

  35294.      returned.  Here TYPE will be the data type of the value, and FNTYPE

  35295.      will be the type of the function doing the returning, or 'NULL' for

  35296.      libcalls.

  35297. 

  35298.      Note that values of mode 'BLKmode' must be explicitly handled by

  35299.      this function.  Also, the option '-fpcc-struct-return' takes effect

  35300.      regardless of this macro.  On most systems, it is possible to leave

  35301.      the hook undefined; this causes a default definition to be used,

  35302.      whose value is the constant 1 for 'BLKmode' values, and 0

  35303.      otherwise.

  35304. 

  35305.      Do not use this hook to indicate that structures and unions should

  35306.      always be returned in memory.  You should instead use

  35307.      'DEFAULT_PCC_STRUCT_RETURN' to indicate this.

  35308. 

  35309.  -- Macro: DEFAULT_PCC_STRUCT_RETURN

  35310.      Define this macro to be 1 if all structure and union return values

  35311.      must be in memory.  Since this results in slower code, this should

  35312.      be defined only if needed for compatibility with other compilers or

  35313.      with an ABI.  If you define this macro to be 0, then the

  35314.      conventions used for structure and union return values are decided

  35315.      by the 'TARGET_RETURN_IN_MEMORY' target hook.

  35316. 

  35317.      If not defined, this defaults to the value 1.

  35318. 

  35319.  -- Target Hook: rtx TARGET_STRUCT_VALUE_RTX (tree FNDECL, int INCOMING)

  35320.      This target hook should return the location of the structure value

  35321.      address (normally a 'mem' or 'reg'), or 0 if the address is passed

  35322.      as an "invisible" first argument.  Note that FNDECL may be 'NULL',

  35323.      for libcalls.  You do not need to define this target hook if the

  35324.      address is always passed as an "invisible" first argument.

  35325. 

  35326.      On some architectures the place where the structure value address

  35327.      is found by the called function is not the same place that the

  35328.      caller put it.  This can be due to register windows, or it could be

  35329.      because the function prologue moves it to a different place.

  35330.      INCOMING is '1' or '2' when the location is needed in the context

  35331.      of the called function, and '0' in the context of the caller.

  35332. 

  35333.      If INCOMING is nonzero and the address is to be found on the stack,

  35334.      return a 'mem' which refers to the frame pointer.  If INCOMING is

  35335.      '2', the result is being used to fetch the structure value address

  35336.      at the beginning of a function.  If you need to emit adjusting

  35337.      code, you should do it at this point.

  35338. 

  35339.  -- Macro: PCC_STATIC_STRUCT_RETURN

  35340.      Define this macro if the usual system convention on the target

  35341.      machine for returning structures and unions is for the called

  35342.      function to return the address of a static variable containing the

  35343.      value.

  35344. 

  35345.      Do not define this if the usual system convention is for the caller

  35346.      to pass an address to the subroutine.

  35347. 

  35348.      This macro has effect in '-fpcc-struct-return' mode, but it does

  35349.      nothing when you use '-freg-struct-return' mode.

  35350. 

  35351.  -- Target Hook: fixed_size_mode TARGET_GET_RAW_RESULT_MODE (int REGNO)

  35352.      This target hook returns the mode to be used when accessing raw

  35353.      return registers in '__builtin_return'.  Define this macro if the

  35354.      value in REG_RAW_MODE is not correct.

  35355. 

  35356.  -- Target Hook: fixed_size_mode TARGET_GET_RAW_ARG_MODE (int REGNO)

  35357.      This target hook returns the mode to be used when accessing raw

  35358.      argument registers in '__builtin_apply_args'.  Define this macro if

  35359.      the value in REG_RAW_MODE is not correct.

  35360. 

  35361.  -- Target Hook: bool TARGET_EMPTY_RECORD_P (const_tree TYPE)

  35362.      This target hook returns true if the type is an empty record.  The

  35363.      default is to return 'false'.

  35364. 

  35365.  -- Target Hook: void TARGET_WARN_PARAMETER_PASSING_ABI

  35366.           (cumulative_args_t CA, tree TYPE)

  35367.      This target hook warns about the change in empty class parameter

  35368.      passing ABI.

  35369. 

  35370. 

  35371. File: gccint.info,  Node: Caller Saves,  Next: Function Entry,  Prev: Aggregate Return,  Up: Stack and Calling

  35372. 

  35373. 18.9.10 Caller-Saves Register Allocation

  35374. ----------------------------------------

  35375. 

  35376. If you enable it, GCC can save registers around function calls.  This

  35377. makes it possible to use call-clobbered registers to hold variables that

  35378. must live across calls.

  35379. 

  35380.  -- Macro: HARD_REGNO_CALLER_SAVE_MODE (REGNO, NREGS)

  35381.      A C expression specifying which mode is required for saving NREGS

  35382.      of a pseudo-register in call-clobbered hard register REGNO.  If

  35383.      REGNO is unsuitable for caller save, 'VOIDmode' should be returned.

  35384.      For most machines this macro need not be defined since GCC will

  35385.      select the smallest suitable mode.

  35386. 

  35387. 

  35388. File: gccint.info,  Node: Function Entry,  Next: Profiling,  Prev: Caller Saves,  Up: Stack and Calling

  35389. 

  35390. 18.9.11 Function Entry and Exit

  35391. -------------------------------

  35392. 

  35393. This section describes the macros that output function entry

  35394. ("prologue") and exit ("epilogue") code.

  35395. 

  35396.  -- Target Hook: void TARGET_ASM_PRINT_PATCHABLE_FUNCTION_ENTRY (FILE

  35397.           *FILE, unsigned HOST_WIDE_INT PATCH_AREA_SIZE, bool RECORD_P)

  35398.      Generate a patchable area at the function start, consisting of

  35399.      PATCH_AREA_SIZE NOP instructions.  If the target supports named

  35400.      sections and if RECORD_P is true, insert a pointer to the current

  35401.      location in the table of patchable functions.  The default

  35402.      implementation of the hook places the table of pointers in the

  35403.      special section named '__patchable_function_entries'.

  35404. 

  35405.  -- Target Hook: void TARGET_ASM_FUNCTION_PROLOGUE (FILE *FILE)

  35406.      If defined, a function that outputs the assembler code for entry to

  35407.      a function.  The prologue is responsible for setting up the stack

  35408.      frame, initializing the frame pointer register, saving registers

  35409.      that must be saved, and allocating SIZE additional bytes of storage

  35410.      for the local variables.  FILE is a stdio stream to which the

  35411.      assembler code should be output.

  35412. 

  35413.      The label for the beginning of the function need not be output by

  35414.      this macro.  That has already been done when the macro is run.

  35415. 

  35416.      To determine which registers to save, the macro can refer to the

  35417.      array 'regs_ever_live': element R is nonzero if hard register R is

  35418.      used anywhere within the function.  This implies the function

  35419.      prologue should save register R, provided it is not one of the

  35420.      call-used registers.  ('TARGET_ASM_FUNCTION_EPILOGUE' must likewise

  35421.      use 'regs_ever_live'.)

  35422. 

  35423.      On machines that have "register windows", the function entry code

  35424.      does not save on the stack the registers that are in the windows,

  35425.      even if they are supposed to be preserved by function calls;

  35426.      instead it takes appropriate steps to "push" the register stack, if

  35427.      any non-call-used registers are used in the function.

  35428. 

  35429.      On machines where functions may or may not have frame-pointers, the

  35430.      function entry code must vary accordingly; it must set up the frame

  35431.      pointer if one is wanted, and not otherwise.  To determine whether

  35432.      a frame pointer is in wanted, the macro can refer to the variable

  35433.      'frame_pointer_needed'.  The variable's value will be 1 at run time

  35434.      in a function that needs a frame pointer.  *Note Elimination::.

  35435. 

  35436.      The function entry code is responsible for allocating any stack

  35437.      space required for the function.  This stack space consists of the

  35438.      regions listed below.  In most cases, these regions are allocated

  35439.      in the order listed, with the last listed region closest to the top

  35440.      of the stack (the lowest address if 'STACK_GROWS_DOWNWARD' is

  35441.      defined, and the highest address if it is not defined).  You can

  35442.      use a different order for a machine if doing so is more convenient

  35443.      or required for compatibility reasons.  Except in cases where

  35444.      required by standard or by a debugger, there is no reason why the

  35445.      stack layout used by GCC need agree with that used by other

  35446.      compilers for a machine.

  35447. 

  35448.  -- Target Hook: void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *FILE)

  35449.      If defined, a function that outputs assembler code at the end of a

  35450.      prologue.  This should be used when the function prologue is being

  35451.      emitted as RTL, and you have some extra assembler that needs to be

  35452.      emitted.  *Note prologue instruction pattern::.

  35453. 

  35454.  -- Target Hook: void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *FILE)

  35455.      If defined, a function that outputs assembler code at the start of

  35456.      an epilogue.  This should be used when the function epilogue is

  35457.      being emitted as RTL, and you have some extra assembler that needs

  35458.      to be emitted.  *Note epilogue instruction pattern::.

  35459. 

  35460.  -- Target Hook: void TARGET_ASM_FUNCTION_EPILOGUE (FILE *FILE)

  35461.      If defined, a function that outputs the assembler code for exit

  35462.      from a function.  The epilogue is responsible for restoring the

  35463.      saved registers and stack pointer to their values when the function

  35464.      was called, and returning control to the caller.  This macro takes

  35465.      the same argument as the macro 'TARGET_ASM_FUNCTION_PROLOGUE', and

  35466.      the registers to restore are determined from 'regs_ever_live' and

  35467.      'CALL_USED_REGISTERS' in the same way.

  35468. 

  35469.      On some machines, there is a single instruction that does all the

  35470.      work of returning from the function.  On these machines, give that

  35471.      instruction the name 'return' and do not define the macro

  35472.      'TARGET_ASM_FUNCTION_EPILOGUE' at all.

  35473. 

  35474.      Do not define a pattern named 'return' if you want the

  35475.      'TARGET_ASM_FUNCTION_EPILOGUE' to be used.  If you want the target

  35476.      switches to control whether return instructions or epilogues are

  35477.      used, define a 'return' pattern with a validity condition that

  35478.      tests the target switches appropriately.  If the 'return' pattern's

  35479.      validity condition is false, epilogues will be used.

  35480. 

  35481.      On machines where functions may or may not have frame-pointers, the

  35482.      function exit code must vary accordingly.  Sometimes the code for

  35483.      these two cases is completely different.  To determine whether a

  35484.      frame pointer is wanted, the macro can refer to the variable

  35485.      'frame_pointer_needed'.  The variable's value will be 1 when

  35486.      compiling a function that needs a frame pointer.

  35487. 

  35488.      Normally, 'TARGET_ASM_FUNCTION_PROLOGUE' and

  35489.      'TARGET_ASM_FUNCTION_EPILOGUE' must treat leaf functions specially.

  35490.      The C variable 'current_function_is_leaf' is nonzero for such a

  35491.      function.  *Note Leaf Functions::.

  35492. 

  35493.      On some machines, some functions pop their arguments on exit while

  35494.      others leave that for the caller to do.  For example, the 68020

  35495.      when given '-mrtd' pops arguments in functions that take a fixed

  35496.      number of arguments.

  35497. 

  35498.      Your definition of the macro 'RETURN_POPS_ARGS' decides which

  35499.      functions pop their own arguments.  'TARGET_ASM_FUNCTION_EPILOGUE'

  35500.      needs to know what was decided.  The number of bytes of the current

  35501.      function's arguments that this function should pop is available in

  35502.      'crtl->args.pops_args'.  *Note Scalar Return::.

  35503. 

  35504.    * A region of 'crtl->args.pretend_args_size' bytes of uninitialized

  35505.      space just underneath the first argument arriving on the stack.

  35506.      (This may not be at the very start of the allocated stack region if

  35507.      the calling sequence has pushed anything else since pushing the

  35508.      stack arguments.  But usually, on such machines, nothing else has

  35509.      been pushed yet, because the function prologue itself does all the

  35510.      pushing.)  This region is used on machines where an argument may be

  35511.      passed partly in registers and partly in memory, and, in some cases

  35512.      to support the features in '<stdarg.h>'.

  35513. 

  35514.    * An area of memory used to save certain registers used by the

  35515.      function.  The size of this area, which may also include space for

  35516.      such things as the return address and pointers to previous stack

  35517.      frames, is machine-specific and usually depends on which registers

  35518.      have been used in the function.  Machines with register windows

  35519.      often do not require a save area.

  35520. 

  35521.    * A region of at least SIZE bytes, possibly rounded up to an

  35522.      allocation boundary, to contain the local variables of the

  35523.      function.  On some machines, this region and the save area may

  35524.      occur in the opposite order, with the save area closer to the top

  35525.      of the stack.

  35526. 

  35527.    * Optionally, when 'ACCUMULATE_OUTGOING_ARGS' is defined, a region of

  35528.      'crtl->outgoing_args_size' bytes to be used for outgoing argument

  35529.      lists of the function.  *Note Stack Arguments::.

  35530. 

  35531.  -- Macro: EXIT_IGNORE_STACK

  35532.      Define this macro as a C expression that is nonzero if the return

  35533.      instruction or the function epilogue ignores the value of the stack

  35534.      pointer; in other words, if it is safe to delete an instruction to

  35535.      adjust the stack pointer before a return from the function.  The

  35536.      default is 0.

  35537. 

  35538.      Note that this macro's value is relevant only for functions for

  35539.      which frame pointers are maintained.  It is never safe to delete a

  35540.      final stack adjustment in a function that has no frame pointer, and

  35541.      the compiler knows this regardless of 'EXIT_IGNORE_STACK'.

  35542. 

  35543.  -- Macro: EPILOGUE_USES (REGNO)

  35544.      Define this macro as a C expression that is nonzero for registers

  35545.      that are used by the epilogue or the 'return' pattern.  The stack

  35546.      and frame pointer registers are already assumed to be used as

  35547.      needed.

  35548. 

  35549.  -- Macro: EH_USES (REGNO)

  35550.      Define this macro as a C expression that is nonzero for registers

  35551.      that are used by the exception handling mechanism, and so should be

  35552.      considered live on entry to an exception edge.

  35553. 

  35554.  -- Target Hook: void TARGET_ASM_OUTPUT_MI_THUNK (FILE *FILE, tree

  35555.           THUNK_FNDECL, HOST_WIDE_INT DELTA, HOST_WIDE_INT VCALL_OFFSET,

  35556.           tree FUNCTION)

  35557.      A function that outputs the assembler code for a thunk function,

  35558.      used to implement C++ virtual function calls with multiple

  35559.      inheritance.  The thunk acts as a wrapper around a virtual

  35560.      function, adjusting the implicit object parameter before handing

  35561.      control off to the real function.

  35562. 

  35563.      First, emit code to add the integer DELTA to the location that

  35564.      contains the incoming first argument.  Assume that this argument

  35565.      contains a pointer, and is the one used to pass the 'this' pointer

  35566.      in C++.  This is the incoming argument _before_ the function

  35567.      prologue, e.g. '%o0' on a sparc.  The addition must preserve the

  35568.      values of all other incoming arguments.

  35569. 

  35570.      Then, if VCALL_OFFSET is nonzero, an additional adjustment should

  35571.      be made after adding 'delta'.  In particular, if P is the adjusted

  35572.      pointer, the following adjustment should be made:

  35573. 

  35574.           p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]

  35575. 

  35576.      After the additions, emit code to jump to FUNCTION, which is a

  35577.      'FUNCTION_DECL'.  This is a direct pure jump, not a call, and does

  35578.      not touch the return address.  Hence returning from FUNCTION will

  35579.      return to whoever called the current 'thunk'.

  35580. 

  35581.      The effect must be as if FUNCTION had been called directly with the

  35582.      adjusted first argument.  This macro is responsible for emitting

  35583.      all of the code for a thunk function;

  35584.      'TARGET_ASM_FUNCTION_PROLOGUE' and 'TARGET_ASM_FUNCTION_EPILOGUE'

  35585.      are not invoked.

  35586. 

  35587.      The THUNK_FNDECL is redundant.  (DELTA and FUNCTION have already

  35588.      been extracted from it.)  It might possibly be useful on some

  35589.      targets, but probably not.

  35590. 

  35591.      If you do not define this macro, the target-independent code in the

  35592.      C++ front end will generate a less efficient heavyweight thunk that

  35593.      calls FUNCTION instead of jumping to it.  The generic approach does

  35594.      not support varargs.

  35595. 

  35596.  -- Target Hook: bool TARGET_ASM_CAN_OUTPUT_MI_THUNK (const_tree

  35597.           THUNK_FNDECL, HOST_WIDE_INT DELTA, HOST_WIDE_INT VCALL_OFFSET,

  35598.           const_tree FUNCTION)

  35599.      A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be

  35600.      able to output the assembler code for the thunk function specified

  35601.      by the arguments it is passed, and false otherwise.  In the latter

  35602.      case, the generic approach will be used by the C++ front end, with

  35603.      the limitations previously exposed.

  35604. 

  35605. 

  35606. File: gccint.info,  Node: Profiling,  Next: Tail Calls,  Prev: Function Entry,  Up: Stack and Calling

  35607. 

  35608. 18.9.12 Generating Code for Profiling

  35609. -------------------------------------

  35610. 

  35611. These macros will help you generate code for profiling.

  35612. 

  35613.  -- Macro: FUNCTION_PROFILER (FILE, LABELNO)

  35614.      A C statement or compound statement to output to FILE some

  35615.      assembler code to call the profiling subroutine 'mcount'.

  35616. 

  35617.      The details of how 'mcount' expects to be called are determined by

  35618.      your operating system environment, not by GCC.  To figure them out,

  35619.      compile a small program for profiling using the system's installed

  35620.      C compiler and look at the assembler code that results.

  35621. 

  35622.      Older implementations of 'mcount' expect the address of a counter

  35623.      variable to be loaded into some register.  The name of this

  35624.      variable is 'LP' followed by the number LABELNO, so you would

  35625.      generate the name using 'LP%d' in a 'fprintf'.

  35626. 

  35627.  -- Macro: PROFILE_HOOK

  35628.      A C statement or compound statement to output to FILE some assembly

  35629.      code to call the profiling subroutine 'mcount' even the target does

  35630.      not support profiling.

  35631. 

  35632.  -- Macro: NO_PROFILE_COUNTERS

  35633.      Define this macro to be an expression with a nonzero value if the

  35634.      'mcount' subroutine on your system does not need a counter variable

  35635.      allocated for each function.  This is true for almost all modern

  35636.      implementations.  If you define this macro, you must not use the

  35637.      LABELNO argument to 'FUNCTION_PROFILER'.

  35638. 

  35639.  -- Macro: PROFILE_BEFORE_PROLOGUE

  35640.      Define this macro if the code for function profiling should come

  35641.      before the function prologue.  Normally, the profiling code comes

  35642.      after.

  35643. 

  35644.  -- Target Hook: bool TARGET_KEEP_LEAF_WHEN_PROFILED (void)

  35645.      This target hook returns true if the target wants the leaf flag for

  35646.      the current function to stay true even if it calls mcount.  This

  35647.      might make sense for targets using the leaf flag only to determine

  35648.      whether a stack frame needs to be generated or not and for which

  35649.      the call to mcount is generated before the function prologue.

  35650. 

  35651. 

  35652. File: gccint.info,  Node: Tail Calls,  Next: Shrink-wrapping separate components,  Prev: Profiling,  Up: Stack and Calling

  35653. 

  35654. 18.9.13 Permitting tail calls

  35655. -----------------------------

  35656. 

  35657.  -- Target Hook: bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree DECL, tree

  35658.           EXP)

  35659.      True if it is OK to do sibling call optimization for the specified

  35660.      call expression EXP.  DECL will be the called function, or 'NULL'

  35661.      if this is an indirect call.

  35662. 

  35663.      It is not uncommon for limitations of calling conventions to

  35664.      prevent tail calls to functions outside the current unit of

  35665.      translation, or during PIC compilation.  The hook is used to

  35666.      enforce these restrictions, as the 'sibcall' md pattern cannot

  35667.      fail, or fall over to a "normal" call.  The criteria for successful

  35668.      sibling call optimization may vary greatly between different

  35669.      architectures.

  35670. 

  35671.  -- Target Hook: void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap REGS)

  35672.      Add any hard registers to REGS that are live on entry to the

  35673.      function.  This hook only needs to be defined to provide registers

  35674.      that cannot be found by examination of FUNCTION_ARG_REGNO_P, the

  35675.      callee saved registers, STATIC_CHAIN_INCOMING_REGNUM,

  35676.      STATIC_CHAIN_REGNUM, TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM,

  35677.      EH_USES, FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the

  35678.      PIC_OFFSET_TABLE_REGNUM.

  35679. 

  35680.  -- Target Hook: void TARGET_SET_UP_BY_PROLOGUE (struct

  35681.           hard_reg_set_container *)

  35682.      This hook should add additional registers that are computed by the

  35683.      prologue to the hard regset for shrink-wrapping optimization

  35684.      purposes.

  35685. 

  35686.  -- Target Hook: bool TARGET_WARN_FUNC_RETURN (tree)

  35687.      True if a function's return statements should be checked for

  35688.      matching the function's return type.  This includes checking for

  35689.      falling off the end of a non-void function.  Return false if no

  35690.      such check should be made.

  35691. 

  35692. 

  35693. File: gccint.info,  Node: Shrink-wrapping separate components,  Next: Stack Smashing Protection,  Prev: Tail Calls,  Up: Stack and Calling

  35694. 

  35695. 18.9.14 Shrink-wrapping separate components

  35696. -------------------------------------------

  35697. 

  35698. The prologue may perform a variety of target dependent tasks such as

  35699. saving callee-saved registers, saving the return address, aligning the

  35700. stack, creating a stack frame, initializing the PIC register, setting up

  35701. the static chain, etc.

  35702. 

  35703.  On some targets some of these tasks may be independent of others and

  35704. thus may be shrink-wrapped separately.  These independent tasks are

  35705. referred to as components and are handled generically by the target

  35706. independent parts of GCC.

  35707. 

  35708.  Using the following hooks those prologue or epilogue components can be

  35709. shrink-wrapped separately, so that the initialization (and possibly

  35710. teardown) those components do is not done as frequently on execution

  35711. paths where this would unnecessary.

  35712. 

  35713.  What exactly those components are is up to the target code; the generic

  35714. code treats them abstractly, as a bit in an 'sbitmap'.  These 'sbitmap's

  35715. are allocated by the 'shrink_wrap.get_separate_components' and

  35716. 'shrink_wrap.components_for_bb' hooks, and deallocated by the generic

  35717. code.

  35718. 

  35719.  -- Target Hook: sbitmap TARGET_SHRINK_WRAP_GET_SEPARATE_COMPONENTS

  35720.           (void)

  35721.      This hook should return an 'sbitmap' with the bits set for those

  35722.      components that can be separately shrink-wrapped in the current

  35723.      function.  Return 'NULL' if the current function should not get any

  35724.      separate shrink-wrapping.  Don't define this hook if it would

  35725.      always return 'NULL'.  If it is defined, the other hooks in this

  35726.      group have to be defined as well.

  35727. 

  35728.  -- Target Hook: sbitmap TARGET_SHRINK_WRAP_COMPONENTS_FOR_BB

  35729.           (basic_block)

  35730.      This hook should return an 'sbitmap' with the bits set for those

  35731.      components where either the prologue component has to be executed

  35732.      before the 'basic_block', or the epilogue component after it, or

  35733.      both.

  35734. 

  35735.  -- Target Hook: void TARGET_SHRINK_WRAP_DISQUALIFY_COMPONENTS (sbitmap

  35736.           COMPONENTS, edge E, sbitmap EDGE_COMPONENTS, bool IS_PROLOGUE)

  35737.      This hook should clear the bits in the COMPONENTS bitmap for those

  35738.      components in EDGE_COMPONENTS that the target cannot handle on edge

  35739.      E, where IS_PROLOGUE says if this is for a prologue or an epilogue

  35740.      instead.

  35741. 

  35742.  -- Target Hook: void TARGET_SHRINK_WRAP_EMIT_PROLOGUE_COMPONENTS

  35743.           (sbitmap)

  35744.      Emit prologue insns for the components indicated by the parameter.

  35745. 

  35746.  -- Target Hook: void TARGET_SHRINK_WRAP_EMIT_EPILOGUE_COMPONENTS

  35747.           (sbitmap)

  35748.      Emit epilogue insns for the components indicated by the parameter.

  35749. 

  35750.  -- Target Hook: void TARGET_SHRINK_WRAP_SET_HANDLED_COMPONENTS

  35751.           (sbitmap)

  35752.      Mark the components in the parameter as handled, so that the

  35753.      'prologue' and 'epilogue' named patterns know to ignore those

  35754.      components.  The target code should not hang on to the 'sbitmap',

  35755.      it will be deleted after this call.

  35756. 

  35757. 

  35758. File: gccint.info,  Node: Stack Smashing Protection,  Next: Miscellaneous Register Hooks,  Prev: Shrink-wrapping separate components,  Up: Stack and Calling

  35759. 

  35760. 18.9.15 Stack smashing protection

  35761. ---------------------------------

  35762. 

  35763.  -- Target Hook: tree TARGET_STACK_PROTECT_GUARD (void)

  35764.      This hook returns a 'DECL' node for the external variable to use

  35765.      for the stack protection guard.  This variable is initialized by

  35766.      the runtime to some random value and is used to initialize the

  35767.      guard value that is placed at the top of the local stack frame.

  35768.      The type of this variable must be 'ptr_type_node'.

  35769. 

  35770.      The default version of this hook creates a variable called

  35771.      '__stack_chk_guard', which is normally defined in 'libgcc2.c'.

  35772. 

  35773.  -- Target Hook: tree TARGET_STACK_PROTECT_FAIL (void)

  35774.      This hook returns a 'CALL_EXPR' that alerts the runtime that the

  35775.      stack protect guard variable has been modified.  This expression

  35776.      should involve a call to a 'noreturn' function.

  35777. 

  35778.      The default version of this hook invokes a function called

  35779.      '__stack_chk_fail', taking no arguments.  This function is normally

  35780.      defined in 'libgcc2.c'.

  35781. 

  35782.  -- Target Hook: bool TARGET_STACK_PROTECT_RUNTIME_ENABLED_P (void)

  35783.      Returns true if the target wants GCC's default stack protect

  35784.      runtime support, otherwise return false.  The default

  35785.      implementation always returns true.

  35786. 

  35787.  -- Common Target Hook: bool TARGET_SUPPORTS_SPLIT_STACK (bool REPORT,

  35788.           struct gcc_options *OPTS)

  35789.      Whether this target supports splitting the stack when the options

  35790.      described in OPTS have been passed.  This is called after options

  35791.      have been parsed, so the target may reject splitting the stack in

  35792.      some configurations.  The default version of this hook returns

  35793.      false.  If REPORT is true, this function may issue a warning or

  35794.      error; if REPORT is false, it must simply return a value

  35795. 

  35796.  -- Common Target Hook: vec<const char *> TARGET_GET_VALID_OPTION_VALUES

  35797.           (int OPTION_CODE, const char *PREFIX)

  35798.      The hook is used for options that have a non-trivial list of

  35799.      possible option values.  OPTION_CODE is option code of opt_code

  35800.      enum type.  PREFIX is used for bash completion and allows an

  35801.      implementation to return more specific completion based on the

  35802.      prefix.  All string values should be allocated from heap memory and

  35803.      consumers should release them.  The result will be pruned to cases

  35804.      with PREFIX if not NULL.

  35805. 

  35806. 

  35807. File: gccint.info,  Node: Miscellaneous Register Hooks,  Prev: Stack Smashing Protection,  Up: Stack and Calling

  35808. 

  35809. 18.9.16 Miscellaneous register hooks

  35810. ------------------------------------

  35811. 

  35812.  -- Target Hook: bool TARGET_CALL_FUSAGE_CONTAINS_NON_CALLEE_CLOBBERS

  35813.      Set to true if each call that binds to a local definition

  35814.      explicitly clobbers or sets all non-fixed registers modified by

  35815.      performing the call.  That is, by the call pattern itself, or by

  35816.      code that might be inserted by the linker (e.g. stubs, veneers,

  35817.      branch islands), but not including those modifiable by the callee.

  35818.      The affected registers may be mentioned explicitly in the call

  35819.      pattern, or included as clobbers in CALL_INSN_FUNCTION_USAGE. The

  35820.      default version of this hook is set to false.  The purpose of this

  35821.      hook is to enable the fipa-ra optimization.

  35822. 

  35823. 

  35824. File: gccint.info,  Node: Varargs,  Next: Trampolines,  Prev: Stack and Calling,  Up: Target Macros

  35825. 

  35826. 18.10 Implementing the Varargs Macros

  35827. =====================================

  35828. 

  35829. GCC comes with an implementation of '<varargs.h>' and '<stdarg.h>' that

  35830. work without change on machines that pass arguments on the stack.  Other

  35831. machines require their own implementations of varargs, and the two

  35832. machine independent header files must have conditionals to include it.

  35833. 

  35834.  ISO '<stdarg.h>' differs from traditional '<varargs.h>' mainly in the

  35835. calling convention for 'va_start'.  The traditional implementation takes

  35836. just one argument, which is the variable in which to store the argument

  35837. pointer.  The ISO implementation of 'va_start' takes an additional

  35838. second argument.  The user is supposed to write the last named argument

  35839. of the function here.

  35840. 

  35841.  However, 'va_start' should not use this argument.  The way to find the

  35842. end of the named arguments is with the built-in functions described

  35843. below.

  35844. 

  35845.  -- Macro: __builtin_saveregs ()

  35846.      Use this built-in function to save the argument registers in memory

  35847.      so that the varargs mechanism can access them.  Both ISO and

  35848.      traditional versions of 'va_start' must use '__builtin_saveregs',

  35849.      unless you use 'TARGET_SETUP_INCOMING_VARARGS' (see below) instead.

  35850. 

  35851.      On some machines, '__builtin_saveregs' is open-coded under the

  35852.      control of the target hook 'TARGET_EXPAND_BUILTIN_SAVEREGS'.  On

  35853.      other machines, it calls a routine written in assembler language,

  35854.      found in 'libgcc2.c'.

  35855. 

  35856.      Code generated for the call to '__builtin_saveregs' appears at the

  35857.      beginning of the function, as opposed to where the call to

  35858.      '__builtin_saveregs' is written, regardless of what the code is.

  35859.      This is because the registers must be saved before the function

  35860.      starts to use them for its own purposes.

  35861. 

  35862.  -- Macro: __builtin_next_arg (LASTARG)

  35863.      This builtin returns the address of the first anonymous stack

  35864.      argument, as type 'void *'.  If 'ARGS_GROW_DOWNWARD', it returns

  35865.      the address of the location above the first anonymous stack

  35866.      argument.  Use it in 'va_start' to initialize the pointer for

  35867.      fetching arguments from the stack.  Also use it in 'va_start' to

  35868.      verify that the second parameter LASTARG is the last named argument

  35869.      of the current function.

  35870. 

  35871.  -- Macro: __builtin_classify_type (OBJECT)

  35872.      Since each machine has its own conventions for which data types are

  35873.      passed in which kind of register, your implementation of 'va_arg'

  35874.      has to embody these conventions.  The easiest way to categorize the

  35875.      specified data type is to use '__builtin_classify_type' together

  35876.      with 'sizeof' and '__alignof__'.

  35877. 

  35878.      '__builtin_classify_type' ignores the value of OBJECT, considering

  35879.      only its data type.  It returns an integer describing what kind of

  35880.      type that is--integer, floating, pointer, structure, and so on.

  35881. 

  35882.      The file 'typeclass.h' defines an enumeration that you can use to

  35883.      interpret the values of '__builtin_classify_type'.

  35884. 

  35885.  These machine description macros help implement varargs:

  35886. 

  35887.  -- Target Hook: rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void)

  35888.      If defined, this hook produces the machine-specific code for a call

  35889.      to '__builtin_saveregs'.  This code will be moved to the very

  35890.      beginning of the function, before any parameter access are made.

  35891.      The return value of this function should be an RTX that contains

  35892.      the value to use as the return of '__builtin_saveregs'.

  35893. 

  35894.  -- Target Hook: void TARGET_SETUP_INCOMING_VARARGS (cumulative_args_t

  35895.           ARGS_SO_FAR, const function_arg_info &ARG, int

  35896.           *PRETEND_ARGS_SIZE, int SECOND_TIME)

  35897.      This target hook offers an alternative to using

  35898.      '__builtin_saveregs' and defining the hook

  35899.      'TARGET_EXPAND_BUILTIN_SAVEREGS'.  Use it to store the anonymous

  35900.      register arguments into the stack so that all the arguments appear

  35901.      to have been passed consecutively on the stack.  Once this is done,

  35902.      you can use the standard implementation of varargs that works for

  35903.      machines that pass all their arguments on the stack.

  35904. 

  35905.      The argument ARGS_SO_FAR points to the 'CUMULATIVE_ARGS' data

  35906.      structure, containing the values that are obtained after processing

  35907.      the named arguments.  The argument ARG describes the last of these

  35908.      named arguments.

  35909. 

  35910.      The target hook should do two things: first, push onto the stack

  35911.      all the argument registers _not_ used for the named arguments, and

  35912.      second, store the size of the data thus pushed into the

  35913.      'int'-valued variable pointed to by PRETEND_ARGS_SIZE.  The value

  35914.      that you store here will serve as additional offset for setting up

  35915.      the stack frame.

  35916. 

  35917.      Because you must generate code to push the anonymous arguments at

  35918.      compile time without knowing their data types,

  35919.      'TARGET_SETUP_INCOMING_VARARGS' is only useful on machines that

  35920.      have just a single category of argument register and use it

  35921.      uniformly for all data types.

  35922. 

  35923.      If the argument SECOND_TIME is nonzero, it means that the arguments

  35924.      of the function are being analyzed for the second time.  This

  35925.      happens for an inline function, which is not actually compiled

  35926.      until the end of the source file.  The hook

  35927.      'TARGET_SETUP_INCOMING_VARARGS' should not generate any

  35928.      instructions in this case.

  35929. 

  35930.  -- Target Hook: bool TARGET_STRICT_ARGUMENT_NAMING (cumulative_args_t

  35931.           CA)

  35932.      Define this hook to return 'true' if the location where a function

  35933.      argument is passed depends on whether or not it is a named

  35934.      argument.

  35935. 

  35936.      This hook controls how the NAMED argument to 'TARGET_FUNCTION_ARG'

  35937.      is set for varargs and stdarg functions.  If this hook returns

  35938.      'true', the NAMED argument is always true for named arguments, and

  35939.      false for unnamed arguments.  If it returns 'false', but

  35940.      'TARGET_PRETEND_OUTGOING_VARARGS_NAMED' returns 'true', then all

  35941.      arguments are treated as named.  Otherwise, all named arguments

  35942.      except the last are treated as named.

  35943. 

  35944.      You need not define this hook if it always returns 'false'.

  35945. 

  35946.  -- Target Hook: void TARGET_CALL_ARGS (rtx, TREE)

  35947.      While generating RTL for a function call, this target hook is

  35948.      invoked once for each argument passed to the function, either a

  35949.      register returned by 'TARGET_FUNCTION_ARG' or a memory location.

  35950.      It is called just before the point where argument registers are

  35951.      stored.  The type of the function to be called is also passed as

  35952.      the second argument; it is 'NULL_TREE' for libcalls.  The

  35953.      'TARGET_END_CALL_ARGS' hook is invoked just after the code to copy

  35954.      the return reg has been emitted.  This functionality can be used to

  35955.      perform special setup of call argument registers if a target needs

  35956.      it.  For functions without arguments, the hook is called once with

  35957.      'pc_rtx' passed instead of an argument register.  Most ports do not

  35958.      need to implement anything for this hook.

  35959. 

  35960.  -- Target Hook: void TARGET_END_CALL_ARGS (void)

  35961.      This target hook is invoked while generating RTL for a function

  35962.      call, just after the point where the return reg is copied into a

  35963.      pseudo.  It signals that all the call argument and return registers

  35964.      for the just emitted call are now no longer in use.  Most ports do

  35965.      not need to implement anything for this hook.

  35966. 

  35967.  -- Target Hook: bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED

  35968.           (cumulative_args_t CA)

  35969.      If you need to conditionally change ABIs so that one works with

  35970.      'TARGET_SETUP_INCOMING_VARARGS', but the other works like neither

  35971.      'TARGET_SETUP_INCOMING_VARARGS' nor 'TARGET_STRICT_ARGUMENT_NAMING'

  35972.      was defined, then define this hook to return 'true' if

  35973.      'TARGET_SETUP_INCOMING_VARARGS' is used, 'false' otherwise.

  35974.      Otherwise, you should not define this hook.

  35975. 

  35976.  -- Target Hook: rtx TARGET_LOAD_BOUNDS_FOR_ARG (rtx SLOT, rtx ARG, rtx

  35977.           SLOT_NO)

  35978.      This hook is used by expand pass to emit insn to load bounds of ARG

  35979.      passed in SLOT.  Expand pass uses this hook in case bounds of ARG

  35980.      are not passed in register.  If SLOT is a memory, then bounds are

  35981.      loaded as for regular pointer loaded from memory.  If SLOT is not a

  35982.      memory then SLOT_NO is an integer constant holding number of the

  35983.      target dependent special slot which should be used to obtain

  35984.      bounds.  Hook returns RTX holding loaded bounds.

  35985. 

  35986.  -- Target Hook: void TARGET_STORE_BOUNDS_FOR_ARG (rtx ARG, rtx SLOT,

  35987.           rtx BOUNDS, rtx SLOT_NO)

  35988.      This hook is used by expand pass to emit insns to store BOUNDS of

  35989.      ARG passed in SLOT.  Expand pass uses this hook in case BOUNDS of

  35990.      ARG are not passed in register.  If SLOT is a memory, then BOUNDS

  35991.      are stored as for regular pointer stored in memory.  If SLOT is not

  35992.      a memory then SLOT_NO is an integer constant holding number of the

  35993.      target dependent special slot which should be used to store BOUNDS.

  35994. 

  35995.  -- Target Hook: rtx TARGET_LOAD_RETURNED_BOUNDS (rtx SLOT)

  35996.      This hook is used by expand pass to emit insn to load bounds

  35997.      returned by function call in SLOT.  Hook returns RTX holding loaded

  35998.      bounds.

  35999. 

  36000.  -- Target Hook: void TARGET_STORE_RETURNED_BOUNDS (rtx SLOT, rtx

  36001.           BOUNDS)

  36002.      This hook is used by expand pass to emit insn to store BOUNDS

  36003.      returned by function call into SLOT.

  36004. 

  36005. 

  36006. File: gccint.info,  Node: Trampolines,  Next: Library Calls,  Prev: Varargs,  Up: Target Macros

  36007. 

  36008. 18.11 Support for Nested Functions

  36009. ==================================

  36010. 

  36011. Taking the address of a nested function requires special compiler

  36012. handling to ensure that the static chain register is loaded when the

  36013. function is invoked via an indirect call.

  36014. 

  36015.  GCC has traditionally supported nested functions by creating an

  36016. executable "trampoline" at run time when the address of a nested

  36017. function is taken.  This is a small piece of code which normally resides

  36018. on the stack, in the stack frame of the containing function.  The

  36019. trampoline loads the static chain register and then jumps to the real

  36020. address of the nested function.

  36021. 

  36022.  The use of trampolines requires an executable stack, which is a

  36023. security risk.  To avoid this problem, GCC also supports another

  36024. strategy: using descriptors for nested functions.  Under this model,

  36025. taking the address of a nested function results in a pointer to a

  36026. non-executable function descriptor object.  Initializing the static

  36027. chain from the descriptor is handled at indirect call sites.

  36028. 

  36029.  On some targets, including HPPA and IA-64, function descriptors may be

  36030. mandated by the ABI or be otherwise handled in a target-specific way by

  36031. the back end in its code generation strategy for indirect calls.  GCC

  36032. also provides its own generic descriptor implementation to support the

  36033. '-fno-trampolines' option.  In this case runtime detection of function

  36034. descriptors at indirect call sites relies on descriptor pointers being

  36035. tagged with a bit that is never set in bare function addresses.  Since

  36036. GCC's generic function descriptors are not ABI-compliant, this option is

  36037. typically used only on a per-language basis (notably by Ada) or when it

  36038. can otherwise be applied to the whole program.

  36039. 

  36040.  Define the following hook if your backend either implements

  36041. ABI-specified descriptor support, or can use GCC's generic descriptor

  36042. implementation for nested functions.

  36043. 

  36044.  -- Target Hook: int TARGET_CUSTOM_FUNCTION_DESCRIPTORS

  36045.      If the target can use GCC's generic descriptor mechanism for nested

  36046.      functions, define this hook to a power of 2 representing an unused

  36047.      bit in function pointers which can be used to differentiate

  36048.      descriptors at run time.  This value gives the number of bytes by

  36049.      which descriptor pointers are misaligned compared to function

  36050.      pointers.  For example, on targets that require functions to be

  36051.      aligned to a 4-byte boundary, a value of either 1 or 2 is

  36052.      appropriate unless the architecture already reserves the bit for

  36053.      another purpose, such as on ARM.

  36054. 

  36055.      Define this hook to 0 if the target implements ABI support for

  36056.      function descriptors in its standard calling sequence, like for

  36057.      example HPPA or IA-64.

  36058. 

  36059.      Using descriptors for nested functions eliminates the need for

  36060.      trampolines that reside on the stack and require it to be made

  36061.      executable.

  36062. 

  36063.  The following macros tell GCC how to generate code to allocate and

  36064. initialize an executable trampoline.  You can also use this interface if

  36065. your back end needs to create ABI-specified non-executable descriptors;

  36066. in this case the "trampoline" created is the descriptor containing data

  36067. only.

  36068. 

  36069.  The instructions in an executable trampoline must do two things: load a

  36070. constant address into the static chain register, and jump to the real

  36071. address of the nested function.  On CISC machines such as the m68k, this

  36072. requires two instructions, a move immediate and a jump.  Then the two

  36073. addresses exist in the trampoline as word-long immediate operands.  On

  36074. RISC machines, it is often necessary to load each address into a

  36075. register in two parts.  Then pieces of each address form separate

  36076. immediate operands.

  36077. 

  36078.  The code generated to initialize the trampoline must store the variable

  36079. parts--the static chain value and the function address--into the

  36080. immediate operands of the instructions.  On a CISC machine, this is

  36081. simply a matter of copying each address to a memory reference at the

  36082. proper offset from the start of the trampoline.  On a RISC machine, it

  36083. may be necessary to take out pieces of the address and store them

  36084. separately.

  36085. 

  36086.  -- Target Hook: void TARGET_ASM_TRAMPOLINE_TEMPLATE (FILE *F)

  36087.      This hook is called by 'assemble_trampoline_template' to output, on

  36088.      the stream F, assembler code for a block of data that contains the

  36089.      constant parts of a trampoline.  This code should not include a

  36090.      label--the label is taken care of automatically.

  36091. 

  36092.      If you do not define this hook, it means no template is needed for

  36093.      the target.  Do not define this hook on systems where the block

  36094.      move code to copy the trampoline into place would be larger than

  36095.      the code to generate it on the spot.

  36096. 

  36097.  -- Macro: TRAMPOLINE_SECTION

  36098.      Return the section into which the trampoline template is to be

  36099.      placed (*note Sections::).  The default value is

  36100.      'readonly_data_section'.

  36101. 

  36102.  -- Macro: TRAMPOLINE_SIZE

  36103.      A C expression for the size in bytes of the trampoline, as an

  36104.      integer.

  36105. 

  36106.  -- Macro: TRAMPOLINE_ALIGNMENT

  36107.      Alignment required for trampolines, in bits.

  36108. 

  36109.      If you don't define this macro, the value of 'FUNCTION_ALIGNMENT'

  36110.      is used for aligning trampolines.

  36111. 

  36112.  -- Target Hook: void TARGET_TRAMPOLINE_INIT (rtx M_TRAMP, tree FNDECL,

  36113.           rtx STATIC_CHAIN)

  36114.      This hook is called to initialize a trampoline.  M_TRAMP is an RTX

  36115.      for the memory block for the trampoline; FNDECL is the

  36116.      'FUNCTION_DECL' for the nested function; STATIC_CHAIN is an RTX for

  36117.      the static chain value that should be passed to the function when

  36118.      it is called.

  36119. 

  36120.      If the target defines 'TARGET_ASM_TRAMPOLINE_TEMPLATE', then the

  36121.      first thing this hook should do is emit a block move into M_TRAMP

  36122.      from the memory block returned by 'assemble_trampoline_template'.

  36123.      Note that the block move need only cover the constant parts of the

  36124.      trampoline.  If the target isolates the variable parts of the

  36125.      trampoline to the end, not all 'TRAMPOLINE_SIZE' bytes need be

  36126.      copied.

  36127. 

  36128.      If the target requires any other actions, such as flushing caches

  36129.      or enabling stack execution, these actions should be performed

  36130.      after initializing the trampoline proper.

  36131. 

  36132.  -- Target Hook: rtx TARGET_TRAMPOLINE_ADJUST_ADDRESS (rtx ADDR)

  36133.      This hook should perform any machine-specific adjustment in the

  36134.      address of the trampoline.  Its argument contains the address of

  36135.      the memory block that was passed to 'TARGET_TRAMPOLINE_INIT'.  In

  36136.      case the address to be used for a function call should be different

  36137.      from the address at which the template was stored, the different

  36138.      address should be returned; otherwise ADDR should be returned

  36139.      unchanged.  If this hook is not defined, ADDR will be used for

  36140.      function calls.

  36141. 

  36142.  Implementing trampolines is difficult on many machines because they

  36143. have separate instruction and data caches.  Writing into a stack

  36144. location fails to clear the memory in the instruction cache, so when the

  36145. program jumps to that location, it executes the old contents.

  36146. 

  36147.  Here are two possible solutions.  One is to clear the relevant parts of

  36148. the instruction cache whenever a trampoline is set up.  The other is to

  36149. make all trampolines identical, by having them jump to a standard

  36150. subroutine.  The former technique makes trampoline execution faster; the

  36151. latter makes initialization faster.

  36152. 

  36153.  To clear the instruction cache when a trampoline is initialized, define

  36154. the following macro.

  36155. 

  36156.  -- Macro: CLEAR_INSN_CACHE (BEG, END)

  36157.      If defined, expands to a C expression clearing the _instruction

  36158.      cache_ in the specified interval.  The definition of this macro

  36159.      would typically be a series of 'asm' statements.  Both BEG and END

  36160.      are both pointer expressions.

  36161. 

  36162.  To use a standard subroutine, define the following macro.  In addition,

  36163. you must make sure that the instructions in a trampoline fill an entire

  36164. cache line with identical instructions, or else ensure that the

  36165. beginning of the trampoline code is always aligned at the same point in

  36166. its cache line.  Look in 'm68k.h' as a guide.

  36167. 

  36168.  -- Macro: TRANSFER_FROM_TRAMPOLINE

  36169.      Define this macro if trampolines need a special subroutine to do

  36170.      their work.  The macro should expand to a series of 'asm'

  36171.      statements which will be compiled with GCC.  They go in a library

  36172.      function named '__transfer_from_trampoline'.

  36173. 

  36174.      If you need to avoid executing the ordinary prologue code of a

  36175.      compiled C function when you jump to the subroutine, you can do so

  36176.      by placing a special label of your own in the assembler code.  Use

  36177.      one 'asm' statement to generate an assembler label, and another to

  36178.      make the label global.  Then trampolines can use that label to jump

  36179.      directly to your special assembler code.

  36180. 

  36181. 

  36182. File: gccint.info,  Node: Library Calls,  Next: Addressing Modes,  Prev: Trampolines,  Up: Target Macros

  36183. 

  36184. 18.12 Implicit Calls to Library Routines

  36185. ========================================

  36186. 

  36187. Here is an explanation of implicit calls to library routines.

  36188. 

  36189.  -- Macro: DECLARE_LIBRARY_RENAMES

  36190.      This macro, if defined, should expand to a piece of C code that

  36191.      will get expanded when compiling functions for libgcc.a.  It can be

  36192.      used to provide alternate names for GCC's internal library

  36193.      functions if there are ABI-mandated names that the compiler should

  36194.      provide.

  36195. 

  36196.  -- Target Hook: void TARGET_INIT_LIBFUNCS (void)

  36197.      This hook should declare additional library routines or rename

  36198.      existing ones, using the functions 'set_optab_libfunc' and

  36199.      'init_one_libfunc' defined in 'optabs.c'.  'init_optabs' calls this

  36200.      macro after initializing all the normal library routines.

  36201. 

  36202.      The default is to do nothing.  Most ports don't need to define this

  36203.      hook.

  36204. 

  36205.  -- Target Hook: bool TARGET_LIBFUNC_GNU_PREFIX

  36206.      If false (the default), internal library routines start with two

  36207.      underscores.  If set to true, these routines start with '__gnu_'

  36208.      instead.  E.g., '__muldi3' changes to '__gnu_muldi3'.  This

  36209.      currently only affects functions defined in 'libgcc2.c'.  If this

  36210.      is set to true, the 'tm.h' file must also '#define

  36211.      LIBGCC2_GNU_PREFIX'.

  36212. 

  36213.  -- Macro: FLOAT_LIB_COMPARE_RETURNS_BOOL (MODE, COMPARISON)

  36214.      This macro should return 'true' if the library routine that

  36215.      implements the floating point comparison operator COMPARISON in

  36216.      mode MODE will return a boolean, and FALSE if it will return a

  36217.      tristate.

  36218. 

  36219.      GCC's own floating point libraries return tristates from the

  36220.      comparison operators, so the default returns false always.  Most

  36221.      ports don't need to define this macro.

  36222. 

  36223.  -- Macro: TARGET_LIB_INT_CMP_BIASED

  36224.      This macro should evaluate to 'true' if the integer comparison

  36225.      functions (like '__cmpdi2') return 0 to indicate that the first

  36226.      operand is smaller than the second, 1 to indicate that they are

  36227.      equal, and 2 to indicate that the first operand is greater than the

  36228.      second.  If this macro evaluates to 'false' the comparison

  36229.      functions return -1, 0, and 1 instead of 0, 1, and 2.  If the

  36230.      target uses the routines in 'libgcc.a', you do not need to define

  36231.      this macro.

  36232. 

  36233.  -- Macro: TARGET_HAS_NO_HW_DIVIDE

  36234.      This macro should be defined if the target has no hardware divide

  36235.      instructions.  If this macro is defined, GCC will use an algorithm

  36236.      which make use of simple logical and arithmetic operations for

  36237.      64-bit division.  If the macro is not defined, GCC will use an

  36238.      algorithm which make use of a 64-bit by 32-bit divide primitive.

  36239. 

  36240.  -- Macro: TARGET_EDOM

  36241.      The value of 'EDOM' on the target machine, as a C integer constant

  36242.      expression.  If you don't define this macro, GCC does not attempt

  36243.      to deposit the value of 'EDOM' into 'errno' directly.  Look in

  36244.      '/usr/include/errno.h' to find the value of 'EDOM' on your system.

  36245. 

  36246.      If you do not define 'TARGET_EDOM', then compiled code reports

  36247.      domain errors by calling the library function and letting it report

  36248.      the error.  If mathematical functions on your system use 'matherr'

  36249.      when there is an error, then you should leave 'TARGET_EDOM'

  36250.      undefined so that 'matherr' is used normally.

  36251. 

  36252.  -- Macro: GEN_ERRNO_RTX

  36253.      Define this macro as a C expression to create an rtl expression

  36254.      that refers to the global "variable" 'errno'.  (On certain systems,

  36255.      'errno' may not actually be a variable.)  If you don't define this

  36256.      macro, a reasonable default is used.

  36257. 

  36258.  -- Target Hook: bool TARGET_LIBC_HAS_FUNCTION (enum function_class

  36259.           FN_CLASS)

  36260.      This hook determines whether a function from a class of functions

  36261.      FN_CLASS is present in the target C library.

  36262. 

  36263.  -- Target Hook: bool TARGET_LIBC_HAS_FAST_FUNCTION (int FCODE)

  36264.      This hook determines whether a function from a class of functions

  36265.      '(enum function_class)'FCODE has a fast implementation.

  36266. 

  36267.  -- Macro: NEXT_OBJC_RUNTIME

  36268.      Set this macro to 1 to use the "NeXT" Objective-C message sending

  36269.      conventions by default.  This calling convention involves passing

  36270.      the object, the selector and the method arguments all at once to

  36271.      the method-lookup library function.  This is the usual setting when

  36272.      targeting Darwin/Mac OS X systems, which have the NeXT runtime

  36273.      installed.

  36274. 

  36275.      If the macro is set to 0, the "GNU" Objective-C message sending

  36276.      convention will be used by default.  This convention passes just

  36277.      the object and the selector to the method-lookup function, which

  36278.      returns a pointer to the method.

  36279. 

  36280.      In either case, it remains possible to select code-generation for

  36281.      the alternate scheme, by means of compiler command line switches.

  36282. 

  36283. 

  36284. File: gccint.info,  Node: Addressing Modes,  Next: Anchored Addresses,  Prev: Library Calls,  Up: Target Macros

  36285. 

  36286. 18.13 Addressing Modes

  36287. ======================

  36288. 

  36289. This is about addressing modes.

  36290. 

  36291.  -- Macro: HAVE_PRE_INCREMENT

  36292.  -- Macro: HAVE_PRE_DECREMENT

  36293.  -- Macro: HAVE_POST_INCREMENT

  36294.  -- Macro: HAVE_POST_DECREMENT

  36295.      A C expression that is nonzero if the machine supports

  36296.      pre-increment, pre-decrement, post-increment, or post-decrement

  36297.      addressing respectively.

  36298. 

  36299.  -- Macro: HAVE_PRE_MODIFY_DISP

  36300.  -- Macro: HAVE_POST_MODIFY_DISP

  36301.      A C expression that is nonzero if the machine supports pre- or

  36302.      post-address side-effect generation involving constants other than

  36303.      the size of the memory operand.

  36304. 

  36305.  -- Macro: HAVE_PRE_MODIFY_REG

  36306.  -- Macro: HAVE_POST_MODIFY_REG

  36307.      A C expression that is nonzero if the machine supports pre- or

  36308.      post-address side-effect generation involving a register

  36309.      displacement.

  36310. 

  36311.  -- Macro: CONSTANT_ADDRESS_P (X)

  36312.      A C expression that is 1 if the RTX X is a constant which is a

  36313.      valid address.  On most machines the default definition of

  36314.      '(CONSTANT_P (X) && GET_CODE (X) != CONST_DOUBLE)' is acceptable,

  36315.      but a few machines are more restrictive as to which constant

  36316.      addresses are supported.

  36317. 

  36318.  -- Macro: CONSTANT_P (X)

  36319.      'CONSTANT_P', which is defined by target-independent code, accepts

  36320.      integer-values expressions whose values are not explicitly known,

  36321.      such as 'symbol_ref', 'label_ref', and 'high' expressions and

  36322.      'const' arithmetic expressions, in addition to 'const_int' and

  36323.      'const_double' expressions.

  36324. 

  36325.  -- Macro: MAX_REGS_PER_ADDRESS

  36326.      A number, the maximum number of registers that can appear in a

  36327.      valid memory address.  Note that it is up to you to specify a value

  36328.      equal to the maximum number that 'TARGET_LEGITIMATE_ADDRESS_P'

  36329.      would ever accept.

  36330. 

  36331.  -- Target Hook: bool TARGET_LEGITIMATE_ADDRESS_P (machine_mode MODE,

  36332.           rtx X, bool STRICT)

  36333.      A function that returns whether X (an RTX) is a legitimate memory

  36334.      address on the target machine for a memory operand of mode MODE.

  36335. 

  36336.      Legitimate addresses are defined in two variants: a strict variant

  36337.      and a non-strict one.  The STRICT parameter chooses which variant

  36338.      is desired by the caller.

  36339. 

  36340.      The strict variant is used in the reload pass.  It must be defined

  36341.      so that any pseudo-register that has not been allocated a hard

  36342.      register is considered a memory reference.  This is because in

  36343.      contexts where some kind of register is required, a pseudo-register

  36344.      with no hard register must be rejected.  For non-hard registers,

  36345.      the strict variant should look up the 'reg_renumber' array; it

  36346.      should then proceed using the hard register number in the array, or

  36347.      treat the pseudo as a memory reference if the array holds '-1'.

  36348. 

  36349.      The non-strict variant is used in other passes.  It must be defined

  36350.      to accept all pseudo-registers in every context where some kind of

  36351.      register is required.

  36352. 

  36353.      Normally, constant addresses which are the sum of a 'symbol_ref'

  36354.      and an integer are stored inside a 'const' RTX to mark them as

  36355.      constant.  Therefore, there is no need to recognize such sums

  36356.      specifically as legitimate addresses.  Normally you would simply

  36357.      recognize any 'const' as legitimate.

  36358. 

  36359.      Usually 'PRINT_OPERAND_ADDRESS' is not prepared to handle constant

  36360.      sums that are not marked with 'const'.  It assumes that a naked

  36361.      'plus' indicates indexing.  If so, then you _must_ reject such

  36362.      naked constant sums as illegitimate addresses, so that none of them

  36363.      will be given to 'PRINT_OPERAND_ADDRESS'.

  36364. 

  36365.      On some machines, whether a symbolic address is legitimate depends

  36366.      on the section that the address refers to.  On these machines,

  36367.      define the target hook 'TARGET_ENCODE_SECTION_INFO' to store the

  36368.      information into the 'symbol_ref', and then check for it here.

  36369.      When you see a 'const', you will have to look inside it to find the

  36370.      'symbol_ref' in order to determine the section.  *Note Assembler

  36371.      Format::.

  36372. 

  36373.      Some ports are still using a deprecated legacy substitute for this

  36374.      hook, the 'GO_IF_LEGITIMATE_ADDRESS' macro.  This macro has this

  36375.      syntax:

  36376. 

  36377.           #define GO_IF_LEGITIMATE_ADDRESS (MODE, X, LABEL)

  36378. 

  36379.      and should 'goto LABEL' if the address X is a valid address on the

  36380.      target machine for a memory operand of mode MODE.

  36381. 

  36382.      Compiler source files that want to use the strict variant of this

  36383.      macro define the macro 'REG_OK_STRICT'.  You should use an '#ifdef

  36384.      REG_OK_STRICT' conditional to define the strict variant in that

  36385.      case and the non-strict variant otherwise.

  36386. 

  36387.      Using the hook is usually simpler because it limits the number of

  36388.      files that are recompiled when changes are made.

  36389. 

  36390.  -- Macro: TARGET_MEM_CONSTRAINT

  36391.      A single character to be used instead of the default ''m''

  36392.      character for general memory addresses.  This defines the

  36393.      constraint letter which matches the memory addresses accepted by

  36394.      'TARGET_LEGITIMATE_ADDRESS_P'.  Define this macro if you want to

  36395.      support new address formats in your back end without changing the

  36396.      semantics of the ''m'' constraint.  This is necessary in order to

  36397.      preserve functionality of inline assembly constructs using the

  36398.      ''m'' constraint.

  36399. 

  36400.  -- Macro: FIND_BASE_TERM (X)

  36401.      A C expression to determine the base term of address X, or to

  36402.      provide a simplified version of X from which 'alias.c' can easily

  36403.      find the base term.  This macro is used in only two places:

  36404.      'find_base_value' and 'find_base_term' in 'alias.c'.

  36405. 

  36406.      It is always safe for this macro to not be defined.  It exists so

  36407.      that alias analysis can understand machine-dependent addresses.

  36408. 

  36409.      The typical use of this macro is to handle addresses containing a

  36410.      label_ref or symbol_ref within an UNSPEC.

  36411. 

  36412.  -- Target Hook: rtx TARGET_LEGITIMIZE_ADDRESS (rtx X, rtx OLDX,

  36413.           machine_mode MODE)

  36414.      This hook is given an invalid memory address X for an operand of

  36415.      mode MODE and should try to return a valid memory address.

  36416. 

  36417.      X will always be the result of a call to 'break_out_memory_refs',

  36418.      and OLDX will be the operand that was given to that function to

  36419.      produce X.

  36420. 

  36421.      The code of the hook should not alter the substructure of X.  If it

  36422.      transforms X into a more legitimate form, it should return the new

  36423.      X.

  36424. 

  36425.      It is not necessary for this hook to come up with a legitimate

  36426.      address, with the exception of native TLS addresses (*note Emulated

  36427.      TLS::).  The compiler has standard ways of doing so in all cases.

  36428.      In fact, if the target supports only emulated TLS, it is safe to

  36429.      omit this hook or make it return X if it cannot find a valid way to

  36430.      legitimize the address.  But often a machine-dependent strategy can

  36431.      generate better code.

  36432. 

  36433.  -- Macro: LEGITIMIZE_RELOAD_ADDRESS (X, MODE, OPNUM, TYPE, IND_LEVELS,

  36434.           WIN)

  36435.      A C compound statement that attempts to replace X, which is an

  36436.      address that needs reloading, with a valid memory address for an

  36437.      operand of mode MODE.  WIN will be a C statement label elsewhere in

  36438.      the code.  It is not necessary to define this macro, but it might

  36439.      be useful for performance reasons.

  36440. 

  36441.      For example, on the i386, it is sometimes possible to use a single

  36442.      reload register instead of two by reloading a sum of two pseudo

  36443.      registers into a register.  On the other hand, for number of RISC

  36444.      processors offsets are limited so that often an intermediate

  36445.      address needs to be generated in order to address a stack slot.  By

  36446.      defining 'LEGITIMIZE_RELOAD_ADDRESS' appropriately, the

  36447.      intermediate addresses generated for adjacent some stack slots can

  36448.      be made identical, and thus be shared.

  36449. 

  36450.      _Note_: This macro should be used with caution.  It is necessary to

  36451.      know something of how reload works in order to effectively use

  36452.      this, and it is quite easy to produce macros that build in too much

  36453.      knowledge of reload internals.

  36454. 

  36455.      _Note_: This macro must be able to reload an address created by a

  36456.      previous invocation of this macro.  If it fails to handle such

  36457.      addresses then the compiler may generate incorrect code or abort.

  36458. 

  36459.      The macro definition should use 'push_reload' to indicate parts

  36460.      that need reloading; OPNUM, TYPE and IND_LEVELS are usually

  36461.      suitable to be passed unaltered to 'push_reload'.

  36462. 

  36463.      The code generated by this macro must not alter the substructure of

  36464.      X.  If it transforms X into a more legitimate form, it should

  36465.      assign X (which will always be a C variable) a new value.  This

  36466.      also applies to parts that you change indirectly by calling

  36467.      'push_reload'.

  36468. 

  36469.      The macro definition may use 'strict_memory_address_p' to test if

  36470.      the address has become legitimate.

  36471. 

  36472.      If you want to change only a part of X, one standard way of doing

  36473.      this is to use 'copy_rtx'.  Note, however, that it unshares only a

  36474.      single level of rtl.  Thus, if the part to be changed is not at the

  36475.      top level, you'll need to replace first the top level.  It is not

  36476.      necessary for this macro to come up with a legitimate address; but

  36477.      often a machine-dependent strategy can generate better code.

  36478. 

  36479.  -- Target Hook: bool TARGET_MODE_DEPENDENT_ADDRESS_P (const_rtx ADDR,

  36480.           addr_space_t ADDRSPACE)

  36481.      This hook returns 'true' if memory address ADDR in address space

  36482.      ADDRSPACE can have different meanings depending on the machine mode

  36483.      of the memory reference it is used for or if the address is valid

  36484.      for some modes but not others.

  36485. 

  36486.      Autoincrement and autodecrement addresses typically have

  36487.      mode-dependent effects because the amount of the increment or

  36488.      decrement is the size of the operand being addressed.  Some

  36489.      machines have other mode-dependent addresses.  Many RISC machines

  36490.      have no mode-dependent addresses.

  36491. 

  36492.      You may assume that ADDR is a valid address for the machine.

  36493. 

  36494.      The default version of this hook returns 'false'.

  36495. 

  36496.  -- Target Hook: bool TARGET_LEGITIMATE_CONSTANT_P (machine_mode MODE,

  36497.           rtx X)

  36498.      This hook returns true if X is a legitimate constant for a

  36499.      MODE-mode immediate operand on the target machine.  You can assume

  36500.      that X satisfies 'CONSTANT_P', so you need not check this.

  36501. 

  36502.      The default definition returns true.

  36503. 

  36504.  -- Target Hook: rtx TARGET_DELEGITIMIZE_ADDRESS (rtx X)

  36505.      This hook is used to undo the possibly obfuscating effects of the

  36506.      'LEGITIMIZE_ADDRESS' and 'LEGITIMIZE_RELOAD_ADDRESS' target macros.

  36507.      Some backend implementations of these macros wrap symbol references

  36508.      inside an 'UNSPEC' rtx to represent PIC or similar addressing

  36509.      modes.  This target hook allows GCC's optimizers to understand the

  36510.      semantics of these opaque 'UNSPEC's by converting them back into

  36511.      their original form.

  36512. 

  36513.  -- Target Hook: bool TARGET_CONST_NOT_OK_FOR_DEBUG_P (rtx X)

  36514.      This hook should return true if X should not be emitted into debug

  36515.      sections.

  36516. 

  36517.  -- Target Hook: bool TARGET_CANNOT_FORCE_CONST_MEM (machine_mode MODE,

  36518.           rtx X)

  36519.      This hook should return true if X is of a form that cannot (or

  36520.      should not) be spilled to the constant pool.  MODE is the mode of

  36521.      X.

  36522. 

  36523.      The default version of this hook returns false.

  36524. 

  36525.      The primary reason to define this hook is to prevent reload from

  36526.      deciding that a non-legitimate constant would be better reloaded

  36527.      from the constant pool instead of spilling and reloading a register

  36528.      holding the constant.  This restriction is often true of addresses

  36529.      of TLS symbols for various targets.

  36530. 

  36531.  -- Target Hook: bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (machine_mode

  36532.           MODE, const_rtx X)

  36533.      This hook should return true if pool entries for constant X can be

  36534.      placed in an 'object_block' structure.  MODE is the mode of X.

  36535. 

  36536.      The default version returns false for all constants.

  36537. 

  36538.  -- Target Hook: bool TARGET_USE_BLOCKS_FOR_DECL_P (const_tree DECL)

  36539.      This hook should return true if pool entries for DECL should be

  36540.      placed in an 'object_block' structure.

  36541. 

  36542.      The default version returns true for all decls.

  36543. 

  36544.  -- Target Hook: tree TARGET_BUILTIN_RECIPROCAL (tree FNDECL)

  36545.      This hook should return the DECL of a function that implements the

  36546.      reciprocal of the machine-specific builtin function FNDECL, or

  36547.      'NULL_TREE' if such a function is not available.

  36548. 

  36549.  -- Target Hook: tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void)

  36550.      This hook should return the DECL of a function F that given an

  36551.      address ADDR as an argument returns a mask M that can be used to

  36552.      extract from two vectors the relevant data that resides in ADDR in

  36553.      case ADDR is not properly aligned.

  36554. 

  36555.      The autovectorizer, when vectorizing a load operation from an

  36556.      address ADDR that may be unaligned, will generate two vector loads

  36557.      from the two aligned addresses around ADDR.  It then generates a

  36558.      'REALIGN_LOAD' operation to extract the relevant data from the two

  36559.      loaded vectors.  The first two arguments to 'REALIGN_LOAD', V1 and

  36560.      V2, are the two vectors, each of size VS, and the third argument,

  36561.      OFF, defines how the data will be extracted from these two vectors:

  36562.      if OFF is 0, then the returned vector is V2; otherwise, the

  36563.      returned vector is composed from the last VS-OFF elements of V1

  36564.      concatenated to the first OFF elements of V2.

  36565. 

  36566.      If this hook is defined, the autovectorizer will generate a call to

  36567.      F (using the DECL tree that this hook returns) and will use the

  36568.      return value of F as the argument OFF to 'REALIGN_LOAD'.

  36569.      Therefore, the mask M returned by F should comply with the

  36570.      semantics expected by 'REALIGN_LOAD' described above.  If this hook

  36571.      is not defined, then ADDR will be used as the argument OFF to

  36572.      'REALIGN_LOAD', in which case the low log2(VS) - 1 bits of ADDR

  36573.      will be considered.

  36574. 

  36575.  -- Target Hook: int TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST (enum

  36576.           vect_cost_for_stmt TYPE_OF_COST, tree VECTYPE, int MISALIGN)

  36577.      Returns cost of different scalar or vector statements for

  36578.      vectorization cost model.  For vector memory operations the cost

  36579.      may depend on type (VECTYPE) and misalignment value (MISALIGN).

  36580. 

  36581.  -- Target Hook: poly_uint64 TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT

  36582.           (const_tree TYPE)

  36583.      This hook returns the preferred alignment in bits for accesses to

  36584.      vectors of type TYPE in vectorized code.  This might be less than

  36585.      or greater than the ABI-defined value returned by

  36586.      'TARGET_VECTOR_ALIGNMENT'.  It can be equal to the alignment of a

  36587.      single element, in which case the vectorizer will not try to

  36588.      optimize for alignment.

  36589. 

  36590.      The default hook returns 'TYPE_ALIGN (TYPE)', which is correct for

  36591.      most targets.

  36592. 

  36593.  -- Target Hook: bool TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE

  36594.           (const_tree TYPE, bool IS_PACKED)

  36595.      Return true if vector alignment is reachable (by peeling N

  36596.      iterations) for the given scalar type TYPE.  IS_PACKED is false if

  36597.      the scalar access using TYPE is known to be naturally aligned.

  36598. 

  36599.  -- Target Hook: bool TARGET_VECTORIZE_VEC_PERM_CONST (machine_mode

  36600.           MODE, rtx OUTPUT, rtx IN0, rtx IN1, const vec_perm_indices

  36601.           &SEL)

  36602.      This hook is used to test whether the target can permute up to two

  36603.      vectors of mode MODE using the permutation vector 'sel', and also

  36604.      to emit such a permutation.  In the former case IN0, IN1 and OUT

  36605.      are all null.  In the latter case IN0 and IN1 are the source

  36606.      vectors and OUT is the destination vector; all three are registers

  36607.      of mode MODE.  IN1 is the same as IN0 if SEL describes a

  36608.      permutation on one vector instead of two.

  36609. 

  36610.      Return true if the operation is possible, emitting instructions for

  36611.      it if rtxes are provided.

  36612. 

  36613.      If the hook returns false for a mode with multibyte elements, GCC

  36614.      will try the equivalent byte operation.  If that also fails, it

  36615.      will try forcing the selector into a register and using the

  36616.      VEC_PERMMODE instruction pattern.  There is no need for the hook to

  36617.      handle these two implementation approaches itself.

  36618. 

  36619.  -- Target Hook: tree TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION

  36620.           (unsigned CODE, tree VEC_TYPE_OUT, tree VEC_TYPE_IN)

  36621.      This hook should return the decl of a function that implements the

  36622.      vectorized variant of the function with the 'combined_fn' code CODE

  36623.      or 'NULL_TREE' if such a function is not available.  The return

  36624.      type of the vectorized function shall be of vector type

  36625.      VEC_TYPE_OUT and the argument types should be VEC_TYPE_IN.

  36626. 

  36627.  -- Target Hook: tree TARGET_VECTORIZE_BUILTIN_MD_VECTORIZED_FUNCTION

  36628.           (tree FNDECL, tree VEC_TYPE_OUT, tree VEC_TYPE_IN)

  36629.      This hook should return the decl of a function that implements the

  36630.      vectorized variant of target built-in function 'fndecl'.  The

  36631.      return type of the vectorized function shall be of vector type

  36632.      VEC_TYPE_OUT and the argument types should be VEC_TYPE_IN.

  36633. 

  36634.  -- Target Hook: bool TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT

  36635.           (machine_mode MODE, const_tree TYPE, int MISALIGNMENT, bool

  36636.           IS_PACKED)

  36637.      This hook should return true if the target supports misaligned

  36638.      vector store/load of a specific factor denoted in the MISALIGNMENT

  36639.      parameter.  The vector store/load should be of machine mode MODE

  36640.      and the elements in the vectors should be of type TYPE.  IS_PACKED

  36641.      parameter is true if the memory access is defined in a packed

  36642.      struct.

  36643. 

  36644.  -- Target Hook: machine_mode TARGET_VECTORIZE_PREFERRED_SIMD_MODE

  36645.           (scalar_mode MODE)

  36646.      This hook should return the preferred mode for vectorizing scalar

  36647.      mode MODE.  The default is equal to 'word_mode', because the

  36648.      vectorizer can do some transformations even in absence of

  36649.      specialized SIMD hardware.

  36650. 

  36651.  -- Target Hook: machine_mode TARGET_VECTORIZE_SPLIT_REDUCTION

  36652.           (machine_mode)

  36653.      This hook should return the preferred mode to split the final

  36654.      reduction step on MODE to.  The reduction is then carried out

  36655.      reducing upper against lower halves of vectors recursively until

  36656.      the specified mode is reached.  The default is MODE which means no

  36657.      splitting.

  36658. 

  36659.  -- Target Hook: unsigned int

  36660.           TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_MODES (vector_modes

  36661.           *MODES, bool ALL)

  36662.      If using the mode returned by

  36663.      'TARGET_VECTORIZE_PREFERRED_SIMD_MODE' is not the only approach

  36664.      worth considering, this hook should add one mode to MODES for each

  36665.      useful alternative approach.  These modes are then passed to

  36666.      'TARGET_VECTORIZE_RELATED_MODE' to obtain the vector mode for a

  36667.      given element mode.

  36668. 

  36669.      The modes returned in MODES should use the smallest element mode

  36670.      possible for the vectorization approach that they represent,

  36671.      preferring integer modes over floating-poing modes in the event of

  36672.      a tie.  The first mode should be the

  36673.      'TARGET_VECTORIZE_PREFERRED_SIMD_MODE' for its element mode.

  36674. 

  36675.      If ALL is true, add suitable vector modes even when they are

  36676.      generally not expected to be worthwhile.

  36677. 

  36678.      The hook returns a bitmask of flags that control how the modes in

  36679.      MODES are used.  The flags are:

  36680.      'VECT_COMPARE_COSTS'

  36681.           Tells the loop vectorizer to try all the provided modes and

  36682.           pick the one with the lowest cost.  By default the vectorizer

  36683.           will choose the first mode that works.

  36684. 

  36685.      The hook does not need to do anything if the vector returned by

  36686.      'TARGET_VECTORIZE_PREFERRED_SIMD_MODE' is the only one relevant for

  36687.      autovectorization.  The default implementation adds no modes and

  36688.      returns 0.

  36689. 

  36690.  -- Target Hook: opt_machine_mode TARGET_VECTORIZE_RELATED_MODE

  36691.           (machine_mode VECTOR_MODE, scalar_mode ELEMENT_MODE,

  36692.           poly_uint64 NUNITS)

  36693.      If a piece of code is using vector mode VECTOR_MODE and also wants

  36694.      to operate on elements of mode ELEMENT_MODE, return the vector mode

  36695.      it should use for those elements.  If NUNITS is nonzero, ensure

  36696.      that the mode has exactly NUNITS elements, otherwise pick whichever

  36697.      vector size pairs the most naturally with VECTOR_MODE.  Return an

  36698.      empty 'opt_machine_mode' if there is no supported vector mode with

  36699.      the required properties.

  36700. 

  36701.      There is no prescribed way of handling the case in which NUNITS is

  36702.      zero.  One common choice is to pick a vector mode with the same

  36703.      size as VECTOR_MODE; this is the natural choice if the target has a

  36704.      fixed vector size.  Another option is to choose a vector mode with

  36705.      the same number of elements as VECTOR_MODE; this is the natural

  36706.      choice if the target has a fixed number of elements.

  36707.      Alternatively, the hook might choose a middle ground, such as

  36708.      trying to keep the number of elements as similar as possible while

  36709.      applying maximum and minimum vector sizes.

  36710. 

  36711.      The default implementation uses 'mode_for_vector' to find the

  36712.      requested mode, returning a mode with the same size as VECTOR_MODE

  36713.      when NUNITS is zero.  This is the correct behavior for most

  36714.      targets.

  36715. 

  36716.  -- Target Hook: opt_machine_mode TARGET_VECTORIZE_GET_MASK_MODE

  36717.           (machine_mode MODE)

  36718.      Return the mode to use for a vector mask that holds one boolean

  36719.      result for each element of vector mode MODE.  The returned mask

  36720.      mode can be a vector of integers (class 'MODE_VECTOR_INT'), a

  36721.      vector of booleans (class 'MODE_VECTOR_BOOL') or a scalar integer

  36722.      (class 'MODE_INT').  Return an empty 'opt_machine_mode' if no such

  36723.      mask mode exists.

  36724. 

  36725.      The default implementation returns a 'MODE_VECTOR_INT' with the

  36726.      same size and number of elements as MODE, if such a mode exists.

  36727. 

  36728.  -- Target Hook: bool TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE (unsigned

  36729.           IFN)

  36730.      This hook returns true if masked internal function IFN (really of

  36731.      type 'internal_fn') should be considered expensive when the mask is

  36732.      all zeros.  GCC can then try to branch around the instruction

  36733.      instead.

  36734. 

  36735.  -- Target Hook: void * TARGET_VECTORIZE_INIT_COST (class loop

  36736.           *LOOP_INFO)

  36737.      This hook should initialize target-specific data structures in

  36738.      preparation for modeling the costs of vectorizing a loop or basic

  36739.      block.  The default allocates three unsigned integers for

  36740.      accumulating costs for the prologue, body, and epilogue of the loop

  36741.      or basic block.  If LOOP_INFO is non-NULL, it identifies the loop

  36742.      being vectorized; otherwise a single block is being vectorized.

  36743. 

  36744.  -- Target Hook: unsigned TARGET_VECTORIZE_ADD_STMT_COST (void *DATA,

  36745.           int COUNT, enum vect_cost_for_stmt KIND, class _stmt_vec_info

  36746.           *STMT_INFO, int MISALIGN, enum vect_cost_model_location WHERE)

  36747.      This hook should update the target-specific DATA in response to

  36748.      adding COUNT copies of the given KIND of statement to a loop or

  36749.      basic block.  The default adds the builtin vectorizer cost for the

  36750.      copies of the statement to the accumulator specified by WHERE, (the

  36751.      prologue, body, or epilogue) and returns the amount added.  The

  36752.      return value should be viewed as a tentative cost that may later be

  36753.      revised.

  36754. 

  36755.  -- Target Hook: void TARGET_VECTORIZE_FINISH_COST (void *DATA, unsigned

  36756.           *PROLOGUE_COST, unsigned *BODY_COST, unsigned *EPILOGUE_COST)

  36757.      This hook should complete calculations of the cost of vectorizing a

  36758.      loop or basic block based on DATA, and return the prologue, body,

  36759.      and epilogue costs as unsigned integers.  The default returns the

  36760.      value of the three accumulators.

  36761. 

  36762.  -- Target Hook: void TARGET_VECTORIZE_DESTROY_COST_DATA (void *DATA)

  36763.      This hook should release DATA and any related data structures

  36764.      allocated by TARGET_VECTORIZE_INIT_COST. The default releases the

  36765.      accumulator.

  36766. 

  36767.  -- Target Hook: tree TARGET_VECTORIZE_BUILTIN_GATHER (const_tree

  36768.           MEM_VECTYPE, const_tree INDEX_TYPE, int SCALE)

  36769.      Target builtin that implements vector gather operation.

  36770.      MEM_VECTYPE is the vector type of the load and INDEX_TYPE is scalar

  36771.      type of the index, scaled by SCALE.  The default is 'NULL_TREE'

  36772.      which means to not vectorize gather loads.

  36773. 

  36774.  -- Target Hook: tree TARGET_VECTORIZE_BUILTIN_SCATTER (const_tree

  36775.           VECTYPE, const_tree INDEX_TYPE, int SCALE)

  36776.      Target builtin that implements vector scatter operation.  VECTYPE

  36777.      is the vector type of the store and INDEX_TYPE is scalar type of

  36778.      the index, scaled by SCALE.  The default is 'NULL_TREE' which means

  36779.      to not vectorize scatter stores.

  36780. 

  36781.  -- Target Hook: int TARGET_SIMD_CLONE_COMPUTE_VECSIZE_AND_SIMDLEN

  36782.           (struct cgraph_node *, struct cgraph_simd_clone *, TREE, INT)

  36783.      This hook should set VECSIZE_MANGLE, VECSIZE_INT, VECSIZE_FLOAT

  36784.      fields in SIMD_CLONE structure pointed by CLONE_INFO argument and

  36785.      also SIMDLEN field if it was previously 0.  The hook should return

  36786.      0 if SIMD clones shouldn't be emitted, or number of VECSIZE_MANGLE

  36787.      variants that should be emitted.

  36788. 

  36789.  -- Target Hook: void TARGET_SIMD_CLONE_ADJUST (struct cgraph_node *)

  36790.      This hook should add implicit 'attribute(target("..."))' attribute

  36791.      to SIMD clone NODE if needed.

  36792. 

  36793.  -- Target Hook: int TARGET_SIMD_CLONE_USABLE (struct cgraph_node *)

  36794.      This hook should return -1 if SIMD clone NODE shouldn't be used in

  36795.      vectorized loops in current function, or non-negative number if it

  36796.      is usable.  In that case, the smaller the number is, the more

  36797.      desirable it is to use it.

  36798. 

  36799.  -- Target Hook: int TARGET_SIMT_VF (void)

  36800.      Return number of threads in SIMT thread group on the target.

  36801. 

  36802.  -- Target Hook: int TARGET_OMP_DEVICE_KIND_ARCH_ISA (enum

  36803.           omp_device_kind_arch_isa TRAIT, const char *NAME)

  36804.      Return 1 if TRAIT NAME is present in the OpenMP context's device

  36805.      trait set, return 0 if not present in any OpenMP context in the

  36806.      whole translation unit, or -1 if not present in the current OpenMP

  36807.      context but might be present in another OpenMP context in the same

  36808.      TU.

  36809. 

  36810.  -- Target Hook: bool TARGET_GOACC_VALIDATE_DIMS (tree DECL, int *DIMS,

  36811.           int FN_LEVEL, unsigned USED)

  36812.      This hook should check the launch dimensions provided for an

  36813.      OpenACC compute region, or routine.  Defaulted values are

  36814.      represented as -1 and non-constant values as 0.  The FN_LEVEL is

  36815.      negative for the function corresponding to the compute region.  For

  36816.      a routine it is the outermost level at which partitioned execution

  36817.      may be spawned.  The hook should verify non-default values.  If

  36818.      DECL is NULL, global defaults are being validated and unspecified

  36819.      defaults should be filled in.  Diagnostics should be issued as

  36820.      appropriate.  Return true, if changes have been made.  You must

  36821.      override this hook to provide dimensions larger than 1.

  36822. 

  36823.  -- Target Hook: int TARGET_GOACC_DIM_LIMIT (int AXIS)

  36824.      This hook should return the maximum size of a particular dimension,

  36825.      or zero if unbounded.

  36826. 

  36827.  -- Target Hook: bool TARGET_GOACC_FORK_JOIN (gcall *CALL, const int

  36828.           *DIMS, bool IS_FORK)

  36829.      This hook can be used to convert IFN_GOACC_FORK and IFN_GOACC_JOIN

  36830.      function calls to target-specific gimple, or indicate whether they

  36831.      should be retained.  It is executed during the oacc_device_lower

  36832.      pass.  It should return true, if the call should be retained.  It

  36833.      should return false, if it is to be deleted (either because

  36834.      target-specific gimple has been inserted before it, or there is no

  36835.      need for it).  The default hook returns false, if there are no RTL

  36836.      expanders for them.

  36837. 

  36838.  -- Target Hook: void TARGET_GOACC_REDUCTION (gcall *CALL)

  36839.      This hook is used by the oacc_transform pass to expand calls to the

  36840.      GOACC_REDUCTION internal function, into a sequence of gimple

  36841.      instructions.  CALL is gimple statement containing the call to the

  36842.      function.  This hook removes statement CALL after the expanded

  36843.      sequence has been inserted.  This hook is also responsible for

  36844.      allocating any storage for reductions when necessary.

  36845. 

  36846.  -- Target Hook: tree TARGET_PREFERRED_ELSE_VALUE (unsigned IFN, tree

  36847.           TYPE, unsigned NOPS, tree *OPS)

  36848.      This hook returns the target's preferred final argument for a call

  36849.      to conditional internal function IFN (really of type

  36850.      'internal_fn').  TYPE specifies the return type of the function and

  36851.      OPS are the operands to the conditional operation, of which there

  36852.      are NOPS.

  36853. 

  36854.      For example, if IFN is 'IFN_COND_ADD', the hook returns a value of

  36855.      type TYPE that should be used when 'OPS[0]' and 'OPS[1]' are

  36856.      conditionally added together.

  36857. 

  36858.      This hook is only relevant if the target supports conditional

  36859.      patterns like 'cond_addM'.  The default implementation returns a

  36860.      zero constant of type TYPE.

  36861. 

  36862. 

  36863. File: gccint.info,  Node: Anchored Addresses,  Next: Condition Code,  Prev: Addressing Modes,  Up: Target Macros

  36864. 

  36865. 18.14 Anchored Addresses

  36866. ========================

  36867. 

  36868. GCC usually addresses every static object as a separate entity.  For

  36869. example, if we have:

  36870. 

  36871.      static int a, b, c;

  36872.      int foo (void) { return a + b + c; }

  36873. 

  36874.  the code for 'foo' will usually calculate three separate symbolic

  36875. addresses: those of 'a', 'b' and 'c'.  On some targets, it would be

  36876. better to calculate just one symbolic address and access the three

  36877. variables relative to it.  The equivalent pseudocode would be something

  36878. like:

  36879. 

  36880.      int foo (void)

  36881.      {

  36882.        register int *xr = &x;

  36883.        return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];

  36884.      }

  36885. 

  36886.  (which isn't valid C). We refer to shared addresses like 'x' as

  36887. "section anchors".  Their use is controlled by '-fsection-anchors'.

  36888. 

  36889.  The hooks below describe the target properties that GCC needs to know

  36890. in order to make effective use of section anchors.  It won't use section

  36891. anchors at all unless either 'TARGET_MIN_ANCHOR_OFFSET' or

  36892. 'TARGET_MAX_ANCHOR_OFFSET' is set to a nonzero value.

  36893. 

  36894.  -- Target Hook: HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET

  36895.      The minimum offset that should be applied to a section anchor.  On

  36896.      most targets, it should be the smallest offset that can be applied

  36897.      to a base register while still giving a legitimate address for

  36898.      every mode.  The default value is 0.

  36899. 

  36900.  -- Target Hook: HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET

  36901.      Like 'TARGET_MIN_ANCHOR_OFFSET', but the maximum (inclusive) offset

  36902.      that should be applied to section anchors.  The default value is 0.

  36903. 

  36904.  -- Target Hook: void TARGET_ASM_OUTPUT_ANCHOR (rtx X)

  36905.      Write the assembly code to define section anchor X, which is a

  36906.      'SYMBOL_REF' for which 'SYMBOL_REF_ANCHOR_P (X)' is true.  The hook

  36907.      is called with the assembly output position set to the beginning of

  36908.      'SYMBOL_REF_BLOCK (X)'.

  36909. 

  36910.      If 'ASM_OUTPUT_DEF' is available, the hook's default definition

  36911.      uses it to define the symbol as '. + SYMBOL_REF_BLOCK_OFFSET (X)'.

  36912.      If 'ASM_OUTPUT_DEF' is not available, the hook's default definition

  36913.      is 'NULL', which disables the use of section anchors altogether.

  36914. 

  36915.  -- Target Hook: bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (const_rtx X)

  36916.      Return true if GCC should attempt to use anchors to access

  36917.      'SYMBOL_REF' X.  You can assume 'SYMBOL_REF_HAS_BLOCK_INFO_P (X)'

  36918.      and '!SYMBOL_REF_ANCHOR_P (X)'.

  36919. 

  36920.      The default version is correct for most targets, but you might need

  36921.      to intercept this hook to handle things like target-specific

  36922.      attributes or target-specific sections.

  36923. 

  36924. 

  36925. File: gccint.info,  Node: Condition Code,  Next: Costs,  Prev: Anchored Addresses,  Up: Target Macros

  36926. 

  36927. 18.15 Condition Code Status

  36928. ===========================

  36929. 

  36930. The macros in this section can be split in two families, according to

  36931. the two ways of representing condition codes in GCC.

  36932. 

  36933.  The first representation is the so called '(cc0)' representation (*note

  36934. Jump Patterns::), where all instructions can have an implicit clobber of

  36935. the condition codes.  The second is the condition code register

  36936. representation, which provides better schedulability for architectures

  36937. that do have a condition code register, but on which most instructions

  36938. do not affect it.  The latter category includes most RISC machines.

  36939. 

  36940.  The implicit clobbering poses a strong restriction on the placement of

  36941. the definition and use of the condition code.  In the past the

  36942. definition and use were always adjacent.  However, recent changes to

  36943. support trapping arithmatic may result in the definition and user being

  36944. in different blocks.  Thus, there may be a 'NOTE_INSN_BASIC_BLOCK'

  36945. between them.  Additionally, the definition may be the source of

  36946. exception handling edges.

  36947. 

  36948.  These restrictions can prevent important optimizations on some

  36949. machines.  For example, on the IBM RS/6000, there is a delay for taken

  36950. branches unless the condition code register is set three instructions

  36951. earlier than the conditional branch.  The instruction scheduler cannot

  36952. perform this optimization if it is not permitted to separate the

  36953. definition and use of the condition code register.

  36954. 

  36955.  For this reason, it is possible and suggested to use a register to

  36956. represent the condition code for new ports.  If there is a specific

  36957. condition code register in the machine, use a hard register.  If the

  36958. condition code or comparison result can be placed in any general

  36959. register, or if there are multiple condition registers, use a pseudo

  36960. register.  Registers used to store the condition code value will usually

  36961. have a mode that is in class 'MODE_CC'.

  36962. 

  36963.  Alternatively, you can use 'BImode' if the comparison operator is

  36964. specified already in the compare instruction.  In this case, you are not

  36965. interested in most macros in this section.

  36966. 

  36967. * Menu:

  36968. 

  36969. * CC0 Condition Codes::      Old style representation of condition codes.

  36970. * MODE_CC Condition Codes::  Modern representation of condition codes.

  36971. 

  36972. 

  36973. File: gccint.info,  Node: CC0 Condition Codes,  Next: MODE_CC Condition Codes,  Up: Condition Code

  36974. 

  36975. 18.15.1 Representation of condition codes using '(cc0)'

  36976. -------------------------------------------------------

  36977. 

  36978. The file 'conditions.h' defines a variable 'cc_status' to describe how

  36979. the condition code was computed (in case the interpretation of the

  36980. condition code depends on the instruction that it was set by).  This

  36981. variable contains the RTL expressions on which the condition code is

  36982. currently based, and several standard flags.

  36983. 

  36984.  Sometimes additional machine-specific flags must be defined in the

  36985. machine description header file.  It can also add additional

  36986. machine-specific information by defining 'CC_STATUS_MDEP'.

  36987. 

  36988.  -- Macro: CC_STATUS_MDEP

  36989.      C code for a data type which is used for declaring the 'mdep'

  36990.      component of 'cc_status'.  It defaults to 'int'.

  36991. 

  36992.      This macro is not used on machines that do not use 'cc0'.

  36993. 

  36994.  -- Macro: CC_STATUS_MDEP_INIT

  36995.      A C expression to initialize the 'mdep' field to "empty".  The

  36996.      default definition does nothing, since most machines don't use the

  36997.      field anyway.  If you want to use the field, you should probably

  36998.      define this macro to initialize it.

  36999. 

  37000.      This macro is not used on machines that do not use 'cc0'.

  37001. 

  37002.  -- Macro: NOTICE_UPDATE_CC (EXP, INSN)

  37003.      A C compound statement to set the components of 'cc_status'

  37004.      appropriately for an insn INSN whose body is EXP.  It is this

  37005.      macro's responsibility to recognize insns that set the condition

  37006.      code as a byproduct of other activity as well as those that

  37007.      explicitly set '(cc0)'.

  37008. 

  37009.      This macro is not used on machines that do not use 'cc0'.

  37010. 

  37011.      If there are insns that do not set the condition code but do alter

  37012.      other machine registers, this macro must check to see whether they

  37013.      invalidate the expressions that the condition code is recorded as

  37014.      reflecting.  For example, on the 68000, insns that store in address

  37015.      registers do not set the condition code, which means that usually

  37016.      'NOTICE_UPDATE_CC' can leave 'cc_status' unaltered for such insns.

  37017.      But suppose that the previous insn set the condition code based on

  37018.      location 'a4@(102)' and the current insn stores a new value in

  37019.      'a4'.  Although the condition code is not changed by this, it will

  37020.      no longer be true that it reflects the contents of 'a4@(102)'.

  37021.      Therefore, 'NOTICE_UPDATE_CC' must alter 'cc_status' in this case

  37022.      to say that nothing is known about the condition code value.

  37023. 

  37024.      The definition of 'NOTICE_UPDATE_CC' must be prepared to deal with

  37025.      the results of peephole optimization: insns whose patterns are

  37026.      'parallel' RTXs containing various 'reg', 'mem' or constants which

  37027.      are just the operands.  The RTL structure of these insns is not

  37028.      sufficient to indicate what the insns actually do.  What

  37029.      'NOTICE_UPDATE_CC' should do when it sees one is just to run

  37030.      'CC_STATUS_INIT'.

  37031. 

  37032.      A possible definition of 'NOTICE_UPDATE_CC' is to call a function

  37033.      that looks at an attribute (*note Insn Attributes::) named, for

  37034.      example, 'cc'.  This avoids having detailed information about

  37035.      patterns in two places, the 'md' file and in 'NOTICE_UPDATE_CC'.

  37036. 

  37037. 

  37038. File: gccint.info,  Node: MODE_CC Condition Codes,  Prev: CC0 Condition Codes,  Up: Condition Code

  37039. 

  37040. 18.15.2 Representation of condition codes using registers

  37041. ---------------------------------------------------------

  37042. 

  37043.  -- Macro: SELECT_CC_MODE (OP, X, Y)

  37044.      On many machines, the condition code may be produced by other

  37045.      instructions than compares, for example the branch can use directly

  37046.      the condition code set by a subtract instruction.  However, on some

  37047.      machines when the condition code is set this way some bits (such as

  37048.      the overflow bit) are not set in the same way as a test

  37049.      instruction, so that a different branch instruction must be used

  37050.      for some conditional branches.  When this happens, use the machine

  37051.      mode of the condition code register to record different formats of

  37052.      the condition code register.  Modes can also be used to record

  37053.      which compare instruction (e.g. a signed or an unsigned comparison)

  37054.      produced the condition codes.

  37055. 

  37056.      If other modes than 'CCmode' are required, add them to

  37057.      'MACHINE-modes.def' and define 'SELECT_CC_MODE' to choose a mode

  37058.      given an operand of a compare.  This is needed because the modes

  37059.      have to be chosen not only during RTL generation but also, for

  37060.      example, by instruction combination.  The result of

  37061.      'SELECT_CC_MODE' should be consistent with the mode used in the

  37062.      patterns; for example to support the case of the add on the SPARC

  37063.      discussed above, we have the pattern

  37064. 

  37065.           (define_insn ""

  37066.             [(set (reg:CCNZ 0)

  37067.                   (compare:CCNZ

  37068.                     (plus:SI (match_operand:SI 0 "register_operand" "%r")

  37069.                              (match_operand:SI 1 "arith_operand" "rI"))

  37070.                     (const_int 0)))]

  37071.             ""

  37072.             "...")

  37073. 

  37074.      together with a 'SELECT_CC_MODE' that returns 'CCNZmode' for

  37075.      comparisons whose argument is a 'plus':

  37076. 

  37077.           #define SELECT_CC_MODE(OP,X,Y) \

  37078.             (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT           \

  37079.              ? ((OP == LT || OP == LE || OP == GT || OP == GE)     \

  37080.                 ? CCFPEmode : CCFPmode)                            \

  37081.              : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS     \

  37082.                  || GET_CODE (X) == NEG || GET_CODE (x) == ASHIFT) \

  37083.                 ? CCNZmode : CCmode))

  37084. 

  37085.      Another reason to use modes is to retain information on which

  37086.      operands were used by the comparison; see 'REVERSIBLE_CC_MODE'

  37087.      later in this section.

  37088. 

  37089.      You should define this macro if and only if you define extra CC

  37090.      modes in 'MACHINE-modes.def'.

  37091. 

  37092.  -- Target Hook: void TARGET_CANONICALIZE_COMPARISON (int *CODE, rtx

  37093.           *OP0, rtx *OP1, bool OP0_PRESERVE_VALUE)

  37094.      On some machines not all possible comparisons are defined, but you

  37095.      can convert an invalid comparison into a valid one.  For example,

  37096.      the Alpha does not have a 'GT' comparison, but you can use an 'LT'

  37097.      comparison instead and swap the order of the operands.

  37098. 

  37099.      On such machines, implement this hook to do any required

  37100.      conversions.  CODE is the initial comparison code and OP0 and OP1

  37101.      are the left and right operands of the comparison, respectively.

  37102.      If OP0_PRESERVE_VALUE is 'true' the implementation is not allowed

  37103.      to change the value of OP0 since the value might be used in RTXs

  37104.      which aren't comparisons.  E.g.  the implementation is not allowed

  37105.      to swap operands in that case.

  37106. 

  37107.      GCC will not assume that the comparison resulting from this macro

  37108.      is valid but will see if the resulting insn matches a pattern in

  37109.      the 'md' file.

  37110. 

  37111.      You need not to implement this hook if it would never change the

  37112.      comparison code or operands.

  37113. 

  37114.  -- Macro: REVERSIBLE_CC_MODE (MODE)

  37115.      A C expression whose value is one if it is always safe to reverse a

  37116.      comparison whose mode is MODE.  If 'SELECT_CC_MODE' can ever return

  37117.      MODE for a floating-point inequality comparison, then

  37118.      'REVERSIBLE_CC_MODE (MODE)' must be zero.

  37119. 

  37120.      You need not define this macro if it would always returns zero or

  37121.      if the floating-point format is anything other than

  37122.      'IEEE_FLOAT_FORMAT'.  For example, here is the definition used on

  37123.      the SPARC, where floating-point inequality comparisons are given

  37124.      either 'CCFPEmode' or 'CCFPmode':

  37125. 

  37126.           #define REVERSIBLE_CC_MODE(MODE) \

  37127.              ((MODE) != CCFPEmode && (MODE) != CCFPmode)

  37128. 

  37129.  -- Macro: REVERSE_CONDITION (CODE, MODE)

  37130.      A C expression whose value is reversed condition code of the CODE

  37131.      for comparison done in CC_MODE MODE.  The macro is used only in

  37132.      case 'REVERSIBLE_CC_MODE (MODE)' is nonzero.  Define this macro in

  37133.      case machine has some non-standard way how to reverse certain

  37134.      conditionals.  For instance in case all floating point conditions

  37135.      are non-trapping, compiler may freely convert unordered compares to

  37136.      ordered ones.  Then definition may look like:

  37137. 

  37138.           #define REVERSE_CONDITION(CODE, MODE) \

  37139.              ((MODE) != CCFPmode ? reverse_condition (CODE) \

  37140.               : reverse_condition_maybe_unordered (CODE))

  37141. 

  37142.  -- Target Hook: bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int

  37143.           *P1, unsigned int *P2)

  37144.      On targets which do not use '(cc0)', and which use a hard register

  37145.      rather than a pseudo-register to hold condition codes, the regular

  37146.      CSE passes are often not able to identify cases in which the hard

  37147.      register is set to a common value.  Use this hook to enable a small

  37148.      pass which optimizes such cases.  This hook should return true to

  37149.      enable this pass, and it should set the integers to which its

  37150.      arguments point to the hard register numbers used for condition

  37151.      codes.  When there is only one such register, as is true on most

  37152.      systems, the integer pointed to by P2 should be set to

  37153.      'INVALID_REGNUM'.

  37154. 

  37155.      The default version of this hook returns false.

  37156. 

  37157.  -- Target Hook: machine_mode TARGET_CC_MODES_COMPATIBLE (machine_mode

  37158.           M1, machine_mode M2)

  37159.      On targets which use multiple condition code modes in class

  37160.      'MODE_CC', it is sometimes the case that a comparison can be

  37161.      validly done in more than one mode.  On such a system, define this

  37162.      target hook to take two mode arguments and to return a mode in

  37163.      which both comparisons may be validly done.  If there is no such

  37164.      mode, return 'VOIDmode'.

  37165. 

  37166.      The default version of this hook checks whether the modes are the

  37167.      same.  If they are, it returns that mode.  If they are different,

  37168.      it returns 'VOIDmode'.

  37169. 

  37170.  -- Target Hook: unsigned int TARGET_FLAGS_REGNUM

  37171.      If the target has a dedicated flags register, and it needs to use

  37172.      the post-reload comparison elimination pass, or the delay slot

  37173.      filler pass, then this value should be set appropriately.

  37174. 

  37175. 

  37176. File: gccint.info,  Node: Costs,  Next: Scheduling,  Prev: Condition Code,  Up: Target Macros

  37177. 

  37178. 18.16 Describing Relative Costs of Operations

  37179. =============================================

  37180. 

  37181. These macros let you describe the relative speed of various operations

  37182. on the target machine.

  37183. 

  37184.  -- Macro: REGISTER_MOVE_COST (MODE, FROM, TO)

  37185.      A C expression for the cost of moving data of mode MODE from a

  37186.      register in class FROM to one in class TO.  The classes are

  37187.      expressed using the enumeration values such as 'GENERAL_REGS'.  A

  37188.      value of 2 is the default; other values are interpreted relative to

  37189.      that.

  37190. 

  37191.      It is not required that the cost always equal 2 when FROM is the

  37192.      same as TO; on some machines it is expensive to move between

  37193.      registers if they are not general registers.

  37194. 

  37195.      If reload sees an insn consisting of a single 'set' between two

  37196.      hard registers, and if 'REGISTER_MOVE_COST' applied to their

  37197.      classes returns a value of 2, reload does not check to ensure that

  37198.      the constraints of the insn are met.  Setting a cost of other than

  37199.      2 will allow reload to verify that the constraints are met.  You

  37200.      should do this if the 'movM' pattern's constraints do not allow

  37201.      such copying.

  37202. 

  37203.      These macros are obsolete, new ports should use the target hook

  37204.      'TARGET_REGISTER_MOVE_COST' instead.

  37205. 

  37206.  -- Target Hook: int TARGET_REGISTER_MOVE_COST (machine_mode MODE,

  37207.           reg_class_t FROM, reg_class_t TO)

  37208.      This target hook should return the cost of moving data of mode MODE

  37209.      from a register in class FROM to one in class TO.  The classes are

  37210.      expressed using the enumeration values such as 'GENERAL_REGS'.  A

  37211.      value of 2 is the default; other values are interpreted relative to

  37212.      that.

  37213. 

  37214.      It is not required that the cost always equal 2 when FROM is the

  37215.      same as TO; on some machines it is expensive to move between

  37216.      registers if they are not general registers.

  37217. 

  37218.      If reload sees an insn consisting of a single 'set' between two

  37219.      hard registers, and if 'TARGET_REGISTER_MOVE_COST' applied to their

  37220.      classes returns a value of 2, reload does not check to ensure that

  37221.      the constraints of the insn are met.  Setting a cost of other than

  37222.      2 will allow reload to verify that the constraints are met.  You

  37223.      should do this if the 'movM' pattern's constraints do not allow

  37224.      such copying.

  37225. 

  37226.      The default version of this function returns 2.

  37227. 

  37228.  -- Macro: MEMORY_MOVE_COST (MODE, CLASS, IN)

  37229.      A C expression for the cost of moving data of mode MODE between a

  37230.      register of class CLASS and memory; IN is zero if the value is to

  37231.      be written to memory, nonzero if it is to be read in.  This cost is

  37232.      relative to those in 'REGISTER_MOVE_COST'.  If moving between

  37233.      registers and memory is more expensive than between two registers,

  37234.      you should define this macro to express the relative cost.

  37235. 

  37236.      If you do not define this macro, GCC uses a default cost of 4 plus

  37237.      the cost of copying via a secondary reload register, if one is

  37238.      needed.  If your machine requires a secondary reload register to

  37239.      copy between memory and a register of CLASS but the reload

  37240.      mechanism is more complex than copying via an intermediate, define

  37241.      this macro to reflect the actual cost of the move.

  37242. 

  37243.      GCC defines the function 'memory_move_secondary_cost' if secondary

  37244.      reloads are needed.  It computes the costs due to copying via a

  37245.      secondary register.  If your machine copies from memory using a

  37246.      secondary register in the conventional way but the default base

  37247.      value of 4 is not correct for your machine, define this macro to

  37248.      add some other value to the result of that function.  The arguments

  37249.      to that function are the same as to this macro.

  37250. 

  37251.      These macros are obsolete, new ports should use the target hook

  37252.      'TARGET_MEMORY_MOVE_COST' instead.

  37253. 

  37254.  -- Target Hook: int TARGET_MEMORY_MOVE_COST (machine_mode MODE,

  37255.           reg_class_t RCLASS, bool IN)

  37256.      This target hook should return the cost of moving data of mode MODE

  37257.      between a register of class RCLASS and memory; IN is 'false' if the

  37258.      value is to be written to memory, 'true' if it is to be read in.

  37259.      This cost is relative to those in 'TARGET_REGISTER_MOVE_COST'.  If

  37260.      moving between registers and memory is more expensive than between

  37261.      two registers, you should add this target hook to express the

  37262.      relative cost.

  37263. 

  37264.      If you do not add this target hook, GCC uses a default cost of 4

  37265.      plus the cost of copying via a secondary reload register, if one is

  37266.      needed.  If your machine requires a secondary reload register to

  37267.      copy between memory and a register of RCLASS but the reload

  37268.      mechanism is more complex than copying via an intermediate, use

  37269.      this target hook to reflect the actual cost of the move.

  37270. 

  37271.      GCC defines the function 'memory_move_secondary_cost' if secondary

  37272.      reloads are needed.  It computes the costs due to copying via a

  37273.      secondary register.  If your machine copies from memory using a

  37274.      secondary register in the conventional way but the default base

  37275.      value of 4 is not correct for your machine, use this target hook to

  37276.      add some other value to the result of that function.  The arguments

  37277.      to that function are the same as to this target hook.

  37278. 

  37279.  -- Macro: BRANCH_COST (SPEED_P, PREDICTABLE_P)

  37280.      A C expression for the cost of a branch instruction.  A value of 1

  37281.      is the default; other values are interpreted relative to that.

  37282.      Parameter SPEED_P is true when the branch in question should be

  37283.      optimized for speed.  When it is false, 'BRANCH_COST' should return

  37284.      a value optimal for code size rather than performance.

  37285.      PREDICTABLE_P is true for well-predicted branches.  On many

  37286.      architectures the 'BRANCH_COST' can be reduced then.

  37287. 

  37288.  Here are additional macros which do not specify precise relative costs,

  37289. but only that certain actions are more expensive than GCC would

  37290. ordinarily expect.

  37291. 

  37292.  -- Macro: SLOW_BYTE_ACCESS

  37293.      Define this macro as a C expression which is nonzero if accessing

  37294.      less than a word of memory (i.e. a 'char' or a 'short') is no

  37295.      faster than accessing a word of memory, i.e., if such access

  37296.      require more than one instruction or if there is no difference in

  37297.      cost between byte and (aligned) word loads.

  37298. 

  37299.      When this macro is not defined, the compiler will access a field by

  37300.      finding the smallest containing object; when it is defined, a

  37301.      fullword load will be used if alignment permits.  Unless bytes

  37302.      accesses are faster than word accesses, using word accesses is

  37303.      preferable since it may eliminate subsequent memory access if

  37304.      subsequent accesses occur to other fields in the same word of the

  37305.      structure, but to different bytes.

  37306. 

  37307.  -- Target Hook: bool TARGET_SLOW_UNALIGNED_ACCESS (machine_mode MODE,

  37308.           unsigned int ALIGN)

  37309.      This hook returns true if memory accesses described by the MODE and

  37310.      ALIGNMENT parameters have a cost many times greater than aligned

  37311.      accesses, for example if they are emulated in a trap handler.  This

  37312.      hook is invoked only for unaligned accesses, i.e. when 'ALIGNMENT <

  37313.      GET_MODE_ALIGNMENT (MODE)'.

  37314. 

  37315.      When this hook returns true, the compiler will act as if

  37316.      'STRICT_ALIGNMENT' were true when generating code for block moves.

  37317.      This can cause significantly more instructions to be produced.

  37318.      Therefore, do not make this hook return true if unaligned accesses

  37319.      only add a cycle or two to the time for a memory access.

  37320. 

  37321.      The hook must return true whenever 'STRICT_ALIGNMENT' is true.  The

  37322.      default implementation returns 'STRICT_ALIGNMENT'.

  37323. 

  37324.  -- Macro: MOVE_RATIO (SPEED)

  37325.      The threshold of number of scalar memory-to-memory move insns,

  37326.      _below_ which a sequence of insns should be generated instead of a

  37327.      string move insn or a library call.  Increasing the value will

  37328.      always make code faster, but eventually incurs high cost in

  37329.      increased code size.

  37330. 

  37331.      Note that on machines where the corresponding move insn is a

  37332.      'define_expand' that emits a sequence of insns, this macro counts

  37333.      the number of such sequences.

  37334. 

  37335.      The parameter SPEED is true if the code is currently being

  37336.      optimized for speed rather than size.

  37337. 

  37338.      If you don't define this, a reasonable default is used.

  37339. 

  37340.  -- Target Hook: bool TARGET_USE_BY_PIECES_INFRASTRUCTURE_P (unsigned

  37341.           HOST_WIDE_INT SIZE, unsigned int ALIGNMENT, enum

  37342.           by_pieces_operation OP, bool SPEED_P)

  37343.      GCC will attempt several strategies when asked to copy between two

  37344.      areas of memory, or to set, clear or store to memory, for example

  37345.      when copying a 'struct'.  The 'by_pieces' infrastructure implements

  37346.      such memory operations as a sequence of load, store or move insns.

  37347.      Alternate strategies are to expand the 'cpymem' or 'setmem' optabs,

  37348.      to emit a library call, or to emit unit-by-unit, loop-based

  37349.      operations.

  37350. 

  37351.      This target hook should return true if, for a memory operation with

  37352.      a given SIZE and ALIGNMENT, using the 'by_pieces' infrastructure is

  37353.      expected to result in better code generation.  Both SIZE and

  37354.      ALIGNMENT are measured in terms of storage units.

  37355. 

  37356.      The parameter OP is one of: 'CLEAR_BY_PIECES', 'MOVE_BY_PIECES',

  37357.      'SET_BY_PIECES', 'STORE_BY_PIECES' or 'COMPARE_BY_PIECES'.  These

  37358.      describe the type of memory operation under consideration.

  37359. 

  37360.      The parameter SPEED_P is true if the code is currently being

  37361.      optimized for speed rather than size.

  37362. 

  37363.      Returning true for higher values of SIZE can improve code

  37364.      generation for speed if the target does not provide an

  37365.      implementation of the 'cpymem' or 'setmem' standard names, if the

  37366.      'cpymem' or 'setmem' implementation would be more expensive than a

  37367.      sequence of insns, or if the overhead of a library call would

  37368.      dominate that of the body of the memory operation.

  37369. 

  37370.      Returning true for higher values of 'size' may also cause an

  37371.      increase in code size, for example where the number of insns

  37372.      emitted to perform a move would be greater than that of a library

  37373.      call.

  37374. 

  37375.  -- Target Hook: int TARGET_COMPARE_BY_PIECES_BRANCH_RATIO (machine_mode

  37376.           MODE)

  37377.      When expanding a block comparison in MODE, gcc can try to reduce

  37378.      the number of branches at the expense of more memory operations.

  37379.      This hook allows the target to override the default choice.  It

  37380.      should return the factor by which branches should be reduced over

  37381.      the plain expansion with one comparison per MODE-sized piece.  A

  37382.      port can also prevent a particular mode from being used for block

  37383.      comparisons by returning a negative number from this hook.

  37384. 

  37385.  -- Macro: MOVE_MAX_PIECES

  37386.      A C expression used by 'move_by_pieces' to determine the largest

  37387.      unit a load or store used to copy memory is.  Defaults to

  37388.      'MOVE_MAX'.

  37389. 

  37390.  -- Macro: STORE_MAX_PIECES

  37391.      A C expression used by 'store_by_pieces' to determine the largest

  37392.      unit a store used to memory is.  Defaults to 'MOVE_MAX_PIECES', or

  37393.      two times the size of 'HOST_WIDE_INT', whichever is smaller.

  37394. 

  37395.  -- Macro: COMPARE_MAX_PIECES

  37396.      A C expression used by 'compare_by_pieces' to determine the largest

  37397.      unit a load or store used to compare memory is.  Defaults to

  37398.      'MOVE_MAX_PIECES'.

  37399. 

  37400.  -- Macro: CLEAR_RATIO (SPEED)

  37401.      The threshold of number of scalar move insns, _below_ which a

  37402.      sequence of insns should be generated to clear memory instead of a

  37403.      string clear insn or a library call.  Increasing the value will

  37404.      always make code faster, but eventually incurs high cost in

  37405.      increased code size.

  37406. 

  37407.      The parameter SPEED is true if the code is currently being

  37408.      optimized for speed rather than size.

  37409. 

  37410.      If you don't define this, a reasonable default is used.

  37411. 

  37412.  -- Macro: SET_RATIO (SPEED)

  37413.      The threshold of number of scalar move insns, _below_ which a

  37414.      sequence of insns should be generated to set memory to a constant

  37415.      value, instead of a block set insn or a library call.  Increasing

  37416.      the value will always make code faster, but eventually incurs high

  37417.      cost in increased code size.

  37418. 

  37419.      The parameter SPEED is true if the code is currently being

  37420.      optimized for speed rather than size.

  37421. 

  37422.      If you don't define this, it defaults to the value of 'MOVE_RATIO'.

  37423. 

  37424.  -- Macro: USE_LOAD_POST_INCREMENT (MODE)

  37425.      A C expression used to determine whether a load postincrement is a

  37426.      good thing to use for a given mode.  Defaults to the value of

  37427.      'HAVE_POST_INCREMENT'.

  37428. 

  37429.  -- Macro: USE_LOAD_POST_DECREMENT (MODE)

  37430.      A C expression used to determine whether a load postdecrement is a

  37431.      good thing to use for a given mode.  Defaults to the value of

  37432.      'HAVE_POST_DECREMENT'.

  37433. 

  37434.  -- Macro: USE_LOAD_PRE_INCREMENT (MODE)

  37435.      A C expression used to determine whether a load preincrement is a

  37436.      good thing to use for a given mode.  Defaults to the value of

  37437.      'HAVE_PRE_INCREMENT'.

  37438. 

  37439.  -- Macro: USE_LOAD_PRE_DECREMENT (MODE)

  37440.      A C expression used to determine whether a load predecrement is a

  37441.      good thing to use for a given mode.  Defaults to the value of

  37442.      'HAVE_PRE_DECREMENT'.

  37443. 

  37444.  -- Macro: USE_STORE_POST_INCREMENT (MODE)

  37445.      A C expression used to determine whether a store postincrement is a

  37446.      good thing to use for a given mode.  Defaults to the value of

  37447.      'HAVE_POST_INCREMENT'.

  37448. 

  37449.  -- Macro: USE_STORE_POST_DECREMENT (MODE)

  37450.      A C expression used to determine whether a store postdecrement is a

  37451.      good thing to use for a given mode.  Defaults to the value of

  37452.      'HAVE_POST_DECREMENT'.

  37453. 

  37454.  -- Macro: USE_STORE_PRE_INCREMENT (MODE)

  37455.      This macro is used to determine whether a store preincrement is a

  37456.      good thing to use for a given mode.  Defaults to the value of

  37457.      'HAVE_PRE_INCREMENT'.

  37458. 

  37459.  -- Macro: USE_STORE_PRE_DECREMENT (MODE)

  37460.      This macro is used to determine whether a store predecrement is a

  37461.      good thing to use for a given mode.  Defaults to the value of

  37462.      'HAVE_PRE_DECREMENT'.

  37463. 

  37464.  -- Macro: NO_FUNCTION_CSE

  37465.      Define this macro to be true if it is as good or better to call a

  37466.      constant function address than to call an address kept in a

  37467.      register.

  37468. 

  37469.  -- Macro: LOGICAL_OP_NON_SHORT_CIRCUIT

  37470.      Define this macro if a non-short-circuit operation produced by

  37471.      'fold_range_test ()' is optimal.  This macro defaults to true if

  37472.      'BRANCH_COST' is greater than or equal to the value 2.

  37473. 

  37474.  -- Target Hook: bool TARGET_OPTAB_SUPPORTED_P (int OP, machine_mode

  37475.           MODE1, machine_mode MODE2, optimization_type OPT_TYPE)

  37476.      Return true if the optimizers should use optab OP with modes MODE1

  37477.      and MODE2 for optimization type OPT_TYPE.  The optab is known to

  37478.      have an associated '.md' instruction whose C condition is true.

  37479.      MODE2 is only meaningful for conversion optabs; for direct optabs

  37480.      it is a copy of MODE1.

  37481. 

  37482.      For example, when called with OP equal to 'rint_optab' and MODE1

  37483.      equal to 'DFmode', the hook should say whether the optimizers

  37484.      should use optab 'rintdf2'.

  37485. 

  37486.      The default hook returns true for all inputs.

  37487. 

  37488.  -- Target Hook: bool TARGET_RTX_COSTS (rtx X, machine_mode MODE, int

  37489.           OUTER_CODE, int OPNO, int *TOTAL, bool SPEED)

  37490.      This target hook describes the relative costs of RTL expressions.

  37491. 

  37492.      The cost may depend on the precise form of the expression, which is

  37493.      available for examination in X, and the fact that X appears as

  37494.      operand OPNO of an expression with rtx code OUTER_CODE.  That is,

  37495.      the hook can assume that there is some rtx Y such that 'GET_CODE

  37496.      (Y) == OUTER_CODE' and such that either (a) 'XEXP (Y, OPNO) == X'

  37497.      or (b) 'XVEC (Y, OPNO)' contains X.

  37498. 

  37499.      MODE is X's machine mode, or for cases like 'const_int' that do not

  37500.      have a mode, the mode in which X is used.

  37501. 

  37502.      In implementing this hook, you can use the construct 'COSTS_N_INSNS

  37503.      (N)' to specify a cost equal to N fast instructions.

  37504. 

  37505.      On entry to the hook, '*TOTAL' contains a default estimate for the

  37506.      cost of the expression.  The hook should modify this value as

  37507.      necessary.  Traditionally, the default costs are 'COSTS_N_INSNS

  37508.      (5)' for multiplications, 'COSTS_N_INSNS (7)' for division and

  37509.      modulus operations, and 'COSTS_N_INSNS (1)' for all other

  37510.      operations.

  37511. 

  37512.      When optimizing for code size, i.e. when 'speed' is false, this

  37513.      target hook should be used to estimate the relative size cost of an

  37514.      expression, again relative to 'COSTS_N_INSNS'.

  37515. 

  37516.      The hook returns true when all subexpressions of X have been

  37517.      processed, and false when 'rtx_cost' should recurse.

  37518. 

  37519.  -- Target Hook: int TARGET_ADDRESS_COST (rtx ADDRESS, machine_mode

  37520.           MODE, addr_space_t AS, bool SPEED)

  37521.      This hook computes the cost of an addressing mode that contains

  37522.      ADDRESS.  If not defined, the cost is computed from the ADDRESS

  37523.      expression and the 'TARGET_RTX_COST' hook.

  37524. 

  37525.      For most CISC machines, the default cost is a good approximation of

  37526.      the true cost of the addressing mode.  However, on RISC machines,

  37527.      all instructions normally have the same length and execution time.

  37528.      Hence all addresses will have equal costs.

  37529. 

  37530.      In cases where more than one form of an address is known, the form

  37531.      with the lowest cost will be used.  If multiple forms have the

  37532.      same, lowest, cost, the one that is the most complex will be used.

  37533. 

  37534.      For example, suppose an address that is equal to the sum of a

  37535.      register and a constant is used twice in the same basic block.

  37536.      When this macro is not defined, the address will be computed in a

  37537.      register and memory references will be indirect through that

  37538.      register.  On machines where the cost of the addressing mode

  37539.      containing the sum is no higher than that of a simple indirect

  37540.      reference, this will produce an additional instruction and possibly

  37541.      require an additional register.  Proper specification of this macro

  37542.      eliminates this overhead for such machines.

  37543. 

  37544.      This hook is never called with an invalid address.

  37545. 

  37546.      On machines where an address involving more than one register is as

  37547.      cheap as an address computation involving only one register,

  37548.      defining 'TARGET_ADDRESS_COST' to reflect this can cause two

  37549.      registers to be live over a region of code where only one would

  37550.      have been if 'TARGET_ADDRESS_COST' were not defined in that manner.

  37551.      This effect should be considered in the definition of this macro.

  37552.      Equivalent costs should probably only be given to addresses with

  37553.      different numbers of registers on machines with lots of registers.

  37554. 

  37555.  -- Target Hook: int TARGET_INSN_COST (rtx_insn *INSN, bool SPEED)

  37556.      This target hook describes the relative costs of RTL instructions.

  37557. 

  37558.      In implementing this hook, you can use the construct 'COSTS_N_INSNS

  37559.      (N)' to specify a cost equal to N fast instructions.

  37560. 

  37561.      When optimizing for code size, i.e. when 'speed' is false, this

  37562.      target hook should be used to estimate the relative size cost of an

  37563.      expression, again relative to 'COSTS_N_INSNS'.

  37564. 

  37565.  -- Target Hook: unsigned int TARGET_MAX_NOCE_IFCVT_SEQ_COST (edge E)

  37566.      This hook returns a value in the same units as 'TARGET_RTX_COSTS',

  37567.      giving the maximum acceptable cost for a sequence generated by the

  37568.      RTL if-conversion pass when conditional execution is not available.

  37569.      The RTL if-conversion pass attempts to convert conditional

  37570.      operations that would require a branch to a series of unconditional

  37571.      operations and 'movMODEcc' insns.  This hook returns the maximum

  37572.      cost of the unconditional instructions and the 'movMODEcc' insns.

  37573.      RTL if-conversion is cancelled if the cost of the converted

  37574.      sequence is greater than the value returned by this hook.

  37575. 

  37576.      'e' is the edge between the basic block containing the conditional

  37577.      branch to the basic block which would be executed if the condition

  37578.      were true.

  37579. 

  37580.      The default implementation of this hook uses the

  37581.      'max-rtl-if-conversion-[un]predictable' parameters if they are set,

  37582.      and uses a multiple of 'BRANCH_COST' otherwise.

  37583. 

  37584.  -- Target Hook: bool TARGET_NOCE_CONVERSION_PROFITABLE_P (rtx_insn

  37585.           *SEQ, struct noce_if_info *IF_INFO)

  37586.      This hook returns true if the instruction sequence 'seq' is a good

  37587.      candidate as a replacement for the if-convertible sequence

  37588.      described in 'if_info'.

  37589. 

  37590.  -- Target Hook: bool TARGET_NO_SPECULATION_IN_DELAY_SLOTS_P (void)

  37591.      This predicate controls the use of the eager delay slot filler to

  37592.      disallow speculatively executed instructions being placed in delay

  37593.      slots.  Targets such as certain MIPS architectures possess both

  37594.      branches with and without delay slots.  As the eager delay slot

  37595.      filler can decrease performance, disabling it is beneficial when

  37596.      ordinary branches are available.  Use of delay slot branches filled

  37597.      using the basic filler is often still desirable as the delay slot

  37598.      can hide a pipeline bubble.

  37599. 

  37600.  -- Target Hook: HOST_WIDE_INT TARGET_ESTIMATED_POLY_VALUE (poly_int64

  37601.           VAL)

  37602.      Return an estimate of the runtime value of VAL, for use in things

  37603.      like cost calculations or profiling frequencies.  The default

  37604.      implementation returns the lowest possible value of VAL.

  37605. 

  37606. 

  37607. File: gccint.info,  Node: Scheduling,  Next: Sections,  Prev: Costs,  Up: Target Macros

  37608. 

  37609. 18.17 Adjusting the Instruction Scheduler

  37610. =========================================

  37611. 

  37612. The instruction scheduler may need a fair amount of machine-specific

  37613. adjustment in order to produce good code.  GCC provides several target

  37614. hooks for this purpose.  It is usually enough to define just a few of

  37615. them: try the first ones in this list first.

  37616. 

  37617.  -- Target Hook: int TARGET_SCHED_ISSUE_RATE (void)

  37618.      This hook returns the maximum number of instructions that can ever

  37619.      issue at the same time on the target machine.  The default is one.

  37620.      Although the insn scheduler can define itself the possibility of

  37621.      issue an insn on the same cycle, the value can serve as an

  37622.      additional constraint to issue insns on the same simulated

  37623.      processor cycle (see hooks 'TARGET_SCHED_REORDER' and

  37624.      'TARGET_SCHED_REORDER2').  This value must be constant over the

  37625.      entire compilation.  If you need it to vary depending on what the

  37626.      instructions are, you must use 'TARGET_SCHED_VARIABLE_ISSUE'.

  37627. 

  37628.  -- Target Hook: int TARGET_SCHED_VARIABLE_ISSUE (FILE *FILE, int

  37629.           VERBOSE, rtx_insn *INSN, int MORE)

  37630.      This hook is executed by the scheduler after it has scheduled an

  37631.      insn from the ready list.  It should return the number of insns

  37632.      which can still be issued in the current cycle.  The default is

  37633.      'MORE - 1' for insns other than 'CLOBBER' and 'USE', which normally

  37634.      are not counted against the issue rate.  You should define this

  37635.      hook if some insns take more machine resources than others, so that

  37636.      fewer insns can follow them in the same cycle.  FILE is either a

  37637.      null pointer, or a stdio stream to write any debug output to.

  37638.      VERBOSE is the verbose level provided by '-fsched-verbose-N'.  INSN

  37639.      is the instruction that was scheduled.

  37640. 

  37641.  -- Target Hook: int TARGET_SCHED_ADJUST_COST (rtx_insn *INSN, int

  37642.           DEP_TYPE1, rtx_insn *DEP_INSN, int COST, unsigned int DW)

  37643.      This function corrects the value of COST based on the relationship

  37644.      between INSN and DEP_INSN through a dependence of type dep_type,

  37645.      and strength DW.  It should return the new value.  The default is

  37646.      to make no adjustment to COST.  This can be used for example to

  37647.      specify to the scheduler using the traditional pipeline description

  37648.      that an output- or anti-dependence does not incur the same cost as

  37649.      a data-dependence.  If the scheduler using the automaton based

  37650.      pipeline description, the cost of anti-dependence is zero and the

  37651.      cost of output-dependence is maximum of one and the difference of

  37652.      latency times of the first and the second insns.  If these values

  37653.      are not acceptable, you could use the hook to modify them too.  See

  37654.      also *note Processor pipeline description::.

  37655. 

  37656.  -- Target Hook: int TARGET_SCHED_ADJUST_PRIORITY (rtx_insn *INSN, int

  37657.           PRIORITY)

  37658.      This hook adjusts the integer scheduling priority PRIORITY of INSN.

  37659.      It should return the new priority.  Increase the priority to

  37660.      execute INSN earlier, reduce the priority to execute INSN later.

  37661.      Do not define this hook if you do not need to adjust the scheduling

  37662.      priorities of insns.

  37663. 

  37664.  -- Target Hook: int TARGET_SCHED_REORDER (FILE *FILE, int VERBOSE,

  37665.           rtx_insn **READY, int *N_READYP, int CLOCK)

  37666.      This hook is executed by the scheduler after it has scheduled the

  37667.      ready list, to allow the machine description to reorder it (for

  37668.      example to combine two small instructions together on 'VLIW'

  37669.      machines).  FILE is either a null pointer, or a stdio stream to

  37670.      write any debug output to.  VERBOSE is the verbose level provided

  37671.      by '-fsched-verbose-N'.  READY is a pointer to the ready list of

  37672.      instructions that are ready to be scheduled.  N_READYP is a pointer

  37673.      to the number of elements in the ready list.  The scheduler reads

  37674.      the ready list in reverse order, starting with READY[*N_READYP - 1]

  37675.      and going to READY[0].  CLOCK is the timer tick of the scheduler.

  37676.      You may modify the ready list and the number of ready insns.  The

  37677.      return value is the number of insns that can issue this cycle;

  37678.      normally this is just 'issue_rate'.  See also

  37679.      'TARGET_SCHED_REORDER2'.

  37680. 

  37681.  -- Target Hook: int TARGET_SCHED_REORDER2 (FILE *FILE, int VERBOSE,

  37682.           rtx_insn **READY, int *N_READYP, int CLOCK)

  37683.      Like 'TARGET_SCHED_REORDER', but called at a different time.  That

  37684.      function is called whenever the scheduler starts a new cycle.  This

  37685.      one is called once per iteration over a cycle, immediately after

  37686.      'TARGET_SCHED_VARIABLE_ISSUE'; it can reorder the ready list and

  37687.      return the number of insns to be scheduled in the same cycle.

  37688.      Defining this hook can be useful if there are frequent situations

  37689.      where scheduling one insn causes other insns to become ready in the

  37690.      same cycle.  These other insns can then be taken into account

  37691.      properly.

  37692. 

  37693.  -- Target Hook: bool TARGET_SCHED_MACRO_FUSION_P (void)

  37694.      This hook is used to check whether target platform supports macro

  37695.      fusion.

  37696. 

  37697.  -- Target Hook: bool TARGET_SCHED_MACRO_FUSION_PAIR_P (rtx_insn *PREV,

  37698.           rtx_insn *CURR)

  37699.      This hook is used to check whether two insns should be macro fused

  37700.      for a target microarchitecture.  If this hook returns true for the

  37701.      given insn pair (PREV and CURR), the scheduler will put them into a

  37702.      sched group, and they will not be scheduled apart.  The two insns

  37703.      will be either two SET insns or a compare and a conditional jump

  37704.      and this hook should validate any dependencies needed to fuse the

  37705.      two insns together.

  37706. 

  37707.  -- Target Hook: void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK

  37708.           (rtx_insn *HEAD, rtx_insn *TAIL)

  37709.      This hook is called after evaluation forward dependencies of insns

  37710.      in chain given by two parameter values (HEAD and TAIL

  37711.      correspondingly) but before insns scheduling of the insn chain.

  37712.      For example, it can be used for better insn classification if it

  37713.      requires analysis of dependencies.  This hook can use backward and

  37714.      forward dependencies of the insn scheduler because they are already

  37715.      calculated.

  37716. 

  37717.  -- Target Hook: void TARGET_SCHED_INIT (FILE *FILE, int VERBOSE, int

  37718.           MAX_READY)

  37719.      This hook is executed by the scheduler at the beginning of each

  37720.      block of instructions that are to be scheduled.  FILE is either a

  37721.      null pointer, or a stdio stream to write any debug output to.

  37722.      VERBOSE is the verbose level provided by '-fsched-verbose-N'.

  37723.      MAX_READY is the maximum number of insns in the current scheduling

  37724.      region that can be live at the same time.  This can be used to

  37725.      allocate scratch space if it is needed, e.g. by

  37726.      'TARGET_SCHED_REORDER'.

  37727. 

  37728.  -- Target Hook: void TARGET_SCHED_FINISH (FILE *FILE, int VERBOSE)

  37729.      This hook is executed by the scheduler at the end of each block of

  37730.      instructions that are to be scheduled.  It can be used to perform

  37731.      cleanup of any actions done by the other scheduling hooks.  FILE is

  37732.      either a null pointer, or a stdio stream to write any debug output

  37733.      to.  VERBOSE is the verbose level provided by '-fsched-verbose-N'.

  37734. 

  37735.  -- Target Hook: void TARGET_SCHED_INIT_GLOBAL (FILE *FILE, int VERBOSE,

  37736.           int OLD_MAX_UID)

  37737.      This hook is executed by the scheduler after function level

  37738.      initializations.  FILE is either a null pointer, or a stdio stream

  37739.      to write any debug output to.  VERBOSE is the verbose level

  37740.      provided by '-fsched-verbose-N'.  OLD_MAX_UID is the maximum insn

  37741.      uid when scheduling begins.

  37742. 

  37743.  -- Target Hook: void TARGET_SCHED_FINISH_GLOBAL (FILE *FILE, int

  37744.           VERBOSE)

  37745.      This is the cleanup hook corresponding to

  37746.      'TARGET_SCHED_INIT_GLOBAL'.  FILE is either a null pointer, or a

  37747.      stdio stream to write any debug output to.  VERBOSE is the verbose

  37748.      level provided by '-fsched-verbose-N'.

  37749. 

  37750.  -- Target Hook: rtx TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)

  37751.      The hook returns an RTL insn.  The automaton state used in the

  37752.      pipeline hazard recognizer is changed as if the insn were scheduled

  37753.      when the new simulated processor cycle starts.  Usage of the hook

  37754.      may simplify the automaton pipeline description for some VLIW

  37755.      processors.  If the hook is defined, it is used only for the

  37756.      automaton based pipeline description.  The default is not to change

  37757.      the state when the new simulated processor cycle starts.

  37758. 

  37759.  -- Target Hook: void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)

  37760.      The hook can be used to initialize data used by the previous hook.

  37761. 

  37762.  -- Target Hook: rtx_insn * TARGET_SCHED_DFA_POST_CYCLE_INSN (void)

  37763.      The hook is analogous to 'TARGET_SCHED_DFA_PRE_CYCLE_INSN' but used

  37764.      to changed the state as if the insn were scheduled when the new

  37765.      simulated processor cycle finishes.

  37766. 

  37767.  -- Target Hook: void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)

  37768.      The hook is analogous to 'TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN' but

  37769.      used to initialize data used by the previous hook.

  37770. 

  37771.  -- Target Hook: void TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE (void)

  37772.      The hook to notify target that the current simulated cycle is about

  37773.      to finish.  The hook is analogous to

  37774.      'TARGET_SCHED_DFA_PRE_CYCLE_INSN' but used to change the state in

  37775.      more complicated situations - e.g., when advancing state on a

  37776.      single insn is not enough.

  37777. 

  37778.  -- Target Hook: void TARGET_SCHED_DFA_POST_ADVANCE_CYCLE (void)

  37779.      The hook to notify target that new simulated cycle has just

  37780.      started.  The hook is analogous to

  37781.      'TARGET_SCHED_DFA_POST_CYCLE_INSN' but used to change the state in

  37782.      more complicated situations - e.g., when advancing state on a

  37783.      single insn is not enough.

  37784. 

  37785.  -- Target Hook: int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD

  37786.           (void)

  37787.      This hook controls better choosing an insn from the ready insn

  37788.      queue for the DFA-based insn scheduler.  Usually the scheduler

  37789.      chooses the first insn from the queue.  If the hook returns a

  37790.      positive value, an additional scheduler code tries all permutations

  37791.      of 'TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()' subsequent

  37792.      ready insns to choose an insn whose issue will result in maximal

  37793.      number of issued insns on the same cycle.  For the VLIW processor,

  37794.      the code could actually solve the problem of packing simple insns

  37795.      into the VLIW insn.  Of course, if the rules of VLIW packing are

  37796.      described in the automaton.

  37797. 

  37798.      This code also could be used for superscalar RISC processors.  Let

  37799.      us consider a superscalar RISC processor with 3 pipelines.  Some

  37800.      insns can be executed in pipelines A or B, some insns can be

  37801.      executed only in pipelines B or C, and one insn can be executed in

  37802.      pipeline B.  The processor may issue the 1st insn into A and the

  37803.      2nd one into B.  In this case, the 3rd insn will wait for freeing B

  37804.      until the next cycle.  If the scheduler issues the 3rd insn the

  37805.      first, the processor could issue all 3 insns per cycle.

  37806. 

  37807.      Actually this code demonstrates advantages of the automaton based

  37808.      pipeline hazard recognizer.  We try quickly and easy many insn

  37809.      schedules to choose the best one.

  37810. 

  37811.      The default is no multipass scheduling.

  37812. 

  37813.  -- Target Hook: int

  37814.           TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD

  37815.           (rtx_insn *INSN, int READY_INDEX)

  37816. 

  37817.      This hook controls what insns from the ready insn queue will be

  37818.      considered for the multipass insn scheduling.  If the hook returns

  37819.      zero for INSN, the insn will be considered in multipass scheduling.

  37820.      Positive return values will remove INSN from consideration on the

  37821.      current round of multipass scheduling.  Negative return values will

  37822.      remove INSN from consideration for given number of cycles.

  37823.      Backends should be careful about returning non-zero for highest

  37824.      priority instruction at position 0 in the ready list.  READY_INDEX

  37825.      is passed to allow backends make correct judgements.

  37826. 

  37827.      The default is that any ready insns can be chosen to be issued.

  37828. 

  37829.  -- Target Hook: void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BEGIN (void

  37830.           *DATA, signed char *READY_TRY, int N_READY, bool

  37831.           FIRST_CYCLE_INSN_P)

  37832.      This hook prepares the target backend for a new round of multipass

  37833.      scheduling.

  37834. 

  37835.  -- Target Hook: void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_ISSUE (void

  37836.           *DATA, signed char *READY_TRY, int N_READY, rtx_insn *INSN,

  37837.           const void *PREV_DATA)

  37838.      This hook is called when multipass scheduling evaluates instruction

  37839.      INSN.

  37840. 

  37841.  -- Target Hook: void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BACKTRACK

  37842.           (const void *DATA, signed char *READY_TRY, int N_READY)

  37843.      This is called when multipass scheduling backtracks from evaluation

  37844.      of an instruction.

  37845. 

  37846.  -- Target Hook: void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_END (const void

  37847.           *DATA)

  37848.      This hook notifies the target about the result of the concluded

  37849.      current round of multipass scheduling.

  37850. 

  37851.  -- Target Hook: void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_INIT (void

  37852.           *DATA)

  37853.      This hook initializes target-specific data used in multipass

  37854.      scheduling.

  37855. 

  37856.  -- Target Hook: void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_FINI (void

  37857.           *DATA)

  37858.      This hook finalizes target-specific data used in multipass

  37859.      scheduling.

  37860. 

  37861.  -- Target Hook: int TARGET_SCHED_DFA_NEW_CYCLE (FILE *DUMP, int

  37862.           VERBOSE, rtx_insn *INSN, int LAST_CLOCK, int CLOCK, int

  37863.           *SORT_P)

  37864.      This hook is called by the insn scheduler before issuing INSN on

  37865.      cycle CLOCK.  If the hook returns nonzero, INSN is not issued on

  37866.      this processor cycle.  Instead, the processor cycle is advanced.

  37867.      If *SORT_P is zero, the insn ready queue is not sorted on the new

  37868.      cycle start as usually.  DUMP and VERBOSE specify the file and

  37869.      verbosity level to use for debugging output.  LAST_CLOCK and CLOCK

  37870.      are, respectively, the processor cycle on which the previous insn

  37871.      has been issued, and the current processor cycle.

  37872. 

  37873.  -- Target Hook: bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (struct _dep

  37874.           *_DEP, int COST, int DISTANCE)

  37875.      This hook is used to define which dependences are considered costly

  37876.      by the target, so costly that it is not advisable to schedule the

  37877.      insns that are involved in the dependence too close to one another.

  37878.      The parameters to this hook are as follows: The first parameter

  37879.      _DEP is the dependence being evaluated.  The second parameter COST

  37880.      is the cost of the dependence as estimated by the scheduler, and

  37881.      the third parameter DISTANCE is the distance in cycles between the

  37882.      two insns.  The hook returns 'true' if considering the distance

  37883.      between the two insns the dependence between them is considered

  37884.      costly by the target, and 'false' otherwise.

  37885. 

  37886.      Defining this hook can be useful in multiple-issue out-of-order

  37887.      machines, where (a) it's practically hopeless to predict the actual

  37888.      data/resource delays, however: (b) there's a better chance to

  37889.      predict the actual grouping that will be formed, and (c) correctly

  37890.      emulating the grouping can be very important.  In such targets one

  37891.      may want to allow issuing dependent insns closer to one

  37892.      another--i.e., closer than the dependence distance; however, not in

  37893.      cases of "costly dependences", which this hooks allows to define.

  37894. 

  37895.  -- Target Hook: void TARGET_SCHED_H_I_D_EXTENDED (void)

  37896.      This hook is called by the insn scheduler after emitting a new

  37897.      instruction to the instruction stream.  The hook notifies a target

  37898.      backend to extend its per instruction data structures.

  37899. 

  37900.  -- Target Hook: void * TARGET_SCHED_ALLOC_SCHED_CONTEXT (void)

  37901.      Return a pointer to a store large enough to hold target scheduling

  37902.      context.

  37903. 

  37904.  -- Target Hook: void TARGET_SCHED_INIT_SCHED_CONTEXT (void *TC, bool

  37905.           CLEAN_P)

  37906.      Initialize store pointed to by TC to hold target scheduling

  37907.      context.  It CLEAN_P is true then initialize TC as if scheduler is

  37908.      at the beginning of the block.  Otherwise, copy the current context

  37909.      into TC.

  37910. 

  37911.  -- Target Hook: void TARGET_SCHED_SET_SCHED_CONTEXT (void *TC)

  37912.      Copy target scheduling context pointed to by TC to the current

  37913.      context.

  37914. 

  37915.  -- Target Hook: void TARGET_SCHED_CLEAR_SCHED_CONTEXT (void *TC)

  37916.      Deallocate internal data in target scheduling context pointed to by

  37917.      TC.

  37918. 

  37919.  -- Target Hook: void TARGET_SCHED_FREE_SCHED_CONTEXT (void *TC)

  37920.      Deallocate a store for target scheduling context pointed to by TC.

  37921. 

  37922.  -- Target Hook: int TARGET_SCHED_SPECULATE_INSN (rtx_insn *INSN,

  37923.           unsigned int DEP_STATUS, rtx *NEW_PAT)

  37924.      This hook is called by the insn scheduler when INSN has only

  37925.      speculative dependencies and therefore can be scheduled

  37926.      speculatively.  The hook is used to check if the pattern of INSN

  37927.      has a speculative version and, in case of successful check, to

  37928.      generate that speculative pattern.  The hook should return 1, if

  37929.      the instruction has a speculative form, or -1, if it doesn't.

  37930.      REQUEST describes the type of requested speculation.  If the return

  37931.      value equals 1 then NEW_PAT is assigned the generated speculative

  37932.      pattern.

  37933. 

  37934.  -- Target Hook: bool TARGET_SCHED_NEEDS_BLOCK_P (unsigned int

  37935.           DEP_STATUS)

  37936.      This hook is called by the insn scheduler during generation of

  37937.      recovery code for INSN.  It should return 'true', if the

  37938.      corresponding check instruction should branch to recovery code, or

  37939.      'false' otherwise.

  37940. 

  37941.  -- Target Hook: rtx TARGET_SCHED_GEN_SPEC_CHECK (rtx_insn *INSN,

  37942.           rtx_insn *LABEL, unsigned int DS)

  37943.      This hook is called by the insn scheduler to generate a pattern for

  37944.      recovery check instruction.  If MUTATE_P is zero, then INSN is a

  37945.      speculative instruction for which the check should be generated.

  37946.      LABEL is either a label of a basic block, where recovery code

  37947.      should be emitted, or a null pointer, when requested check doesn't

  37948.      branch to recovery code (a simple check).  If MUTATE_P is nonzero,

  37949.      then a pattern for a branchy check corresponding to a simple check

  37950.      denoted by INSN should be generated.  In this case LABEL can't be

  37951.      null.

  37952. 

  37953.  -- Target Hook: void TARGET_SCHED_SET_SCHED_FLAGS (struct spec_info_def

  37954.           *SPEC_INFO)

  37955.      This hook is used by the insn scheduler to find out what features

  37956.      should be enabled/used.  The structure *SPEC_INFO should be filled

  37957.      in by the target.  The structure describes speculation types that

  37958.      can be used in the scheduler.

  37959. 

  37960.  -- Target Hook: bool TARGET_SCHED_CAN_SPECULATE_INSN (rtx_insn *INSN)

  37961.      Some instructions should never be speculated by the schedulers,

  37962.      usually because the instruction is too expensive to get this wrong.

  37963.      Often such instructions have long latency, and often they are not

  37964.      fully modeled in the pipeline descriptions.  This hook should

  37965.      return 'false' if INSN should not be speculated.

  37966. 

  37967.  -- Target Hook: int TARGET_SCHED_SMS_RES_MII (struct ddg *G)

  37968.      This hook is called by the swing modulo scheduler to calculate a

  37969.      resource-based lower bound which is based on the resources

  37970.      available in the machine and the resources required by each

  37971.      instruction.  The target backend can use G to calculate such bound.

  37972.      A very simple lower bound will be used in case this hook is not

  37973.      implemented: the total number of instructions divided by the issue

  37974.      rate.

  37975. 

  37976.  -- Target Hook: bool TARGET_SCHED_DISPATCH (rtx_insn *INSN, int X)

  37977.      This hook is called by Haifa Scheduler.  It returns true if

  37978.      dispatch scheduling is supported in hardware and the condition

  37979.      specified in the parameter is true.

  37980. 

  37981.  -- Target Hook: void TARGET_SCHED_DISPATCH_DO (rtx_insn *INSN, int X)

  37982.      This hook is called by Haifa Scheduler.  It performs the operation

  37983.      specified in its second parameter.

  37984. 

  37985.  -- Target Hook: bool TARGET_SCHED_EXPOSED_PIPELINE

  37986.      True if the processor has an exposed pipeline, which means that not

  37987.      just the order of instructions is important for correctness when

  37988.      scheduling, but also the latencies of operations.

  37989. 

  37990.  -- Target Hook: int TARGET_SCHED_REASSOCIATION_WIDTH (unsigned int OPC,

  37991.           machine_mode MODE)

  37992.      This hook is called by tree reassociator to determine a level of

  37993.      parallelism required in output calculations chain.

  37994. 

  37995.  -- Target Hook: void TARGET_SCHED_FUSION_PRIORITY (rtx_insn *INSN, int

  37996.           MAX_PRI, int *FUSION_PRI, int *PRI)

  37997.      This hook is called by scheduling fusion pass.  It calculates

  37998.      fusion priorities for each instruction passed in by parameter.  The

  37999.      priorities are returned via pointer parameters.

  38000. 

  38001.      INSN is the instruction whose priorities need to be calculated.

  38002.      MAX_PRI is the maximum priority can be returned in any cases.

  38003.      FUSION_PRI is the pointer parameter through which INSN's fusion

  38004.      priority should be calculated and returned.  PRI is the pointer

  38005.      parameter through which INSN's priority should be calculated and

  38006.      returned.

  38007. 

  38008.      Same FUSION_PRI should be returned for instructions which should be

  38009.      scheduled together.  Different PRI should be returned for

  38010.      instructions with same FUSION_PRI.  FUSION_PRI is the major sort

  38011.      key, PRI is the minor sort key.  All instructions will be scheduled

  38012.      according to the two priorities.  All priorities calculated should

  38013.      be between 0 (exclusive) and MAX_PRI (inclusive).  To avoid false

  38014.      dependencies, FUSION_PRI of instructions which need to be scheduled

  38015.      together should be smaller than FUSION_PRI of irrelevant

  38016.      instructions.

  38017. 

  38018.      Given below example:

  38019. 

  38020.               ldr r10, [r1, 4]

  38021.               add r4, r4, r10

  38022.               ldr r15, [r2, 8]

  38023.               sub r5, r5, r15

  38024.               ldr r11, [r1, 0]

  38025.               add r4, r4, r11

  38026.               ldr r16, [r2, 12]

  38027.               sub r5, r5, r16

  38028. 

  38029.      On targets like ARM/AArch64, the two pairs of consecutive loads

  38030.      should be merged.  Since peephole2 pass can't help in this case

  38031.      unless consecutive loads are actually next to each other in

  38032.      instruction flow.  That's where this scheduling fusion pass works.

  38033.      This hook calculates priority for each instruction based on its

  38034.      fustion type, like:

  38035. 

  38036.               ldr r10, [r1, 4]  ; fusion_pri=99,  pri=96

  38037.               add r4, r4, r10   ; fusion_pri=100, pri=100

  38038.               ldr r15, [r2, 8]  ; fusion_pri=98,  pri=92

  38039.               sub r5, r5, r15   ; fusion_pri=100, pri=100

  38040.               ldr r11, [r1, 0]  ; fusion_pri=99,  pri=100

  38041.               add r4, r4, r11   ; fusion_pri=100, pri=100

  38042.               ldr r16, [r2, 12] ; fusion_pri=98,  pri=88

  38043.               sub r5, r5, r16   ; fusion_pri=100, pri=100

  38044. 

  38045.      Scheduling fusion pass then sorts all ready to issue instructions

  38046.      according to the priorities.  As a result, instructions of same

  38047.      fusion type will be pushed together in instruction flow, like:

  38048. 

  38049.               ldr r11, [r1, 0]

  38050.               ldr r10, [r1, 4]

  38051.               ldr r15, [r2, 8]

  38052.               ldr r16, [r2, 12]

  38053.               add r4, r4, r10

  38054.               sub r5, r5, r15

  38055.               add r4, r4, r11

  38056.               sub r5, r5, r16

  38057. 

  38058.      Now peephole2 pass can simply merge the two pairs of loads.

  38059. 

  38060.      Since scheduling fusion pass relies on peephole2 to do real fusion

  38061.      work, it is only enabled by default when peephole2 is in effect.

  38062. 

  38063.      This is firstly introduced on ARM/AArch64 targets, please refer to

  38064.      the hook implementation for how different fusion types are

  38065.      supported.

  38066. 

  38067.  -- Target Hook: void TARGET_EXPAND_DIVMOD_LIBFUNC (rtx LIBFUNC,

  38068.           machine_mode MODE, rtx OP0, rtx OP1, rtx *QUOT, rtx *REM)

  38069.      Define this hook for enabling divmod transform if the port does not

  38070.      have hardware divmod insn but defines target-specific divmod

  38071.      libfuncs.

  38072. 

  38073. 

  38074. File: gccint.info,  Node: Sections,  Next: PIC,  Prev: Scheduling,  Up: Target Macros

  38075. 

  38076. 18.18 Dividing the Output into Sections (Texts, Data, ...)

  38077. ==========================================================

  38078. 

  38079. An object file is divided into sections containing different types of

  38080. data.  In the most common case, there are three sections: the "text

  38081. section", which holds instructions and read-only data; the "data

  38082. section", which holds initialized writable data; and the "bss section",

  38083. which holds uninitialized data.  Some systems have other kinds of

  38084. sections.

  38085. 

  38086.  'varasm.c' provides several well-known sections, such as

  38087. 'text_section', 'data_section' and 'bss_section'.  The normal way of

  38088. controlling a 'FOO_section' variable is to define the associated

  38089. 'FOO_SECTION_ASM_OP' macro, as described below.  The macros are only

  38090. read once, when 'varasm.c' initializes itself, so their values must be

  38091. run-time constants.  They may however depend on command-line flags.

  38092. 

  38093.  _Note:_ Some run-time files, such 'crtstuff.c', also make use of the

  38094. 'FOO_SECTION_ASM_OP' macros, and expect them to be string literals.

  38095. 

  38096.  Some assemblers require a different string to be written every time a

  38097. section is selected.  If your assembler falls into this category, you

  38098. should define the 'TARGET_ASM_INIT_SECTIONS' hook and use

  38099. 'get_unnamed_section' to set up the sections.

  38100. 

  38101.  You must always create a 'text_section', either by defining

  38102. 'TEXT_SECTION_ASM_OP' or by initializing 'text_section' in

  38103. 'TARGET_ASM_INIT_SECTIONS'.  The same is true of 'data_section' and

  38104. 'DATA_SECTION_ASM_OP'.  If you do not create a distinct

  38105. 'readonly_data_section', the default is to reuse 'text_section'.

  38106. 

  38107.  All the other 'varasm.c' sections are optional, and are null if the

  38108. target does not provide them.

  38109. 

  38110.  -- Macro: TEXT_SECTION_ASM_OP

  38111.      A C expression whose value is a string, including spacing,

  38112.      containing the assembler operation that should precede instructions

  38113.      and read-only data.  Normally '"\t.text"' is right.

  38114. 

  38115.  -- Macro: HOT_TEXT_SECTION_NAME

  38116.      If defined, a C string constant for the name of the section

  38117.      containing most frequently executed functions of the program.  If

  38118.      not defined, GCC will provide a default definition if the target

  38119.      supports named sections.

  38120. 

  38121.  -- Macro: UNLIKELY_EXECUTED_TEXT_SECTION_NAME

  38122.      If defined, a C string constant for the name of the section

  38123.      containing unlikely executed functions in the program.

  38124. 

  38125.  -- Macro: DATA_SECTION_ASM_OP

  38126.      A C expression whose value is a string, including spacing,

  38127.      containing the assembler operation to identify the following data

  38128.      as writable initialized data.  Normally '"\t.data"' is right.

  38129. 

  38130.  -- Macro: SDATA_SECTION_ASM_OP

  38131.      If defined, a C expression whose value is a string, including

  38132.      spacing, containing the assembler operation to identify the

  38133.      following data as initialized, writable small data.

  38134. 

  38135.  -- Macro: READONLY_DATA_SECTION_ASM_OP

  38136.      A C expression whose value is a string, including spacing,

  38137.      containing the assembler operation to identify the following data

  38138.      as read-only initialized data.

  38139. 

  38140.  -- Macro: BSS_SECTION_ASM_OP

  38141.      If defined, a C expression whose value is a string, including

  38142.      spacing, containing the assembler operation to identify the

  38143.      following data as uninitialized global data.  If not defined, and

  38144.      'ASM_OUTPUT_ALIGNED_BSS' not defined, uninitialized global data

  38145.      will be output in the data section if '-fno-common' is passed,

  38146.      otherwise 'ASM_OUTPUT_COMMON' will be used.

  38147. 

  38148.  -- Macro: SBSS_SECTION_ASM_OP

  38149.      If defined, a C expression whose value is a string, including

  38150.      spacing, containing the assembler operation to identify the

  38151.      following data as uninitialized, writable small data.

  38152. 

  38153.  -- Macro: TLS_COMMON_ASM_OP

  38154.      If defined, a C expression whose value is a string containing the

  38155.      assembler operation to identify the following data as thread-local

  38156.      common data.  The default is '".tls_common"'.

  38157. 

  38158.  -- Macro: TLS_SECTION_ASM_FLAG

  38159.      If defined, a C expression whose value is a character constant

  38160.      containing the flag used to mark a section as a TLS section.  The

  38161.      default is ''T''.

  38162. 

  38163.  -- Macro: INIT_SECTION_ASM_OP

  38164.      If defined, a C expression whose value is a string, including

  38165.      spacing, containing the assembler operation to identify the

  38166.      following data as initialization code.  If not defined, GCC will

  38167.      assume such a section does not exist.  This section has no

  38168.      corresponding 'init_section' variable; it is used entirely in

  38169.      runtime code.

  38170. 

  38171.  -- Macro: FINI_SECTION_ASM_OP

  38172.      If defined, a C expression whose value is a string, including

  38173.      spacing, containing the assembler operation to identify the

  38174.      following data as finalization code.  If not defined, GCC will

  38175.      assume such a section does not exist.  This section has no

  38176.      corresponding 'fini_section' variable; it is used entirely in

  38177.      runtime code.

  38178. 

  38179.  -- Macro: INIT_ARRAY_SECTION_ASM_OP

  38180.      If defined, a C expression whose value is a string, including

  38181.      spacing, containing the assembler operation to identify the

  38182.      following data as part of the '.init_array' (or equivalent)

  38183.      section.  If not defined, GCC will assume such a section does not

  38184.      exist.  Do not define both this macro and 'INIT_SECTION_ASM_OP'.

  38185. 

  38186.  -- Macro: FINI_ARRAY_SECTION_ASM_OP

  38187.      If defined, a C expression whose value is a string, including

  38188.      spacing, containing the assembler operation to identify the

  38189.      following data as part of the '.fini_array' (or equivalent)

  38190.      section.  If not defined, GCC will assume such a section does not

  38191.      exist.  Do not define both this macro and 'FINI_SECTION_ASM_OP'.

  38192. 

  38193.  -- Macro: MACH_DEP_SECTION_ASM_FLAG

  38194.      If defined, a C expression whose value is a character constant

  38195.      containing the flag used to mark a machine-dependent section.  This

  38196.      corresponds to the 'SECTION_MACH_DEP' section flag.

  38197. 

  38198.  -- Macro: CRT_CALL_STATIC_FUNCTION (SECTION_OP, FUNCTION)

  38199.      If defined, an ASM statement that switches to a different section

  38200.      via SECTION_OP, calls FUNCTION, and switches back to the text

  38201.      section.  This is used in 'crtstuff.c' if 'INIT_SECTION_ASM_OP' or

  38202.      'FINI_SECTION_ASM_OP' to calls to initialization and finalization

  38203.      functions from the init and fini sections.  By default, this macro

  38204.      uses a simple function call.  Some ports need hand-crafted assembly

  38205.      code to avoid dependencies on registers initialized in the function

  38206.      prologue or to ensure that constant pools don't end up too far way

  38207.      in the text section.

  38208. 

  38209.  -- Macro: TARGET_LIBGCC_SDATA_SECTION

  38210.      If defined, a string which names the section into which small

  38211.      variables defined in crtstuff and libgcc should go.  This is useful

  38212.      when the target has options for optimizing access to small data,

  38213.      and you want the crtstuff and libgcc routines to be conservative in

  38214.      what they expect of your application yet liberal in what your

  38215.      application expects.  For example, for targets with a '.sdata'

  38216.      section (like MIPS), you could compile crtstuff with '-G 0' so that

  38217.      it doesn't require small data support from your application, but

  38218.      use this macro to put small data into '.sdata' so that your

  38219.      application can access these variables whether it uses small data

  38220.      or not.

  38221. 

  38222.  -- Macro: FORCE_CODE_SECTION_ALIGN

  38223.      If defined, an ASM statement that aligns a code section to some

  38224.      arbitrary boundary.  This is used to force all fragments of the

  38225.      '.init' and '.fini' sections to have to same alignment and thus

  38226.      prevent the linker from having to add any padding.

  38227. 

  38228.  -- Macro: JUMP_TABLES_IN_TEXT_SECTION

  38229.      Define this macro to be an expression with a nonzero value if jump

  38230.      tables (for 'tablejump' insns) should be output in the text

  38231.      section, along with the assembler instructions.  Otherwise, the

  38232.      readonly data section is used.

  38233. 

  38234.      This macro is irrelevant if there is no separate readonly data

  38235.      section.

  38236. 

  38237.  -- Target Hook: void TARGET_ASM_INIT_SECTIONS (void)

  38238.      Define this hook if you need to do something special to set up the

  38239.      'varasm.c' sections, or if your target has some special sections of

  38240.      its own that you need to create.

  38241. 

  38242.      GCC calls this hook after processing the command line, but before

  38243.      writing any assembly code, and before calling any of the

  38244.      section-returning hooks described below.

  38245. 

  38246.  -- Target Hook: int TARGET_ASM_RELOC_RW_MASK (void)

  38247.      Return a mask describing how relocations should be treated when

  38248.      selecting sections.  Bit 1 should be set if global relocations

  38249.      should be placed in a read-write section; bit 0 should be set if

  38250.      local relocations should be placed in a read-write section.

  38251. 

  38252.      The default version of this function returns 3 when '-fpic' is in

  38253.      effect, and 0 otherwise.  The hook is typically redefined when the

  38254.      target cannot support (some kinds of) dynamic relocations in

  38255.      read-only sections even in executables.

  38256. 

  38257.  -- Target Hook: bool TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC (void)

  38258.      Return true to generate ADDR_DIF_VEC table or false to generate

  38259.      ADDR_VEC table for jumps in case of -fPIC.

  38260. 

  38261.      The default version of this function returns true if flag_pic

  38262.      equals true and false otherwise

  38263. 

  38264.  -- Target Hook: section * TARGET_ASM_SELECT_SECTION (tree EXP, int

  38265.           RELOC, unsigned HOST_WIDE_INT ALIGN)

  38266.      Return the section into which EXP should be placed.  You can assume

  38267.      that EXP is either a 'VAR_DECL' node or a constant of some sort.

  38268.      RELOC indicates whether the initial value of EXP requires link-time

  38269.      relocations.  Bit 0 is set when variable contains local relocations

  38270.      only, while bit 1 is set for global relocations.  ALIGN is the

  38271.      constant alignment in bits.

  38272. 

  38273.      The default version of this function takes care of putting

  38274.      read-only variables in 'readonly_data_section'.

  38275. 

  38276.      See also USE_SELECT_SECTION_FOR_FUNCTIONS.

  38277. 

  38278.  -- Macro: USE_SELECT_SECTION_FOR_FUNCTIONS

  38279.      Define this macro if you wish TARGET_ASM_SELECT_SECTION to be

  38280.      called for 'FUNCTION_DECL's as well as for variables and constants.

  38281. 

  38282.      In the case of a 'FUNCTION_DECL', RELOC will be zero if the

  38283.      function has been determined to be likely to be called, and nonzero

  38284.      if it is unlikely to be called.

  38285. 

  38286.  -- Target Hook: void TARGET_ASM_UNIQUE_SECTION (tree DECL, int RELOC)

  38287.      Build up a unique section name, expressed as a 'STRING_CST' node,

  38288.      and assign it to 'DECL_SECTION_NAME (DECL)'.  As with

  38289.      'TARGET_ASM_SELECT_SECTION', RELOC indicates whether the initial

  38290.      value of EXP requires link-time relocations.

  38291. 

  38292.      The default version of this function appends the symbol name to the

  38293.      ELF section name that would normally be used for the symbol.  For

  38294.      example, the function 'foo' would be placed in '.text.foo'.

  38295.      Whatever the actual target object format, this is often good

  38296.      enough.

  38297. 

  38298.  -- Target Hook: section * TARGET_ASM_FUNCTION_RODATA_SECTION (tree

  38299.           DECL)

  38300.      Return the readonly data section associated with 'DECL_SECTION_NAME

  38301.      (DECL)'.  The default version of this function selects

  38302.      '.gnu.linkonce.r.name' if the function's section is

  38303.      '.gnu.linkonce.t.name', '.rodata.name' if function is in

  38304.      '.text.name', and the normal readonly-data section otherwise.

  38305. 

  38306.  -- Target Hook: const char * TARGET_ASM_MERGEABLE_RODATA_PREFIX

  38307.      Usually, the compiler uses the prefix '".rodata"' to construct

  38308.      section names for mergeable constant data.  Define this macro to

  38309.      override the string if a different section name should be used.

  38310. 

  38311.  -- Target Hook: section * TARGET_ASM_TM_CLONE_TABLE_SECTION (void)

  38312.      Return the section that should be used for transactional memory

  38313.      clone tables.

  38314. 

  38315.  -- Target Hook: section * TARGET_ASM_SELECT_RTX_SECTION (machine_mode

  38316.           MODE, rtx X, unsigned HOST_WIDE_INT ALIGN)

  38317.      Return the section into which a constant X, of mode MODE, should be

  38318.      placed.  You can assume that X is some kind of constant in RTL.

  38319.      The argument MODE is redundant except in the case of a 'const_int'

  38320.      rtx.  ALIGN is the constant alignment in bits.

  38321. 

  38322.      The default version of this function takes care of putting symbolic

  38323.      constants in 'flag_pic' mode in 'data_section' and everything else

  38324.      in 'readonly_data_section'.

  38325. 

  38326.  -- Target Hook: tree TARGET_MANGLE_DECL_ASSEMBLER_NAME (tree DECL, tree

  38327.           ID)

  38328.      Define this hook if you need to postprocess the assembler name

  38329.      generated by target-independent code.  The ID provided to this hook

  38330.      will be the computed name (e.g., the macro 'DECL_NAME' of the DECL

  38331.      in C, or the mangled name of the DECL in C++).  The return value of

  38332.      the hook is an 'IDENTIFIER_NODE' for the appropriate mangled name

  38333.      on your target system.  The default implementation of this hook

  38334.      just returns the ID provided.

  38335. 

  38336.  -- Target Hook: void TARGET_ENCODE_SECTION_INFO (tree DECL, rtx RTL,

  38337.           int NEW_DECL_P)

  38338.      Define this hook if references to a symbol or a constant must be

  38339.      treated differently depending on something about the variable or

  38340.      function named by the symbol (such as what section it is in).

  38341. 

  38342.      The hook is executed immediately after rtl has been created for

  38343.      DECL, which may be a variable or function declaration or an entry

  38344.      in the constant pool.  In either case, RTL is the rtl in question.

  38345.      Do _not_ use 'DECL_RTL (DECL)' in this hook; that field may not

  38346.      have been initialized yet.

  38347. 

  38348.      In the case of a constant, it is safe to assume that the rtl is a

  38349.      'mem' whose address is a 'symbol_ref'.  Most decls will also have

  38350.      this form, but that is not guaranteed.  Global register variables,

  38351.      for instance, will have a 'reg' for their rtl.  (Normally the right

  38352.      thing to do with such unusual rtl is leave it alone.)

  38353. 

  38354.      The NEW_DECL_P argument will be true if this is the first time that

  38355.      'TARGET_ENCODE_SECTION_INFO' has been invoked on this decl.  It

  38356.      will be false for subsequent invocations, which will happen for

  38357.      duplicate declarations.  Whether or not anything must be done for

  38358.      the duplicate declaration depends on whether the hook examines

  38359.      'DECL_ATTRIBUTES'.  NEW_DECL_P is always true when the hook is

  38360.      called for a constant.

  38361. 

  38362.      The usual thing for this hook to do is to record flags in the

  38363.      'symbol_ref', using 'SYMBOL_REF_FLAG' or 'SYMBOL_REF_FLAGS'.

  38364.      Historically, the name string was modified if it was necessary to

  38365.      encode more than one bit of information, but this practice is now

  38366.      discouraged; use 'SYMBOL_REF_FLAGS'.

  38367. 

  38368.      The default definition of this hook, 'default_encode_section_info'

  38369.      in 'varasm.c', sets a number of commonly-useful bits in

  38370.      'SYMBOL_REF_FLAGS'.  Check whether the default does what you need

  38371.      before overriding it.

  38372. 

  38373.  -- Target Hook: const char * TARGET_STRIP_NAME_ENCODING (const char

  38374.           *NAME)

  38375.      Decode NAME and return the real name part, sans the characters that

  38376.      'TARGET_ENCODE_SECTION_INFO' may have added.

  38377. 

  38378.  -- Target Hook: bool TARGET_IN_SMALL_DATA_P (const_tree EXP)

  38379.      Returns true if EXP should be placed into a "small data" section.

  38380.      The default version of this hook always returns false.

  38381. 

  38382.  -- Target Hook: bool TARGET_HAVE_SRODATA_SECTION

  38383.      Contains the value true if the target places read-only "small data"

  38384.      into a separate section.  The default value is false.

  38385. 

  38386.  -- Target Hook: bool TARGET_PROFILE_BEFORE_PROLOGUE (void)

  38387.      It returns true if target wants profile code emitted before

  38388.      prologue.

  38389. 

  38390.      The default version of this hook use the target macro

  38391.      'PROFILE_BEFORE_PROLOGUE'.

  38392. 

  38393.  -- Target Hook: bool TARGET_BINDS_LOCAL_P (const_tree EXP)

  38394.      Returns true if EXP names an object for which name resolution rules

  38395.      must resolve to the current "module" (dynamic shared library or

  38396.      executable image).

  38397. 

  38398.      The default version of this hook implements the name resolution

  38399.      rules for ELF, which has a looser model of global name binding than

  38400.      other currently supported object file formats.

  38401. 

  38402.  -- Target Hook: bool TARGET_HAVE_TLS

  38403.      Contains the value true if the target supports thread-local

  38404.      storage.  The default value is false.

  38405. 

  38406. 

  38407. File: gccint.info,  Node: PIC,  Next: Assembler Format,  Prev: Sections,  Up: Target Macros

  38408. 

  38409. 18.19 Position Independent Code

  38410. ===============================

  38411. 

  38412. This section describes macros that help implement generation of position

  38413. independent code.  Simply defining these macros is not enough to

  38414. generate valid PIC; you must also add support to the hook

  38415. 'TARGET_LEGITIMATE_ADDRESS_P' and to the macro 'PRINT_OPERAND_ADDRESS',

  38416. as well as 'LEGITIMIZE_ADDRESS'.  You must modify the definition of

  38417. 'movsi' to do something appropriate when the source operand contains a

  38418. symbolic address.  You may also need to alter the handling of switch

  38419. statements so that they use relative addresses.

  38420. 

  38421.  -- Macro: PIC_OFFSET_TABLE_REGNUM

  38422.      The register number of the register used to address a table of

  38423.      static data addresses in memory.  In some cases this register is

  38424.      defined by a processor's "application binary interface" (ABI).

  38425.      When this macro is defined, RTL is generated for this register

  38426.      once, as with the stack pointer and frame pointer registers.  If

  38427.      this macro is not defined, it is up to the machine-dependent files

  38428.      to allocate such a register (if necessary).  Note that this

  38429.      register must be fixed when in use (e.g. when 'flag_pic' is true).

  38430. 

  38431.  -- Macro: PIC_OFFSET_TABLE_REG_CALL_CLOBBERED

  38432.      A C expression that is nonzero if the register defined by

  38433.      'PIC_OFFSET_TABLE_REGNUM' is clobbered by calls.  If not defined,

  38434.      the default is zero.  Do not define this macro if

  38435.      'PIC_OFFSET_TABLE_REGNUM' is not defined.

  38436. 

  38437.  -- Macro: LEGITIMATE_PIC_OPERAND_P (X)

  38438.      A C expression that is nonzero if X is a legitimate immediate

  38439.      operand on the target machine when generating position independent

  38440.      code.  You can assume that X satisfies 'CONSTANT_P', so you need

  38441.      not check this.  You can also assume FLAG_PIC is true, so you need

  38442.      not check it either.  You need not define this macro if all

  38443.      constants (including 'SYMBOL_REF') can be immediate operands when

  38444.      generating position independent code.

  38445. 

  38446. 

  38447. File: gccint.info,  Node: Assembler Format,  Next: Debugging Info,  Prev: PIC,  Up: Target Macros

  38448. 

  38449. 18.20 Defining the Output Assembler Language

  38450. ============================================

  38451. 

  38452. This section describes macros whose principal purpose is to describe how

  38453. to write instructions in assembler language--rather than what the

  38454. instructions do.

  38455. 

  38456. * Menu:

  38457. 

  38458. * File Framework::       Structural information for the assembler file.

  38459. * Data Output::          Output of constants (numbers, strings, addresses).

  38460. * Uninitialized Data::   Output of uninitialized variables.

  38461. * Label Output::         Output and generation of labels.

  38462. * Initialization::       General principles of initialization

  38463.                          and termination routines.

  38464. * Macros for Initialization::

  38465.                          Specific macros that control the handling of

  38466.                          initialization and termination routines.

  38467. * Instruction Output::   Output of actual instructions.

  38468. * Dispatch Tables::      Output of jump tables.

  38469. * Exception Region Output:: Output of exception region code.

  38470. * Alignment Output::     Pseudo ops for alignment and skipping data.

  38471. 

  38472. 

  38473. File: gccint.info,  Node: File Framework,  Next: Data Output,  Up: Assembler Format

  38474. 

  38475. 18.20.1 The Overall Framework of an Assembler File

  38476. --------------------------------------------------

  38477. 

  38478. This describes the overall framework of an assembly file.

  38479. 

  38480.  -- Target Hook: void TARGET_ASM_FILE_START (void)

  38481.      Output to 'asm_out_file' any text which the assembler expects to

  38482.      find at the beginning of a file.  The default behavior is

  38483.      controlled by two flags, documented below.  Unless your target's

  38484.      assembler is quite unusual, if you override the default, you should

  38485.      call 'default_file_start' at some point in your target hook.  This

  38486.      lets other target files rely on these variables.

  38487. 

  38488.  -- Target Hook: bool TARGET_ASM_FILE_START_APP_OFF

  38489.      If this flag is true, the text of the macro 'ASM_APP_OFF' will be

  38490.      printed as the very first line in the assembly file, unless

  38491.      '-fverbose-asm' is in effect.  (If that macro has been defined to

  38492.      the empty string, this variable has no effect.)  With the normal

  38493.      definition of 'ASM_APP_OFF', the effect is to notify the GNU

  38494.      assembler that it need not bother stripping comments or extra

  38495.      whitespace from its input.  This allows it to work a bit faster.

  38496. 

  38497.      The default is false.  You should not set it to true unless you

  38498.      have verified that your port does not generate any extra whitespace

  38499.      or comments that will cause GAS to issue errors in NO_APP mode.

  38500. 

  38501.  -- Target Hook: bool TARGET_ASM_FILE_START_FILE_DIRECTIVE

  38502.      If this flag is true, 'output_file_directive' will be called for

  38503.      the primary source file, immediately after printing 'ASM_APP_OFF'

  38504.      (if that is enabled).  Most ELF assemblers expect this to be done.

  38505.      The default is false.

  38506. 

  38507.  -- Target Hook: void TARGET_ASM_FILE_END (void)

  38508.      Output to 'asm_out_file' any text which the assembler expects to

  38509.      find at the end of a file.  The default is to output nothing.

  38510. 

  38511.  -- Function: void file_end_indicate_exec_stack ()

  38512.      Some systems use a common convention, the '.note.GNU-stack' special

  38513.      section, to indicate whether or not an object file relies on the

  38514.      stack being executable.  If your system uses this convention, you

  38515.      should define 'TARGET_ASM_FILE_END' to this function.  If you need

  38516.      to do other things in that hook, have your hook function call this

  38517.      function.

  38518. 

  38519.  -- Target Hook: void TARGET_ASM_LTO_START (void)

  38520.      Output to 'asm_out_file' any text which the assembler expects to

  38521.      find at the start of an LTO section.  The default is to output

  38522.      nothing.

  38523. 

  38524.  -- Target Hook: void TARGET_ASM_LTO_END (void)

  38525.      Output to 'asm_out_file' any text which the assembler expects to

  38526.      find at the end of an LTO section.  The default is to output

  38527.      nothing.

  38528. 

  38529.  -- Target Hook: void TARGET_ASM_CODE_END (void)

  38530.      Output to 'asm_out_file' any text which is needed before emitting

  38531.      unwind info and debug info at the end of a file.  Some targets emit

  38532.      here PIC setup thunks that cannot be emitted at the end of file,

  38533.      because they couldn't have unwind info then.  The default is to

  38534.      output nothing.

  38535. 

  38536.  -- Macro: ASM_COMMENT_START

  38537.      A C string constant describing how to begin a comment in the target

  38538.      assembler language.  The compiler assumes that the comment will end

  38539.      at the end of the line.

  38540. 

  38541.  -- Macro: ASM_APP_ON

  38542.      A C string constant for text to be output before each 'asm'

  38543.      statement or group of consecutive ones.  Normally this is '"#APP"',

  38544.      which is a comment that has no effect on most assemblers but tells

  38545.      the GNU assembler that it must check the lines that follow for all

  38546.      valid assembler constructs.

  38547. 

  38548.  -- Macro: ASM_APP_OFF

  38549.      A C string constant for text to be output after each 'asm'

  38550.      statement or group of consecutive ones.  Normally this is

  38551.      '"#NO_APP"', which tells the GNU assembler to resume making the

  38552.      time-saving assumptions that are valid for ordinary compiler

  38553.      output.

  38554. 

  38555.  -- Macro: ASM_OUTPUT_SOURCE_FILENAME (STREAM, NAME)

  38556.      A C statement to output COFF information or DWARF debugging

  38557.      information which indicates that filename NAME is the current

  38558.      source file to the stdio stream STREAM.

  38559. 

  38560.      This macro need not be defined if the standard form of output for

  38561.      the file format in use is appropriate.

  38562. 

  38563.  -- Target Hook: void TARGET_ASM_OUTPUT_SOURCE_FILENAME (FILE *FILE,

  38564.           const char *NAME)

  38565.      Output DWARF debugging information which indicates that filename

  38566.      NAME is the current source file to the stdio stream FILE.

  38567. 

  38568.      This target hook need not be defined if the standard form of output

  38569.      for the file format in use is appropriate.

  38570. 

  38571.  -- Target Hook: void TARGET_ASM_OUTPUT_IDENT (const char *NAME)

  38572.      Output a string based on NAME, suitable for the '#ident' directive,

  38573.      or the equivalent directive or pragma in non-C-family languages.

  38574.      If this hook is not defined, nothing is output for the '#ident'

  38575.      directive.

  38576. 

  38577.  -- Macro: OUTPUT_QUOTED_STRING (STREAM, STRING)

  38578.      A C statement to output the string STRING to the stdio stream

  38579.      STREAM.  If you do not call the function 'output_quoted_string' in

  38580.      your config files, GCC will only call it to output filenames to the

  38581.      assembler source.  So you can use it to canonicalize the format of

  38582.      the filename using this macro.

  38583. 

  38584.  -- Target Hook: void TARGET_ASM_NAMED_SECTION (const char *NAME,

  38585.           unsigned int FLAGS, tree DECL)

  38586.      Output assembly directives to switch to section NAME.  The section

  38587.      should have attributes as specified by FLAGS, which is a bit mask

  38588.      of the 'SECTION_*' flags defined in 'output.h'.  If DECL is

  38589.      non-NULL, it is the 'VAR_DECL' or 'FUNCTION_DECL' with which this

  38590.      section is associated.

  38591. 

  38592.  -- Target Hook: bool TARGET_ASM_ELF_FLAGS_NUMERIC (unsigned int FLAGS,

  38593.           unsigned int *NUM)

  38594.      This hook can be used to encode ELF section flags for which no

  38595.      letter code has been defined in the assembler.  It is called by

  38596.      'default_asm_named_section' whenever the section flags need to be

  38597.      emitted in the assembler output.  If the hook returns true, then

  38598.      the numerical value for ELF section flags should be calculated from

  38599.      FLAGS and saved in *NUM; the value is printed out instead of the

  38600.      normal sequence of letter codes.  If the hook is not defined, or if

  38601.      it returns false, then NUM is ignored and the traditional letter

  38602.      sequence is emitted.

  38603. 

  38604.  -- Target Hook: section * TARGET_ASM_FUNCTION_SECTION (tree DECL, enum

  38605.           node_frequency FREQ, bool STARTUP, bool EXIT)

  38606.      Return preferred text (sub)section for function DECL.  Main purpose

  38607.      of this function is to separate cold, normal and hot functions.

  38608.      STARTUP is true when function is known to be used only at startup

  38609.      (from static constructors or it is 'main()').  EXIT is true when

  38610.      function is known to be used only at exit (from static

  38611.      destructors).  Return NULL if function should go to default text

  38612.      section.

  38613. 

  38614.  -- Target Hook: void TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS (FILE

  38615.           *FILE, tree DECL, bool NEW_IS_COLD)

  38616.      Used by the target to emit any assembler directives or additional

  38617.      labels needed when a function is partitioned between different

  38618.      sections.  Output should be written to FILE.  The function decl is

  38619.      available as DECL and the new section is 'cold' if NEW_IS_COLD is

  38620.      'true'.

  38621. 

  38622.  -- Common Target Hook: bool TARGET_HAVE_NAMED_SECTIONS

  38623.      This flag is true if the target supports

  38624.      'TARGET_ASM_NAMED_SECTION'.  It must not be modified by

  38625.      command-line option processing.

  38626. 

  38627.  -- Target Hook: bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS

  38628.      This flag is true if we can create zeroed data by switching to a

  38629.      BSS section and then using 'ASM_OUTPUT_SKIP' to allocate the space.

  38630.      This is true on most ELF targets.

  38631. 

  38632.  -- Target Hook: unsigned int TARGET_SECTION_TYPE_FLAGS (tree DECL,

  38633.           const char *NAME, int RELOC)

  38634.      Choose a set of section attributes for use by

  38635.      'TARGET_ASM_NAMED_SECTION' based on a variable or function decl, a

  38636.      section name, and whether or not the declaration's initializer may

  38637.      contain runtime relocations.  DECL may be null, in which case

  38638.      read-write data should be assumed.

  38639. 

  38640.      The default version of this function handles choosing code vs data,

  38641.      read-only vs read-write data, and 'flag_pic'.  You should only need

  38642.      to override this if your target has special flags that might be set

  38643.      via '__attribute__'.

  38644. 

  38645.  -- Target Hook: int TARGET_ASM_RECORD_GCC_SWITCHES (print_switch_type

  38646.           TYPE, const char *TEXT)

  38647.      Provides the target with the ability to record the gcc command line

  38648.      switches that have been passed to the compiler, and options that

  38649.      are enabled.  The TYPE argument specifies what is being recorded.

  38650.      It can take the following values:

  38651. 

  38652.      'SWITCH_TYPE_PASSED'

  38653.           TEXT is a command line switch that has been set by the user.

  38654. 

  38655.      'SWITCH_TYPE_ENABLED'

  38656.           TEXT is an option which has been enabled.  This might be as a

  38657.           direct result of a command line switch, or because it is

  38658.           enabled by default or because it has been enabled as a side

  38659.           effect of a different command line switch.  For example, the

  38660.           '-O2' switch enables various different individual optimization

  38661.           passes.

  38662. 

  38663.      'SWITCH_TYPE_DESCRIPTIVE'

  38664.           TEXT is either NULL or some descriptive text which should be

  38665.           ignored.  If TEXT is NULL then it is being used to warn the

  38666.           target hook that either recording is starting or ending.  The

  38667.           first time TYPE is SWITCH_TYPE_DESCRIPTIVE and TEXT is NULL,

  38668.           the warning is for start up and the second time the warning is

  38669.           for wind down.  This feature is to allow the target hook to

  38670.           make any necessary preparations before it starts to record

  38671.           switches and to perform any necessary tidying up after it has

  38672.           finished recording switches.

  38673. 

  38674.      'SWITCH_TYPE_LINE_START'

  38675.           This option can be ignored by this target hook.

  38676. 

  38677.      'SWITCH_TYPE_LINE_END'

  38678.           This option can be ignored by this target hook.

  38679. 

  38680.      The hook's return value must be zero.  Other return values may be

  38681.      supported in the future.

  38682. 

  38683.      By default this hook is set to NULL, but an example implementation

  38684.      is provided for ELF based targets.  Called ELF_RECORD_GCC_SWITCHES,

  38685.      it records the switches as ASCII text inside a new, string

  38686.      mergeable section in the assembler output file.  The name of the

  38687.      new section is provided by the

  38688.      'TARGET_ASM_RECORD_GCC_SWITCHES_SECTION' target hook.

  38689. 

  38690.  -- Target Hook: const char * TARGET_ASM_RECORD_GCC_SWITCHES_SECTION

  38691.      This is the name of the section that will be created by the example

  38692.      ELF implementation of the 'TARGET_ASM_RECORD_GCC_SWITCHES' target

  38693.      hook.

  38694. 

  38695. 

  38696. File: gccint.info,  Node: Data Output,  Next: Uninitialized Data,  Prev: File Framework,  Up: Assembler Format

  38697. 

  38698. 18.20.2 Output of Data

  38699. ----------------------

  38700. 

  38701.  -- Target Hook: const char * TARGET_ASM_BYTE_OP

  38702.  -- Target Hook: const char * TARGET_ASM_ALIGNED_HI_OP

  38703.  -- Target Hook: const char * TARGET_ASM_ALIGNED_PSI_OP

  38704.  -- Target Hook: const char * TARGET_ASM_ALIGNED_SI_OP

  38705.  -- Target Hook: const char * TARGET_ASM_ALIGNED_PDI_OP

  38706.  -- Target Hook: const char * TARGET_ASM_ALIGNED_DI_OP

  38707.  -- Target Hook: const char * TARGET_ASM_ALIGNED_PTI_OP

  38708.  -- Target Hook: const char * TARGET_ASM_ALIGNED_TI_OP

  38709.  -- Target Hook: const char * TARGET_ASM_UNALIGNED_HI_OP

  38710.  -- Target Hook: const char * TARGET_ASM_UNALIGNED_PSI_OP

  38711.  -- Target Hook: const char * TARGET_ASM_UNALIGNED_SI_OP

  38712.  -- Target Hook: const char * TARGET_ASM_UNALIGNED_PDI_OP

  38713.  -- Target Hook: const char * TARGET_ASM_UNALIGNED_DI_OP

  38714.  -- Target Hook: const char * TARGET_ASM_UNALIGNED_PTI_OP

  38715.  -- Target Hook: const char * TARGET_ASM_UNALIGNED_TI_OP

  38716.      These hooks specify assembly directives for creating certain kinds

  38717.      of integer object.  The 'TARGET_ASM_BYTE_OP' directive creates a

  38718.      byte-sized object, the 'TARGET_ASM_ALIGNED_HI_OP' one creates an

  38719.      aligned two-byte object, and so on.  Any of the hooks may be

  38720.      'NULL', indicating that no suitable directive is available.

  38721. 

  38722.      The compiler will print these strings at the start of a new line,

  38723.      followed immediately by the object's initial value.  In most cases,

  38724.      the string should contain a tab, a pseudo-op, and then another tab.

  38725. 

  38726.  -- Target Hook: bool TARGET_ASM_INTEGER (rtx X, unsigned int SIZE, int

  38727.           ALIGNED_P)

  38728.      The 'assemble_integer' function uses this hook to output an integer

  38729.      object.  X is the object's value, SIZE is its size in bytes and

  38730.      ALIGNED_P indicates whether it is aligned.  The function should

  38731.      return 'true' if it was able to output the object.  If it returns

  38732.      false, 'assemble_integer' will try to split the object into smaller

  38733.      parts.

  38734. 

  38735.      The default implementation of this hook will use the

  38736.      'TARGET_ASM_BYTE_OP' family of strings, returning 'false' when the

  38737.      relevant string is 'NULL'.

  38738. 

  38739.  -- Target Hook: void TARGET_ASM_DECL_END (void)

  38740.      Define this hook if the target assembler requires a special marker

  38741.      to terminate an initialized variable declaration.

  38742. 

  38743.  -- Target Hook: bool TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA (FILE *FILE,

  38744.           rtx X)

  38745.      A target hook to recognize RTX patterns that 'output_addr_const'

  38746.      can't deal with, and output assembly code to FILE corresponding to

  38747.      the pattern X.  This may be used to allow machine-dependent

  38748.      'UNSPEC's to appear within constants.

  38749. 

  38750.      If target hook fails to recognize a pattern, it must return

  38751.      'false', so that a standard error message is printed.  If it prints

  38752.      an error message itself, by calling, for example,

  38753.      'output_operand_lossage', it may just return 'true'.

  38754. 

  38755.  -- Macro: ASM_OUTPUT_ASCII (STREAM, PTR, LEN)

  38756.      A C statement to output to the stdio stream STREAM an assembler

  38757.      instruction to assemble a string constant containing the LEN bytes

  38758.      at PTR.  PTR will be a C expression of type 'char *' and LEN a C

  38759.      expression of type 'int'.

  38760. 

  38761.      If the assembler has a '.ascii' pseudo-op as found in the Berkeley

  38762.      Unix assembler, do not define the macro 'ASM_OUTPUT_ASCII'.

  38763. 

  38764.  -- Macro: ASM_OUTPUT_FDESC (STREAM, DECL, N)

  38765.      A C statement to output word N of a function descriptor for DECL.

  38766.      This must be defined if 'TARGET_VTABLE_USES_DESCRIPTORS' is

  38767.      defined, and is otherwise unused.

  38768. 

  38769.  -- Macro: CONSTANT_POOL_BEFORE_FUNCTION

  38770.      You may define this macro as a C expression.  You should define the

  38771.      expression to have a nonzero value if GCC should output the

  38772.      constant pool for a function before the code for the function, or a

  38773.      zero value if GCC should output the constant pool after the

  38774.      function.  If you do not define this macro, the usual case, GCC

  38775.      will output the constant pool before the function.

  38776. 

  38777.  -- Macro: ASM_OUTPUT_POOL_PROLOGUE (FILE, FUNNAME, FUNDECL, SIZE)

  38778.      A C statement to output assembler commands to define the start of

  38779.      the constant pool for a function.  FUNNAME is a string giving the

  38780.      name of the function.  Should the return type of the function be

  38781.      required, it can be obtained via FUNDECL.  SIZE is the size, in

  38782.      bytes, of the constant pool that will be written immediately after

  38783.      this call.

  38784. 

  38785.      If no constant-pool prefix is required, the usual case, this macro

  38786.      need not be defined.

  38787. 

  38788.  -- Macro: ASM_OUTPUT_SPECIAL_POOL_ENTRY (FILE, X, MODE, ALIGN, LABELNO,

  38789.           JUMPTO)

  38790.      A C statement (with or without semicolon) to output a constant in

  38791.      the constant pool, if it needs special treatment.  (This macro need

  38792.      not do anything for RTL expressions that can be output normally.)

  38793. 

  38794.      The argument FILE is the standard I/O stream to output the

  38795.      assembler code on.  X is the RTL expression for the constant to

  38796.      output, and MODE is the machine mode (in case X is a 'const_int').

  38797.      ALIGN is the required alignment for the value X; you should output

  38798.      an assembler directive to force this much alignment.

  38799. 

  38800.      The argument LABELNO is a number to use in an internal label for

  38801.      the address of this pool entry.  The definition of this macro is

  38802.      responsible for outputting the label definition at the proper

  38803.      place.  Here is how to do this:

  38804. 

  38805.           (*targetm.asm_out.internal_label) (FILE, "LC", LABELNO);

  38806. 

  38807.      When you output a pool entry specially, you should end with a

  38808.      'goto' to the label JUMPTO.  This will prevent the same pool entry

  38809.      from being output a second time in the usual manner.

  38810. 

  38811.      You need not define this macro if it would do nothing.

  38812. 

  38813.  -- Macro: ASM_OUTPUT_POOL_EPILOGUE (FILE FUNNAME FUNDECL SIZE)

  38814.      A C statement to output assembler commands to at the end of the

  38815.      constant pool for a function.  FUNNAME is a string giving the name

  38816.      of the function.  Should the return type of the function be

  38817.      required, you can obtain it via FUNDECL.  SIZE is the size, in

  38818.      bytes, of the constant pool that GCC wrote immediately before this

  38819.      call.

  38820. 

  38821.      If no constant-pool epilogue is required, the usual case, you need

  38822.      not define this macro.

  38823. 

  38824.  -- Macro: IS_ASM_LOGICAL_LINE_SEPARATOR (C, STR)

  38825.      Define this macro as a C expression which is nonzero if C is used

  38826.      as a logical line separator by the assembler.  STR points to the

  38827.      position in the string where C was found; this can be used if a

  38828.      line separator uses multiple characters.

  38829. 

  38830.      If you do not define this macro, the default is that only the

  38831.      character ';' is treated as a logical line separator.

  38832. 

  38833.  -- Target Hook: const char * TARGET_ASM_OPEN_PAREN

  38834.  -- Target Hook: const char * TARGET_ASM_CLOSE_PAREN

  38835.      These target hooks are C string constants, describing the syntax in

  38836.      the assembler for grouping arithmetic expressions.  If not

  38837.      overridden, they default to normal parentheses, which is correct

  38838.      for most assemblers.

  38839. 

  38840.  These macros are provided by 'real.h' for writing the definitions of

  38841. 'ASM_OUTPUT_DOUBLE' and the like:

  38842. 

  38843.  -- Macro: REAL_VALUE_TO_TARGET_SINGLE (X, L)

  38844.  -- Macro: REAL_VALUE_TO_TARGET_DOUBLE (X, L)

  38845.  -- Macro: REAL_VALUE_TO_TARGET_LONG_DOUBLE (X, L)

  38846.  -- Macro: REAL_VALUE_TO_TARGET_DECIMAL32 (X, L)

  38847.  -- Macro: REAL_VALUE_TO_TARGET_DECIMAL64 (X, L)

  38848.  -- Macro: REAL_VALUE_TO_TARGET_DECIMAL128 (X, L)

  38849.      These translate X, of type 'REAL_VALUE_TYPE', to the target's

  38850.      floating point representation, and store its bit pattern in the

  38851.      variable L.  For 'REAL_VALUE_TO_TARGET_SINGLE' and

  38852.      'REAL_VALUE_TO_TARGET_DECIMAL32', this variable should be a simple

  38853.      'long int'.  For the others, it should be an array of 'long int'.

  38854.      The number of elements in this array is determined by the size of

  38855.      the desired target floating point data type: 32 bits of it go in

  38856.      each 'long int' array element.  Each array element holds 32 bits of

  38857.      the result, even if 'long int' is wider than 32 bits on the host

  38858.      machine.

  38859. 

  38860.      The array element values are designed so that you can print them

  38861.      out using 'fprintf' in the order they should appear in the target

  38862.      machine's memory.

  38863. 

  38864. 

  38865. File: gccint.info,  Node: Uninitialized Data,  Next: Label Output,  Prev: Data Output,  Up: Assembler Format

  38866. 

  38867. 18.20.3 Output of Uninitialized Variables

  38868. -----------------------------------------

  38869. 

  38870. Each of the macros in this section is used to do the whole job of

  38871. outputting a single uninitialized variable.

  38872. 

  38873.  -- Macro: ASM_OUTPUT_COMMON (STREAM, NAME, SIZE, ROUNDED)

  38874.      A C statement (sans semicolon) to output to the stdio stream STREAM

  38875.      the assembler definition of a common-label named NAME whose size is

  38876.      SIZE bytes.  The variable ROUNDED is the size rounded up to

  38877.      whatever alignment the caller wants.  It is possible that SIZE may

  38878.      be zero, for instance if a struct with no other member than a

  38879.      zero-length array is defined.  In this case, the backend must

  38880.      output a symbol definition that allocates at least one byte, both

  38881.      so that the address of the resulting object does not compare equal

  38882.      to any other, and because some object formats cannot even express

  38883.      the concept of a zero-sized common symbol, as that is how they

  38884.      represent an ordinary undefined external.

  38885. 

  38886.      Use the expression 'assemble_name (STREAM, NAME)' to output the

  38887.      name itself; before and after that, output the additional assembler

  38888.      syntax for defining the name, and a newline.

  38889. 

  38890.      This macro controls how the assembler definitions of uninitialized

  38891.      common global variables are output.

  38892. 

  38893.  -- Macro: ASM_OUTPUT_ALIGNED_COMMON (STREAM, NAME, SIZE, ALIGNMENT)

  38894.      Like 'ASM_OUTPUT_COMMON' except takes the required alignment as a

  38895.      separate, explicit argument.  If you define this macro, it is used

  38896.      in place of 'ASM_OUTPUT_COMMON', and gives you more flexibility in

  38897.      handling the required alignment of the variable.  The alignment is

  38898.      specified as the number of bits.

  38899. 

  38900.  -- Macro: ASM_OUTPUT_ALIGNED_DECL_COMMON (STREAM, DECL, NAME, SIZE,

  38901.           ALIGNMENT)

  38902.      Like 'ASM_OUTPUT_ALIGNED_COMMON' except that DECL of the variable

  38903.      to be output, if there is one, or 'NULL_TREE' if there is no

  38904.      corresponding variable.  If you define this macro, GCC will use it

  38905.      in place of both 'ASM_OUTPUT_COMMON' and

  38906.      'ASM_OUTPUT_ALIGNED_COMMON'.  Define this macro when you need to

  38907.      see the variable's decl in order to chose what to output.

  38908. 

  38909.  -- Macro: ASM_OUTPUT_ALIGNED_BSS (STREAM, DECL, NAME, SIZE, ALIGNMENT)

  38910.      A C statement (sans semicolon) to output to the stdio stream STREAM

  38911.      the assembler definition of uninitialized global DECL named NAME

  38912.      whose size is SIZE bytes.  The variable ALIGNMENT is the alignment

  38913.      specified as the number of bits.

  38914. 

  38915.      Try to use function 'asm_output_aligned_bss' defined in file

  38916.      'varasm.c' when defining this macro.  If unable, use the expression

  38917.      'assemble_name (STREAM, NAME)' to output the name itself; before

  38918.      and after that, output the additional assembler syntax for defining

  38919.      the name, and a newline.

  38920. 

  38921.      There are two ways of handling global BSS.  One is to define this

  38922.      macro.  The other is to have 'TARGET_ASM_SELECT_SECTION' return a

  38923.      switchable BSS section (*note

  38924.      TARGET_HAVE_SWITCHABLE_BSS_SECTIONS::).  You do not need to do

  38925.      both.

  38926. 

  38927.      Some languages do not have 'common' data, and require a non-common

  38928.      form of global BSS in order to handle uninitialized globals

  38929.      efficiently.  C++ is one example of this.  However, if the target

  38930.      does not support global BSS, the front end may choose to make

  38931.      globals common in order to save space in the object file.

  38932. 

  38933.  -- Macro: ASM_OUTPUT_LOCAL (STREAM, NAME, SIZE, ROUNDED)

  38934.      A C statement (sans semicolon) to output to the stdio stream STREAM

  38935.      the assembler definition of a local-common-label named NAME whose

  38936.      size is SIZE bytes.  The variable ROUNDED is the size rounded up to

  38937.      whatever alignment the caller wants.

  38938. 

  38939.      Use the expression 'assemble_name (STREAM, NAME)' to output the

  38940.      name itself; before and after that, output the additional assembler

  38941.      syntax for defining the name, and a newline.

  38942. 

  38943.      This macro controls how the assembler definitions of uninitialized

  38944.      static variables are output.

  38945. 

  38946.  -- Macro: ASM_OUTPUT_ALIGNED_LOCAL (STREAM, NAME, SIZE, ALIGNMENT)

  38947.      Like 'ASM_OUTPUT_LOCAL' except takes the required alignment as a

  38948.      separate, explicit argument.  If you define this macro, it is used

  38949.      in place of 'ASM_OUTPUT_LOCAL', and gives you more flexibility in

  38950.      handling the required alignment of the variable.  The alignment is

  38951.      specified as the number of bits.

  38952. 

  38953.  -- Macro: ASM_OUTPUT_ALIGNED_DECL_LOCAL (STREAM, DECL, NAME, SIZE,

  38954.           ALIGNMENT)

  38955.      Like 'ASM_OUTPUT_ALIGNED_LOCAL' except that DECL of the variable to

  38956.      be output, if there is one, or 'NULL_TREE' if there is no

  38957.      corresponding variable.  If you define this macro, GCC will use it

  38958.      in place of both 'ASM_OUTPUT_LOCAL' and 'ASM_OUTPUT_ALIGNED_LOCAL'.

  38959.      Define this macro when you need to see the variable's decl in order

  38960.      to chose what to output.

  38961. 

  38962. 

  38963. File: gccint.info,  Node: Label Output,  Next: Initialization,  Prev: Uninitialized Data,  Up: Assembler Format

  38964. 

  38965. 18.20.4 Output and Generation of Labels

  38966. ---------------------------------------

  38967. 

  38968. This is about outputting labels.

  38969. 

  38970.  -- Macro: ASM_OUTPUT_LABEL (STREAM, NAME)

  38971.      A C statement (sans semicolon) to output to the stdio stream STREAM

  38972.      the assembler definition of a label named NAME.  Use the expression

  38973.      'assemble_name (STREAM, NAME)' to output the name itself; before

  38974.      and after that, output the additional assembler syntax for defining

  38975.      the name, and a newline.  A default definition of this macro is

  38976.      provided which is correct for most systems.

  38977. 

  38978.  -- Macro: ASM_OUTPUT_FUNCTION_LABEL (STREAM, NAME, DECL)

  38979.      A C statement (sans semicolon) to output to the stdio stream STREAM

  38980.      the assembler definition of a label named NAME of a function.  Use

  38981.      the expression 'assemble_name (STREAM, NAME)' to output the name

  38982.      itself; before and after that, output the additional assembler

  38983.      syntax for defining the name, and a newline.  A default definition

  38984.      of this macro is provided which is correct for most systems.

  38985. 

  38986.      If this macro is not defined, then the function name is defined in

  38987.      the usual manner as a label (by means of 'ASM_OUTPUT_LABEL').

  38988. 

  38989.  -- Macro: ASM_OUTPUT_INTERNAL_LABEL (STREAM, NAME)

  38990.      Identical to 'ASM_OUTPUT_LABEL', except that NAME is known to refer

  38991.      to a compiler-generated label.  The default definition uses

  38992.      'assemble_name_raw', which is like 'assemble_name' except that it

  38993.      is more efficient.

  38994. 

  38995.  -- Macro: SIZE_ASM_OP

  38996.      A C string containing the appropriate assembler directive to

  38997.      specify the size of a symbol, without any arguments.  On systems

  38998.      that use ELF, the default (in 'config/elfos.h') is '"\t.size\t"';

  38999.      on other systems, the default is not to define this macro.

  39000. 

  39001.      Define this macro only if it is correct to use the default

  39002.      definitions of 'ASM_OUTPUT_SIZE_DIRECTIVE' and

  39003.      'ASM_OUTPUT_MEASURED_SIZE' for your system.  If you need your own

  39004.      custom definitions of those macros, or if you do not need explicit

  39005.      symbol sizes at all, do not define this macro.

  39006. 

  39007.  -- Macro: ASM_OUTPUT_SIZE_DIRECTIVE (STREAM, NAME, SIZE)

  39008.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39009.      a directive telling the assembler that the size of the symbol NAME

  39010.      is SIZE.  SIZE is a 'HOST_WIDE_INT'.  If you define 'SIZE_ASM_OP',

  39011.      a default definition of this macro is provided.

  39012. 

  39013.  -- Macro: ASM_OUTPUT_MEASURED_SIZE (STREAM, NAME)

  39014.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39015.      a directive telling the assembler to calculate the size of the

  39016.      symbol NAME by subtracting its address from the current address.

  39017. 

  39018.      If you define 'SIZE_ASM_OP', a default definition of this macro is

  39019.      provided.  The default assumes that the assembler recognizes a

  39020.      special '.' symbol as referring to the current address, and can

  39021.      calculate the difference between this and another symbol.  If your

  39022.      assembler does not recognize '.' or cannot do calculations with it,

  39023.      you will need to redefine 'ASM_OUTPUT_MEASURED_SIZE' to use some

  39024.      other technique.

  39025. 

  39026.  -- Macro: NO_DOLLAR_IN_LABEL

  39027.      Define this macro if the assembler does not accept the character

  39028.      '$' in label names.  By default constructors and destructors in G++

  39029.      have '$' in the identifiers.  If this macro is defined, '.' is used

  39030.      instead.

  39031. 

  39032.  -- Macro: NO_DOT_IN_LABEL

  39033.      Define this macro if the assembler does not accept the character

  39034.      '.' in label names.  By default constructors and destructors in G++

  39035.      have names that use '.'.  If this macro is defined, these names are

  39036.      rewritten to avoid '.'.

  39037. 

  39038.  -- Macro: TYPE_ASM_OP

  39039.      A C string containing the appropriate assembler directive to

  39040.      specify the type of a symbol, without any arguments.  On systems

  39041.      that use ELF, the default (in 'config/elfos.h') is '"\t.type\t"';

  39042.      on other systems, the default is not to define this macro.

  39043. 

  39044.      Define this macro only if it is correct to use the default

  39045.      definition of 'ASM_OUTPUT_TYPE_DIRECTIVE' for your system.  If you

  39046.      need your own custom definition of this macro, or if you do not

  39047.      need explicit symbol types at all, do not define this macro.

  39048. 

  39049.  -- Macro: TYPE_OPERAND_FMT

  39050.      A C string which specifies (using 'printf' syntax) the format of

  39051.      the second operand to 'TYPE_ASM_OP'.  On systems that use ELF, the

  39052.      default (in 'config/elfos.h') is '"@%s"'; on other systems, the

  39053.      default is not to define this macro.

  39054. 

  39055.      Define this macro only if it is correct to use the default

  39056.      definition of 'ASM_OUTPUT_TYPE_DIRECTIVE' for your system.  If you

  39057.      need your own custom definition of this macro, or if you do not

  39058.      need explicit symbol types at all, do not define this macro.

  39059. 

  39060.  -- Macro: ASM_OUTPUT_TYPE_DIRECTIVE (STREAM, TYPE)

  39061.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39062.      a directive telling the assembler that the type of the symbol NAME

  39063.      is TYPE.  TYPE is a C string; currently, that string is always

  39064.      either '"function"' or '"object"', but you should not count on

  39065.      this.

  39066. 

  39067.      If you define 'TYPE_ASM_OP' and 'TYPE_OPERAND_FMT', a default

  39068.      definition of this macro is provided.

  39069. 

  39070.  -- Macro: ASM_DECLARE_FUNCTION_NAME (STREAM, NAME, DECL)

  39071.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39072.      any text necessary for declaring the name NAME of a function which

  39073.      is being defined.  This macro is responsible for outputting the

  39074.      label definition (perhaps using 'ASM_OUTPUT_FUNCTION_LABEL').  The

  39075.      argument DECL is the 'FUNCTION_DECL' tree node representing the

  39076.      function.

  39077. 

  39078.      If this macro is not defined, then the function name is defined in

  39079.      the usual manner as a label (by means of

  39080.      'ASM_OUTPUT_FUNCTION_LABEL').

  39081. 

  39082.      You may wish to use 'ASM_OUTPUT_TYPE_DIRECTIVE' in the definition

  39083.      of this macro.

  39084. 

  39085.  -- Macro: ASM_DECLARE_FUNCTION_SIZE (STREAM, NAME, DECL)

  39086.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39087.      any text necessary for declaring the size of a function which is

  39088.      being defined.  The argument NAME is the name of the function.  The

  39089.      argument DECL is the 'FUNCTION_DECL' tree node representing the

  39090.      function.

  39091. 

  39092.      If this macro is not defined, then the function size is not

  39093.      defined.

  39094. 

  39095.      You may wish to use 'ASM_OUTPUT_MEASURED_SIZE' in the definition of

  39096.      this macro.

  39097. 

  39098.  -- Macro: ASM_DECLARE_COLD_FUNCTION_NAME (STREAM, NAME, DECL)

  39099.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39100.      any text necessary for declaring the name NAME of a cold function

  39101.      partition which is being defined.  This macro is responsible for

  39102.      outputting the label definition (perhaps using

  39103.      'ASM_OUTPUT_FUNCTION_LABEL').  The argument DECL is the

  39104.      'FUNCTION_DECL' tree node representing the function.

  39105. 

  39106.      If this macro is not defined, then the cold partition name is

  39107.      defined in the usual manner as a label (by means of

  39108.      'ASM_OUTPUT_LABEL').

  39109. 

  39110.      You may wish to use 'ASM_OUTPUT_TYPE_DIRECTIVE' in the definition

  39111.      of this macro.

  39112. 

  39113.  -- Macro: ASM_DECLARE_COLD_FUNCTION_SIZE (STREAM, NAME, DECL)

  39114.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39115.      any text necessary for declaring the size of a cold function

  39116.      partition which is being defined.  The argument NAME is the name of

  39117.      the cold partition of the function.  The argument DECL is the

  39118.      'FUNCTION_DECL' tree node representing the function.

  39119. 

  39120.      If this macro is not defined, then the partition size is not

  39121.      defined.

  39122. 

  39123.      You may wish to use 'ASM_OUTPUT_MEASURED_SIZE' in the definition of

  39124.      this macro.

  39125. 

  39126.  -- Macro: ASM_DECLARE_OBJECT_NAME (STREAM, NAME, DECL)

  39127.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39128.      any text necessary for declaring the name NAME of an initialized

  39129.      variable which is being defined.  This macro must output the label

  39130.      definition (perhaps using 'ASM_OUTPUT_LABEL').  The argument DECL

  39131.      is the 'VAR_DECL' tree node representing the variable.

  39132. 

  39133.      If this macro is not defined, then the variable name is defined in

  39134.      the usual manner as a label (by means of 'ASM_OUTPUT_LABEL').

  39135. 

  39136.      You may wish to use 'ASM_OUTPUT_TYPE_DIRECTIVE' and/or

  39137.      'ASM_OUTPUT_SIZE_DIRECTIVE' in the definition of this macro.

  39138. 

  39139.  -- Target Hook: void TARGET_ASM_DECLARE_CONSTANT_NAME (FILE *FILE,

  39140.           const char *NAME, const_tree EXPR, HOST_WIDE_INT SIZE)

  39141.      A target hook to output to the stdio stream FILE any text necessary

  39142.      for declaring the name NAME of a constant which is being defined.

  39143.      This target hook is responsible for outputting the label definition

  39144.      (perhaps using 'assemble_label').  The argument EXP is the value of

  39145.      the constant, and SIZE is the size of the constant in bytes.  The

  39146.      NAME will be an internal label.

  39147. 

  39148.      The default version of this target hook, define the NAME in the

  39149.      usual manner as a label (by means of 'assemble_label').

  39150. 

  39151.      You may wish to use 'ASM_OUTPUT_TYPE_DIRECTIVE' in this target

  39152.      hook.

  39153. 

  39154.  -- Macro: ASM_DECLARE_REGISTER_GLOBAL (STREAM, DECL, REGNO, NAME)

  39155.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39156.      any text necessary for claiming a register REGNO for a global

  39157.      variable DECL with name NAME.

  39158. 

  39159.      If you don't define this macro, that is equivalent to defining it

  39160.      to do nothing.

  39161. 

  39162.  -- Macro: ASM_FINISH_DECLARE_OBJECT (STREAM, DECL, TOPLEVEL, ATEND)

  39163.      A C statement (sans semicolon) to finish up declaring a variable

  39164.      name once the compiler has processed its initializer fully and thus

  39165.      has had a chance to determine the size of an array when controlled

  39166.      by an initializer.  This is used on systems where it's necessary to

  39167.      declare something about the size of the object.

  39168. 

  39169.      If you don't define this macro, that is equivalent to defining it

  39170.      to do nothing.

  39171. 

  39172.      You may wish to use 'ASM_OUTPUT_SIZE_DIRECTIVE' and/or

  39173.      'ASM_OUTPUT_MEASURED_SIZE' in the definition of this macro.

  39174. 

  39175.  -- Target Hook: void TARGET_ASM_GLOBALIZE_LABEL (FILE *STREAM, const

  39176.           char *NAME)

  39177.      This target hook is a function to output to the stdio stream STREAM

  39178.      some commands that will make the label NAME global; that is,

  39179.      available for reference from other files.

  39180. 

  39181.      The default implementation relies on a proper definition of

  39182.      'GLOBAL_ASM_OP'.

  39183. 

  39184.  -- Target Hook: void TARGET_ASM_GLOBALIZE_DECL_NAME (FILE *STREAM, tree

  39185.           DECL)

  39186.      This target hook is a function to output to the stdio stream STREAM

  39187.      some commands that will make the name associated with DECL global;

  39188.      that is, available for reference from other files.

  39189. 

  39190.      The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL

  39191.      target hook.

  39192. 

  39193.  -- Target Hook: void TARGET_ASM_ASSEMBLE_UNDEFINED_DECL (FILE *STREAM,

  39194.           const char *NAME, const_tree DECL)

  39195.      This target hook is a function to output to the stdio stream STREAM

  39196.      some commands that will declare the name associated with DECL which

  39197.      is not defined in the current translation unit.  Most assemblers do

  39198.      not require anything to be output in this case.

  39199. 

  39200.  -- Macro: ASM_WEAKEN_LABEL (STREAM, NAME)

  39201.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39202.      some commands that will make the label NAME weak; that is,

  39203.      available for reference from other files but only used if no other

  39204.      definition is available.  Use the expression 'assemble_name

  39205.      (STREAM, NAME)' to output the name itself; before and after that,

  39206.      output the additional assembler syntax for making that name weak,

  39207.      and a newline.

  39208. 

  39209.      If you don't define this macro or 'ASM_WEAKEN_DECL', GCC will not

  39210.      support weak symbols and you should not define the 'SUPPORTS_WEAK'

  39211.      macro.

  39212. 

  39213.  -- Macro: ASM_WEAKEN_DECL (STREAM, DECL, NAME, VALUE)

  39214.      Combines (and replaces) the function of 'ASM_WEAKEN_LABEL' and

  39215.      'ASM_OUTPUT_WEAK_ALIAS', allowing access to the associated function

  39216.      or variable decl.  If VALUE is not 'NULL', this C statement should

  39217.      output to the stdio stream STREAM assembler code which defines

  39218.      (equates) the weak symbol NAME to have the value VALUE.  If VALUE

  39219.      is 'NULL', it should output commands to make NAME weak.

  39220. 

  39221.  -- Macro: ASM_OUTPUT_WEAKREF (STREAM, DECL, NAME, VALUE)

  39222.      Outputs a directive that enables NAME to be used to refer to symbol

  39223.      VALUE with weak-symbol semantics.  'decl' is the declaration of

  39224.      'name'.

  39225. 

  39226.  -- Macro: SUPPORTS_WEAK

  39227.      A preprocessor constant expression which evaluates to true if the

  39228.      target supports weak symbols.

  39229. 

  39230.      If you don't define this macro, 'defaults.h' provides a default

  39231.      definition.  If either 'ASM_WEAKEN_LABEL' or 'ASM_WEAKEN_DECL' is

  39232.      defined, the default definition is '1'; otherwise, it is '0'.

  39233. 

  39234.  -- Macro: TARGET_SUPPORTS_WEAK

  39235.      A C expression which evaluates to true if the target supports weak

  39236.      symbols.

  39237. 

  39238.      If you don't define this macro, 'defaults.h' provides a default

  39239.      definition.  The default definition is '(SUPPORTS_WEAK)'.  Define

  39240.      this macro if you want to control weak symbol support with a

  39241.      compiler flag such as '-melf'.

  39242. 

  39243.  -- Macro: MAKE_DECL_ONE_ONLY (DECL)

  39244.      A C statement (sans semicolon) to mark DECL to be emitted as a

  39245.      public symbol such that extra copies in multiple translation units

  39246.      will be discarded by the linker.  Define this macro if your object

  39247.      file format provides support for this concept, such as the 'COMDAT'

  39248.      section flags in the Microsoft Windows PE/COFF format, and this

  39249.      support requires changes to DECL, such as putting it in a separate

  39250.      section.

  39251. 

  39252.  -- Macro: SUPPORTS_ONE_ONLY

  39253.      A C expression which evaluates to true if the target supports

  39254.      one-only semantics.

  39255. 

  39256.      If you don't define this macro, 'varasm.c' provides a default

  39257.      definition.  If 'MAKE_DECL_ONE_ONLY' is defined, the default

  39258.      definition is '1'; otherwise, it is '0'.  Define this macro if you

  39259.      want to control one-only symbol support with a compiler flag, or if

  39260.      setting the 'DECL_ONE_ONLY' flag is enough to mark a declaration to

  39261.      be emitted as one-only.

  39262. 

  39263.  -- Target Hook: void TARGET_ASM_ASSEMBLE_VISIBILITY (tree DECL, int

  39264.           VISIBILITY)

  39265.      This target hook is a function to output to ASM_OUT_FILE some

  39266.      commands that will make the symbol(s) associated with DECL have

  39267.      hidden, protected or internal visibility as specified by

  39268.      VISIBILITY.

  39269. 

  39270.  -- Macro: TARGET_WEAK_NOT_IN_ARCHIVE_TOC

  39271.      A C expression that evaluates to true if the target's linker

  39272.      expects that weak symbols do not appear in a static archive's table

  39273.      of contents.  The default is '0'.

  39274. 

  39275.      Leaving weak symbols out of an archive's table of contents means

  39276.      that, if a symbol will only have a definition in one translation

  39277.      unit and will have undefined references from other translation

  39278.      units, that symbol should not be weak.  Defining this macro to be

  39279.      nonzero will thus have the effect that certain symbols that would

  39280.      normally be weak (explicit template instantiations, and vtables for

  39281.      polymorphic classes with noninline key methods) will instead be

  39282.      nonweak.

  39283. 

  39284.      The C++ ABI requires this macro to be zero.  Define this macro for

  39285.      targets where full C++ ABI compliance is impossible and where

  39286.      linker restrictions require weak symbols to be left out of a static

  39287.      archive's table of contents.

  39288. 

  39289.  -- Macro: ASM_OUTPUT_EXTERNAL (STREAM, DECL, NAME)

  39290.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39291.      any text necessary for declaring the name of an external symbol

  39292.      named NAME which is referenced in this compilation but not defined.

  39293.      The value of DECL is the tree node for the declaration.

  39294. 

  39295.      This macro need not be defined if it does not need to output

  39296.      anything.  The GNU assembler and most Unix assemblers don't require

  39297.      anything.

  39298. 

  39299.  -- Target Hook: void TARGET_ASM_EXTERNAL_LIBCALL (rtx SYMREF)

  39300.      This target hook is a function to output to ASM_OUT_FILE an

  39301.      assembler pseudo-op to declare a library function name external.

  39302.      The name of the library function is given by SYMREF, which is a

  39303.      'symbol_ref'.

  39304. 

  39305.  -- Target Hook: void TARGET_ASM_MARK_DECL_PRESERVED (const char

  39306.           *SYMBOL)

  39307.      This target hook is a function to output to ASM_OUT_FILE an

  39308.      assembler directive to annotate SYMBOL as used.  The Darwin target

  39309.      uses the .no_dead_code_strip directive.

  39310. 

  39311.  -- Macro: ASM_OUTPUT_LABELREF (STREAM, NAME)

  39312.      A C statement (sans semicolon) to output to the stdio stream STREAM

  39313.      a reference in assembler syntax to a label named NAME.  This should

  39314.      add '_' to the front of the name, if that is customary on your

  39315.      operating system, as it is in most Berkeley Unix systems.  This

  39316.      macro is used in 'assemble_name'.

  39317. 

  39318.  -- Target Hook: tree TARGET_MANGLE_ASSEMBLER_NAME (const char *NAME)

  39319.      Given a symbol NAME, perform same mangling as 'varasm.c''s

  39320.      'assemble_name', but in memory rather than to a file stream,

  39321.      returning result as an 'IDENTIFIER_NODE'.  Required for correct LTO

  39322.      symtabs.  The default implementation calls the

  39323.      'TARGET_STRIP_NAME_ENCODING' hook and then prepends the

  39324.      'USER_LABEL_PREFIX', if any.

  39325. 

  39326.  -- Macro: ASM_OUTPUT_SYMBOL_REF (STREAM, SYM)

  39327.      A C statement (sans semicolon) to output a reference to

  39328.      'SYMBOL_REF' SYM.  If not defined, 'assemble_name' will be used to

  39329.      output the name of the symbol.  This macro may be used to modify

  39330.      the way a symbol is referenced depending on information encoded by

  39331.      'TARGET_ENCODE_SECTION_INFO'.

  39332. 

  39333.  -- Macro: ASM_OUTPUT_LABEL_REF (STREAM, BUF)

  39334.      A C statement (sans semicolon) to output a reference to BUF, the

  39335.      result of 'ASM_GENERATE_INTERNAL_LABEL'.  If not defined,

  39336.      'assemble_name' will be used to output the name of the symbol.

  39337.      This macro is not used by 'output_asm_label', or the '%l' specifier

  39338.      that calls it; the intention is that this macro should be set when

  39339.      it is necessary to output a label differently when its address is

  39340.      being taken.

  39341. 

  39342.  -- Target Hook: void TARGET_ASM_INTERNAL_LABEL (FILE *STREAM, const

  39343.           char *PREFIX, unsigned long LABELNO)

  39344.      A function to output to the stdio stream STREAM a label whose name

  39345.      is made from the string PREFIX and the number LABELNO.

  39346. 

  39347.      It is absolutely essential that these labels be distinct from the

  39348.      labels used for user-level functions and variables.  Otherwise,

  39349.      certain programs will have name conflicts with internal labels.

  39350. 

  39351.      It is desirable to exclude internal labels from the symbol table of

  39352.      the object file.  Most assemblers have a naming convention for

  39353.      labels that should be excluded; on many systems, the letter 'L' at

  39354.      the beginning of a label has this effect.  You should find out what

  39355.      convention your system uses, and follow it.

  39356. 

  39357.      The default version of this function utilizes

  39358.      'ASM_GENERATE_INTERNAL_LABEL'.

  39359. 

  39360.  -- Macro: ASM_OUTPUT_DEBUG_LABEL (STREAM, PREFIX, NUM)

  39361.      A C statement to output to the stdio stream STREAM a debug info

  39362.      label whose name is made from the string PREFIX and the number NUM.

  39363.      This is useful for VLIW targets, where debug info labels may need

  39364.      to be treated differently than branch target labels.  On some

  39365.      systems, branch target labels must be at the beginning of

  39366.      instruction bundles, but debug info labels can occur in the middle

  39367.      of instruction bundles.

  39368. 

  39369.      If this macro is not defined, then

  39370.      '(*targetm.asm_out.internal_label)' will be used.

  39371. 

  39372.  -- Macro: ASM_GENERATE_INTERNAL_LABEL (STRING, PREFIX, NUM)

  39373.      A C statement to store into the string STRING a label whose name is

  39374.      made from the string PREFIX and the number NUM.

  39375. 

  39376.      This string, when output subsequently by 'assemble_name', should

  39377.      produce the output that '(*targetm.asm_out.internal_label)' would

  39378.      produce with the same PREFIX and NUM.

  39379. 

  39380.      If the string begins with '*', then 'assemble_name' will output the

  39381.      rest of the string unchanged.  It is often convenient for

  39382.      'ASM_GENERATE_INTERNAL_LABEL' to use '*' in this way.  If the

  39383.      string doesn't start with '*', then 'ASM_OUTPUT_LABELREF' gets to

  39384.      output the string, and may change it.  (Of course,

  39385.      'ASM_OUTPUT_LABELREF' is also part of your machine description, so

  39386.      you should know what it does on your machine.)

  39387. 

  39388.  -- Macro: ASM_FORMAT_PRIVATE_NAME (OUTVAR, NAME, NUMBER)

  39389.      A C expression to assign to OUTVAR (which is a variable of type

  39390.      'char *') a newly allocated string made from the string NAME and

  39391.      the number NUMBER, with some suitable punctuation added.  Use

  39392.      'alloca' to get space for the string.

  39393. 

  39394.      The string will be used as an argument to 'ASM_OUTPUT_LABELREF' to

  39395.      produce an assembler label for an internal static variable whose

  39396.      name is NAME.  Therefore, the string must be such as to result in

  39397.      valid assembler code.  The argument NUMBER is different each time

  39398.      this macro is executed; it prevents conflicts between

  39399.      similarly-named internal static variables in different scopes.

  39400. 

  39401.      Ideally this string should not be a valid C identifier, to prevent

  39402.      any conflict with the user's own symbols.  Most assemblers allow

  39403.      periods or percent signs in assembler symbols; putting at least one

  39404.      of these between the name and the number will suffice.

  39405. 

  39406.      If this macro is not defined, a default definition will be provided

  39407.      which is correct for most systems.

  39408. 

  39409.  -- Macro: ASM_OUTPUT_DEF (STREAM, NAME, VALUE)

  39410.      A C statement to output to the stdio stream STREAM assembler code

  39411.      which defines (equates) the symbol NAME to have the value VALUE.

  39412. 

  39413.      If 'SET_ASM_OP' is defined, a default definition is provided which

  39414.      is correct for most systems.

  39415. 

  39416.  -- Macro: ASM_OUTPUT_DEF_FROM_DECLS (STREAM, DECL_OF_NAME,

  39417.           DECL_OF_VALUE)

  39418.      A C statement to output to the stdio stream STREAM assembler code

  39419.      which defines (equates) the symbol whose tree node is DECL_OF_NAME

  39420.      to have the value of the tree node DECL_OF_VALUE.  This macro will

  39421.      be used in preference to 'ASM_OUTPUT_DEF' if it is defined and if

  39422.      the tree nodes are available.

  39423. 

  39424.      If 'SET_ASM_OP' is defined, a default definition is provided which

  39425.      is correct for most systems.

  39426. 

  39427.  -- Macro: TARGET_DEFERRED_OUTPUT_DEFS (DECL_OF_NAME, DECL_OF_VALUE)

  39428.      A C statement that evaluates to true if the assembler code which

  39429.      defines (equates) the symbol whose tree node is DECL_OF_NAME to

  39430.      have the value of the tree node DECL_OF_VALUE should be emitted

  39431.      near the end of the current compilation unit.  The default is to

  39432.      not defer output of defines.  This macro affects defines output by

  39433.      'ASM_OUTPUT_DEF' and 'ASM_OUTPUT_DEF_FROM_DECLS'.

  39434. 

  39435.  -- Macro: ASM_OUTPUT_WEAK_ALIAS (STREAM, NAME, VALUE)

  39436.      A C statement to output to the stdio stream STREAM assembler code

  39437.      which defines (equates) the weak symbol NAME to have the value

  39438.      VALUE.  If VALUE is 'NULL', it defines NAME as an undefined weak

  39439.      symbol.

  39440. 

  39441.      Define this macro if the target only supports weak aliases; define

  39442.      'ASM_OUTPUT_DEF' instead if possible.

  39443. 

  39444.  -- Macro: OBJC_GEN_METHOD_LABEL (BUF, IS_INST, CLASS_NAME, CAT_NAME,

  39445.           SEL_NAME)

  39446.      Define this macro to override the default assembler names used for

  39447.      Objective-C methods.

  39448. 

  39449.      The default name is a unique method number followed by the name of

  39450.      the class (e.g. '_1_Foo').  For methods in categories, the name of

  39451.      the category is also included in the assembler name (e.g.

  39452.      '_1_Foo_Bar').

  39453. 

  39454.      These names are safe on most systems, but make debugging difficult

  39455.      since the method's selector is not present in the name.  Therefore,

  39456.      particular systems define other ways of computing names.

  39457. 

  39458.      BUF is an expression of type 'char *' which gives you a buffer in

  39459.      which to store the name; its length is as long as CLASS_NAME,

  39460.      CAT_NAME and SEL_NAME put together, plus 50 characters extra.

  39461. 

  39462.      The argument IS_INST specifies whether the method is an instance

  39463.      method or a class method; CLASS_NAME is the name of the class;

  39464.      CAT_NAME is the name of the category (or 'NULL' if the method is

  39465.      not in a category); and SEL_NAME is the name of the selector.

  39466. 

  39467.      On systems where the assembler can handle quoted names, you can use

  39468.      this macro to provide more human-readable names.

  39469. 

  39470. 

  39471. File: gccint.info,  Node: Initialization,  Next: Macros for Initialization,  Prev: Label Output,  Up: Assembler Format

  39472. 

  39473. 18.20.5 How Initialization Functions Are Handled

  39474. ------------------------------------------------

  39475. 

  39476. The compiled code for certain languages includes "constructors" (also

  39477. called "initialization routines")--functions to initialize data in the

  39478. program when the program is started.  These functions need to be called

  39479. before the program is "started"--that is to say, before 'main' is

  39480. called.

  39481. 

  39482.  Compiling some languages generates "destructors" (also called

  39483. "termination routines") that should be called when the program

  39484. terminates.

  39485. 

  39486.  To make the initialization and termination functions work, the compiler

  39487. must output something in the assembler code to cause those functions to

  39488. be called at the appropriate time.  When you port the compiler to a new

  39489. system, you need to specify how to do this.

  39490. 

  39491.  There are two major ways that GCC currently supports the execution of

  39492. initialization and termination functions.  Each way has two variants.

  39493. Much of the structure is common to all four variations.

  39494. 

  39495.  The linker must build two lists of these functions--a list of

  39496. initialization functions, called '__CTOR_LIST__', and a list of

  39497. termination functions, called '__DTOR_LIST__'.

  39498. 

  39499.  Each list always begins with an ignored function pointer (which may

  39500. hold 0, -1, or a count of the function pointers after it, depending on

  39501. the environment).  This is followed by a series of zero or more function

  39502. pointers to constructors (or destructors), followed by a function

  39503. pointer containing zero.

  39504. 

  39505.  Depending on the operating system and its executable file format,

  39506. either 'crtstuff.c' or 'libgcc2.c' traverses these lists at startup time

  39507. and exit time.  Constructors are called in reverse order of the list;

  39508. destructors in forward order.

  39509. 

  39510.  The best way to handle static constructors works only for object file

  39511. formats which provide arbitrarily-named sections.  A section is set

  39512. aside for a list of constructors, and another for a list of destructors.

  39513. Traditionally these are called '.ctors' and '.dtors'.  Each object file

  39514. that defines an initialization function also puts a word in the

  39515. constructor section to point to that function.  The linker accumulates

  39516. all these words into one contiguous '.ctors' section.  Termination

  39517. functions are handled similarly.

  39518. 

  39519.  This method will be chosen as the default by 'target-def.h' if

  39520. 'TARGET_ASM_NAMED_SECTION' is defined.  A target that does not support

  39521. arbitrary sections, but does support special designated constructor and

  39522. destructor sections may define 'CTORS_SECTION_ASM_OP' and

  39523. 'DTORS_SECTION_ASM_OP' to achieve the same effect.

  39524. 

  39525.  When arbitrary sections are available, there are two variants,

  39526. depending upon how the code in 'crtstuff.c' is called.  On systems that

  39527. support a ".init" section which is executed at program startup, parts of

  39528. 'crtstuff.c' are compiled into that section.  The program is linked by

  39529. the 'gcc' driver like this:

  39530. 

  39531.      ld -o OUTPUT_FILE crti.o crtbegin.o ... -lgcc crtend.o crtn.o

  39532. 

  39533.  The prologue of a function ('__init') appears in the '.init' section of

  39534. 'crti.o'; the epilogue appears in 'crtn.o'.  Likewise for the function

  39535. '__fini' in the ".fini" section.  Normally these files are provided by

  39536. the operating system or by the GNU C library, but are provided by GCC

  39537. for a few targets.

  39538. 

  39539.  The objects 'crtbegin.o' and 'crtend.o' are (for most targets) compiled

  39540. from 'crtstuff.c'.  They contain, among other things, code fragments

  39541. within the '.init' and '.fini' sections that branch to routines in the

  39542. '.text' section.  The linker will pull all parts of a section together,

  39543. which results in a complete '__init' function that invokes the routines

  39544. we need at startup.

  39545. 

  39546.  To use this variant, you must define the 'INIT_SECTION_ASM_OP' macro

  39547. properly.

  39548. 

  39549.  If no init section is available, when GCC compiles any function called

  39550. 'main' (or more accurately, any function designated as a program entry

  39551. point by the language front end calling 'expand_main_function'), it

  39552. inserts a procedure call to '__main' as the first executable code after

  39553. the function prologue.  The '__main' function is defined in 'libgcc2.c'

  39554. and runs the global constructors.

  39555. 

  39556.  In file formats that don't support arbitrary sections, there are again

  39557. two variants.  In the simplest variant, the GNU linker (GNU 'ld') and an

  39558. 'a.out' format must be used.  In this case, 'TARGET_ASM_CONSTRUCTOR' is

  39559. defined to produce a '.stabs' entry of type 'N_SETT', referencing the

  39560. name '__CTOR_LIST__', and with the address of the void function

  39561. containing the initialization code as its value.  The GNU linker

  39562. recognizes this as a request to add the value to a "set"; the values are

  39563. accumulated, and are eventually placed in the executable as a vector in

  39564. the format described above, with a leading (ignored) count and a

  39565. trailing zero element.  'TARGET_ASM_DESTRUCTOR' is handled similarly.

  39566. Since no init section is available, the absence of 'INIT_SECTION_ASM_OP'

  39567. causes the compilation of 'main' to call '__main' as above, starting the

  39568. initialization process.

  39569. 

  39570.  The last variant uses neither arbitrary sections nor the GNU linker.

  39571. This is preferable when you want to do dynamic linking and when using

  39572. file formats which the GNU linker does not support, such as 'ECOFF'.  In

  39573. this case, 'TARGET_HAVE_CTORS_DTORS' is false, initialization and

  39574. termination functions are recognized simply by their names.  This

  39575. requires an extra program in the linkage step, called 'collect2'.  This

  39576. program pretends to be the linker, for use with GCC; it does its job by

  39577. running the ordinary linker, but also arranges to include the vectors of

  39578. initialization and termination functions.  These functions are called

  39579. via '__main' as described above.  In order to use this method,

  39580. 'use_collect2' must be defined in the target in 'config.gcc'.

  39581. 

  39582.  The following section describes the specific macros that control and

  39583. customize the handling of initialization and termination functions.

  39584. 

  39585. 

  39586. File: gccint.info,  Node: Macros for Initialization,  Next: Instruction Output,  Prev: Initialization,  Up: Assembler Format

  39587. 

  39588. 18.20.6 Macros Controlling Initialization Routines

  39589. --------------------------------------------------

  39590. 

  39591. Here are the macros that control how the compiler handles initialization

  39592. and termination functions:

  39593. 

  39594.  -- Macro: INIT_SECTION_ASM_OP

  39595.      If defined, a C string constant, including spacing, for the

  39596.      assembler operation to identify the following data as

  39597.      initialization code.  If not defined, GCC will assume such a

  39598.      section does not exist.  When you are using special sections for

  39599.      initialization and termination functions, this macro also controls

  39600.      how 'crtstuff.c' and 'libgcc2.c' arrange to run the initialization

  39601.      functions.

  39602. 

  39603.  -- Macro: HAS_INIT_SECTION

  39604.      If defined, 'main' will not call '__main' as described above.  This

  39605.      macro should be defined for systems that control start-up code on a

  39606.      symbol-by-symbol basis, such as OSF/1, and should not be defined

  39607.      explicitly for systems that support 'INIT_SECTION_ASM_OP'.

  39608. 

  39609.  -- Macro: LD_INIT_SWITCH

  39610.      If defined, a C string constant for a switch that tells the linker

  39611.      that the following symbol is an initialization routine.

  39612. 

  39613.  -- Macro: LD_FINI_SWITCH

  39614.      If defined, a C string constant for a switch that tells the linker

  39615.      that the following symbol is a finalization routine.

  39616. 

  39617.  -- Macro: COLLECT_SHARED_INIT_FUNC (STREAM, FUNC)

  39618.      If defined, a C statement that will write a function that can be

  39619.      automatically called when a shared library is loaded.  The function

  39620.      should call FUNC, which takes no arguments.  If not defined, and

  39621.      the object format requires an explicit initialization function,

  39622.      then a function called '_GLOBAL__DI' will be generated.

  39623. 

  39624.      This function and the following one are used by collect2 when

  39625.      linking a shared library that needs constructors or destructors, or

  39626.      has DWARF2 exception tables embedded in the code.

  39627. 

  39628.  -- Macro: COLLECT_SHARED_FINI_FUNC (STREAM, FUNC)

  39629.      If defined, a C statement that will write a function that can be

  39630.      automatically called when a shared library is unloaded.  The

  39631.      function should call FUNC, which takes no arguments.  If not

  39632.      defined, and the object format requires an explicit finalization

  39633.      function, then a function called '_GLOBAL__DD' will be generated.

  39634. 

  39635.  -- Macro: INVOKE__main

  39636.      If defined, 'main' will call '__main' despite the presence of

  39637.      'INIT_SECTION_ASM_OP'.  This macro should be defined for systems

  39638.      where the init section is not actually run automatically, but is

  39639.      still useful for collecting the lists of constructors and

  39640.      destructors.

  39641. 

  39642.  -- Macro: SUPPORTS_INIT_PRIORITY

  39643.      If nonzero, the C++ 'init_priority' attribute is supported and the

  39644.      compiler should emit instructions to control the order of

  39645.      initialization of objects.  If zero, the compiler will issue an

  39646.      error message upon encountering an 'init_priority' attribute.

  39647. 

  39648.  -- Target Hook: bool TARGET_HAVE_CTORS_DTORS

  39649.      This value is true if the target supports some "native" method of

  39650.      collecting constructors and destructors to be run at startup and

  39651.      exit.  It is false if we must use 'collect2'.

  39652. 

  39653.  -- Target Hook: void TARGET_ASM_CONSTRUCTOR (rtx SYMBOL, int PRIORITY)

  39654.      If defined, a function that outputs assembler code to arrange to

  39655.      call the function referenced by SYMBOL at initialization time.

  39656. 

  39657.      Assume that SYMBOL is a 'SYMBOL_REF' for a function taking no

  39658.      arguments and with no return value.  If the target supports

  39659.      initialization priorities, PRIORITY is a value between 0 and

  39660.      'MAX_INIT_PRIORITY'; otherwise it must be 'DEFAULT_INIT_PRIORITY'.

  39661. 

  39662.      If this macro is not defined by the target, a suitable default will

  39663.      be chosen if (1) the target supports arbitrary section names, (2)

  39664.      the target defines 'CTORS_SECTION_ASM_OP', or (3) 'USE_COLLECT2' is

  39665.      not defined.

  39666. 

  39667.  -- Target Hook: void TARGET_ASM_DESTRUCTOR (rtx SYMBOL, int PRIORITY)

  39668.      This is like 'TARGET_ASM_CONSTRUCTOR' but used for termination

  39669.      functions rather than initialization functions.

  39670. 

  39671.  If 'TARGET_HAVE_CTORS_DTORS' is true, the initialization routine

  39672. generated for the generated object file will have static linkage.

  39673. 

  39674.  If your system uses 'collect2' as the means of processing constructors,

  39675. then that program normally uses 'nm' to scan an object file for

  39676. constructor functions to be called.

  39677. 

  39678.  On certain kinds of systems, you can define this macro to make

  39679. 'collect2' work faster (and, in some cases, make it work at all):

  39680. 

  39681.  -- Macro: OBJECT_FORMAT_COFF

  39682.      Define this macro if the system uses COFF (Common Object File

  39683.      Format) object files, so that 'collect2' can assume this format and

  39684.      scan object files directly for dynamic constructor/destructor

  39685.      functions.

  39686. 

  39687.      This macro is effective only in a native compiler; 'collect2' as

  39688.      part of a cross compiler always uses 'nm' for the target machine.

  39689. 

  39690.  -- Macro: REAL_NM_FILE_NAME

  39691.      Define this macro as a C string constant containing the file name

  39692.      to use to execute 'nm'.  The default is to search the path normally

  39693.      for 'nm'.

  39694. 

  39695.  -- Macro: NM_FLAGS

  39696.      'collect2' calls 'nm' to scan object files for static constructors

  39697.      and destructors and LTO info.  By default, '-n' is passed.  Define

  39698.      'NM_FLAGS' to a C string constant if other options are needed to

  39699.      get the same output format as GNU 'nm -n' produces.

  39700. 

  39701.  If your system supports shared libraries and has a program to list the

  39702. dynamic dependencies of a given library or executable, you can define

  39703. these macros to enable support for running initialization and

  39704. termination functions in shared libraries:

  39705. 

  39706.  -- Macro: LDD_SUFFIX

  39707.      Define this macro to a C string constant containing the name of the

  39708.      program which lists dynamic dependencies, like 'ldd' under SunOS 4.

  39709. 

  39710.  -- Macro: PARSE_LDD_OUTPUT (PTR)

  39711.      Define this macro to be C code that extracts filenames from the

  39712.      output of the program denoted by 'LDD_SUFFIX'.  PTR is a variable

  39713.      of type 'char *' that points to the beginning of a line of output

  39714.      from 'LDD_SUFFIX'.  If the line lists a dynamic dependency, the

  39715.      code must advance PTR to the beginning of the filename on that

  39716.      line.  Otherwise, it must set PTR to 'NULL'.

  39717. 

  39718.  -- Macro: SHLIB_SUFFIX

  39719.      Define this macro to a C string constant containing the default

  39720.      shared library extension of the target (e.g., '".so"').  'collect2'

  39721.      strips version information after this suffix when generating global

  39722.      constructor and destructor names.  This define is only needed on

  39723.      targets that use 'collect2' to process constructors and

  39724.      destructors.

  39725. 

  39726. 

  39727. File: gccint.info,  Node: Instruction Output,  Next: Dispatch Tables,  Prev: Macros for Initialization,  Up: Assembler Format

  39728. 

  39729. 18.20.7 Output of Assembler Instructions

  39730. ----------------------------------------

  39731. 

  39732. This describes assembler instruction output.

  39733. 

  39734.  -- Macro: REGISTER_NAMES

  39735.      A C initializer containing the assembler's names for the machine

  39736.      registers, each one as a C string constant.  This is what

  39737.      translates register numbers in the compiler into assembler

  39738.      language.

  39739. 

  39740.  -- Macro: ADDITIONAL_REGISTER_NAMES

  39741.      If defined, a C initializer for an array of structures containing a

  39742.      name and a register number.  This macro defines additional names

  39743.      for hard registers, thus allowing the 'asm' option in declarations

  39744.      to refer to registers using alternate names.

  39745. 

  39746.  -- Macro: OVERLAPPING_REGISTER_NAMES

  39747.      If defined, a C initializer for an array of structures containing a

  39748.      name, a register number and a count of the number of consecutive

  39749.      machine registers the name overlaps.  This macro defines additional

  39750.      names for hard registers, thus allowing the 'asm' option in

  39751.      declarations to refer to registers using alternate names.  Unlike

  39752.      'ADDITIONAL_REGISTER_NAMES', this macro should be used when the

  39753.      register name implies multiple underlying registers.

  39754. 

  39755.      This macro should be used when it is important that a clobber in an

  39756.      'asm' statement clobbers all the underlying values implied by the

  39757.      register name.  For example, on ARM, clobbering the

  39758.      double-precision VFP register "d0" implies clobbering both

  39759.      single-precision registers "s0" and "s1".

  39760. 

  39761.  -- Macro: ASM_OUTPUT_OPCODE (STREAM, PTR)

  39762.      Define this macro if you are using an unusual assembler that

  39763.      requires different names for the machine instructions.

  39764. 

  39765.      The definition is a C statement or statements which output an

  39766.      assembler instruction opcode to the stdio stream STREAM.  The

  39767.      macro-operand PTR is a variable of type 'char *' which points to

  39768.      the opcode name in its "internal" form--the form that is written in

  39769.      the machine description.  The definition should output the opcode

  39770.      name to STREAM, performing any translation you desire, and

  39771.      increment the variable PTR to point at the end of the opcode so

  39772.      that it will not be output twice.

  39773. 

  39774.      In fact, your macro definition may process less than the entire

  39775.      opcode name, or more than the opcode name; but if you want to

  39776.      process text that includes '%'-sequences to substitute operands,

  39777.      you must take care of the substitution yourself.  Just be sure to

  39778.      increment PTR over whatever text should not be output normally.

  39779. 

  39780.      If you need to look at the operand values, they can be found as the

  39781.      elements of 'recog_data.operand'.

  39782. 

  39783.      If the macro definition does nothing, the instruction is output in

  39784.      the usual way.

  39785. 

  39786.  -- Macro: FINAL_PRESCAN_INSN (INSN, OPVEC, NOPERANDS)

  39787.      If defined, a C statement to be executed just prior to the output

  39788.      of assembler code for INSN, to modify the extracted operands so

  39789.      they will be output differently.

  39790. 

  39791.      Here the argument OPVEC is the vector containing the operands

  39792.      extracted from INSN, and NOPERANDS is the number of elements of the

  39793.      vector which contain meaningful data for this insn.  The contents

  39794.      of this vector are what will be used to convert the insn template

  39795.      into assembler code, so you can change the assembler output by

  39796.      changing the contents of the vector.

  39797. 

  39798.      This macro is useful when various assembler syntaxes share a single

  39799.      file of instruction patterns; by defining this macro differently,

  39800.      you can cause a large class of instructions to be output

  39801.      differently (such as with rearranged operands).  Naturally,

  39802.      variations in assembler syntax affecting individual insn patterns

  39803.      ought to be handled by writing conditional output routines in those

  39804.      patterns.

  39805. 

  39806.      If this macro is not defined, it is equivalent to a null statement.

  39807. 

  39808.  -- Target Hook: void TARGET_ASM_FINAL_POSTSCAN_INSN (FILE *FILE,

  39809.           rtx_insn *INSN, rtx *OPVEC, int NOPERANDS)

  39810.      If defined, this target hook is a function which is executed just

  39811.      after the output of assembler code for INSN, to change the mode of

  39812.      the assembler if necessary.

  39813. 

  39814.      Here the argument OPVEC is the vector containing the operands

  39815.      extracted from INSN, and NOPERANDS is the number of elements of the

  39816.      vector which contain meaningful data for this insn.  The contents

  39817.      of this vector are what was used to convert the insn template into

  39818.      assembler code, so you can change the assembler mode by checking

  39819.      the contents of the vector.

  39820. 

  39821.  -- Macro: PRINT_OPERAND (STREAM, X, CODE)

  39822.      A C compound statement to output to stdio stream STREAM the

  39823.      assembler syntax for an instruction operand X.  X is an RTL

  39824.      expression.

  39825. 

  39826.      CODE is a value that can be used to specify one of several ways of

  39827.      printing the operand.  It is used when identical operands must be

  39828.      printed differently depending on the context.  CODE comes from the

  39829.      '%' specification that was used to request printing of the operand.

  39830.      If the specification was just '%DIGIT' then CODE is 0; if the

  39831.      specification was '%LTR DIGIT' then CODE is the ASCII code for LTR.

  39832. 

  39833.      If X is a register, this macro should print the register's name.

  39834.      The names can be found in an array 'reg_names' whose type is 'char

  39835.      *[]'.  'reg_names' is initialized from 'REGISTER_NAMES'.

  39836. 

  39837.      When the machine description has a specification '%PUNCT' (a '%'

  39838.      followed by a punctuation character), this macro is called with a

  39839.      null pointer for X and the punctuation character for CODE.

  39840. 

  39841.  -- Macro: PRINT_OPERAND_PUNCT_VALID_P (CODE)

  39842.      A C expression which evaluates to true if CODE is a valid

  39843.      punctuation character for use in the 'PRINT_OPERAND' macro.  If

  39844.      'PRINT_OPERAND_PUNCT_VALID_P' is not defined, it means that no

  39845.      punctuation characters (except for the standard one, '%') are used

  39846.      in this way.

  39847. 

  39848.  -- Macro: PRINT_OPERAND_ADDRESS (STREAM, X)

  39849.      A C compound statement to output to stdio stream STREAM the

  39850.      assembler syntax for an instruction operand that is a memory

  39851.      reference whose address is X.  X is an RTL expression.

  39852. 

  39853.      On some machines, the syntax for a symbolic address depends on the

  39854.      section that the address refers to.  On these machines, define the

  39855.      hook 'TARGET_ENCODE_SECTION_INFO' to store the information into the

  39856.      'symbol_ref', and then check for it here.  *Note Assembler

  39857.      Format::.

  39858. 

  39859.  -- Macro: DBR_OUTPUT_SEQEND (FILE)

  39860.      A C statement, to be executed after all slot-filler instructions

  39861.      have been output.  If necessary, call 'dbr_sequence_length' to

  39862.      determine the number of slots filled in a sequence (zero if not

  39863.      currently outputting a sequence), to decide how many no-ops to

  39864.      output, or whatever.

  39865. 

  39866.      Don't define this macro if it has nothing to do, but it is helpful

  39867.      in reading assembly output if the extent of the delay sequence is

  39868.      made explicit (e.g. with white space).

  39869. 

  39870.  Note that output routines for instructions with delay slots must be

  39871. prepared to deal with not being output as part of a sequence (i.e. when

  39872. the scheduling pass is not run, or when no slot fillers could be found.)

  39873. The variable 'final_sequence' is null when not processing a sequence,

  39874. otherwise it contains the 'sequence' rtx being output.

  39875. 

  39876.  -- Macro: REGISTER_PREFIX

  39877.  -- Macro: LOCAL_LABEL_PREFIX

  39878.  -- Macro: USER_LABEL_PREFIX

  39879.  -- Macro: IMMEDIATE_PREFIX

  39880.      If defined, C string expressions to be used for the '%R', '%L',

  39881.      '%U', and '%I' options of 'asm_fprintf' (see 'final.c').  These are

  39882.      useful when a single 'md' file must support multiple assembler

  39883.      formats.  In that case, the various 'tm.h' files can define these

  39884.      macros differently.

  39885. 

  39886.  -- Macro: ASM_FPRINTF_EXTENSIONS (FILE, ARGPTR, FORMAT)

  39887.      If defined this macro should expand to a series of 'case'

  39888.      statements which will be parsed inside the 'switch' statement of

  39889.      the 'asm_fprintf' function.  This allows targets to define extra

  39890.      printf formats which may useful when generating their assembler

  39891.      statements.  Note that uppercase letters are reserved for future

  39892.      generic extensions to asm_fprintf, and so are not available to

  39893.      target specific code.  The output file is given by the parameter

  39894.      FILE.  The varargs input pointer is ARGPTR and the rest of the

  39895.      format string, starting the character after the one that is being

  39896.      switched upon, is pointed to by FORMAT.

  39897. 

  39898.  -- Macro: ASSEMBLER_DIALECT

  39899.      If your target supports multiple dialects of assembler language

  39900.      (such as different opcodes), define this macro as a C expression

  39901.      that gives the numeric index of the assembler language dialect to

  39902.      use, with zero as the first variant.

  39903. 

  39904.      If this macro is defined, you may use constructs of the form

  39905.           '{option0|option1|option2...}'

  39906.      in the output templates of patterns (*note Output Template::) or in

  39907.      the first argument of 'asm_fprintf'.  This construct outputs

  39908.      'option0', 'option1', 'option2', etc., if the value of

  39909.      'ASSEMBLER_DIALECT' is zero, one, two, etc.  Any special characters

  39910.      within these strings retain their usual meaning.  If there are

  39911.      fewer alternatives within the braces than the value of

  39912.      'ASSEMBLER_DIALECT', the construct outputs nothing.  If it's needed

  39913.      to print curly braces or '|' character in assembler output

  39914.      directly, '%{', '%}' and '%|' can be used.

  39915. 

  39916.      If you do not define this macro, the characters '{', '|' and '}' do

  39917.      not have any special meaning when used in templates or operands to

  39918.      'asm_fprintf'.

  39919. 

  39920.      Define the macros 'REGISTER_PREFIX', 'LOCAL_LABEL_PREFIX',

  39921.      'USER_LABEL_PREFIX' and 'IMMEDIATE_PREFIX' if you can express the

  39922.      variations in assembler language syntax with that mechanism.

  39923.      Define 'ASSEMBLER_DIALECT' and use the '{option0|option1}' syntax

  39924.      if the syntax variant are larger and involve such things as

  39925.      different opcodes or operand order.

  39926. 

  39927.  -- Macro: ASM_OUTPUT_REG_PUSH (STREAM, REGNO)

  39928.      A C expression to output to STREAM some assembler code which will

  39929.      push hard register number REGNO onto the stack.  The code need not

  39930.      be optimal, since this macro is used only when profiling.

  39931. 

  39932.  -- Macro: ASM_OUTPUT_REG_POP (STREAM, REGNO)

  39933.      A C expression to output to STREAM some assembler code which will

  39934.      pop hard register number REGNO off of the stack.  The code need not

  39935.      be optimal, since this macro is used only when profiling.

  39936. 

  39937. 

  39938. File: gccint.info,  Node: Dispatch Tables,  Next: Exception Region Output,  Prev: Instruction Output,  Up: Assembler Format

  39939. 

  39940. 18.20.8 Output of Dispatch Tables

  39941. ---------------------------------

  39942. 

  39943. This concerns dispatch tables.

  39944. 

  39945.  -- Macro: ASM_OUTPUT_ADDR_DIFF_ELT (STREAM, BODY, VALUE, REL)

  39946.      A C statement to output to the stdio stream STREAM an assembler

  39947.      pseudo-instruction to generate a difference between two labels.

  39948.      VALUE and REL are the numbers of two internal labels.  The

  39949.      definitions of these labels are output using

  39950.      '(*targetm.asm_out.internal_label)', and they must be printed in

  39951.      the same way here.  For example,

  39952. 

  39953.           fprintf (STREAM, "\t.word L%d-L%d\n",

  39954.                    VALUE, REL)

  39955. 

  39956.      You must provide this macro on machines where the addresses in a

  39957.      dispatch table are relative to the table's own address.  If

  39958.      defined, GCC will also use this macro on all machines when

  39959.      producing PIC.  BODY is the body of the 'ADDR_DIFF_VEC'; it is

  39960.      provided so that the mode and flags can be read.

  39961. 

  39962.  -- Macro: ASM_OUTPUT_ADDR_VEC_ELT (STREAM, VALUE)

  39963.      This macro should be provided on machines where the addresses in a

  39964.      dispatch table are absolute.

  39965. 

  39966.      The definition should be a C statement to output to the stdio

  39967.      stream STREAM an assembler pseudo-instruction to generate a

  39968.      reference to a label.  VALUE is the number of an internal label

  39969.      whose definition is output using

  39970.      '(*targetm.asm_out.internal_label)'.  For example,

  39971. 

  39972.           fprintf (STREAM, "\t.word L%d\n", VALUE)

  39973. 

  39974.  -- Macro: ASM_OUTPUT_CASE_LABEL (STREAM, PREFIX, NUM, TABLE)

  39975.      Define this if the label before a jump-table needs to be output

  39976.      specially.  The first three arguments are the same as for

  39977.      '(*targetm.asm_out.internal_label)'; the fourth argument is the

  39978.      jump-table which follows (a 'jump_table_data' containing an

  39979.      'addr_vec' or 'addr_diff_vec').

  39980. 

  39981.      This feature is used on system V to output a 'swbeg' statement for

  39982.      the table.

  39983. 

  39984.      If this macro is not defined, these labels are output with

  39985.      '(*targetm.asm_out.internal_label)'.

  39986. 

  39987.  -- Macro: ASM_OUTPUT_CASE_END (STREAM, NUM, TABLE)

  39988.      Define this if something special must be output at the end of a

  39989.      jump-table.  The definition should be a C statement to be executed

  39990.      after the assembler code for the table is written.  It should write

  39991.      the appropriate code to stdio stream STREAM.  The argument TABLE is

  39992.      the jump-table insn, and NUM is the label-number of the preceding

  39993.      label.

  39994. 

  39995.      If this macro is not defined, nothing special is output at the end

  39996.      of the jump-table.

  39997. 

  39998.  -- Target Hook: void TARGET_ASM_POST_CFI_STARTPROC (FILE *, TREE)

  39999.      This target hook is used to emit assembly strings required by the

  40000.      target after the .cfi_startproc directive.  The first argument is

  40001.      the file stream to write the strings to and the second argument is

  40002.      the function's declaration.  The expected use is to add more .cfi_*

  40003.      directives.

  40004. 

  40005.      The default is to not output any assembly strings.

  40006. 

  40007.  -- Target Hook: void TARGET_ASM_EMIT_UNWIND_LABEL (FILE *STREAM, tree

  40008.           DECL, int FOR_EH, int EMPTY)

  40009.      This target hook emits a label at the beginning of each FDE.  It

  40010.      should be defined on targets where FDEs need special labels, and it

  40011.      should write the appropriate label, for the FDE associated with the

  40012.      function declaration DECL, to the stdio stream STREAM.  The third

  40013.      argument, FOR_EH, is a boolean: true if this is for an exception

  40014.      table.  The fourth argument, EMPTY, is a boolean: true if this is a

  40015.      placeholder label for an omitted FDE.

  40016. 

  40017.      The default is that FDEs are not given nonlocal labels.

  40018. 

  40019.  -- Target Hook: void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (FILE *STREAM)

  40020.      This target hook emits a label at the beginning of the exception

  40021.      table.  It should be defined on targets where it is desirable for

  40022.      the table to be broken up according to function.

  40023. 

  40024.      The default is that no label is emitted.

  40025. 

  40026.  -- Target Hook: void TARGET_ASM_EMIT_EXCEPT_PERSONALITY (rtx

  40027.           PERSONALITY)

  40028.      If the target implements 'TARGET_ASM_UNWIND_EMIT', this hook may be

  40029.      used to emit a directive to install a personality hook into the

  40030.      unwind info.  This hook should not be used if dwarf2 unwind info is

  40031.      used.

  40032. 

  40033.  -- Target Hook: void TARGET_ASM_UNWIND_EMIT (FILE *STREAM, rtx_insn

  40034.           *INSN)

  40035.      This target hook emits assembly directives required to unwind the

  40036.      given instruction.  This is only used when

  40037.      'TARGET_EXCEPT_UNWIND_INFO' returns 'UI_TARGET'.

  40038. 

  40039.  -- Target Hook: bool TARGET_ASM_UNWIND_EMIT_BEFORE_INSN

  40040.      True if the 'TARGET_ASM_UNWIND_EMIT' hook should be called before

  40041.      the assembly for INSN has been emitted, false if the hook should be

  40042.      called afterward.

  40043. 

  40044. 

  40045. File: gccint.info,  Node: Exception Region Output,  Next: Alignment Output,  Prev: Dispatch Tables,  Up: Assembler Format

  40046. 

  40047. 18.20.9 Assembler Commands for Exception Regions

  40048. ------------------------------------------------

  40049. 

  40050. This describes commands marking the start and the end of an exception

  40051. region.

  40052. 

  40053.  -- Macro: EH_FRAME_SECTION_NAME

  40054.      If defined, a C string constant for the name of the section

  40055.      containing exception handling frame unwind information.  If not

  40056.      defined, GCC will provide a default definition if the target

  40057.      supports named sections.  'crtstuff.c' uses this macro to switch to

  40058.      the appropriate section.

  40059. 

  40060.      You should define this symbol if your target supports DWARF 2 frame

  40061.      unwind information and the default definition does not work.

  40062. 

  40063.  -- Macro: EH_FRAME_THROUGH_COLLECT2

  40064.      If defined, DWARF 2 frame unwind information will identified by

  40065.      specially named labels.  The collect2 process will locate these

  40066.      labels and generate code to register the frames.

  40067. 

  40068.      This might be necessary, for instance, if the system linker will

  40069.      not place the eh_frames in-between the sentinals from 'crtstuff.c',

  40070.      or if the system linker does garbage collection and sections cannot

  40071.      be marked as not to be collected.

  40072. 

  40073.  -- Macro: EH_TABLES_CAN_BE_READ_ONLY

  40074.      Define this macro to 1 if your target is such that no frame unwind

  40075.      information encoding used with non-PIC code will ever require a

  40076.      runtime relocation, but the linker may not support merging

  40077.      read-only and read-write sections into a single read-write section.

  40078. 

  40079.  -- Macro: MASK_RETURN_ADDR

  40080.      An rtx used to mask the return address found via 'RETURN_ADDR_RTX',

  40081.      so that it does not contain any extraneous set bits in it.

  40082. 

  40083.  -- Macro: DWARF2_UNWIND_INFO

  40084.      Define this macro to 0 if your target supports DWARF 2 frame unwind

  40085.      information, but it does not yet work with exception handling.

  40086.      Otherwise, if your target supports this information (if it defines

  40087.      'INCOMING_RETURN_ADDR_RTX' and 'OBJECT_FORMAT_ELF'), GCC will

  40088.      provide a default definition of 1.

  40089. 

  40090.  -- Common Target Hook: enum unwind_info_type TARGET_EXCEPT_UNWIND_INFO

  40091.           (struct gcc_options *OPTS)

  40092.      This hook defines the mechanism that will be used for exception

  40093.      handling by the target.  If the target has ABI specified unwind

  40094.      tables, the hook should return 'UI_TARGET'.  If the target is to

  40095.      use the 'setjmp'/'longjmp'-based exception handling scheme, the

  40096.      hook should return 'UI_SJLJ'.  If the target supports DWARF 2 frame

  40097.      unwind information, the hook should return 'UI_DWARF2'.

  40098. 

  40099.      A target may, if exceptions are disabled, choose to return

  40100.      'UI_NONE'.  This may end up simplifying other parts of

  40101.      target-specific code.  The default implementation of this hook

  40102.      never returns 'UI_NONE'.

  40103. 

  40104.      Note that the value returned by this hook should be constant.  It

  40105.      should not depend on anything except the command-line switches

  40106.      described by OPTS.  In particular, the setting 'UI_SJLJ' must be

  40107.      fixed at compiler start-up as C pre-processor macros and builtin

  40108.      functions related to exception handling are set up depending on

  40109.      this setting.

  40110. 

  40111.      The default implementation of the hook first honors the

  40112.      '--enable-sjlj-exceptions' configure option, then

  40113.      'DWARF2_UNWIND_INFO', and finally defaults to 'UI_SJLJ'.  If

  40114.      'DWARF2_UNWIND_INFO' depends on command-line options, the target

  40115.      must define this hook so that OPTS is used correctly.

  40116. 

  40117.  -- Common Target Hook: bool TARGET_UNWIND_TABLES_DEFAULT

  40118.      This variable should be set to 'true' if the target ABI requires

  40119.      unwinding tables even when exceptions are not used.  It must not be

  40120.      modified by command-line option processing.

  40121. 

  40122.  -- Macro: DONT_USE_BUILTIN_SETJMP

  40123.      Define this macro to 1 if the 'setjmp'/'longjmp'-based scheme

  40124.      should use the 'setjmp'/'longjmp' functions from the C library

  40125.      instead of the '__builtin_setjmp'/'__builtin_longjmp' machinery.

  40126. 

  40127.  -- Macro: JMP_BUF_SIZE

  40128.      This macro has no effect unless 'DONT_USE_BUILTIN_SETJMP' is also

  40129.      defined.  Define this macro if the default size of 'jmp_buf' buffer

  40130.      for the 'setjmp'/'longjmp'-based exception handling mechanism is

  40131.      not large enough, or if it is much too large.  The default size is

  40132.      'FIRST_PSEUDO_REGISTER * sizeof(void *)'.

  40133. 

  40134.  -- Macro: DWARF_CIE_DATA_ALIGNMENT

  40135.      This macro need only be defined if the target might save registers

  40136.      in the function prologue at an offset to the stack pointer that is

  40137.      not aligned to 'UNITS_PER_WORD'.  The definition should be the

  40138.      negative minimum alignment if 'STACK_GROWS_DOWNWARD' is true, and

  40139.      the positive minimum alignment otherwise.  *Note DWARF::.  Only

  40140.      applicable if the target supports DWARF 2 frame unwind information.

  40141. 

  40142.  -- Target Hook: bool TARGET_TERMINATE_DW2_EH_FRAME_INFO

  40143.      Contains the value true if the target should add a zero word onto

  40144.      the end of a Dwarf-2 frame info section when used for exception

  40145.      handling.  Default value is false if 'EH_FRAME_SECTION_NAME' is

  40146.      defined, and true otherwise.

  40147. 

  40148.  -- Target Hook: rtx TARGET_DWARF_REGISTER_SPAN (rtx REG)

  40149.      Given a register, this hook should return a parallel of registers

  40150.      to represent where to find the register pieces.  Define this hook

  40151.      if the register and its mode are represented in Dwarf in

  40152.      non-contiguous locations, or if the register should be represented

  40153.      in more than one register in Dwarf.  Otherwise, this hook should

  40154.      return 'NULL_RTX'.  If not defined, the default is to return

  40155.      'NULL_RTX'.

  40156. 

  40157.  -- Target Hook: machine_mode TARGET_DWARF_FRAME_REG_MODE (int REGNO)

  40158.      Given a register, this hook should return the mode which the

  40159.      corresponding Dwarf frame register should have.  This is normally

  40160.      used to return a smaller mode than the raw mode to prevent call

  40161.      clobbered parts of a register altering the frame register size

  40162. 

  40163.  -- Target Hook: void TARGET_INIT_DWARF_REG_SIZES_EXTRA (tree ADDRESS)

  40164.      If some registers are represented in Dwarf-2 unwind information in

  40165.      multiple pieces, define this hook to fill in information about the

  40166.      sizes of those pieces in the table used by the unwinder at runtime.

  40167.      It will be called by 'expand_builtin_init_dwarf_reg_sizes' after

  40168.      filling in a single size corresponding to each hard register;

  40169.      ADDRESS is the address of the table.

  40170. 

  40171.  -- Target Hook: bool TARGET_ASM_TTYPE (rtx SYM)

  40172.      This hook is used to output a reference from a frame unwinding

  40173.      table to the type_info object identified by SYM.  It should return

  40174.      'true' if the reference was output.  Returning 'false' will cause

  40175.      the reference to be output using the normal Dwarf2 routines.

  40176. 

  40177.  -- Target Hook: bool TARGET_ARM_EABI_UNWINDER

  40178.      This flag should be set to 'true' on targets that use an ARM EABI

  40179.      based unwinding library, and 'false' on other targets.  This

  40180.      effects the format of unwinding tables, and how the unwinder in

  40181.      entered after running a cleanup.  The default is 'false'.

  40182. 

  40183. 

  40184. File: gccint.info,  Node: Alignment Output,  Prev: Exception Region Output,  Up: Assembler Format

  40185. 

  40186. 18.20.10 Assembler Commands for Alignment

  40187. -----------------------------------------

  40188. 

  40189. This describes commands for alignment.

  40190. 

  40191.  -- Macro: JUMP_ALIGN (LABEL)

  40192.      The alignment (log base 2) to put in front of LABEL, which is a

  40193.      common destination of jumps and has no fallthru incoming edge.

  40194. 

  40195.      This macro need not be defined if you don't want any special

  40196.      alignment to be done at such a time.  Most machine descriptions do

  40197.      not currently define the macro.

  40198. 

  40199.      Unless it's necessary to inspect the LABEL parameter, it is better

  40200.      to set the variable ALIGN_JUMPS in the target's

  40201.      'TARGET_OPTION_OVERRIDE'.  Otherwise, you should try to honor the

  40202.      user's selection in ALIGN_JUMPS in a 'JUMP_ALIGN' implementation.

  40203. 

  40204.  -- Macro: LABEL_ALIGN_AFTER_BARRIER (LABEL)

  40205.      The alignment (log base 2) to put in front of LABEL, which follows

  40206.      a 'BARRIER'.

  40207. 

  40208.      This macro need not be defined if you don't want any special

  40209.      alignment to be done at such a time.  Most machine descriptions do

  40210.      not currently define the macro.

  40211. 

  40212.  -- Macro: LOOP_ALIGN (LABEL)

  40213.      The alignment (log base 2) to put in front of LABEL that heads a

  40214.      frequently executed basic block (usually the header of a loop).

  40215. 

  40216.      This macro need not be defined if you don't want any special

  40217.      alignment to be done at such a time.  Most machine descriptions do

  40218.      not currently define the macro.

  40219. 

  40220.      Unless it's necessary to inspect the LABEL parameter, it is better

  40221.      to set the variable 'align_loops' in the target's

  40222.      'TARGET_OPTION_OVERRIDE'.  Otherwise, you should try to honor the

  40223.      user's selection in 'align_loops' in a 'LOOP_ALIGN' implementation.

  40224. 

  40225.  -- Macro: LABEL_ALIGN (LABEL)

  40226.      The alignment (log base 2) to put in front of LABEL.  If

  40227.      'LABEL_ALIGN_AFTER_BARRIER' / 'LOOP_ALIGN' specify a different

  40228.      alignment, the maximum of the specified values is used.

  40229. 

  40230.      Unless it's necessary to inspect the LABEL parameter, it is better

  40231.      to set the variable 'align_labels' in the target's

  40232.      'TARGET_OPTION_OVERRIDE'.  Otherwise, you should try to honor the

  40233.      user's selection in 'align_labels' in a 'LABEL_ALIGN'

  40234.      implementation.

  40235. 

  40236.  -- Macro: ASM_OUTPUT_SKIP (STREAM, NBYTES)

  40237.      A C statement to output to the stdio stream STREAM an assembler

  40238.      instruction to advance the location counter by NBYTES bytes.  Those

  40239.      bytes should be zero when loaded.  NBYTES will be a C expression of

  40240.      type 'unsigned HOST_WIDE_INT'.

  40241. 

  40242.  -- Macro: ASM_NO_SKIP_IN_TEXT

  40243.      Define this macro if 'ASM_OUTPUT_SKIP' should not be used in the

  40244.      text section because it fails to put zeros in the bytes that are

  40245.      skipped.  This is true on many Unix systems, where the pseudo-op to

  40246.      skip bytes produces no-op instructions rather than zeros when used

  40247.      in the text section.

  40248. 

  40249.  -- Macro: ASM_OUTPUT_ALIGN (STREAM, POWER)

  40250.      A C statement to output to the stdio stream STREAM an assembler

  40251.      command to advance the location counter to a multiple of 2 to the

  40252.      POWER bytes.  POWER will be a C expression of type 'int'.

  40253. 

  40254.  -- Macro: ASM_OUTPUT_ALIGN_WITH_NOP (STREAM, POWER)

  40255.      Like 'ASM_OUTPUT_ALIGN', except that the "nop" instruction is used

  40256.      for padding, if necessary.

  40257. 

  40258.  -- Macro: ASM_OUTPUT_MAX_SKIP_ALIGN (STREAM, POWER, MAX_SKIP)

  40259.      A C statement to output to the stdio stream STREAM an assembler

  40260.      command to advance the location counter to a multiple of 2 to the

  40261.      POWER bytes, but only if MAX_SKIP or fewer bytes are needed to

  40262.      satisfy the alignment request.  POWER and MAX_SKIP will be a C

  40263.      expression of type 'int'.

  40264. 

  40265. 

  40266. File: gccint.info,  Node: Debugging Info,  Next: Floating Point,  Prev: Assembler Format,  Up: Target Macros

  40267. 

  40268. 18.21 Controlling Debugging Information Format

  40269. ==============================================

  40270. 

  40271. This describes how to specify debugging information.

  40272. 

  40273. * Menu:

  40274. 

  40275. * All Debuggers::      Macros that affect all debugging formats uniformly.

  40276. * DBX Options::        Macros enabling specific options in DBX format.

  40277. * DBX Hooks::          Hook macros for varying DBX format.

  40278. * File Names and DBX:: Macros controlling output of file names in DBX format.

  40279. * DWARF::              Macros for DWARF format.

  40280. * VMS Debug::          Macros for VMS debug format.

  40281. 

  40282. 

  40283. File: gccint.info,  Node: All Debuggers,  Next: DBX Options,  Up: Debugging Info

  40284. 

  40285. 18.21.1 Macros Affecting All Debugging Formats

  40286. ----------------------------------------------

  40287. 

  40288. These macros affect all debugging formats.

  40289. 

  40290.  -- Macro: DBX_REGISTER_NUMBER (REGNO)

  40291.      A C expression that returns the DBX register number for the

  40292.      compiler register number REGNO.  In the default macro provided, the

  40293.      value of this expression will be REGNO itself.  But sometimes there

  40294.      are some registers that the compiler knows about and DBX does not,

  40295.      or vice versa.  In such cases, some register may need to have one

  40296.      number in the compiler and another for DBX.

  40297. 

  40298.      If two registers have consecutive numbers inside GCC, and they can

  40299.      be used as a pair to hold a multiword value, then they _must_ have

  40300.      consecutive numbers after renumbering with 'DBX_REGISTER_NUMBER'.

  40301.      Otherwise, debuggers will be unable to access such a pair, because

  40302.      they expect register pairs to be consecutive in their own numbering

  40303.      scheme.

  40304. 

  40305.      If you find yourself defining 'DBX_REGISTER_NUMBER' in way that

  40306.      does not preserve register pairs, then what you must do instead is

  40307.      redefine the actual register numbering scheme.

  40308. 

  40309.  -- Macro: DEBUGGER_AUTO_OFFSET (X)

  40310.      A C expression that returns the integer offset value for an

  40311.      automatic variable having address X (an RTL expression).  The

  40312.      default computation assumes that X is based on the frame-pointer

  40313.      and gives the offset from the frame-pointer.  This is required for

  40314.      targets that produce debugging output for DBX and allow the

  40315.      frame-pointer to be eliminated when the '-g' option is used.

  40316. 

  40317.  -- Macro: DEBUGGER_ARG_OFFSET (OFFSET, X)

  40318.      A C expression that returns the integer offset value for an

  40319.      argument having address X (an RTL expression).  The nominal offset

  40320.      is OFFSET.

  40321. 

  40322.  -- Macro: PREFERRED_DEBUGGING_TYPE

  40323.      A C expression that returns the type of debugging output GCC should

  40324.      produce when the user specifies just '-g'.  Define this if you have

  40325.      arranged for GCC to support more than one format of debugging

  40326.      output.  Currently, the allowable values are 'DBX_DEBUG',

  40327.      'DWARF2_DEBUG', 'XCOFF_DEBUG', 'VMS_DEBUG', and

  40328.      'VMS_AND_DWARF2_DEBUG'.

  40329. 

  40330.      When the user specifies '-ggdb', GCC normally also uses the value

  40331.      of this macro to select the debugging output format, but with two

  40332.      exceptions.  If 'DWARF2_DEBUGGING_INFO' is defined, GCC uses the

  40333.      value 'DWARF2_DEBUG'.  Otherwise, if 'DBX_DEBUGGING_INFO' is

  40334.      defined, GCC uses 'DBX_DEBUG'.

  40335. 

  40336.      The value of this macro only affects the default debugging output;

  40337.      the user can always get a specific type of output by using

  40338.      '-gstabs', '-gdwarf-2', '-gxcoff', or '-gvms'.

  40339. 

  40340. 

  40341. File: gccint.info,  Node: DBX Options,  Next: DBX Hooks,  Prev: All Debuggers,  Up: Debugging Info

  40342. 

  40343. 18.21.2 Specific Options for DBX Output

  40344. ---------------------------------------

  40345. 

  40346. These are specific options for DBX output.

  40347. 

  40348.  -- Macro: DBX_DEBUGGING_INFO

  40349.      Define this macro if GCC should produce debugging output for DBX in

  40350.      response to the '-g' option.

  40351. 

  40352.  -- Macro: XCOFF_DEBUGGING_INFO

  40353.      Define this macro if GCC should produce XCOFF format debugging

  40354.      output in response to the '-g' option.  This is a variant of DBX

  40355.      format.

  40356. 

  40357.  -- Macro: DEFAULT_GDB_EXTENSIONS

  40358.      Define this macro to control whether GCC should by default generate

  40359.      GDB's extended version of DBX debugging information (assuming

  40360.      DBX-format debugging information is enabled at all).  If you don't

  40361.      define the macro, the default is 1: always generate the extended

  40362.      information if there is any occasion to.

  40363. 

  40364.  -- Macro: DEBUG_SYMS_TEXT

  40365.      Define this macro if all '.stabs' commands should be output while

  40366.      in the text section.

  40367. 

  40368.  -- Macro: ASM_STABS_OP

  40369.      A C string constant, including spacing, naming the assembler pseudo

  40370.      op to use instead of '"\t.stabs\t"' to define an ordinary debugging

  40371.      symbol.  If you don't define this macro, '"\t.stabs\t"' is used.

  40372.      This macro applies only to DBX debugging information format.

  40373. 

  40374.  -- Macro: ASM_STABD_OP

  40375.      A C string constant, including spacing, naming the assembler pseudo

  40376.      op to use instead of '"\t.stabd\t"' to define a debugging symbol

  40377.      whose value is the current location.  If you don't define this

  40378.      macro, '"\t.stabd\t"' is used.  This macro applies only to DBX

  40379.      debugging information format.

  40380. 

  40381.  -- Macro: ASM_STABN_OP

  40382.      A C string constant, including spacing, naming the assembler pseudo

  40383.      op to use instead of '"\t.stabn\t"' to define a debugging symbol

  40384.      with no name.  If you don't define this macro, '"\t.stabn\t"' is

  40385.      used.  This macro applies only to DBX debugging information format.

  40386. 

  40387.  -- Macro: DBX_NO_XREFS

  40388.      Define this macro if DBX on your system does not support the

  40389.      construct 'xsTAGNAME'.  On some systems, this construct is used to

  40390.      describe a forward reference to a structure named TAGNAME.  On

  40391.      other systems, this construct is not supported at all.

  40392. 

  40393.  -- Macro: DBX_CONTIN_LENGTH

  40394.      A symbol name in DBX-format debugging information is normally

  40395.      continued (split into two separate '.stabs' directives) when it

  40396.      exceeds a certain length (by default, 80 characters).  On some

  40397.      operating systems, DBX requires this splitting; on others,

  40398.      splitting must not be done.  You can inhibit splitting by defining

  40399.      this macro with the value zero.  You can override the default

  40400.      splitting-length by defining this macro as an expression for the

  40401.      length you desire.

  40402. 

  40403.  -- Macro: DBX_CONTIN_CHAR

  40404.      Normally continuation is indicated by adding a '\' character to the

  40405.      end of a '.stabs' string when a continuation follows.  To use a

  40406.      different character instead, define this macro as a character

  40407.      constant for the character you want to use.  Do not define this

  40408.      macro if backslash is correct for your system.

  40409. 

  40410.  -- Macro: DBX_STATIC_STAB_DATA_SECTION

  40411.      Define this macro if it is necessary to go to the data section

  40412.      before outputting the '.stabs' pseudo-op for a non-global static

  40413.      variable.

  40414. 

  40415.  -- Macro: DBX_TYPE_DECL_STABS_CODE

  40416.      The value to use in the "code" field of the '.stabs' directive for

  40417.      a typedef.  The default is 'N_LSYM'.

  40418. 

  40419.  -- Macro: DBX_STATIC_CONST_VAR_CODE

  40420.      The value to use in the "code" field of the '.stabs' directive for

  40421.      a static variable located in the text section.  DBX format does not

  40422.      provide any "right" way to do this.  The default is 'N_FUN'.

  40423. 

  40424.  -- Macro: DBX_REGPARM_STABS_CODE

  40425.      The value to use in the "code" field of the '.stabs' directive for

  40426.      a parameter passed in registers.  DBX format does not provide any

  40427.      "right" way to do this.  The default is 'N_RSYM'.

  40428. 

  40429.  -- Macro: DBX_REGPARM_STABS_LETTER

  40430.      The letter to use in DBX symbol data to identify a symbol as a

  40431.      parameter passed in registers.  DBX format does not customarily

  40432.      provide any way to do this.  The default is ''P''.

  40433. 

  40434.  -- Macro: DBX_FUNCTION_FIRST

  40435.      Define this macro if the DBX information for a function and its

  40436.      arguments should precede the assembler code for the function.

  40437.      Normally, in DBX format, the debugging information entirely follows

  40438.      the assembler code.

  40439. 

  40440.  -- Macro: DBX_BLOCKS_FUNCTION_RELATIVE

  40441.      Define this macro, with value 1, if the value of a symbol

  40442.      describing the scope of a block ('N_LBRAC' or 'N_RBRAC') should be

  40443.      relative to the start of the enclosing function.  Normally, GCC

  40444.      uses an absolute address.

  40445. 

  40446.  -- Macro: DBX_LINES_FUNCTION_RELATIVE

  40447.      Define this macro, with value 1, if the value of a symbol

  40448.      indicating the current line number ('N_SLINE') should be relative

  40449.      to the start of the enclosing function.  Normally, GCC uses an

  40450.      absolute address.

  40451. 

  40452.  -- Macro: DBX_USE_BINCL

  40453.      Define this macro if GCC should generate 'N_BINCL' and 'N_EINCL'

  40454.      stabs for included header files, as on Sun systems.  This macro

  40455.      also directs GCC to output a type number as a pair of a file number

  40456.      and a type number within the file.  Normally, GCC does not generate

  40457.      'N_BINCL' or 'N_EINCL' stabs, and it outputs a single number for a

  40458.      type number.

  40459. 

  40460. 

  40461. File: gccint.info,  Node: DBX Hooks,  Next: File Names and DBX,  Prev: DBX Options,  Up: Debugging Info

  40462. 

  40463. 18.21.3 Open-Ended Hooks for DBX Format

  40464. ---------------------------------------

  40465. 

  40466. These are hooks for DBX format.

  40467. 

  40468.  -- Macro: DBX_OUTPUT_SOURCE_LINE (STREAM, LINE, COUNTER)

  40469.      A C statement to output DBX debugging information before code for

  40470.      line number LINE of the current source file to the stdio stream

  40471.      STREAM.  COUNTER is the number of time the macro was invoked,

  40472.      including the current invocation; it is intended to generate unique

  40473.      labels in the assembly output.

  40474. 

  40475.      This macro should not be defined if the default output is correct,

  40476.      or if it can be made correct by defining

  40477.      'DBX_LINES_FUNCTION_RELATIVE'.

  40478. 

  40479.  -- Macro: NO_DBX_FUNCTION_END

  40480.      Some stabs encapsulation formats (in particular ECOFF), cannot

  40481.      handle the '.stabs "",N_FUN,,0,0,Lscope-function-1' gdb dbx

  40482.      extension construct.  On those machines, define this macro to turn

  40483.      this feature off without disturbing the rest of the gdb extensions.

  40484. 

  40485.  -- Macro: NO_DBX_BNSYM_ENSYM

  40486.      Some assemblers cannot handle the '.stabd BNSYM/ENSYM,0,0' gdb dbx

  40487.      extension construct.  On those machines, define this macro to turn

  40488.      this feature off without disturbing the rest of the gdb extensions.

  40489. 

  40490. 

  40491. File: gccint.info,  Node: File Names and DBX,  Next: DWARF,  Prev: DBX Hooks,  Up: Debugging Info

  40492. 

  40493. 18.21.4 File Names in DBX Format

  40494. --------------------------------

  40495. 

  40496. This describes file names in DBX format.

  40497. 

  40498.  -- Macro: DBX_OUTPUT_MAIN_SOURCE_FILENAME (STREAM, NAME)

  40499.      A C statement to output DBX debugging information to the stdio

  40500.      stream STREAM, which indicates that file NAME is the main source

  40501.      file--the file specified as the input file for compilation.  This

  40502.      macro is called only once, at the beginning of compilation.

  40503. 

  40504.      This macro need not be defined if the standard form of output for

  40505.      DBX debugging information is appropriate.

  40506. 

  40507.      It may be necessary to refer to a label equal to the beginning of

  40508.      the text section.  You can use 'assemble_name (stream,

  40509.      ltext_label_name)' to do so.  If you do this, you must also set the

  40510.      variable USED_LTEXT_LABEL_NAME to 'true'.

  40511. 

  40512.  -- Macro: NO_DBX_MAIN_SOURCE_DIRECTORY

  40513.      Define this macro, with value 1, if GCC should not emit an

  40514.      indication of the current directory for compilation and current

  40515.      source language at the beginning of the file.

  40516. 

  40517.  -- Macro: NO_DBX_GCC_MARKER

  40518.      Define this macro, with value 1, if GCC should not emit an

  40519.      indication that this object file was compiled by GCC.  The default

  40520.      is to emit an 'N_OPT' stab at the beginning of every source file,

  40521.      with 'gcc2_compiled.' for the string and value 0.

  40522. 

  40523.  -- Macro: DBX_OUTPUT_MAIN_SOURCE_FILE_END (STREAM, NAME)

  40524.      A C statement to output DBX debugging information at the end of

  40525.      compilation of the main source file NAME.  Output should be written

  40526.      to the stdio stream STREAM.

  40527. 

  40528.      If you don't define this macro, nothing special is output at the

  40529.      end of compilation, which is correct for most machines.

  40530. 

  40531.  -- Macro: DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END

  40532.      Define this macro _instead of_ defining

  40533.      'DBX_OUTPUT_MAIN_SOURCE_FILE_END', if what needs to be output at

  40534.      the end of compilation is an 'N_SO' stab with an empty string,

  40535.      whose value is the highest absolute text address in the file.

  40536. 

  40537. 

  40538. File: gccint.info,  Node: DWARF,  Next: VMS Debug,  Prev: File Names and DBX,  Up: Debugging Info

  40539. 

  40540. 18.21.5 Macros for DWARF Output

  40541. -------------------------------

  40542. 

  40543. Here are macros for DWARF output.

  40544. 

  40545.  -- Macro: DWARF2_DEBUGGING_INFO

  40546.      Define this macro if GCC should produce dwarf version 2 format

  40547.      debugging output in response to the '-g' option.

  40548. 

  40549.       -- Target Hook: int TARGET_DWARF_CALLING_CONVENTION (const_tree

  40550.                FUNCTION)

  40551.           Define this to enable the dwarf attribute

  40552.           'DW_AT_calling_convention' to be emitted for each function.

  40553.           Instead of an integer return the enum value for the 'DW_CC_'

  40554.           tag.

  40555. 

  40556.      To support optional call frame debugging information, you must also

  40557.      define 'INCOMING_RETURN_ADDR_RTX' and either set

  40558.      'RTX_FRAME_RELATED_P' on the prologue insns if you use RTL for the

  40559.      prologue, or call 'dwarf2out_def_cfa' and 'dwarf2out_reg_save' as

  40560.      appropriate from 'TARGET_ASM_FUNCTION_PROLOGUE' if you don't.

  40561. 

  40562.  -- Macro: DWARF2_FRAME_INFO

  40563.      Define this macro to a nonzero value if GCC should always output

  40564.      Dwarf 2 frame information.  If 'TARGET_EXCEPT_UNWIND_INFO' (*note

  40565.      Exception Region Output::) returns 'UI_DWARF2', and exceptions are

  40566.      enabled, GCC will output this information not matter how you define

  40567.      'DWARF2_FRAME_INFO'.

  40568. 

  40569.  -- Target Hook: enum unwind_info_type TARGET_DEBUG_UNWIND_INFO (void)

  40570.      This hook defines the mechanism that will be used for describing

  40571.      frame unwind information to the debugger.  Normally the hook will

  40572.      return 'UI_DWARF2' if DWARF 2 debug information is enabled, and

  40573.      return 'UI_NONE' otherwise.

  40574. 

  40575.      A target may return 'UI_DWARF2' even when DWARF 2 debug information

  40576.      is disabled in order to always output DWARF 2 frame information.

  40577. 

  40578.      A target may return 'UI_TARGET' if it has ABI specified unwind

  40579.      tables.  This will suppress generation of the normal debug frame

  40580.      unwind information.

  40581. 

  40582.  -- Macro: DWARF2_ASM_LINE_DEBUG_INFO

  40583.      Define this macro to be a nonzero value if the assembler can

  40584.      generate Dwarf 2 line debug info sections.  This will result in

  40585.      much more compact line number tables, and hence is desirable if it

  40586.      works.

  40587. 

  40588.  -- Macro: DWARF2_ASM_VIEW_DEBUG_INFO

  40589.      Define this macro to be a nonzero value if the assembler supports

  40590.      view assignment and verification in '.loc'.  If it does not, but

  40591.      the user enables location views, the compiler may have to fallback

  40592.      to internal line number tables.

  40593. 

  40594.  -- Target Hook: int TARGET_RESET_LOCATION_VIEW (rtx_insn *)

  40595.      This hook, if defined, enables -ginternal-reset-location-views, and

  40596.      uses its result to override cases in which the estimated min insn

  40597.      length might be nonzero even when a PC advance (i.e., a view reset)

  40598.      cannot be taken for granted.

  40599. 

  40600.      If the hook is defined, it must return a positive value to indicate

  40601.      the insn definitely advances the PC, and so the view number can be

  40602.      safely assumed to be reset; a negative value to mean the insn

  40603.      definitely does not advance the PC, and os the view number must not

  40604.      be reset; or zero to decide based on the estimated insn length.

  40605. 

  40606.      If insn length is to be regarded as reliable, set the hook to

  40607.      'hook_int_rtx_insn_0'.

  40608. 

  40609.  -- Target Hook: bool TARGET_WANT_DEBUG_PUB_SECTIONS

  40610.      True if the '.debug_pubtypes' and '.debug_pubnames' sections should

  40611.      be emitted.  These sections are not used on most platforms, and in

  40612.      particular GDB does not use them.

  40613. 

  40614.  -- Target Hook: bool TARGET_DELAY_SCHED2

  40615.      True if sched2 is not to be run at its normal place.  This usually

  40616.      means it will be run as part of machine-specific reorg.

  40617. 

  40618.  -- Target Hook: bool TARGET_DELAY_VARTRACK

  40619.      True if vartrack is not to be run at its normal place.  This

  40620.      usually means it will be run as part of machine-specific reorg.

  40621. 

  40622.  -- Target Hook: bool TARGET_NO_REGISTER_ALLOCATION

  40623.      True if register allocation and the passes following it should not

  40624.      be run.  Usually true only for virtual assembler targets.

  40625. 

  40626.  -- Macro: ASM_OUTPUT_DWARF_DELTA (STREAM, SIZE, LABEL1, LABEL2)

  40627.      A C statement to issue assembly directives that create a difference

  40628.      LAB1 minus LAB2, using an integer of the given SIZE.

  40629. 

  40630.  -- Macro: ASM_OUTPUT_DWARF_VMS_DELTA (STREAM, SIZE, LABEL1, LABEL2)

  40631.      A C statement to issue assembly directives that create a difference

  40632.      between the two given labels in system defined units, e.g.

  40633.      instruction slots on IA64 VMS, using an integer of the given size.

  40634. 

  40635.  -- Macro: ASM_OUTPUT_DWARF_OFFSET (STREAM, SIZE, LABEL, OFFSET,

  40636.           SECTION)

  40637.      A C statement to issue assembly directives that create a

  40638.      section-relative reference to the given LABEL plus OFFSET, using an

  40639.      integer of the given SIZE.  The label is known to be defined in the

  40640.      given SECTION.

  40641. 

  40642.  -- Macro: ASM_OUTPUT_DWARF_PCREL (STREAM, SIZE, LABEL)

  40643.      A C statement to issue assembly directives that create a

  40644.      self-relative reference to the given LABEL, using an integer of the

  40645.      given SIZE.

  40646. 

  40647.  -- Macro: ASM_OUTPUT_DWARF_DATAREL (STREAM, SIZE, LABEL)

  40648.      A C statement to issue assembly directives that create a reference

  40649.      to the given LABEL relative to the dbase, using an integer of the

  40650.      given SIZE.

  40651. 

  40652.  -- Macro: ASM_OUTPUT_DWARF_TABLE_REF (LABEL)

  40653.      A C statement to issue assembly directives that create a reference

  40654.      to the DWARF table identifier LABEL from the current section.  This

  40655.      is used on some systems to avoid garbage collecting a DWARF table

  40656.      which is referenced by a function.

  40657. 

  40658.  -- Target Hook: void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *FILE, int

  40659.           SIZE, rtx X)

  40660.      If defined, this target hook is a function which outputs a

  40661.      DTP-relative reference to the given TLS symbol of the specified

  40662.      size.

  40663. 

  40664. 

  40665. File: gccint.info,  Node: VMS Debug,  Prev: DWARF,  Up: Debugging Info

  40666. 

  40667. 18.21.6 Macros for VMS Debug Format

  40668. -----------------------------------

  40669. 

  40670. Here are macros for VMS debug format.

  40671. 

  40672.  -- Macro: VMS_DEBUGGING_INFO

  40673.      Define this macro if GCC should produce debugging output for VMS in

  40674.      response to the '-g' option.  The default behavior for VMS is to

  40675.      generate minimal debug info for a traceback in the absence of '-g'

  40676.      unless explicitly overridden with '-g0'.  This behavior is

  40677.      controlled by 'TARGET_OPTION_OPTIMIZATION' and

  40678.      'TARGET_OPTION_OVERRIDE'.

  40679. 

  40680. 

  40681. File: gccint.info,  Node: Floating Point,  Next: Mode Switching,  Prev: Debugging Info,  Up: Target Macros

  40682. 

  40683. 18.22 Cross Compilation and Floating Point

  40684. ==========================================

  40685. 

  40686. While all modern machines use twos-complement representation for

  40687. integers, there are a variety of representations for floating point

  40688. numbers.  This means that in a cross-compiler the representation of

  40689. floating point numbers in the compiled program may be different from

  40690. that used in the machine doing the compilation.

  40691. 

  40692.  Because different representation systems may offer different amounts of

  40693. range and precision, all floating point constants must be represented in

  40694. the target machine's format.  Therefore, the cross compiler cannot

  40695. safely use the host machine's floating point arithmetic; it must emulate

  40696. the target's arithmetic.  To ensure consistency, GCC always uses

  40697. emulation to work with floating point values, even when the host and

  40698. target floating point formats are identical.

  40699. 

  40700.  The following macros are provided by 'real.h' for the compiler to use.

  40701. All parts of the compiler which generate or optimize floating-point

  40702. calculations must use these macros.  They may evaluate their operands

  40703. more than once, so operands must not have side effects.

  40704. 

  40705.  -- Macro: REAL_VALUE_TYPE

  40706.      The C data type to be used to hold a floating point value in the

  40707.      target machine's format.  Typically this is a 'struct' containing

  40708.      an array of 'HOST_WIDE_INT', but all code should treat it as an

  40709.      opaque quantity.

  40710. 

  40711.  -- Macro: HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE X)

  40712.      Truncates X to a signed integer, rounding toward zero.

  40713. 

  40714.  -- Macro: unsigned HOST_WIDE_INT REAL_VALUE_UNSIGNED_FIX

  40715.           (REAL_VALUE_TYPE X)

  40716.      Truncates X to an unsigned integer, rounding toward zero.  If X is

  40717.      negative, returns zero.

  40718. 

  40719.  -- Macro: REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *STRING,

  40720.           machine_mode MODE)

  40721.      Converts STRING into a floating point number in the target

  40722.      machine's representation for mode MODE.  This routine can handle

  40723.      both decimal and hexadecimal floating point constants, using the

  40724.      syntax defined by the C language for both.

  40725. 

  40726.  -- Macro: int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE X)

  40727.      Returns 1 if X is negative (including negative zero), 0 otherwise.

  40728. 

  40729.  -- Macro: int REAL_VALUE_ISINF (REAL_VALUE_TYPE X)

  40730.      Determines whether X represents infinity (positive or negative).

  40731. 

  40732.  -- Macro: int REAL_VALUE_ISNAN (REAL_VALUE_TYPE X)

  40733.      Determines whether X represents a "NaN" (not-a-number).

  40734. 

  40735.  -- Macro: REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE X)

  40736.      Returns the negative of the floating point value X.

  40737. 

  40738.  -- Macro: REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE X)

  40739.      Returns the absolute value of X.

  40740. 

  40741. 

  40742. File: gccint.info,  Node: Mode Switching,  Next: Target Attributes,  Prev: Floating Point,  Up: Target Macros

  40743. 

  40744. 18.23 Mode Switching Instructions

  40745. =================================

  40746. 

  40747. The following macros control mode switching optimizations:

  40748. 

  40749.  -- Macro: OPTIMIZE_MODE_SWITCHING (ENTITY)

  40750.      Define this macro if the port needs extra instructions inserted for

  40751.      mode switching in an optimizing compilation.

  40752. 

  40753.      For an example, the SH4 can perform both single and double

  40754.      precision floating point operations, but to perform a single

  40755.      precision operation, the FPSCR PR bit has to be cleared, while for

  40756.      a double precision operation, this bit has to be set.  Changing the

  40757.      PR bit requires a general purpose register as a scratch register,

  40758.      hence these FPSCR sets have to be inserted before reload, i.e. you

  40759.      cannot put this into instruction emitting or

  40760.      'TARGET_MACHINE_DEPENDENT_REORG'.

  40761. 

  40762.      You can have multiple entities that are mode-switched, and select

  40763.      at run time which entities actually need it.

  40764.      'OPTIMIZE_MODE_SWITCHING' should return nonzero for any ENTITY that

  40765.      needs mode-switching.  If you define this macro, you also have to

  40766.      define 'NUM_MODES_FOR_MODE_SWITCHING', 'TARGET_MODE_NEEDED',

  40767.      'TARGET_MODE_PRIORITY' and 'TARGET_MODE_EMIT'.

  40768.      'TARGET_MODE_AFTER', 'TARGET_MODE_ENTRY', and 'TARGET_MODE_EXIT'

  40769.      are optional.

  40770. 

  40771.  -- Macro: NUM_MODES_FOR_MODE_SWITCHING

  40772.      If you define 'OPTIMIZE_MODE_SWITCHING', you have to define this as

  40773.      initializer for an array of integers.  Each initializer element N

  40774.      refers to an entity that needs mode switching, and specifies the

  40775.      number of different modes that might need to be set for this

  40776.      entity.  The position of the initializer in the

  40777.      initializer--starting counting at zero--determines the integer that

  40778.      is used to refer to the mode-switched entity in question.  In

  40779.      macros that take mode arguments / yield a mode result, modes are

  40780.      represented as numbers 0 ... N - 1.  N is used to specify that no

  40781.      mode switch is needed / supplied.

  40782. 

  40783.  -- Target Hook: void TARGET_MODE_EMIT (int ENTITY, int MODE, int

  40784.           PREV_MODE, HARD_REG_SET REGS_LIVE)

  40785.      Generate one or more insns to set ENTITY to MODE.  HARD_REG_LIVE is

  40786.      the set of hard registers live at the point where the insn(s) are

  40787.      to be inserted.  PREV_MOXDE indicates the mode to switch from.

  40788.      Sets of a lower numbered entity will be emitted before sets of a

  40789.      higher numbered entity to a mode of the same or lower priority.

  40790. 

  40791.  -- Target Hook: int TARGET_MODE_NEEDED (int ENTITY, rtx_insn *INSN)

  40792.      ENTITY is an integer specifying a mode-switched entity.  If

  40793.      'OPTIMIZE_MODE_SWITCHING' is defined, you must define this macro to

  40794.      return an integer value not larger than the corresponding element

  40795.      in 'NUM_MODES_FOR_MODE_SWITCHING', to denote the mode that ENTITY

  40796.      must be switched into prior to the execution of INSN.

  40797. 

  40798.  -- Target Hook: int TARGET_MODE_AFTER (int ENTITY, int MODE, rtx_insn

  40799.           *INSN)

  40800.      ENTITY is an integer specifying a mode-switched entity.  If this

  40801.      macro is defined, it is evaluated for every INSN during mode

  40802.      switching.  It determines the mode that an insn results in (if

  40803.      different from the incoming mode).

  40804. 

  40805.  -- Target Hook: int TARGET_MODE_ENTRY (int ENTITY)

  40806.      If this macro is defined, it is evaluated for every ENTITY that

  40807.      needs mode switching.  It should evaluate to an integer, which is a

  40808.      mode that ENTITY is assumed to be switched to at function entry.

  40809.      If 'TARGET_MODE_ENTRY' is defined then 'TARGET_MODE_EXIT' must be

  40810.      defined.

  40811. 

  40812.  -- Target Hook: int TARGET_MODE_EXIT (int ENTITY)

  40813.      If this macro is defined, it is evaluated for every ENTITY that

  40814.      needs mode switching.  It should evaluate to an integer, which is a

  40815.      mode that ENTITY is assumed to be switched to at function exit.  If

  40816.      'TARGET_MODE_EXIT' is defined then 'TARGET_MODE_ENTRY' must be

  40817.      defined.

  40818. 

  40819.  -- Target Hook: int TARGET_MODE_PRIORITY (int ENTITY, int N)

  40820.      This macro specifies the order in which modes for ENTITY are

  40821.      processed.  0 is the highest priority,

  40822.      'NUM_MODES_FOR_MODE_SWITCHING[ENTITY] - 1' the lowest.  The value

  40823.      of the macro should be an integer designating a mode for ENTITY.

  40824.      For any fixed ENTITY, 'mode_priority' (ENTITY, N) shall be a

  40825.      bijection in 0 ... 'num_modes_for_mode_switching[ENTITY] - 1'.

  40826. 

  40827. 

  40828. File: gccint.info,  Node: Target Attributes,  Next: Emulated TLS,  Prev: Mode Switching,  Up: Target Macros

  40829. 

  40830. 18.24 Defining target-specific uses of '__attribute__'

  40831. ======================================================

  40832. 

  40833. Target-specific attributes may be defined for functions, data and types.

  40834. These are described using the following target hooks; they also need to

  40835. be documented in 'extend.texi'.

  40836. 

  40837.  -- Target Hook: const struct attribute_spec * TARGET_ATTRIBUTE_TABLE

  40838.      If defined, this target hook points to an array of 'struct

  40839.      attribute_spec' (defined in 'tree-core.h') specifying the machine

  40840.      specific attributes for this target and some of the restrictions on

  40841.      the entities to which these attributes are applied and the

  40842.      arguments they take.

  40843. 

  40844.  -- Target Hook: bool TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P (const_tree

  40845.           NAME)

  40846.      If defined, this target hook is a function which returns true if

  40847.      the machine-specific attribute named NAME expects an identifier

  40848.      given as its first argument to be passed on as a plain identifier,

  40849.      not subjected to name lookup.  If this is not defined, the default

  40850.      is false for all machine-specific attributes.

  40851. 

  40852.  -- Target Hook: int TARGET_COMP_TYPE_ATTRIBUTES (const_tree TYPE1,

  40853.           const_tree TYPE2)

  40854.      If defined, this target hook is a function which returns zero if

  40855.      the attributes on TYPE1 and TYPE2 are incompatible, one if they are

  40856.      compatible, and two if they are nearly compatible (which causes a

  40857.      warning to be generated).  If this is not defined, machine-specific

  40858.      attributes are supposed always to be compatible.

  40859. 

  40860.  -- Target Hook: void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree TYPE)

  40861.      If defined, this target hook is a function which assigns default

  40862.      attributes to the newly defined TYPE.

  40863. 

  40864.  -- Target Hook: tree TARGET_MERGE_TYPE_ATTRIBUTES (tree TYPE1, tree

  40865.           TYPE2)

  40866.      Define this target hook if the merging of type attributes needs

  40867.      special handling.  If defined, the result is a list of the combined

  40868.      'TYPE_ATTRIBUTES' of TYPE1 and TYPE2.  It is assumed that

  40869.      'comptypes' has already been called and returned 1.  This function

  40870.      may call 'merge_attributes' to handle machine-independent merging.

  40871. 

  40872.  -- Target Hook: tree TARGET_MERGE_DECL_ATTRIBUTES (tree OLDDECL, tree

  40873.           NEWDECL)

  40874.      Define this target hook if the merging of decl attributes needs

  40875.      special handling.  If defined, the result is a list of the combined

  40876.      'DECL_ATTRIBUTES' of OLDDECL and NEWDECL.  NEWDECL is a duplicate

  40877.      declaration of OLDDECL.  Examples of when this is needed are when

  40878.      one attribute overrides another, or when an attribute is nullified

  40879.      by a subsequent definition.  This function may call

  40880.      'merge_attributes' to handle machine-independent merging.

  40881. 

  40882.      If the only target-specific handling you require is 'dllimport' for

  40883.      Microsoft Windows targets, you should define the macro

  40884.      'TARGET_DLLIMPORT_DECL_ATTRIBUTES' to '1'.  The compiler will then

  40885.      define a function called 'merge_dllimport_decl_attributes' which

  40886.      can then be defined as the expansion of

  40887.      'TARGET_MERGE_DECL_ATTRIBUTES'.  You can also add

  40888.      'handle_dll_attribute' in the attribute table for your port to

  40889.      perform initial processing of the 'dllimport' and 'dllexport'

  40890.      attributes.  This is done in 'i386/cygwin.h' and 'i386/i386.c', for

  40891.      example.

  40892. 

  40893.  -- Target Hook: bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (const_tree

  40894.           DECL)

  40895.      DECL is a variable or function with '__attribute__((dllimport))'

  40896.      specified.  Use this hook if the target needs to add extra

  40897.      validation checks to 'handle_dll_attribute'.

  40898. 

  40899.  -- Macro: TARGET_DECLSPEC

  40900.      Define this macro to a nonzero value if you want to treat

  40901.      '__declspec(X)' as equivalent to '__attribute((X))'.  By default,

  40902.      this behavior is enabled only for targets that define

  40903.      'TARGET_DLLIMPORT_DECL_ATTRIBUTES'.  The current implementation of

  40904.      '__declspec' is via a built-in macro, but you should not rely on

  40905.      this implementation detail.

  40906. 

  40907.  -- Target Hook: void TARGET_INSERT_ATTRIBUTES (tree NODE, tree

  40908.           *ATTR_PTR)

  40909.      Define this target hook if you want to be able to add attributes to

  40910.      a decl when it is being created.  This is normally useful for back

  40911.      ends which wish to implement a pragma by using the attributes which

  40912.      correspond to the pragma's effect.  The NODE argument is the decl

  40913.      which is being created.  The ATTR_PTR argument is a pointer to the

  40914.      attribute list for this decl.  The list itself should not be

  40915.      modified, since it may be shared with other decls, but attributes

  40916.      may be chained on the head of the list and '*ATTR_PTR' modified to

  40917.      point to the new attributes, or a copy of the list may be made if

  40918.      further changes are needed.

  40919. 

  40920.  -- Target Hook: tree TARGET_HANDLE_GENERIC_ATTRIBUTE (tree *NODE, tree

  40921.           NAME, tree ARGS, int FLAGS, bool *NO_ADD_ATTRS)

  40922.      Define this target hook if you want to be able to perform

  40923.      additional target-specific processing of an attribute which is

  40924.      handled generically by a front end.  The arguments are the same as

  40925.      those which are passed to attribute handlers.  So far this only

  40926.      affects the NOINIT and SECTION attribute.

  40927. 

  40928.  -- Target Hook: bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (const_tree

  40929.           FNDECL)

  40930.      This target hook returns 'true' if it is OK to inline FNDECL into

  40931.      the current function, despite its having target-specific

  40932.      attributes, 'false' otherwise.  By default, if a function has a

  40933.      target specific attribute attached to it, it will not be inlined.

  40934. 

  40935.  -- Target Hook: bool TARGET_OPTION_VALID_ATTRIBUTE_P (tree FNDECL, tree

  40936.           NAME, tree ARGS, int FLAGS)

  40937.      This hook is called to parse 'attribute(target("..."))', which

  40938.      allows setting target-specific options on individual functions.

  40939.      These function-specific options may differ from the options

  40940.      specified on the command line.  The hook should return 'true' if

  40941.      the options are valid.

  40942. 

  40943.      The hook should set the 'DECL_FUNCTION_SPECIFIC_TARGET' field in

  40944.      the function declaration to hold a pointer to a target-specific

  40945.      'struct cl_target_option' structure.

  40946. 

  40947.  -- Target Hook: void TARGET_OPTION_SAVE (struct cl_target_option *PTR,

  40948.           struct gcc_options *OPTS)

  40949.      This hook is called to save any additional target-specific

  40950.      information in the 'struct cl_target_option' structure for

  40951.      function-specific options from the 'struct gcc_options' structure.

  40952.      *Note Option file format::.

  40953. 

  40954.  -- Target Hook: void TARGET_OPTION_RESTORE (struct gcc_options *OPTS,

  40955.           struct cl_target_option *PTR)

  40956.      This hook is called to restore any additional target-specific

  40957.      information in the 'struct cl_target_option' structure for

  40958.      function-specific options to the 'struct gcc_options' structure.

  40959. 

  40960.  -- Target Hook: void TARGET_OPTION_POST_STREAM_IN (struct

  40961.           cl_target_option *PTR)

  40962.      This hook is called to update target-specific information in the

  40963.      'struct cl_target_option' structure after it is streamed in from

  40964.      LTO bytecode.

  40965. 

  40966.  -- Target Hook: void TARGET_OPTION_PRINT (FILE *FILE, int INDENT,

  40967.           struct cl_target_option *PTR)

  40968.      This hook is called to print any additional target-specific

  40969.      information in the 'struct cl_target_option' structure for

  40970.      function-specific options.

  40971. 

  40972.  -- Target Hook: bool TARGET_OPTION_PRAGMA_PARSE (tree ARGS, tree

  40973.           POP_TARGET)

  40974.      This target hook parses the options for '#pragma GCC target', which

  40975.      sets the target-specific options for functions that occur later in

  40976.      the input stream.  The options accepted should be the same as those

  40977.      handled by the 'TARGET_OPTION_VALID_ATTRIBUTE_P' hook.

  40978. 

  40979.  -- Target Hook: void TARGET_OPTION_OVERRIDE (void)

  40980.      Sometimes certain combinations of command options do not make sense

  40981.      on a particular target machine.  You can override the hook

  40982.      'TARGET_OPTION_OVERRIDE' to take account of this.  This hooks is

  40983.      called once just after all the command options have been parsed.

  40984. 

  40985.      Don't use this hook to turn on various extra optimizations for

  40986.      '-O'.  That is what 'TARGET_OPTION_OPTIMIZATION' is for.

  40987. 

  40988.      If you need to do something whenever the optimization level is

  40989.      changed via the optimize attribute or pragma, see

  40990.      'TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE'

  40991. 

  40992.  -- Target Hook: bool TARGET_OPTION_FUNCTION_VERSIONS (tree DECL1, tree

  40993.           DECL2)

  40994.      This target hook returns 'true' if DECL1 and DECL2 are versions of

  40995.      the same function.  DECL1 and DECL2 are function versions if and

  40996.      only if they have the same function signature and different target

  40997.      specific attributes, that is, they are compiled for different

  40998.      target machines.

  40999. 

  41000.  -- Target Hook: bool TARGET_CAN_INLINE_P (tree CALLER, tree CALLEE)

  41001.      This target hook returns 'false' if the CALLER function cannot

  41002.      inline CALLEE, based on target specific information.  By default,

  41003.      inlining is not allowed if the callee function has function

  41004.      specific target options and the caller does not use the same

  41005.      options.

  41006. 

  41007.  -- Target Hook: void TARGET_RELAYOUT_FUNCTION (tree FNDECL)

  41008.      This target hook fixes function FNDECL after attributes are

  41009.      processed.  Default does nothing.  On ARM, the default function's

  41010.      alignment is updated with the attribute target.

  41011. 

  41012. 

  41013. File: gccint.info,  Node: Emulated TLS,  Next: MIPS Coprocessors,  Prev: Target Attributes,  Up: Target Macros

  41014. 

  41015. 18.25 Emulating TLS

  41016. ===================

  41017. 

  41018. For targets whose psABI does not provide Thread Local Storage via

  41019. specific relocations and instruction sequences, an emulation layer is

  41020. used.  A set of target hooks allows this emulation layer to be

  41021. configured for the requirements of a particular target.  For instance

  41022. the psABI may in fact specify TLS support in terms of an emulation

  41023. layer.

  41024. 

  41025.  The emulation layer works by creating a control object for every TLS

  41026. object.  To access the TLS object, a lookup function is provided which,

  41027. when given the address of the control object, will return the address of

  41028. the current thread's instance of the TLS object.

  41029. 

  41030.  -- Target Hook: const char * TARGET_EMUTLS_GET_ADDRESS

  41031.      Contains the name of the helper function that uses a TLS control

  41032.      object to locate a TLS instance.  The default causes libgcc's

  41033.      emulated TLS helper function to be used.

  41034. 

  41035.  -- Target Hook: const char * TARGET_EMUTLS_REGISTER_COMMON

  41036.      Contains the name of the helper function that should be used at

  41037.      program startup to register TLS objects that are implicitly

  41038.      initialized to zero.  If this is 'NULL', all TLS objects will have

  41039.      explicit initializers.  The default causes libgcc's emulated TLS

  41040.      registration function to be used.

  41041. 

  41042.  -- Target Hook: const char * TARGET_EMUTLS_VAR_SECTION

  41043.      Contains the name of the section in which TLS control variables

  41044.      should be placed.  The default of 'NULL' allows these to be placed

  41045.      in any section.

  41046. 

  41047.  -- Target Hook: const char * TARGET_EMUTLS_TMPL_SECTION

  41048.      Contains the name of the section in which TLS initializers should

  41049.      be placed.  The default of 'NULL' allows these to be placed in any

  41050.      section.

  41051. 

  41052.  -- Target Hook: const char * TARGET_EMUTLS_VAR_PREFIX

  41053.      Contains the prefix to be prepended to TLS control variable names.

  41054.      The default of 'NULL' uses a target-specific prefix.

  41055. 

  41056.  -- Target Hook: const char * TARGET_EMUTLS_TMPL_PREFIX

  41057.      Contains the prefix to be prepended to TLS initializer objects.

  41058.      The default of 'NULL' uses a target-specific prefix.

  41059. 

  41060.  -- Target Hook: tree TARGET_EMUTLS_VAR_FIELDS (tree TYPE, tree *NAME)

  41061.      Specifies a function that generates the FIELD_DECLs for a TLS

  41062.      control object type.  TYPE is the RECORD_TYPE the fields are for

  41063.      and NAME should be filled with the structure tag, if the default of

  41064.      '__emutls_object' is unsuitable.  The default creates a type

  41065.      suitable for libgcc's emulated TLS function.

  41066. 

  41067.  -- Target Hook: tree TARGET_EMUTLS_VAR_INIT (tree VAR, tree DECL, tree

  41068.           TMPL_ADDR)

  41069.      Specifies a function that generates the CONSTRUCTOR to initialize a

  41070.      TLS control object.  VAR is the TLS control object, DECL is the TLS

  41071.      object and TMPL_ADDR is the address of the initializer.  The

  41072.      default initializes libgcc's emulated TLS control object.

  41073. 

  41074.  -- Target Hook: bool TARGET_EMUTLS_VAR_ALIGN_FIXED

  41075.      Specifies whether the alignment of TLS control variable objects is

  41076.      fixed and should not be increased as some backends may do to

  41077.      optimize single objects.  The default is false.

  41078. 

  41079.  -- Target Hook: bool TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS

  41080.      Specifies whether a DWARF 'DW_OP_form_tls_address' location

  41081.      descriptor may be used to describe emulated TLS control objects.

  41082. 

  41083. 

  41084. File: gccint.info,  Node: MIPS Coprocessors,  Next: PCH Target,  Prev: Emulated TLS,  Up: Target Macros

  41085. 

  41086. 18.26 Defining coprocessor specifics for MIPS targets.

  41087. ======================================================

  41088. 

  41089. The MIPS specification allows MIPS implementations to have as many as 4

  41090. coprocessors, each with as many as 32 private registers.  GCC supports

  41091. accessing these registers and transferring values between the registers

  41092. and memory using asm-ized variables.  For example:

  41093. 

  41094.        register unsigned int cp0count asm ("c0r1");

  41095.        unsigned int d;

  41096. 

  41097.        d = cp0count + 3;

  41098. 

  41099.  ("c0r1" is the default name of register 1 in coprocessor 0; alternate

  41100. names may be added as described below, or the default names may be

  41101. overridden entirely in 'SUBTARGET_CONDITIONAL_REGISTER_USAGE'.)

  41102. 

  41103.  Coprocessor registers are assumed to be epilogue-used; sets to them

  41104. will be preserved even if it does not appear that the register is used

  41105. again later in the function.

  41106. 

  41107.  Another note: according to the MIPS spec, coprocessor 1 (if present) is

  41108. the FPU.  One accesses COP1 registers through standard mips

  41109. floating-point support; they are not included in this mechanism.

  41110. 

  41111. 

  41112. File: gccint.info,  Node: PCH Target,  Next: C++ ABI,  Prev: MIPS Coprocessors,  Up: Target Macros

  41113. 

  41114. 18.27 Parameters for Precompiled Header Validity Checking

  41115. =========================================================

  41116. 

  41117.  -- Target Hook: void * TARGET_GET_PCH_VALIDITY (size_t *SZ)

  41118.      This hook returns a pointer to the data needed by

  41119.      'TARGET_PCH_VALID_P' and sets '*SZ' to the size of the data in

  41120.      bytes.

  41121. 

  41122.  -- Target Hook: const char * TARGET_PCH_VALID_P (const void *DATA,

  41123.           size_t SZ)

  41124.      This hook checks whether the options used to create a PCH file are

  41125.      compatible with the current settings.  It returns 'NULL' if so and

  41126.      a suitable error message if not.  Error messages will be presented

  41127.      to the user and must be localized using '_(MSG)'.

  41128. 

  41129.      DATA is the data that was returned by 'TARGET_GET_PCH_VALIDITY'

  41130.      when the PCH file was created and SZ is the size of that data in

  41131.      bytes.  It's safe to assume that the data was created by the same

  41132.      version of the compiler, so no format checking is needed.

  41133. 

  41134.      The default definition of 'default_pch_valid_p' should be suitable

  41135.      for most targets.

  41136. 

  41137.  -- Target Hook: const char * TARGET_CHECK_PCH_TARGET_FLAGS (int

  41138.           PCH_FLAGS)

  41139.      If this hook is nonnull, the default implementation of

  41140.      'TARGET_PCH_VALID_P' will use it to check for compatible values of

  41141.      'target_flags'.  PCH_FLAGS specifies the value that 'target_flags'

  41142.      had when the PCH file was created.  The return value is the same as

  41143.      for 'TARGET_PCH_VALID_P'.

  41144. 

  41145.  -- Target Hook: void TARGET_PREPARE_PCH_SAVE (void)

  41146.      Called before writing out a PCH file.  If the target has some

  41147.      garbage-collected data that needs to be in a particular state on

  41148.      PCH loads, it can use this hook to enforce that state.  Very few

  41149.      targets need to do anything here.

  41150. 

  41151. 

  41152. File: gccint.info,  Node: C++ ABI,  Next: D Language and ABI,  Prev: PCH Target,  Up: Target Macros

  41153. 

  41154. 18.28 C++ ABI parameters

  41155. ========================

  41156. 

  41157.  -- Target Hook: tree TARGET_CXX_GUARD_TYPE (void)

  41158.      Define this hook to override the integer type used for guard

  41159.      variables.  These are used to implement one-time construction of

  41160.      static objects.  The default is long_long_integer_type_node.

  41161. 

  41162.  -- Target Hook: bool TARGET_CXX_GUARD_MASK_BIT (void)

  41163.      This hook determines how guard variables are used.  It should

  41164.      return 'false' (the default) if the first byte should be used.  A

  41165.      return value of 'true' indicates that only the least significant

  41166.      bit should be used.

  41167. 

  41168.  -- Target Hook: tree TARGET_CXX_GET_COOKIE_SIZE (tree TYPE)

  41169.      This hook returns the size of the cookie to use when allocating an

  41170.      array whose elements have the indicated TYPE.  Assumes that it is

  41171.      already known that a cookie is needed.  The default is 'max(sizeof

  41172.      (size_t), alignof(type))', as defined in section 2.7 of the

  41173.      IA64/Generic C++ ABI.

  41174. 

  41175.  -- Target Hook: bool TARGET_CXX_COOKIE_HAS_SIZE (void)

  41176.      This hook should return 'true' if the element size should be stored

  41177.      in array cookies.  The default is to return 'false'.

  41178. 

  41179.  -- Target Hook: int TARGET_CXX_IMPORT_EXPORT_CLASS (tree TYPE, int

  41180.           IMPORT_EXPORT)

  41181.      If defined by a backend this hook allows the decision made to

  41182.      export class TYPE to be overruled.  Upon entry IMPORT_EXPORT will

  41183.      contain 1 if the class is going to be exported, -1 if it is going

  41184.      to be imported and 0 otherwise.  This function should return the

  41185.      modified value and perform any other actions necessary to support

  41186.      the backend's targeted operating system.

  41187. 

  41188.  -- Target Hook: bool TARGET_CXX_CDTOR_RETURNS_THIS (void)

  41189.      This hook should return 'true' if constructors and destructors

  41190.      return the address of the object created/destroyed.  The default is

  41191.      to return 'false'.

  41192. 

  41193.  -- Target Hook: bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void)

  41194.      This hook returns true if the key method for a class (i.e., the

  41195.      method which, if defined in the current translation unit, causes

  41196.      the virtual table to be emitted) may be an inline function.  Under

  41197.      the standard Itanium C++ ABI the key method may be an inline

  41198.      function so long as the function is not declared inline in the

  41199.      class definition.  Under some variants of the ABI, an inline

  41200.      function can never be the key method.  The default is to return

  41201.      'true'.

  41202. 

  41203.  -- Target Hook: void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree

  41204.           DECL)

  41205.      DECL is a virtual table, virtual table table, typeinfo object, or

  41206.      other similar implicit class data object that will be emitted with

  41207.      external linkage in this translation unit.  No ELF visibility has

  41208.      been explicitly specified.  If the target needs to specify a

  41209.      visibility other than that of the containing class, use this hook

  41210.      to set 'DECL_VISIBILITY' and 'DECL_VISIBILITY_SPECIFIED'.

  41211. 

  41212.  -- Target Hook: bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void)

  41213.      This hook returns true (the default) if virtual tables and other

  41214.      similar implicit class data objects are always COMDAT if they have

  41215.      external linkage.  If this hook returns false, then class data for

  41216.      classes whose virtual table will be emitted in only one translation

  41217.      unit will not be COMDAT.

  41218. 

  41219.  -- Target Hook: bool TARGET_CXX_LIBRARY_RTTI_COMDAT (void)

  41220.      This hook returns true (the default) if the RTTI information for

  41221.      the basic types which is defined in the C++ runtime should always

  41222.      be COMDAT, false if it should not be COMDAT.

  41223. 

  41224.  -- Target Hook: bool TARGET_CXX_USE_AEABI_ATEXIT (void)

  41225.      This hook returns true if '__aeabi_atexit' (as defined by the ARM

  41226.      EABI) should be used to register static destructors when

  41227.      '-fuse-cxa-atexit' is in effect.  The default is to return false to

  41228.      use '__cxa_atexit'.

  41229. 

  41230.  -- Target Hook: bool TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT (void)

  41231.      This hook returns true if the target 'atexit' function can be used

  41232.      in the same manner as '__cxa_atexit' to register C++ static

  41233.      destructors.  This requires that 'atexit'-registered functions in

  41234.      shared libraries are run in the correct order when the libraries

  41235.      are unloaded.  The default is to return false.

  41236. 

  41237.  -- Target Hook: void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree TYPE)

  41238.      TYPE is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just

  41239.      been defined.  Use this hook to make adjustments to the class (eg,

  41240.      tweak visibility or perform any other required target

  41241.      modifications).

  41242. 

  41243.  -- Target Hook: tree TARGET_CXX_DECL_MANGLING_CONTEXT (const_tree DECL)

  41244.      Return target-specific mangling context of DECL or 'NULL_TREE'.

  41245. 

  41246. 

  41247. File: gccint.info,  Node: D Language and ABI,  Next: Named Address Spaces,  Prev: C++ ABI,  Up: Target Macros

  41248. 

  41249. 18.29 D ABI parameters

  41250. ======================

  41251. 

  41252.  -- D Target Hook: void TARGET_D_CPU_VERSIONS (void)

  41253.      Declare all environmental version identifiers relating to the

  41254.      target CPU using the function 'builtin_version', which takes a

  41255.      string representing the name of the version.  Version identifiers

  41256.      predefined by this hook apply to all modules that are being

  41257.      compiled and imported.

  41258. 

  41259.  -- D Target Hook: void TARGET_D_OS_VERSIONS (void)

  41260.      Similarly to 'TARGET_D_CPU_VERSIONS', but is used for versions

  41261.      relating to the target operating system.

  41262. 

  41263.  -- D Target Hook: unsigned TARGET_D_CRITSEC_SIZE (void)

  41264.      Returns the size of the data structure used by the target operating

  41265.      system for critical sections and monitors.  For example, on

  41266.      Microsoft Windows this would return the 'sizeof(CRITICAL_SECTION)',

  41267.      while other platforms that implement pthreads would return

  41268.      'sizeof(pthread_mutex_t)'.

  41269. 

  41270. 

  41271. File: gccint.info,  Node: Named Address Spaces,  Next: Misc,  Prev: D Language and ABI,  Up: Target Macros

  41272. 

  41273. 18.30 Adding support for named address spaces

  41274. =============================================

  41275. 

  41276. The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275 standards

  41277. committee, 'Programming Languages - C - Extensions to support embedded

  41278. processors', specifies a syntax for embedded processors to specify

  41279. alternate address spaces.  You can configure a GCC port to support

  41280. section 5.1 of the draft report to add support for address spaces other

  41281. than the default address space.  These address spaces are new keywords

  41282. that are similar to the 'volatile' and 'const' type attributes.

  41283. 

  41284.  Pointers to named address spaces can have a different size than

  41285. pointers to the generic address space.

  41286. 

  41287.  For example, the SPU port uses the '__ea' address space to refer to

  41288. memory in the host processor, rather than memory local to the SPU

  41289. processor.  Access to memory in the '__ea' address space involves

  41290. issuing DMA operations to move data between the host processor and the

  41291. local processor memory address space.  Pointers in the '__ea' address

  41292. space are either 32 bits or 64 bits based on the '-mea32' or '-mea64'

  41293. switches (native SPU pointers are always 32 bits).

  41294. 

  41295.  Internally, address spaces are represented as a small integer in the

  41296. range 0 to 15 with address space 0 being reserved for the generic

  41297. address space.

  41298. 

  41299.  To register a named address space qualifier keyword with the C front

  41300. end, the target may call the 'c_register_addr_space' routine.  For

  41301. example, the SPU port uses the following to declare '__ea' as the

  41302. keyword for named address space #1:

  41303.      #define ADDR_SPACE_EA 1

  41304.      c_register_addr_space ("__ea", ADDR_SPACE_EA);

  41305. 

  41306.  -- Target Hook: scalar_int_mode TARGET_ADDR_SPACE_POINTER_MODE

  41307.           (addr_space_t ADDRESS_SPACE)

  41308.      Define this to return the machine mode to use for pointers to

  41309.      ADDRESS_SPACE if the target supports named address spaces.  The

  41310.      default version of this hook returns 'ptr_mode'.

  41311. 

  41312.  -- Target Hook: scalar_int_mode TARGET_ADDR_SPACE_ADDRESS_MODE

  41313.           (addr_space_t ADDRESS_SPACE)

  41314.      Define this to return the machine mode to use for addresses in

  41315.      ADDRESS_SPACE if the target supports named address spaces.  The

  41316.      default version of this hook returns 'Pmode'.

  41317. 

  41318.  -- Target Hook: bool TARGET_ADDR_SPACE_VALID_POINTER_MODE

  41319.           (scalar_int_mode MODE, addr_space_t AS)

  41320.      Define this to return nonzero if the port can handle pointers with

  41321.      machine mode MODE to address space AS.  This target hook is the

  41322.      same as the 'TARGET_VALID_POINTER_MODE' target hook, except that it

  41323.      includes explicit named address space support.  The default version

  41324.      of this hook returns true for the modes returned by either the

  41325.      'TARGET_ADDR_SPACE_POINTER_MODE' or

  41326.      'TARGET_ADDR_SPACE_ADDRESS_MODE' target hooks for the given address

  41327.      space.

  41328. 

  41329.  -- Target Hook: bool TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P

  41330.           (machine_mode MODE, rtx EXP, bool STRICT, addr_space_t AS)

  41331.      Define this to return true if EXP is a valid address for mode MODE

  41332.      in the named address space AS.  The STRICT parameter says whether

  41333.      strict addressing is in effect after reload has finished.  This

  41334.      target hook is the same as the 'TARGET_LEGITIMATE_ADDRESS_P' target

  41335.      hook, except that it includes explicit named address space support.

  41336. 

  41337.  -- Target Hook: rtx TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS (rtx X, rtx

  41338.           OLDX, machine_mode MODE, addr_space_t AS)

  41339.      Define this to modify an invalid address X to be a valid address

  41340.      with mode MODE in the named address space AS.  This target hook is

  41341.      the same as the 'TARGET_LEGITIMIZE_ADDRESS' target hook, except

  41342.      that it includes explicit named address space support.

  41343. 

  41344.  -- Target Hook: bool TARGET_ADDR_SPACE_SUBSET_P (addr_space_t SUBSET,

  41345.           addr_space_t SUPERSET)

  41346.      Define this to return whether the SUBSET named address space is

  41347.      contained within the SUPERSET named address space.  Pointers to a

  41348.      named address space that is a subset of another named address space

  41349.      will be converted automatically without a cast if used together in

  41350.      arithmetic operations.  Pointers to a superset address space can be

  41351.      converted to pointers to a subset address space via explicit casts.

  41352. 

  41353.  -- Target Hook: bool TARGET_ADDR_SPACE_ZERO_ADDRESS_VALID (addr_space_t

  41354.           AS)

  41355.      Define this to modify the default handling of address 0 for the

  41356.      address space.  Return true if 0 should be considered a valid

  41357.      address.

  41358. 

  41359.  -- Target Hook: rtx TARGET_ADDR_SPACE_CONVERT (rtx OP, tree FROM_TYPE,

  41360.           tree TO_TYPE)

  41361.      Define this to convert the pointer expression represented by the

  41362.      RTL OP with type FROM_TYPE that points to a named address space to

  41363.      a new pointer expression with type TO_TYPE that points to a

  41364.      different named address space.  When this hook it called, it is

  41365.      guaranteed that one of the two address spaces is a subset of the

  41366.      other, as determined by the 'TARGET_ADDR_SPACE_SUBSET_P' target

  41367.      hook.

  41368. 

  41369.  -- Target Hook: int TARGET_ADDR_SPACE_DEBUG (addr_space_t AS)

  41370.      Define this to define how the address space is encoded in dwarf.

  41371.      The result is the value to be used with 'DW_AT_address_class'.

  41372. 

  41373.  -- Target Hook: void TARGET_ADDR_SPACE_DIAGNOSE_USAGE (addr_space_t AS,

  41374.           location_t LOC)

  41375.      Define this hook if the availability of an address space depends on

  41376.      command line options and some diagnostics should be printed when

  41377.      the address space is used.  This hook is called during parsing and

  41378.      allows to emit a better diagnostic compared to the case where the

  41379.      address space was not registered with 'c_register_addr_space'.  AS

  41380.      is the address space as registered with 'c_register_addr_space'.

  41381.      LOC is the location of the address space qualifier token.  The

  41382.      default implementation does nothing.

  41383. 

  41384. 

  41385. File: gccint.info,  Node: Misc,  Prev: Named Address Spaces,  Up: Target Macros

  41386. 

  41387. 18.31 Miscellaneous Parameters

  41388. ==============================

  41389. 

  41390. Here are several miscellaneous parameters.

  41391. 

  41392.  -- Macro: HAS_LONG_COND_BRANCH

  41393.      Define this boolean macro to indicate whether or not your

  41394.      architecture has conditional branches that can span all of memory.

  41395.      It is used in conjunction with an optimization that partitions hot

  41396.      and cold basic blocks into separate sections of the executable.  If

  41397.      this macro is set to false, gcc will convert any conditional

  41398.      branches that attempt to cross between sections into unconditional

  41399.      branches or indirect jumps.

  41400. 

  41401.  -- Macro: HAS_LONG_UNCOND_BRANCH

  41402.      Define this boolean macro to indicate whether or not your

  41403.      architecture has unconditional branches that can span all of

  41404.      memory.  It is used in conjunction with an optimization that

  41405.      partitions hot and cold basic blocks into separate sections of the

  41406.      executable.  If this macro is set to false, gcc will convert any

  41407.      unconditional branches that attempt to cross between sections into

  41408.      indirect jumps.

  41409. 

  41410.  -- Macro: CASE_VECTOR_MODE

  41411.      An alias for a machine mode name.  This is the machine mode that

  41412.      elements of a jump-table should have.

  41413. 

  41414.  -- Macro: CASE_VECTOR_SHORTEN_MODE (MIN_OFFSET, MAX_OFFSET, BODY)

  41415.      Optional: return the preferred mode for an 'addr_diff_vec' when the

  41416.      minimum and maximum offset are known.  If you define this, it

  41417.      enables extra code in branch shortening to deal with

  41418.      'addr_diff_vec'.  To make this work, you also have to define

  41419.      'INSN_ALIGN' and make the alignment for 'addr_diff_vec' explicit.

  41420.      The BODY argument is provided so that the offset_unsigned and scale

  41421.      flags can be updated.

  41422. 

  41423.  -- Macro: CASE_VECTOR_PC_RELATIVE

  41424.      Define this macro to be a C expression to indicate when jump-tables

  41425.      should contain relative addresses.  You need not define this macro

  41426.      if jump-tables never contain relative addresses, or jump-tables

  41427.      should contain relative addresses only when '-fPIC' or '-fPIC' is

  41428.      in effect.

  41429. 

  41430.  -- Target Hook: unsigned int TARGET_CASE_VALUES_THRESHOLD (void)

  41431.      This function return the smallest number of different values for

  41432.      which it is best to use a jump-table instead of a tree of

  41433.      conditional branches.  The default is four for machines with a

  41434.      'casesi' instruction and five otherwise.  This is best for most

  41435.      machines.

  41436. 

  41437.  -- Macro: WORD_REGISTER_OPERATIONS

  41438.      Define this macro to 1 if operations between registers with

  41439.      integral mode smaller than a word are always performed on the

  41440.      entire register.  To be more explicit, if you start with a pair of

  41441.      'word_mode' registers with known values and you do a subword, for

  41442.      example 'QImode', addition on the low part of the registers, then

  41443.      the compiler may consider that the result has a known value in

  41444.      'word_mode' too if the macro is defined to 1.  Most RISC machines

  41445.      have this property and most CISC machines do not.

  41446. 

  41447.  -- Target Hook: unsigned int TARGET_MIN_ARITHMETIC_PRECISION (void)

  41448.      On some RISC architectures with 64-bit registers, the processor

  41449.      also maintains 32-bit condition codes that make it possible to do

  41450.      real 32-bit arithmetic, although the operations are performed on

  41451.      the full registers.

  41452. 

  41453.      On such architectures, defining this hook to 32 tells the compiler

  41454.      to try using 32-bit arithmetical operations setting the condition

  41455.      codes instead of doing full 64-bit arithmetic.

  41456. 

  41457.      More generally, define this hook on RISC architectures if you want

  41458.      the compiler to try using arithmetical operations setting the

  41459.      condition codes with a precision lower than the word precision.

  41460. 

  41461.      You need not define this hook if 'WORD_REGISTER_OPERATIONS' is not

  41462.      defined to 1.

  41463. 

  41464.  -- Macro: LOAD_EXTEND_OP (MEM_MODE)

  41465.      Define this macro to be a C expression indicating when insns that

  41466.      read memory in MEM_MODE, an integral mode narrower than a word, set

  41467.      the bits outside of MEM_MODE to be either the sign-extension or the

  41468.      zero-extension of the data read.  Return 'SIGN_EXTEND' for values

  41469.      of MEM_MODE for which the insn sign-extends, 'ZERO_EXTEND' for

  41470.      which it zero-extends, and 'UNKNOWN' for other modes.

  41471. 

  41472.      This macro is not called with MEM_MODE non-integral or with a width

  41473.      greater than or equal to 'BITS_PER_WORD', so you may return any

  41474.      value in this case.  Do not define this macro if it would always

  41475.      return 'UNKNOWN'.  On machines where this macro is defined, you

  41476.      will normally define it as the constant 'SIGN_EXTEND' or

  41477.      'ZERO_EXTEND'.

  41478. 

  41479.      You may return a non-'UNKNOWN' value even if for some hard

  41480.      registers the sign extension is not performed, if for the

  41481.      'REGNO_REG_CLASS' of these hard registers

  41482.      'TARGET_CAN_CHANGE_MODE_CLASS' returns false when the FROM mode is

  41483.      MEM_MODE and the TO mode is any integral mode larger than this but

  41484.      not larger than 'word_mode'.

  41485. 

  41486.      You must return 'UNKNOWN' if for some hard registers that allow

  41487.      this mode, 'TARGET_CAN_CHANGE_MODE_CLASS' says that they cannot

  41488.      change to 'word_mode', but that they can change to another integral

  41489.      mode that is larger then MEM_MODE but still smaller than

  41490.      'word_mode'.

  41491. 

  41492.  -- Macro: SHORT_IMMEDIATES_SIGN_EXTEND

  41493.      Define this macro to 1 if loading short immediate values into

  41494.      registers sign extends.

  41495. 

  41496.  -- Target Hook: unsigned int TARGET_MIN_DIVISIONS_FOR_RECIP_MUL

  41497.           (machine_mode MODE)

  41498.      When '-ffast-math' is in effect, GCC tries to optimize divisions by

  41499.      the same divisor, by turning them into multiplications by the

  41500.      reciprocal.  This target hook specifies the minimum number of

  41501.      divisions that should be there for GCC to perform the optimization

  41502.      for a variable of mode MODE.  The default implementation returns 3

  41503.      if the machine has an instruction for the division, and 2 if it

  41504.      does not.

  41505. 

  41506.  -- Macro: MOVE_MAX

  41507.      The maximum number of bytes that a single instruction can move

  41508.      quickly between memory and registers or between two memory

  41509.      locations.

  41510. 

  41511.  -- Macro: MAX_MOVE_MAX

  41512.      The maximum number of bytes that a single instruction can move

  41513.      quickly between memory and registers or between two memory

  41514.      locations.  If this is undefined, the default is 'MOVE_MAX'.

  41515.      Otherwise, it is the constant value that is the largest value that

  41516.      'MOVE_MAX' can have at run-time.

  41517. 

  41518.  -- Macro: SHIFT_COUNT_TRUNCATED

  41519.      A C expression that is nonzero if on this machine the number of

  41520.      bits actually used for the count of a shift operation is equal to

  41521.      the number of bits needed to represent the size of the object being

  41522.      shifted.  When this macro is nonzero, the compiler will assume that

  41523.      it is safe to omit a sign-extend, zero-extend, and certain bitwise

  41524.      'and' instructions that truncates the count of a shift operation.

  41525.      On machines that have instructions that act on bit-fields at

  41526.      variable positions, which may include 'bit test' instructions, a

  41527.      nonzero 'SHIFT_COUNT_TRUNCATED' also enables deletion of

  41528.      truncations of the values that serve as arguments to bit-field

  41529.      instructions.

  41530. 

  41531.      If both types of instructions truncate the count (for shifts) and

  41532.      position (for bit-field operations), or if no variable-position

  41533.      bit-field instructions exist, you should define this macro.

  41534. 

  41535.      However, on some machines, such as the 80386 and the 680x0,

  41536.      truncation only applies to shift operations and not the (real or

  41537.      pretended) bit-field operations.  Define 'SHIFT_COUNT_TRUNCATED' to

  41538.      be zero on such machines.  Instead, add patterns to the 'md' file

  41539.      that include the implied truncation of the shift instructions.

  41540. 

  41541.      You need not define this macro if it would always have the value of

  41542.      zero.

  41543. 

  41544.  -- Target Hook: unsigned HOST_WIDE_INT TARGET_SHIFT_TRUNCATION_MASK

  41545.           (machine_mode MODE)

  41546.      This function describes how the standard shift patterns for MODE

  41547.      deal with shifts by negative amounts or by more than the width of

  41548.      the mode.  *Note shift patterns::.

  41549. 

  41550.      On many machines, the shift patterns will apply a mask M to the

  41551.      shift count, meaning that a fixed-width shift of X by Y is

  41552.      equivalent to an arbitrary-width shift of X by Y & M.  If this is

  41553.      true for mode MODE, the function should return M, otherwise it

  41554.      should return 0.  A return value of 0 indicates that no particular

  41555.      behavior is guaranteed.

  41556. 

  41557.      Note that, unlike 'SHIFT_COUNT_TRUNCATED', this function does _not_

  41558.      apply to general shift rtxes; it applies only to instructions that

  41559.      are generated by the named shift patterns.

  41560. 

  41561.      The default implementation of this function returns

  41562.      'GET_MODE_BITSIZE (MODE) - 1' if 'SHIFT_COUNT_TRUNCATED' and 0

  41563.      otherwise.  This definition is always safe, but if

  41564.      'SHIFT_COUNT_TRUNCATED' is false, and some shift patterns

  41565.      nevertheless truncate the shift count, you may get better code by

  41566.      overriding it.

  41567. 

  41568.  -- Target Hook: bool TARGET_TRULY_NOOP_TRUNCATION (poly_uint64 OUTPREC,

  41569.           poly_uint64 INPREC)

  41570.      This hook returns true if it is safe to "convert" a value of INPREC

  41571.      bits to one of OUTPREC bits (where OUTPREC is smaller than INPREC)

  41572.      by merely operating on it as if it had only OUTPREC bits.  The

  41573.      default returns true unconditionally, which is correct for most

  41574.      machines.

  41575. 

  41576.      If 'TARGET_MODES_TIEABLE_P' returns false for a pair of modes,

  41577.      suboptimal code can result if this hook returns true for the

  41578.      corresponding mode sizes.  Making this hook return false in such

  41579.      cases may improve things.

  41580. 

  41581.  -- Target Hook: int TARGET_MODE_REP_EXTENDED (scalar_int_mode MODE,

  41582.           scalar_int_mode REP_MODE)

  41583.      The representation of an integral mode can be such that the values

  41584.      are always extended to a wider integral mode.  Return 'SIGN_EXTEND'

  41585.      if values of MODE are represented in sign-extended form to

  41586.      REP_MODE.  Return 'UNKNOWN' otherwise.  (Currently, none of the

  41587.      targets use zero-extended representation this way so unlike

  41588.      'LOAD_EXTEND_OP', 'TARGET_MODE_REP_EXTENDED' is expected to return

  41589.      either 'SIGN_EXTEND' or 'UNKNOWN'.  Also no target extends MODE to

  41590.      REP_MODE so that REP_MODE is not the next widest integral mode and

  41591.      currently we take advantage of this fact.)

  41592. 

  41593.      Similarly to 'LOAD_EXTEND_OP' you may return a non-'UNKNOWN' value

  41594.      even if the extension is not performed on certain hard registers as

  41595.      long as for the 'REGNO_REG_CLASS' of these hard registers

  41596.      'TARGET_CAN_CHANGE_MODE_CLASS' returns false.

  41597. 

  41598.      Note that 'TARGET_MODE_REP_EXTENDED' and 'LOAD_EXTEND_OP' describe

  41599.      two related properties.  If you define 'TARGET_MODE_REP_EXTENDED

  41600.      (mode, word_mode)' you probably also want to define 'LOAD_EXTEND_OP

  41601.      (mode)' to return the same type of extension.

  41602. 

  41603.      In order to enforce the representation of 'mode',

  41604.      'TARGET_TRULY_NOOP_TRUNCATION' should return false when truncating

  41605.      to 'mode'.

  41606. 

  41607.  -- Target Hook: bool TARGET_SETJMP_PRESERVES_NONVOLATILE_REGS_P (void)

  41608.      On some targets, it is assumed that the compiler will spill all

  41609.      pseudos that are live across a call to 'setjmp', while other

  41610.      targets treat 'setjmp' calls as normal function calls.

  41611. 

  41612.      This hook returns false if 'setjmp' calls do not preserve all

  41613.      non-volatile registers so that gcc that must spill all pseudos that

  41614.      are live across 'setjmp' calls.  Define this to return true if the

  41615.      target does not need to spill all pseudos live across 'setjmp'

  41616.      calls.  The default implementation conservatively assumes all

  41617.      pseudos must be spilled across 'setjmp' calls.

  41618. 

  41619.  -- Macro: STORE_FLAG_VALUE

  41620.      A C expression describing the value returned by a comparison

  41621.      operator with an integral mode and stored by a store-flag

  41622.      instruction ('cstoreMODE4') when the condition is true.  This

  41623.      description must apply to _all_ the 'cstoreMODE4' patterns and all

  41624.      the comparison operators whose results have a 'MODE_INT' mode.

  41625. 

  41626.      A value of 1 or -1 means that the instruction implementing the

  41627.      comparison operator returns exactly 1 or -1 when the comparison is

  41628.      true and 0 when the comparison is false.  Otherwise, the value

  41629.      indicates which bits of the result are guaranteed to be 1 when the

  41630.      comparison is true.  This value is interpreted in the mode of the

  41631.      comparison operation, which is given by the mode of the first

  41632.      operand in the 'cstoreMODE4' pattern.  Either the low bit or the

  41633.      sign bit of 'STORE_FLAG_VALUE' be on.  Presently, only those bits

  41634.      are used by the compiler.

  41635. 

  41636.      If 'STORE_FLAG_VALUE' is neither 1 or -1, the compiler will

  41637.      generate code that depends only on the specified bits.  It can also

  41638.      replace comparison operators with equivalent operations if they

  41639.      cause the required bits to be set, even if the remaining bits are

  41640.      undefined.  For example, on a machine whose comparison operators

  41641.      return an 'SImode' value and where 'STORE_FLAG_VALUE' is defined as

  41642.      '0x80000000', saying that just the sign bit is relevant, the

  41643.      expression

  41644. 

  41645.           (ne:SI (and:SI X (const_int POWER-OF-2)) (const_int 0))

  41646. 

  41647.      can be converted to

  41648. 

  41649.           (ashift:SI X (const_int N))

  41650. 

  41651.      where N is the appropriate shift count to move the bit being tested

  41652.      into the sign bit.

  41653. 

  41654.      There is no way to describe a machine that always sets the

  41655.      low-order bit for a true value, but does not guarantee the value of

  41656.      any other bits, but we do not know of any machine that has such an

  41657.      instruction.  If you are trying to port GCC to such a machine,

  41658.      include an instruction to perform a logical-and of the result with

  41659.      1 in the pattern for the comparison operators and let us know at

  41660.      <gcc@gcc.gnu.org>.

  41661. 

  41662.      Often, a machine will have multiple instructions that obtain a

  41663.      value from a comparison (or the condition codes).  Here are rules

  41664.      to guide the choice of value for 'STORE_FLAG_VALUE', and hence the

  41665.      instructions to be used:

  41666. 

  41667.         * Use the shortest sequence that yields a valid definition for

  41668.           'STORE_FLAG_VALUE'.  It is more efficient for the compiler to

  41669.           "normalize" the value (convert it to, e.g., 1 or 0) than for

  41670.           the comparison operators to do so because there may be

  41671.           opportunities to combine the normalization with other

  41672.           operations.

  41673. 

  41674.         * For equal-length sequences, use a value of 1 or -1, with -1

  41675.           being slightly preferred on machines with expensive jumps and

  41676.           1 preferred on other machines.

  41677. 

  41678.         * As a second choice, choose a value of '0x80000001' if

  41679.           instructions exist that set both the sign and low-order bits

  41680.           but do not define the others.

  41681. 

  41682.         * Otherwise, use a value of '0x80000000'.

  41683. 

  41684.      Many machines can produce both the value chosen for

  41685.      'STORE_FLAG_VALUE' and its negation in the same number of

  41686.      instructions.  On those machines, you should also define a pattern

  41687.      for those cases, e.g., one matching

  41688. 

  41689.           (set A (neg:M (ne:M B C)))

  41690. 

  41691.      Some machines can also perform 'and' or 'plus' operations on

  41692.      condition code values with less instructions than the corresponding

  41693.      'cstoreMODE4' insn followed by 'and' or 'plus'.  On those machines,

  41694.      define the appropriate patterns.  Use the names 'incscc' and

  41695.      'decscc', respectively, for the patterns which perform 'plus' or

  41696.      'minus' operations on condition code values.  See 'rs6000.md' for

  41697.      some examples.  The GNU Superoptimizer can be used to find such

  41698.      instruction sequences on other machines.

  41699. 

  41700.      If this macro is not defined, the default value, 1, is used.  You

  41701.      need not define 'STORE_FLAG_VALUE' if the machine has no store-flag

  41702.      instructions, or if the value generated by these instructions is 1.

  41703. 

  41704.  -- Macro: FLOAT_STORE_FLAG_VALUE (MODE)

  41705.      A C expression that gives a nonzero 'REAL_VALUE_TYPE' value that is

  41706.      returned when comparison operators with floating-point results are

  41707.      true.  Define this macro on machines that have comparison

  41708.      operations that return floating-point values.  If there are no such

  41709.      operations, do not define this macro.

  41710. 

  41711.  -- Macro: VECTOR_STORE_FLAG_VALUE (MODE)

  41712.      A C expression that gives a rtx representing the nonzero true

  41713.      element for vector comparisons.  The returned rtx should be valid

  41714.      for the inner mode of MODE which is guaranteed to be a vector mode.

  41715.      Define this macro on machines that have vector comparison

  41716.      operations that return a vector result.  If there are no such

  41717.      operations, do not define this macro.  Typically, this macro is

  41718.      defined as 'const1_rtx' or 'constm1_rtx'.  This macro may return

  41719.      'NULL_RTX' to prevent the compiler optimizing such vector

  41720.      comparison operations for the given mode.

  41721. 

  41722.  -- Macro: CLZ_DEFINED_VALUE_AT_ZERO (MODE, VALUE)

  41723.  -- Macro: CTZ_DEFINED_VALUE_AT_ZERO (MODE, VALUE)

  41724.      A C expression that indicates whether the architecture defines a

  41725.      value for 'clz' or 'ctz' with a zero operand.  A result of '0'

  41726.      indicates the value is undefined.  If the value is defined for only

  41727.      the RTL expression, the macro should evaluate to '1'; if the value

  41728.      applies also to the corresponding optab entry (which is normally

  41729.      the case if it expands directly into the corresponding RTL), then

  41730.      the macro should evaluate to '2'.  In the cases where the value is

  41731.      defined, VALUE should be set to this value.

  41732. 

  41733.      If this macro is not defined, the value of 'clz' or 'ctz' at zero

  41734.      is assumed to be undefined.

  41735. 

  41736.      This macro must be defined if the target's expansion for 'ffs'

  41737.      relies on a particular value to get correct results.  Otherwise it

  41738.      is not necessary, though it may be used to optimize some corner

  41739.      cases, and to provide a default expansion for the 'ffs' optab.

  41740. 

  41741.      Note that regardless of this macro the "definedness" of 'clz' and

  41742.      'ctz' at zero do _not_ extend to the builtin functions visible to

  41743.      the user.  Thus one may be free to adjust the value at will to

  41744.      match the target expansion of these operations without fear of

  41745.      breaking the API.

  41746. 

  41747.  -- Macro: Pmode

  41748.      An alias for the machine mode for pointers.  On most machines,

  41749.      define this to be the integer mode corresponding to the width of a

  41750.      hardware pointer; 'SImode' on 32-bit machine or 'DImode' on 64-bit

  41751.      machines.  On some machines you must define this to be one of the

  41752.      partial integer modes, such as 'PSImode'.

  41753. 

  41754.      The width of 'Pmode' must be at least as large as the value of

  41755.      'POINTER_SIZE'.  If it is not equal, you must define the macro

  41756.      'POINTERS_EXTEND_UNSIGNED' to specify how pointers are extended to

  41757.      'Pmode'.

  41758. 

  41759.  -- Macro: FUNCTION_MODE

  41760.      An alias for the machine mode used for memory references to

  41761.      functions being called, in 'call' RTL expressions.  On most CISC

  41762.      machines, where an instruction can begin at any byte address, this

  41763.      should be 'QImode'.  On most RISC machines, where all instructions

  41764.      have fixed size and alignment, this should be a mode with the same

  41765.      size and alignment as the machine instruction words - typically

  41766.      'SImode' or 'HImode'.

  41767. 

  41768.  -- Macro: STDC_0_IN_SYSTEM_HEADERS

  41769.      In normal operation, the preprocessor expands '__STDC__' to the

  41770.      constant 1, to signify that GCC conforms to ISO Standard C.  On

  41771.      some hosts, like Solaris, the system compiler uses a different

  41772.      convention, where '__STDC__' is normally 0, but is 1 if the user

  41773.      specifies strict conformance to the C Standard.

  41774. 

  41775.      Defining 'STDC_0_IN_SYSTEM_HEADERS' makes GNU CPP follows the host

  41776.      convention when processing system header files, but when processing

  41777.      user files '__STDC__' will always expand to 1.

  41778. 

  41779.  -- C Target Hook: const char * TARGET_C_PREINCLUDE (void)

  41780.      Define this hook to return the name of a header file to be included

  41781.      at the start of all compilations, as if it had been included with

  41782.      '#include <FILE>'.  If this hook returns 'NULL', or is not defined,

  41783.      or the header is not found, or if the user specifies

  41784.      '-ffreestanding' or '-nostdinc', no header is included.

  41785. 

  41786.      This hook can be used together with a header provided by the system

  41787.      C library to implement ISO C requirements for certain macros to be

  41788.      predefined that describe properties of the whole implementation

  41789.      rather than just the compiler.

  41790. 

  41791.  -- C Target Hook: bool TARGET_CXX_IMPLICIT_EXTERN_C (const char*)

  41792.      Define this hook to add target-specific C++ implicit extern C

  41793.      functions.  If this function returns true for the name of a

  41794.      file-scope function, that function implicitly gets extern "C"

  41795.      linkage rather than whatever language linkage the declaration would

  41796.      normally have.  An example of such function is WinMain on Win32

  41797.      targets.

  41798. 

  41799.  -- Macro: SYSTEM_IMPLICIT_EXTERN_C

  41800.      Define this macro if the system header files do not support C++.

  41801.      This macro handles system header files by pretending that system

  41802.      header files are enclosed in 'extern "C" {...}'.

  41803. 

  41804.  -- Macro: REGISTER_TARGET_PRAGMAS ()

  41805.      Define this macro if you want to implement any target-specific

  41806.      pragmas.  If defined, it is a C expression which makes a series of

  41807.      calls to 'c_register_pragma' or 'c_register_pragma_with_expansion'

  41808.      for each pragma.  The macro may also do any setup required for the

  41809.      pragmas.

  41810. 

  41811.      The primary reason to define this macro is to provide compatibility

  41812.      with other compilers for the same target.  In general, we

  41813.      discourage definition of target-specific pragmas for GCC.

  41814. 

  41815.      If the pragma can be implemented by attributes then you should

  41816.      consider defining the target hook 'TARGET_INSERT_ATTRIBUTES' as

  41817.      well.

  41818. 

  41819.      Preprocessor macros that appear on pragma lines are not expanded.

  41820.      All '#pragma' directives that do not match any registered pragma

  41821.      are silently ignored, unless the user specifies

  41822.      '-Wunknown-pragmas'.

  41823. 

  41824.  -- Function: void c_register_pragma (const char *SPACE, const char

  41825.           *NAME, void (*CALLBACK) (struct cpp_reader *))

  41826.  -- Function: void c_register_pragma_with_expansion (const char *SPACE,

  41827.           const char *NAME, void (*CALLBACK) (struct cpp_reader *))

  41828. 

  41829.      Each call to 'c_register_pragma' or

  41830.      'c_register_pragma_with_expansion' establishes one pragma.  The

  41831.      CALLBACK routine will be called when the preprocessor encounters a

  41832.      pragma of the form

  41833. 

  41834.           #pragma [SPACE] NAME ...

  41835. 

  41836.      SPACE is the case-sensitive namespace of the pragma, or 'NULL' to

  41837.      put the pragma in the global namespace.  The callback routine

  41838.      receives PFILE as its first argument, which can be passed on to

  41839.      cpplib's functions if necessary.  You can lex tokens after the NAME

  41840.      by calling 'pragma_lex'.  Tokens that are not read by the callback

  41841.      will be silently ignored.  The end of the line is indicated by a

  41842.      token of type 'CPP_EOF'.  Macro expansion occurs on the arguments

  41843.      of pragmas registered with 'c_register_pragma_with_expansion' but

  41844.      not on the arguments of pragmas registered with

  41845.      'c_register_pragma'.

  41846. 

  41847.      Note that the use of 'pragma_lex' is specific to the C and C++

  41848.      compilers.  It will not work in the Java or Fortran compilers, or

  41849.      any other language compilers for that matter.  Thus if 'pragma_lex'

  41850.      is going to be called from target-specific code, it must only be

  41851.      done so when building the C and C++ compilers.  This can be done by

  41852.      defining the variables 'c_target_objs' and 'cxx_target_objs' in the

  41853.      target entry in the 'config.gcc' file.  These variables should name

  41854.      the target-specific, language-specific object file which contains

  41855.      the code that uses 'pragma_lex'.  Note it will also be necessary to

  41856.      add a rule to the makefile fragment pointed to by 'tmake_file' that

  41857.      shows how to build this object file.

  41858. 

  41859.  -- Macro: HANDLE_PRAGMA_PACK_WITH_EXPANSION

  41860.      Define this macro if macros should be expanded in the arguments of

  41861.      '#pragma pack'.

  41862. 

  41863.  -- Macro: TARGET_DEFAULT_PACK_STRUCT

  41864.      If your target requires a structure packing default other than 0

  41865.      (meaning the machine default), define this macro to the necessary

  41866.      value (in bytes).  This must be a value that would also be valid to

  41867.      use with '#pragma pack()' (that is, a small power of two).

  41868. 

  41869.  -- Macro: DOLLARS_IN_IDENTIFIERS

  41870.      Define this macro to control use of the character '$' in identifier

  41871.      names for the C family of languages.  0 means '$' is not allowed by

  41872.      default; 1 means it is allowed.  1 is the default; there is no need

  41873.      to define this macro in that case.

  41874. 

  41875.  -- Macro: INSN_SETS_ARE_DELAYED (INSN)

  41876.      Define this macro as a C expression that is nonzero if it is safe

  41877.      for the delay slot scheduler to place instructions in the delay

  41878.      slot of INSN, even if they appear to use a resource set or

  41879.      clobbered in INSN.  INSN is always a 'jump_insn' or an 'insn'; GCC

  41880.      knows that every 'call_insn' has this behavior.  On machines where

  41881.      some 'insn' or 'jump_insn' is really a function call and hence has

  41882.      this behavior, you should define this macro.

  41883. 

  41884.      You need not define this macro if it would always return zero.

  41885. 

  41886.  -- Macro: INSN_REFERENCES_ARE_DELAYED (INSN)

  41887.      Define this macro as a C expression that is nonzero if it is safe

  41888.      for the delay slot scheduler to place instructions in the delay

  41889.      slot of INSN, even if they appear to set or clobber a resource

  41890.      referenced in INSN.  INSN is always a 'jump_insn' or an 'insn'.  On

  41891.      machines where some 'insn' or 'jump_insn' is really a function call

  41892.      and its operands are registers whose use is actually in the

  41893.      subroutine it calls, you should define this macro.  Doing so allows

  41894.      the delay slot scheduler to move instructions which copy arguments

  41895.      into the argument registers into the delay slot of INSN.

  41896. 

  41897.      You need not define this macro if it would always return zero.

  41898. 

  41899.  -- Macro: MULTIPLE_SYMBOL_SPACES

  41900.      Define this macro as a C expression that is nonzero if, in some

  41901.      cases, global symbols from one translation unit may not be bound to

  41902.      undefined symbols in another translation unit without user

  41903.      intervention.  For instance, under Microsoft Windows symbols must

  41904.      be explicitly imported from shared libraries (DLLs).

  41905. 

  41906.      You need not define this macro if it would always evaluate to zero.

  41907. 

  41908.  -- Target Hook: rtx_insn * TARGET_MD_ASM_ADJUST (vec<rtx>& OUTPUTS,

  41909.           vec<rtx>& INPUTS, vec<const char *>& CONSTRAINTS, vec<rtx>&

  41910.           CLOBBERS, HARD_REG_SET& CLOBBERED_REGS)

  41911.      This target hook may add "clobbers" to CLOBBERS and CLOBBERED_REGS

  41912.      for any hard regs the port wishes to automatically clobber for an

  41913.      asm.  The OUTPUTS and INPUTS may be inspected to avoid clobbering a

  41914.      register that is already used by the asm.

  41915. 

  41916.      It may modify the OUTPUTS, INPUTS, and CONSTRAINTS as necessary for

  41917.      other pre-processing.  In this case the return value is a sequence

  41918.      of insns to emit after the asm.

  41919. 

  41920.  -- Macro: MATH_LIBRARY

  41921.      Define this macro as a C string constant for the linker argument to

  41922.      link in the system math library, minus the initial '"-l"', or '""'

  41923.      if the target does not have a separate math library.

  41924. 

  41925.      You need only define this macro if the default of '"m"' is wrong.

  41926. 

  41927.  -- Macro: LIBRARY_PATH_ENV

  41928.      Define this macro as a C string constant for the environment

  41929.      variable that specifies where the linker should look for libraries.

  41930. 

  41931.      You need only define this macro if the default of '"LIBRARY_PATH"'

  41932.      is wrong.

  41933. 

  41934.  -- Macro: TARGET_POSIX_IO

  41935.      Define this macro if the target supports the following POSIX file

  41936.      functions, access, mkdir and file locking with fcntl / F_SETLKW.

  41937.      Defining 'TARGET_POSIX_IO' will enable the test coverage code to

  41938.      use file locking when exiting a program, which avoids race

  41939.      conditions if the program has forked.  It will also create

  41940.      directories at run-time for cross-profiling.

  41941. 

  41942.  -- Macro: MAX_CONDITIONAL_EXECUTE

  41943. 

  41944.      A C expression for the maximum number of instructions to execute

  41945.      via conditional execution instructions instead of a branch.  A

  41946.      value of 'BRANCH_COST'+1 is the default if the machine does not use

  41947.      cc0, and 1 if it does use cc0.

  41948. 

  41949.  -- Macro: IFCVT_MODIFY_TESTS (CE_INFO, TRUE_EXPR, FALSE_EXPR)

  41950.      Used if the target needs to perform machine-dependent modifications

  41951.      on the conditionals used for turning basic blocks into

  41952.      conditionally executed code.  CE_INFO points to a data structure,

  41953.      'struct ce_if_block', which contains information about the

  41954.      currently processed blocks.  TRUE_EXPR and FALSE_EXPR are the tests

  41955.      that are used for converting the then-block and the else-block,

  41956.      respectively.  Set either TRUE_EXPR or FALSE_EXPR to a null pointer

  41957.      if the tests cannot be converted.

  41958. 

  41959.  -- Macro: IFCVT_MODIFY_MULTIPLE_TESTS (CE_INFO, BB, TRUE_EXPR,

  41960.           FALSE_EXPR)

  41961.      Like 'IFCVT_MODIFY_TESTS', but used when converting more

  41962.      complicated if-statements into conditions combined by 'and' and

  41963.      'or' operations.  BB contains the basic block that contains the

  41964.      test that is currently being processed and about to be turned into

  41965.      a condition.

  41966. 

  41967.  -- Macro: IFCVT_MODIFY_INSN (CE_INFO, PATTERN, INSN)

  41968.      A C expression to modify the PATTERN of an INSN that is to be

  41969.      converted to conditional execution format.  CE_INFO points to a

  41970.      data structure, 'struct ce_if_block', which contains information

  41971.      about the currently processed blocks.

  41972. 

  41973.  -- Macro: IFCVT_MODIFY_FINAL (CE_INFO)

  41974.      A C expression to perform any final machine dependent modifications

  41975.      in converting code to conditional execution.  The involved basic

  41976.      blocks can be found in the 'struct ce_if_block' structure that is

  41977.      pointed to by CE_INFO.

  41978. 

  41979.  -- Macro: IFCVT_MODIFY_CANCEL (CE_INFO)

  41980.      A C expression to cancel any machine dependent modifications in

  41981.      converting code to conditional execution.  The involved basic

  41982.      blocks can be found in the 'struct ce_if_block' structure that is

  41983.      pointed to by CE_INFO.

  41984. 

  41985.  -- Macro: IFCVT_MACHDEP_INIT (CE_INFO)

  41986.      A C expression to initialize any machine specific data for

  41987.      if-conversion of the if-block in the 'struct ce_if_block' structure

  41988.      that is pointed to by CE_INFO.

  41989. 

  41990.  -- Target Hook: void TARGET_MACHINE_DEPENDENT_REORG (void)

  41991.      If non-null, this hook performs a target-specific pass over the

  41992.      instruction stream.  The compiler will run it at all optimization

  41993.      levels, just before the point at which it normally does

  41994.      delayed-branch scheduling.

  41995. 

  41996.      The exact purpose of the hook varies from target to target.  Some

  41997.      use it to do transformations that are necessary for correctness,

  41998.      such as laying out in-function constant pools or avoiding hardware

  41999.      hazards.  Others use it as an opportunity to do some

  42000.      machine-dependent optimizations.

  42001. 

  42002.      You need not implement the hook if it has nothing to do.  The

  42003.      default definition is null.

  42004. 

  42005.  -- Target Hook: void TARGET_INIT_BUILTINS (void)

  42006.      Define this hook if you have any machine-specific built-in

  42007.      functions that need to be defined.  It should be a function that

  42008.      performs the necessary setup.

  42009. 

  42010.      Machine specific built-in functions can be useful to expand special

  42011.      machine instructions that would otherwise not normally be generated

  42012.      because they have no equivalent in the source language (for

  42013.      example, SIMD vector instructions or prefetch instructions).

  42014. 

  42015.      To create a built-in function, call the function

  42016.      'lang_hooks.builtin_function' which is defined by the language

  42017.      front end.  You can use any type nodes set up by

  42018.      'build_common_tree_nodes'; only language front ends that use those

  42019.      two functions will call 'TARGET_INIT_BUILTINS'.

  42020. 

  42021.  -- Target Hook: tree TARGET_BUILTIN_DECL (unsigned CODE, bool

  42022.           INITIALIZE_P)

  42023.      Define this hook if you have any machine-specific built-in

  42024.      functions that need to be defined.  It should be a function that

  42025.      returns the builtin function declaration for the builtin function

  42026.      code CODE.  If there is no such builtin and it cannot be

  42027.      initialized at this time if INITIALIZE_P is true the function

  42028.      should return 'NULL_TREE'.  If CODE is out of range the function

  42029.      should return 'error_mark_node'.

  42030. 

  42031.  -- Target Hook: rtx TARGET_EXPAND_BUILTIN (tree EXP, rtx TARGET, rtx

  42032.           SUBTARGET, machine_mode MODE, int IGNORE)

  42033. 

  42034.      Expand a call to a machine specific built-in function that was set

  42035.      up by 'TARGET_INIT_BUILTINS'.  EXP is the expression for the

  42036.      function call; the result should go to TARGET if that is

  42037.      convenient, and have mode MODE if that is convenient.  SUBTARGET

  42038.      may be used as the target for computing one of EXP's operands.

  42039.      IGNORE is nonzero if the value is to be ignored.  This function

  42040.      should return the result of the call to the built-in function.

  42041. 

  42042.  -- Target Hook: tree TARGET_RESOLVE_OVERLOADED_BUILTIN (unsigned int

  42043.           LOC, tree FNDECL, void *ARGLIST)

  42044.      Select a replacement for a machine specific built-in function that

  42045.      was set up by 'TARGET_INIT_BUILTINS'.  This is done _before_

  42046.      regular type checking, and so allows the target to implement a

  42047.      crude form of function overloading.  FNDECL is the declaration of

  42048.      the built-in function.  ARGLIST is the list of arguments passed to

  42049.      the built-in function.  The result is a complete expression that

  42050.      implements the operation, usually another 'CALL_EXPR'.  ARGLIST

  42051.      really has type 'VEC(tree,gc)*'

  42052. 

  42053.  -- Target Hook: bool TARGET_CHECK_BUILTIN_CALL (location_t LOC,

  42054.           vec<location_t> ARG_LOC, tree FNDECL, tree ORIG_FNDECL,

  42055.           unsigned int NARGS, tree *ARGS)

  42056.      Perform semantic checking on a call to a machine-specific built-in

  42057.      function after its arguments have been constrained to the function

  42058.      signature.  Return true if the call is valid, otherwise report an

  42059.      error and return false.

  42060. 

  42061.      This hook is called after 'TARGET_RESOLVE_OVERLOADED_BUILTIN'.  The

  42062.      call was originally to built-in function ORIG_FNDECL, but after the

  42063.      optional 'TARGET_RESOLVE_OVERLOADED_BUILTIN' step is now to

  42064.      built-in function FNDECL.  LOC is the location of the call and ARGS

  42065.      is an array of function arguments, of which there are NARGS.

  42066.      ARG_LOC specifies the location of each argument.

  42067. 

  42068.  -- Target Hook: tree TARGET_FOLD_BUILTIN (tree FNDECL, int N_ARGS, tree

  42069.           *ARGP, bool IGNORE)

  42070.      Fold a call to a machine specific built-in function that was set up

  42071.      by 'TARGET_INIT_BUILTINS'.  FNDECL is the declaration of the

  42072.      built-in function.  N_ARGS is the number of arguments passed to the

  42073.      function; the arguments themselves are pointed to by ARGP.  The

  42074.      result is another tree, valid for both GIMPLE and GENERIC,

  42075.      containing a simplified expression for the call's result.  If

  42076.      IGNORE is true the value will be ignored.

  42077. 

  42078.  -- Target Hook: bool TARGET_GIMPLE_FOLD_BUILTIN (gimple_stmt_iterator

  42079.           *GSI)

  42080.      Fold a call to a machine specific built-in function that was set up

  42081.      by 'TARGET_INIT_BUILTINS'.  GSI points to the gimple statement

  42082.      holding the function call.  Returns true if any change was made to

  42083.      the GIMPLE stream.

  42084. 

  42085.  -- Target Hook: int TARGET_COMPARE_VERSION_PRIORITY (tree DECL1, tree

  42086.           DECL2)

  42087.      This hook is used to compare the target attributes in two functions

  42088.      to determine which function's features get higher priority.  This

  42089.      is used during function multi-versioning to figure out the order in

  42090.      which two versions must be dispatched.  A function version with a

  42091.      higher priority is checked for dispatching earlier.  DECL1 and

  42092.      DECL2 are the two function decls that will be compared.

  42093. 

  42094.  -- Target Hook: tree TARGET_GET_FUNCTION_VERSIONS_DISPATCHER (void

  42095.           *DECL)

  42096.      This hook is used to get the dispatcher function for a set of

  42097.      function versions.  The dispatcher function is called to invoke the

  42098.      right function version at run-time.  DECL is one version from a set

  42099.      of semantically identical versions.

  42100. 

  42101.  -- Target Hook: tree TARGET_GENERATE_VERSION_DISPATCHER_BODY (void

  42102.           *ARG)

  42103.      This hook is used to generate the dispatcher logic to invoke the

  42104.      right function version at run-time for a given set of function

  42105.      versions.  ARG points to the callgraph node of the dispatcher

  42106.      function whose body must be generated.

  42107. 

  42108.  -- Target Hook: bool TARGET_PREDICT_DOLOOP_P (class loop *LOOP)

  42109.      Return true if we can predict it is possible to use a low-overhead

  42110.      loop for a particular loop.  The parameter LOOP is a pointer to the

  42111.      loop.  This target hook is required only when the target supports

  42112.      low-overhead loops, and will help ivopts to make some decisions.

  42113.      The default version of this hook returns false.

  42114. 

  42115.  -- Target Hook: bool TARGET_HAVE_COUNT_REG_DECR_P

  42116.      Return true if the target supports hardware count register for

  42117.      decrement and branch.  The default value is false.

  42118. 

  42119.  -- Target Hook: int64_t TARGET_DOLOOP_COST_FOR_GENERIC

  42120.      One IV candidate dedicated for doloop is introduced in IVOPTs, we

  42121.      can calculate the computation cost of adopting it to any generic IV

  42122.      use by function get_computation_cost as before.  But for targets

  42123.      which have hardware count register support for decrement and

  42124.      branch, it may have to move IV value from hardware count register

  42125.      to general purpose register while doloop IV candidate is used for

  42126.      generic IV uses.  It probably takes expensive penalty.  This hook

  42127.      allows target owners to define the cost for this especially for

  42128.      generic IV uses.  The default value is zero.

  42129. 

  42130.  -- Target Hook: int64_t TARGET_DOLOOP_COST_FOR_ADDRESS

  42131.      One IV candidate dedicated for doloop is introduced in IVOPTs, we

  42132.      can calculate the computation cost of adopting it to any address IV

  42133.      use by function get_computation_cost as before.  But for targets

  42134.      which have hardware count register support for decrement and

  42135.      branch, it may have to move IV value from hardware count register

  42136.      to general purpose register while doloop IV candidate is used for

  42137.      address IV uses.  It probably takes expensive penalty.  This hook

  42138.      allows target owners to define the cost for this escpecially for

  42139.      address IV uses.  The default value is zero.

  42140. 

  42141.  -- Target Hook: bool TARGET_CAN_USE_DOLOOP_P (const widest_int

  42142.           &ITERATIONS, const widest_int &ITERATIONS_MAX, unsigned int

  42143.           LOOP_DEPTH, bool ENTERED_AT_TOP)

  42144.      Return true if it is possible to use low-overhead loops

  42145.      ('doloop_end' and 'doloop_begin') for a particular loop.

  42146.      ITERATIONS gives the exact number of iterations, or 0 if not known.

  42147.      ITERATIONS_MAX gives the maximum number of iterations, or 0 if not

  42148.      known.  LOOP_DEPTH is the nesting depth of the loop, with 1 for

  42149.      innermost loops, 2 for loops that contain innermost loops, and so

  42150.      on.  ENTERED_AT_TOP is true if the loop is only entered from the

  42151.      top.

  42152. 

  42153.      This hook is only used if 'doloop_end' is available.  The default

  42154.      implementation returns true.  You can use

  42155.      'can_use_doloop_if_innermost' if the loop must be the innermost,

  42156.      and if there are no other restrictions.

  42157. 

  42158.  -- Target Hook: const char * TARGET_INVALID_WITHIN_DOLOOP (const

  42159.           rtx_insn *INSN)

  42160. 

  42161.      Take an instruction in INSN and return NULL if it is valid within a

  42162.      low-overhead loop, otherwise return a string explaining why doloop

  42163.      could not be applied.

  42164. 

  42165.      Many targets use special registers for low-overhead looping.  For

  42166.      any instruction that clobbers these this function should return a

  42167.      string indicating the reason why the doloop could not be applied.

  42168.      By default, the RTL loop optimizer does not use a present doloop

  42169.      pattern for loops containing function calls or branch on table

  42170.      instructions.

  42171. 

  42172.  -- Target Hook: bool TARGET_LEGITIMATE_COMBINED_INSN (rtx_insn *INSN)

  42173.      Take an instruction in INSN and return 'false' if the instruction

  42174.      is not appropriate as a combination of two or more instructions.

  42175.      The default is to accept all instructions.

  42176. 

  42177.  -- Target Hook: bool TARGET_CAN_FOLLOW_JUMP (const rtx_insn *FOLLOWER,

  42178.           const rtx_insn *FOLLOWEE)

  42179.      FOLLOWER and FOLLOWEE are JUMP_INSN instructions; return true if

  42180.      FOLLOWER may be modified to follow FOLLOWEE; false, if it can't.

  42181.      For example, on some targets, certain kinds of branches can't be

  42182.      made to follow through a hot/cold partitioning.

  42183. 

  42184.  -- Target Hook: bool TARGET_COMMUTATIVE_P (const_rtx X, int OUTER_CODE)

  42185.      This target hook returns 'true' if X is considered to be

  42186.      commutative.  Usually, this is just COMMUTATIVE_P (X), but the HP

  42187.      PA doesn't consider PLUS to be commutative inside a MEM.

  42188.      OUTER_CODE is the rtx code of the enclosing rtl, if known,

  42189.      otherwise it is UNKNOWN.

  42190. 

  42191.  -- Target Hook: rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx HARD_REG)

  42192. 

  42193.      When the initial value of a hard register has been copied in a

  42194.      pseudo register, it is often not necessary to actually allocate

  42195.      another register to this pseudo register, because the original hard

  42196.      register or a stack slot it has been saved into can be used.

  42197.      'TARGET_ALLOCATE_INITIAL_VALUE' is called at the start of register

  42198.      allocation once for each hard register that had its initial value

  42199.      copied by using 'get_func_hard_reg_initial_val' or

  42200.      'get_hard_reg_initial_val'.  Possible values are 'NULL_RTX', if you

  42201.      don't want to do any special allocation, a 'REG' rtx--that would

  42202.      typically be the hard register itself, if it is known not to be

  42203.      clobbered--or a 'MEM'.  If you are returning a 'MEM', this is only

  42204.      a hint for the allocator; it might decide to use another register

  42205.      anyways.  You may use 'current_function_is_leaf' or 'REG_N_SETS' in

  42206.      the hook to determine if the hard register in question will not be

  42207.      clobbered.  The default value of this hook is 'NULL', which

  42208.      disables any special allocation.

  42209. 

  42210.  -- Target Hook: int TARGET_UNSPEC_MAY_TRAP_P (const_rtx X, unsigned

  42211.           FLAGS)

  42212.      This target hook returns nonzero if X, an 'unspec' or

  42213.      'unspec_volatile' operation, might cause a trap.  Targets can use

  42214.      this hook to enhance precision of analysis for 'unspec' and

  42215.      'unspec_volatile' operations.  You may call 'may_trap_p_1' to

  42216.      analyze inner elements of X in which case FLAGS should be passed

  42217.      along.

  42218. 

  42219.  -- Target Hook: void TARGET_SET_CURRENT_FUNCTION (tree DECL)

  42220.      The compiler invokes this hook whenever it changes its current

  42221.      function context ('cfun').  You can define this function if the

  42222.      back end needs to perform any initialization or reset actions on a

  42223.      per-function basis.  For example, it may be used to implement

  42224.      function attributes that affect register usage or code generation

  42225.      patterns.  The argument DECL is the declaration for the new

  42226.      function context, and may be null to indicate that the compiler has

  42227.      left a function context and is returning to processing at the top

  42228.      level.  The default hook function does nothing.

  42229. 

  42230.      GCC sets 'cfun' to a dummy function context during initialization

  42231.      of some parts of the back end.  The hook function is not invoked in

  42232.      this situation; you need not worry about the hook being invoked

  42233.      recursively, or when the back end is in a partially-initialized

  42234.      state.  'cfun' might be 'NULL' to indicate processing at top level,

  42235.      outside of any function scope.

  42236. 

  42237.  -- Macro: TARGET_OBJECT_SUFFIX

  42238.      Define this macro to be a C string representing the suffix for

  42239.      object files on your target machine.  If you do not define this

  42240.      macro, GCC will use '.o' as the suffix for object files.

  42241. 

  42242.  -- Macro: TARGET_EXECUTABLE_SUFFIX

  42243.      Define this macro to be a C string representing the suffix to be

  42244.      automatically added to executable files on your target machine.  If

  42245.      you do not define this macro, GCC will use the null string as the

  42246.      suffix for executable files.

  42247. 

  42248.  -- Macro: COLLECT_EXPORT_LIST

  42249.      If defined, 'collect2' will scan the individual object files

  42250.      specified on its command line and create an export list for the

  42251.      linker.  Define this macro for systems like AIX, where the linker

  42252.      discards object files that are not referenced from 'main' and uses

  42253.      export lists.

  42254. 

  42255.  -- Target Hook: bool TARGET_CANNOT_MODIFY_JUMPS_P (void)

  42256.      This target hook returns 'true' past the point in which new jump

  42257.      instructions could be created.  On machines that require a register

  42258.      for every jump such as the SHmedia ISA of SH5, this point would

  42259.      typically be reload, so this target hook should be defined to a

  42260.      function such as:

  42261. 

  42262.           static bool

  42263.           cannot_modify_jumps_past_reload_p ()

  42264.           {

  42265.             return (reload_completed || reload_in_progress);

  42266.           }

  42267. 

  42268.  -- Target Hook: bool TARGET_HAVE_CONDITIONAL_EXECUTION (void)

  42269.      This target hook returns true if the target supports conditional

  42270.      execution.  This target hook is required only when the target has

  42271.      several different modes and they have different conditional

  42272.      execution capability, such as ARM.

  42273. 

  42274.  -- Target Hook: rtx TARGET_GEN_CCMP_FIRST (rtx_insn **PREP_SEQ,

  42275.           rtx_insn **GEN_SEQ, int CODE, tree OP0, tree OP1)

  42276.      This function prepares to emit a comparison insn for the first

  42277.      compare in a sequence of conditional comparisions.  It returns an

  42278.      appropriate comparison with 'CC' for passing to 'gen_ccmp_next' or

  42279.      'cbranch_optab'.  The insns to prepare the compare are saved in

  42280.      PREP_SEQ and the compare insns are saved in GEN_SEQ.  They will be

  42281.      emitted when all the compares in the conditional comparision are

  42282.      generated without error.  CODE is the 'rtx_code' of the compare for

  42283.      OP0 and OP1.

  42284. 

  42285.  -- Target Hook: rtx TARGET_GEN_CCMP_NEXT (rtx_insn **PREP_SEQ, rtx_insn

  42286.           **GEN_SEQ, rtx PREV, int CMP_CODE, tree OP0, tree OP1, int

  42287.           BIT_CODE)

  42288.      This function prepares to emit a conditional comparison within a

  42289.      sequence of conditional comparisons.  It returns an appropriate

  42290.      comparison with 'CC' for passing to 'gen_ccmp_next' or

  42291.      'cbranch_optab'.  The insns to prepare the compare are saved in

  42292.      PREP_SEQ and the compare insns are saved in GEN_SEQ.  They will be

  42293.      emitted when all the compares in the conditional comparision are

  42294.      generated without error.  The PREV expression is the result of a

  42295.      prior call to 'gen_ccmp_first' or 'gen_ccmp_next'.  It may return

  42296.      'NULL' if the combination of PREV and this comparison is not

  42297.      supported, otherwise the result must be appropriate for passing to

  42298.      'gen_ccmp_next' or 'cbranch_optab'.  CODE is the 'rtx_code' of the

  42299.      compare for OP0 and OP1.  BIT_CODE is 'AND' or 'IOR', which is the

  42300.      op on the compares.

  42301. 

  42302.  -- Target Hook: unsigned TARGET_LOOP_UNROLL_ADJUST (unsigned NUNROLL,

  42303.           class loop *LOOP)

  42304.      This target hook returns a new value for the number of times LOOP

  42305.      should be unrolled.  The parameter NUNROLL is the number of times

  42306.      the loop is to be unrolled.  The parameter LOOP is a pointer to the

  42307.      loop, which is going to be checked for unrolling.  This target hook

  42308.      is required only when the target has special constraints like

  42309.      maximum number of memory accesses.

  42310. 

  42311.  -- Macro: POWI_MAX_MULTS

  42312.      If defined, this macro is interpreted as a signed integer C

  42313.      expression that specifies the maximum number of floating point

  42314.      multiplications that should be emitted when expanding

  42315.      exponentiation by an integer constant inline.  When this value is

  42316.      defined, exponentiation requiring more than this number of

  42317.      multiplications is implemented by calling the system library's

  42318.      'pow', 'powf' or 'powl' routines.  The default value places no

  42319.      upper bound on the multiplication count.

  42320. 

  42321.  -- Macro: void TARGET_EXTRA_INCLUDES (const char *SYSROOT, const char

  42322.           *IPREFIX, int STDINC)

  42323.      This target hook should register any extra include files for the

  42324.      target.  The parameter STDINC indicates if normal include files are

  42325.      present.  The parameter SYSROOT is the system root directory.  The

  42326.      parameter IPREFIX is the prefix for the gcc directory.

  42327. 

  42328.  -- Macro: void TARGET_EXTRA_PRE_INCLUDES (const char *SYSROOT, const

  42329.           char *IPREFIX, int STDINC)

  42330.      This target hook should register any extra include files for the

  42331.      target before any standard headers.  The parameter STDINC indicates

  42332.      if normal include files are present.  The parameter SYSROOT is the

  42333.      system root directory.  The parameter IPREFIX is the prefix for the

  42334.      gcc directory.

  42335. 

  42336.  -- Macro: void TARGET_OPTF (char *PATH)

  42337.      This target hook should register special include paths for the

  42338.      target.  The parameter PATH is the include to register.  On Darwin

  42339.      systems, this is used for Framework includes, which have semantics

  42340.      that are different from '-I'.

  42341. 

  42342.  -- Macro: bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree FNDECL)

  42343.      This target macro returns 'true' if it is safe to use a local alias

  42344.      for a virtual function FNDECL when constructing thunks, 'false'

  42345.      otherwise.  By default, the macro returns 'true' for all functions,

  42346.      if a target supports aliases (i.e. defines 'ASM_OUTPUT_DEF'),

  42347.      'false' otherwise,

  42348. 

  42349.  -- Macro: TARGET_FORMAT_TYPES

  42350.      If defined, this macro is the name of a global variable containing

  42351.      target-specific format checking information for the '-Wformat'

  42352.      option.  The default is to have no target-specific format checks.

  42353. 

  42354.  -- Macro: TARGET_N_FORMAT_TYPES

  42355.      If defined, this macro is the number of entries in

  42356.      'TARGET_FORMAT_TYPES'.

  42357. 

  42358.  -- Macro: TARGET_OVERRIDES_FORMAT_ATTRIBUTES

  42359.      If defined, this macro is the name of a global variable containing

  42360.      target-specific format overrides for the '-Wformat' option.  The

  42361.      default is to have no target-specific format overrides.  If

  42362.      defined, 'TARGET_FORMAT_TYPES' must be defined, too.

  42363. 

  42364.  -- Macro: TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT

  42365.      If defined, this macro specifies the number of entries in

  42366.      'TARGET_OVERRIDES_FORMAT_ATTRIBUTES'.

  42367. 

  42368.  -- Macro: TARGET_OVERRIDES_FORMAT_INIT

  42369.      If defined, this macro specifies the optional initialization

  42370.      routine for target specific customizations of the system printf and

  42371.      scanf formatter settings.

  42372. 

  42373.  -- Target Hook: const char * TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN

  42374.           (const_tree TYPELIST, const_tree FUNCDECL, const_tree VAL)

  42375.      If defined, this macro returns the diagnostic message when it is

  42376.      illegal to pass argument VAL to function FUNCDECL with prototype

  42377.      TYPELIST.

  42378. 

  42379.  -- Target Hook: const char * TARGET_INVALID_CONVERSION (const_tree

  42380.           FROMTYPE, const_tree TOTYPE)

  42381.      If defined, this macro returns the diagnostic message when it is

  42382.      invalid to convert from FROMTYPE to TOTYPE, or 'NULL' if validity

  42383.      should be determined by the front end.

  42384. 

  42385.  -- Target Hook: const char * TARGET_INVALID_UNARY_OP (int OP,

  42386.           const_tree TYPE)

  42387.      If defined, this macro returns the diagnostic message when it is

  42388.      invalid to apply operation OP (where unary plus is denoted by

  42389.      'CONVERT_EXPR') to an operand of type TYPE, or 'NULL' if validity

  42390.      should be determined by the front end.

  42391. 

  42392.  -- Target Hook: const char * TARGET_INVALID_BINARY_OP (int OP,

  42393.           const_tree TYPE1, const_tree TYPE2)

  42394.      If defined, this macro returns the diagnostic message when it is

  42395.      invalid to apply operation OP to operands of types TYPE1 and TYPE2,

  42396.      or 'NULL' if validity should be determined by the front end.

  42397. 

  42398.  -- Target Hook: tree TARGET_PROMOTED_TYPE (const_tree TYPE)

  42399.      If defined, this target hook returns the type to which values of

  42400.      TYPE should be promoted when they appear in expressions, analogous

  42401.      to the integer promotions, or 'NULL_TREE' to use the front end's

  42402.      normal promotion rules.  This hook is useful when there are

  42403.      target-specific types with special promotion rules.  This is

  42404.      currently used only by the C and C++ front ends.

  42405. 

  42406.  -- Target Hook: tree TARGET_CONVERT_TO_TYPE (tree TYPE, tree EXPR)

  42407.      If defined, this hook returns the result of converting EXPR to

  42408.      TYPE.  It should return the converted expression, or 'NULL_TREE' to

  42409.      apply the front end's normal conversion rules.  This hook is useful

  42410.      when there are target-specific types with special conversion rules.

  42411.      This is currently used only by the C and C++ front ends.

  42412. 

  42413.  -- Target Hook: bool TARGET_VERIFY_TYPE_CONTEXT (location_t LOC,

  42414.           type_context_kind CONTEXT, const_tree TYPE, bool SILENT_P)

  42415.      If defined, this hook returns false if there is a target-specific

  42416.      reason why type TYPE cannot be used in the source language context

  42417.      described by CONTEXT.  When SILENT_P is false, the hook also

  42418.      reports an error against LOC for invalid uses of TYPE.

  42419. 

  42420.      Calls to this hook should be made through the global function

  42421.      'verify_type_context', which makes the SILENT_P parameter default

  42422.      to false and also handles 'error_mark_node'.

  42423. 

  42424.      The default implementation always returns true.

  42425. 

  42426.  -- Macro: OBJC_JBLEN

  42427.      This macro determines the size of the objective C jump buffer for

  42428.      the NeXT runtime.  By default, OBJC_JBLEN is defined to an

  42429.      innocuous value.

  42430. 

  42431.  -- Macro: LIBGCC2_UNWIND_ATTRIBUTE

  42432.      Define this macro if any target-specific attributes need to be

  42433.      attached to the functions in 'libgcc' that provide low-level

  42434.      support for call stack unwinding.  It is used in declarations in

  42435.      'unwind-generic.h' and the associated definitions of those

  42436.      functions.

  42437. 

  42438.  -- Target Hook: void TARGET_UPDATE_STACK_BOUNDARY (void)

  42439.      Define this macro to update the current function stack boundary if

  42440.      necessary.

  42441. 

  42442.  -- Target Hook: rtx TARGET_GET_DRAP_RTX (void)

  42443.      This hook should return an rtx for Dynamic Realign Argument Pointer

  42444.      (DRAP) if a different argument pointer register is needed to access

  42445.      the function's argument list due to stack realignment.  Return

  42446.      'NULL' if no DRAP is needed.

  42447. 

  42448.  -- Target Hook: bool TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS (void)

  42449.      When optimization is disabled, this hook indicates whether or not

  42450.      arguments should be allocated to stack slots.  Normally, GCC

  42451.      allocates stacks slots for arguments when not optimizing in order

  42452.      to make debugging easier.  However, when a function is declared

  42453.      with '__attribute__((naked))', there is no stack frame, and the

  42454.      compiler cannot safely move arguments from the registers in which

  42455.      they are passed to the stack.  Therefore, this hook should return

  42456.      true in general, but false for naked functions.  The default

  42457.      implementation always returns true.

  42458. 

  42459.  -- Target Hook: unsigned HOST_WIDE_INT TARGET_CONST_ANCHOR

  42460.      On some architectures it can take multiple instructions to

  42461.      synthesize a constant.  If there is another constant already in a

  42462.      register that is close enough in value then it is preferable that

  42463.      the new constant is computed from this register using immediate

  42464.      addition or subtraction.  We accomplish this through CSE. Besides

  42465.      the value of the constant we also add a lower and an upper constant

  42466.      anchor to the available expressions.  These are then queried when

  42467.      encountering new constants.  The anchors are computed by rounding

  42468.      the constant up and down to a multiple of the value of

  42469.      'TARGET_CONST_ANCHOR'.  'TARGET_CONST_ANCHOR' should be the maximum

  42470.      positive value accepted by immediate-add plus one.  We currently

  42471.      assume that the value of 'TARGET_CONST_ANCHOR' is a power of 2.

  42472.      For example, on MIPS, where add-immediate takes a 16-bit signed

  42473.      value, 'TARGET_CONST_ANCHOR' is set to '0x8000'.  The default value

  42474.      is zero, which disables this optimization.

  42475. 

  42476.  -- Target Hook: unsigned HOST_WIDE_INT TARGET_ASAN_SHADOW_OFFSET (void)

  42477.      Return the offset bitwise ored into shifted address to get

  42478.      corresponding Address Sanitizer shadow memory address.  NULL if

  42479.      Address Sanitizer is not supported by the target.

  42480. 

  42481.  -- Target Hook: unsigned HOST_WIDE_INT TARGET_MEMMODEL_CHECK (unsigned

  42482.           HOST_WIDE_INT VAL)

  42483.      Validate target specific memory model mask bits.  When NULL no

  42484.      target specific memory model bits are allowed.

  42485. 

  42486.  -- Target Hook: unsigned char TARGET_ATOMIC_TEST_AND_SET_TRUEVAL

  42487.      This value should be set if the result written by

  42488.      'atomic_test_and_set' is not exactly 1, i.e. the 'bool' 'true'.

  42489. 

  42490.  -- Target Hook: bool TARGET_HAS_IFUNC_P (void)

  42491.      It returns true if the target supports GNU indirect functions.  The

  42492.      support includes the assembler, linker and dynamic linker.  The

  42493.      default value of this hook is based on target's libc.

  42494. 

  42495.  -- Target Hook: unsigned int TARGET_ATOMIC_ALIGN_FOR_MODE (machine_mode

  42496.           MODE)

  42497.      If defined, this function returns an appropriate alignment in bits

  42498.      for an atomic object of machine_mode MODE.  If 0 is returned then

  42499.      the default alignment for the specified mode is used.

  42500. 

  42501.  -- Target Hook: void TARGET_ATOMIC_ASSIGN_EXPAND_FENV (tree *HOLD, tree

  42502.           *CLEAR, tree *UPDATE)

  42503.      ISO C11 requires atomic compound assignments that may raise

  42504.      floating-point exceptions to raise exceptions corresponding to the

  42505.      arithmetic operation whose result was successfully stored in a

  42506.      compare-and-exchange sequence.  This requires code equivalent to

  42507.      calls to 'feholdexcept', 'feclearexcept' and 'feupdateenv' to be

  42508.      generated at appropriate points in the compare-and-exchange

  42509.      sequence.  This hook should set '*HOLD' to an expression equivalent

  42510.      to the call to 'feholdexcept', '*CLEAR' to an expression equivalent

  42511.      to the call to 'feclearexcept' and '*UPDATE' to an expression

  42512.      equivalent to the call to 'feupdateenv'.  The three expressions are

  42513.      'NULL_TREE' on entry to the hook and may be left as 'NULL_TREE' if

  42514.      no code is required in a particular place.  The default

  42515.      implementation leaves all three expressions as 'NULL_TREE'.  The

  42516.      '__atomic_feraiseexcept' function from 'libatomic' may be of use as

  42517.      part of the code generated in '*UPDATE'.

  42518. 

  42519.  -- Target Hook: void TARGET_RECORD_OFFLOAD_SYMBOL (tree)

  42520.      Used when offloaded functions are seen in the compilation unit and

  42521.      no named sections are available.  It is called once for each symbol

  42522.      that must be recorded in the offload function and variable table.

  42523. 

  42524.  -- Target Hook: char * TARGET_OFFLOAD_OPTIONS (void)

  42525.      Used when writing out the list of options into an LTO file.  It

  42526.      should translate any relevant target-specific options (such as the

  42527.      ABI in use) into one of the '-foffload' options that exist as a

  42528.      common interface to express such options.  It should return a

  42529.      string containing these options, separated by spaces, which the

  42530.      caller will free.

  42531. 

  42532.  -- Macro: TARGET_SUPPORTS_WIDE_INT

  42533. 

  42534.      On older ports, large integers are stored in 'CONST_DOUBLE' rtl

  42535.      objects.  Newer ports define 'TARGET_SUPPORTS_WIDE_INT' to be

  42536.      nonzero to indicate that large integers are stored in

  42537.      'CONST_WIDE_INT' rtl objects.  The 'CONST_WIDE_INT' allows very

  42538.      large integer constants to be represented.  'CONST_DOUBLE' is

  42539.      limited to twice the size of the host's 'HOST_WIDE_INT'

  42540.      representation.

  42541. 

  42542.      Converting a port mostly requires looking for the places where

  42543.      'CONST_DOUBLE's are used with 'VOIDmode' and replacing that code

  42544.      with code that accesses 'CONST_WIDE_INT's.  '"grep -i

  42545.      const_double"' at the port level gets you to 95% of the changes

  42546.      that need to be made.  There are a few places that require a deeper

  42547.      look.

  42548. 

  42549.         * There is no equivalent to 'hval' and 'lval' for

  42550.           'CONST_WIDE_INT's.  This would be difficult to express in the

  42551.           md language since there are a variable number of elements.

  42552. 

  42553.           Most ports only check that 'hval' is either 0 or -1 to see if

  42554.           the value is small.  As mentioned above, this will no longer

  42555.           be necessary since small constants are always 'CONST_INT'.  Of

  42556.           course there are still a few exceptions, the alpha's

  42557.           constraint used by the zap instruction certainly requires

  42558.           careful examination by C code.  However, all the current code

  42559.           does is pass the hval and lval to C code, so evolving the c

  42560.           code to look at the 'CONST_WIDE_INT' is not really a large

  42561.           change.

  42562. 

  42563.         * Because there is no standard template that ports use to

  42564.           materialize constants, there is likely to be some futzing that

  42565.           is unique to each port in this code.

  42566. 

  42567.         * The rtx costs may have to be adjusted to properly account for

  42568.           larger constants that are represented as 'CONST_WIDE_INT'.

  42569. 

  42570.      All and all it does not take long to convert ports that the

  42571.      maintainer is familiar with.

  42572. 

  42573.  -- Target Hook: bool TARGET_HAVE_SPECULATION_SAFE_VALUE (bool ACTIVE)

  42574.      This hook is used to determine the level of target support for

  42575.      '__builtin_speculation_safe_value'.  If called with an argument of

  42576.      false, it returns true if the target has been modified to support

  42577.      this builtin.  If called with an argument of true, it returns true

  42578.      if the target requires active mitigation execution might be

  42579.      speculative.

  42580. 

  42581.      The default implementation returns false if the target does not

  42582.      define a pattern named 'speculation_barrier'.  Else it returns true

  42583.      for the first case and whether the pattern is enabled for the

  42584.      current compilation for the second case.

  42585. 

  42586.      For targets that have no processors that can execute instructions

  42587.      speculatively an alternative implemenation of this hook is

  42588.      available: simply redefine this hook to

  42589.      'speculation_safe_value_not_needed' along with your other target

  42590.      hooks.

  42591. 

  42592.  -- Target Hook: rtx TARGET_SPECULATION_SAFE_VALUE (machine_mode MODE,

  42593.           rtx RESULT, rtx VAL, rtx FAILVAL)

  42594.      This target hook can be used to generate a target-specific code

  42595.      sequence that implements the '__builtin_speculation_safe_value'

  42596.      built-in function.  The function must always return VAL in RESULT

  42597.      in mode MODE when the cpu is not executing speculatively, but must

  42598.      never return that when speculating until it is known that the

  42599.      speculation will not be unwound.  The hook supports two primary

  42600.      mechanisms for implementing the requirements.  The first is to emit

  42601.      a speculation barrier which forces the processor to wait until all

  42602.      prior speculative operations have been resolved; the second is to

  42603.      use a target-specific mechanism that can track the speculation

  42604.      state and to return FAILVAL if it can determine that speculation

  42605.      must be unwound at a later time.

  42606. 

  42607.      The default implementation simply copies VAL to RESULT and emits a

  42608.      'speculation_barrier' instruction if that is defined.

  42609. 

  42610.  -- Target Hook: void TARGET_RUN_TARGET_SELFTESTS (void)

  42611.      If selftests are enabled, run any selftests for this target.

  42612. 

  42613. 

  42614. File: gccint.info,  Node: Host Config,  Next: Fragments,  Prev: Target Macros,  Up: Top

  42615. 

  42616. 19 Host Configuration

  42617. *********************

  42618. 

  42619. Most details about the machine and system on which the compiler is

  42620. actually running are detected by the 'configure' script.  Some things

  42621. are impossible for 'configure' to detect; these are described in two

  42622. ways, either by macros defined in a file named 'xm-MACHINE.h' or by hook

  42623. functions in the file specified by the OUT_HOST_HOOK_OBJ variable in

  42624. 'config.gcc'.  (The intention is that very few hosts will need a header

  42625. file but nearly every fully supported host will need to override some

  42626. hooks.)

  42627. 

  42628.  If you need to define only a few macros, and they have simple

  42629. definitions, consider using the 'xm_defines' variable in your

  42630. 'config.gcc' entry instead of creating a host configuration header.

  42631. *Note System Config::.

  42632. 

  42633. * Menu:

  42634. 

  42635. * Host Common::         Things every host probably needs implemented.

  42636. * Filesystem::          Your host cannot have the letter 'a' in filenames?

  42637. * Host Misc::           Rare configuration options for hosts.

  42638. 

  42639. 

  42640. File: gccint.info,  Node: Host Common,  Next: Filesystem,  Up: Host Config

  42641. 

  42642. 19.1 Host Common

  42643. ================

  42644. 

  42645. Some things are just not portable, even between similar operating

  42646. systems, and are too difficult for autoconf to detect.  They get

  42647. implemented using hook functions in the file specified by the

  42648. HOST_HOOK_OBJ variable in 'config.gcc'.

  42649. 

  42650.  -- Host Hook: void HOST_HOOKS_EXTRA_SIGNALS (void)

  42651.      This host hook is used to set up handling for extra signals.  The

  42652.      most common thing to do in this hook is to detect stack overflow.

  42653. 

  42654.  -- Host Hook: void * HOST_HOOKS_GT_PCH_GET_ADDRESS (size_t SIZE, int

  42655.           FD)

  42656.      This host hook returns the address of some space that is likely to

  42657.      be free in some subsequent invocation of the compiler.  We intend

  42658.      to load the PCH data at this address such that the data need not be

  42659.      relocated.  The area should be able to hold SIZE bytes.  If the

  42660.      host uses 'mmap', FD is an open file descriptor that can be used

  42661.      for probing.

  42662. 

  42663.  -- Host Hook: int HOST_HOOKS_GT_PCH_USE_ADDRESS (void * ADDRESS, size_t

  42664.           SIZE, int FD, size_t OFFSET)

  42665.      This host hook is called when a PCH file is about to be loaded.  We

  42666.      want to load SIZE bytes from FD at OFFSET into memory at ADDRESS.

  42667.      The given address will be the result of a previous invocation of

  42668.      'HOST_HOOKS_GT_PCH_GET_ADDRESS'.  Return -1 if we couldn't allocate

  42669.      SIZE bytes at ADDRESS.  Return 0 if the memory is allocated but the

  42670.      data is not loaded.  Return 1 if the hook has performed everything.

  42671. 

  42672.      If the implementation uses reserved address space, free any

  42673.      reserved space beyond SIZE, regardless of the return value.  If no

  42674.      PCH will be loaded, this hook may be called with SIZE zero, in

  42675.      which case all reserved address space should be freed.

  42676. 

  42677.      Do not try to handle values of ADDRESS that could not have been

  42678.      returned by this executable; just return -1.  Such values usually

  42679.      indicate an out-of-date PCH file (built by some other GCC

  42680.      executable), and such a PCH file won't work.

  42681. 

  42682.  -- Host Hook: size_t HOST_HOOKS_GT_PCH_ALLOC_GRANULARITY (void);

  42683.      This host hook returns the alignment required for allocating

  42684.      virtual memory.  Usually this is the same as getpagesize, but on

  42685.      some hosts the alignment for reserving memory differs from the

  42686.      pagesize for committing memory.

  42687. 

  42688. 

  42689. File: gccint.info,  Node: Filesystem,  Next: Host Misc,  Prev: Host Common,  Up: Host Config

  42690. 

  42691. 19.2 Host Filesystem

  42692. ====================

  42693. 

  42694. GCC needs to know a number of things about the semantics of the host

  42695. machine's filesystem.  Filesystems with Unix and MS-DOS semantics are

  42696. automatically detected.  For other systems, you can define the following

  42697. macros in 'xm-MACHINE.h'.

  42698. 

  42699. 'HAVE_DOS_BASED_FILE_SYSTEM'

  42700.      This macro is automatically defined by 'system.h' if the host file

  42701.      system obeys the semantics defined by MS-DOS instead of Unix.  DOS

  42702.      file systems are case insensitive, file specifications may begin

  42703.      with a drive letter, and both forward slash and backslash ('/' and

  42704.      '\') are directory separators.

  42705. 

  42706. 'DIR_SEPARATOR'

  42707. 'DIR_SEPARATOR_2'

  42708.      If defined, these macros expand to character constants specifying

  42709.      separators for directory names within a file specification.

  42710.      'system.h' will automatically give them appropriate values on Unix

  42711.      and MS-DOS file systems.  If your file system is neither of these,

  42712.      define one or both appropriately in 'xm-MACHINE.h'.

  42713. 

  42714.      However, operating systems like VMS, where constructing a pathname

  42715.      is more complicated than just stringing together directory names

  42716.      separated by a special character, should not define either of these

  42717.      macros.

  42718. 

  42719. 'PATH_SEPARATOR'

  42720.      If defined, this macro should expand to a character constant

  42721.      specifying the separator for elements of search paths.  The default

  42722.      value is a colon (':').  DOS-based systems usually, but not always,

  42723.      use semicolon (';').

  42724. 

  42725. 'VMS'

  42726.      Define this macro if the host system is VMS.

  42727. 

  42728. 'HOST_OBJECT_SUFFIX'

  42729.      Define this macro to be a C string representing the suffix for

  42730.      object files on your host machine.  If you do not define this

  42731.      macro, GCC will use '.o' as the suffix for object files.

  42732. 

  42733. 'HOST_EXECUTABLE_SUFFIX'

  42734.      Define this macro to be a C string representing the suffix for

  42735.      executable files on your host machine.  If you do not define this

  42736.      macro, GCC will use the null string as the suffix for executable

  42737.      files.

  42738. 

  42739. 'HOST_BIT_BUCKET'

  42740.      A pathname defined by the host operating system, which can be

  42741.      opened as a file and written to, but all the information written is

  42742.      discarded.  This is commonly known as a "bit bucket" or "null

  42743.      device".  If you do not define this macro, GCC will use '/dev/null'

  42744.      as the bit bucket.  If the host does not support a bit bucket,

  42745.      define this macro to an invalid filename.

  42746. 

  42747. 'UPDATE_PATH_HOST_CANONICALIZE (PATH)'

  42748.      If defined, a C statement (sans semicolon) that performs

  42749.      host-dependent canonicalization when a path used in a compilation

  42750.      driver or preprocessor is canonicalized.  PATH is a malloc-ed path

  42751.      to be canonicalized.  If the C statement does canonicalize PATH

  42752.      into a different buffer, the old path should be freed and the new

  42753.      buffer should have been allocated with malloc.

  42754. 

  42755. 'DUMPFILE_FORMAT'

  42756.      Define this macro to be a C string representing the format to use

  42757.      for constructing the index part of debugging dump file names.  The

  42758.      resultant string must fit in fifteen bytes.  The full filename will

  42759.      be the concatenation of: the prefix of the assembler file name, the

  42760.      string resulting from applying this format to an index number, and

  42761.      a string unique to each dump file kind, e.g. 'rtl'.

  42762. 

  42763.      If you do not define this macro, GCC will use '.%02d.'.  You should

  42764.      define this macro if using the default will create an invalid file

  42765.      name.

  42766. 

  42767. 'DELETE_IF_ORDINARY'

  42768.      Define this macro to be a C statement (sans semicolon) that

  42769.      performs host-dependent removal of ordinary temp files in the

  42770.      compilation driver.

  42771. 

  42772.      If you do not define this macro, GCC will use the default version.

  42773.      You should define this macro if the default version does not

  42774.      reliably remove the temp file as, for example, on VMS which allows

  42775.      multiple versions of a file.

  42776. 

  42777. 'HOST_LACKS_INODE_NUMBERS'

  42778.      Define this macro if the host filesystem does not report meaningful

  42779.      inode numbers in struct stat.

  42780. 

  42781. 

  42782. File: gccint.info,  Node: Host Misc,  Prev: Filesystem,  Up: Host Config

  42783. 

  42784. 19.3 Host Misc

  42785. ==============

  42786. 

  42787. 'FATAL_EXIT_CODE'

  42788.      A C expression for the status code to be returned when the compiler

  42789.      exits after serious errors.  The default is the system-provided

  42790.      macro 'EXIT_FAILURE', or '1' if the system doesn't define that

  42791.      macro.  Define this macro only if these defaults are incorrect.

  42792. 

  42793. 'SUCCESS_EXIT_CODE'

  42794.      A C expression for the status code to be returned when the compiler

  42795.      exits without serious errors.  (Warnings are not serious errors.)

  42796.      The default is the system-provided macro 'EXIT_SUCCESS', or '0' if

  42797.      the system doesn't define that macro.  Define this macro only if

  42798.      these defaults are incorrect.

  42799. 

  42800. 'USE_C_ALLOCA'

  42801.      Define this macro if GCC should use the C implementation of

  42802.      'alloca' provided by 'libiberty.a'.  This only affects how some

  42803.      parts of the compiler itself allocate memory.  It does not change

  42804.      code generation.

  42805. 

  42806.      When GCC is built with a compiler other than itself, the C 'alloca'

  42807.      is always used.  This is because most other implementations have

  42808.      serious bugs.  You should define this macro only on a system where

  42809.      no stack-based 'alloca' can possibly work.  For instance, if a

  42810.      system has a small limit on the size of the stack, GCC's builtin

  42811.      'alloca' will not work reliably.

  42812. 

  42813. 'COLLECT2_HOST_INITIALIZATION'

  42814.      If defined, a C statement (sans semicolon) that performs

  42815.      host-dependent initialization when 'collect2' is being initialized.

  42816. 

  42817. 'GCC_DRIVER_HOST_INITIALIZATION'

  42818.      If defined, a C statement (sans semicolon) that performs

  42819.      host-dependent initialization when a compilation driver is being

  42820.      initialized.

  42821. 

  42822. 'HOST_LONG_LONG_FORMAT'

  42823.      If defined, the string used to indicate an argument of type 'long

  42824.      long' to functions like 'printf'.  The default value is '"ll"'.

  42825. 

  42826. 'HOST_LONG_FORMAT'

  42827.      If defined, the string used to indicate an argument of type 'long'

  42828.      to functions like 'printf'.  The default value is '"l"'.

  42829. 

  42830. 'HOST_PTR_PRINTF'

  42831.      If defined, the string used to indicate an argument of type 'void

  42832.      *' to functions like 'printf'.  The default value is '"%p"'.

  42833. 

  42834.  In addition, if 'configure' generates an incorrect definition of any of

  42835. the macros in 'auto-host.h', you can override that definition in a host

  42836. configuration header.  If you need to do this, first see if it is

  42837. possible to fix 'configure'.

  42838. 

  42839. 

  42840. File: gccint.info,  Node: Fragments,  Next: Collect2,  Prev: Host Config,  Up: Top

  42841. 

  42842. 20 Makefile Fragments

  42843. *********************

  42844. 

  42845. When you configure GCC using the 'configure' script, it will construct

  42846. the file 'Makefile' from the template file 'Makefile.in'.  When it does

  42847. this, it can incorporate makefile fragments from the 'config' directory.

  42848. These are used to set Makefile parameters that are not amenable to being

  42849. calculated by autoconf.  The list of fragments to incorporate is set by

  42850. 'config.gcc' (and occasionally 'config.build' and 'config.host'); *Note

  42851. System Config::.

  42852. 

  42853.  Fragments are named either 't-TARGET' or 'x-HOST', depending on whether

  42854. they are relevant to configuring GCC to produce code for a particular

  42855. target, or to configuring GCC to run on a particular host.  Here TARGET

  42856. and HOST are mnemonics which usually have some relationship to the

  42857. canonical system name, but no formal connection.

  42858. 

  42859.  If these files do not exist, it means nothing needs to be added for a

  42860. given target or host.  Most targets need a few 't-TARGET' fragments, but

  42861. needing 'x-HOST' fragments is rare.

  42862. 

  42863. * Menu:

  42864. 

  42865. * Target Fragment:: Writing 't-TARGET' files.

  42866. * Host Fragment::   Writing 'x-HOST' files.

  42867. 

  42868. 

  42869. File: gccint.info,  Node: Target Fragment,  Next: Host Fragment,  Up: Fragments

  42870. 

  42871. 20.1 Target Makefile Fragments

  42872. ==============================

  42873. 

  42874. Target makefile fragments can set these Makefile variables.

  42875. 

  42876. 'LIBGCC2_CFLAGS'

  42877.      Compiler flags to use when compiling 'libgcc2.c'.

  42878. 

  42879. 'LIB2FUNCS_EXTRA'

  42880.      A list of source file names to be compiled or assembled and

  42881.      inserted into 'libgcc.a'.

  42882. 

  42883. 'CRTSTUFF_T_CFLAGS'

  42884.      Special flags used when compiling 'crtstuff.c'.  *Note

  42885.      Initialization::.

  42886. 

  42887. 'CRTSTUFF_T_CFLAGS_S'

  42888.      Special flags used when compiling 'crtstuff.c' for shared linking.

  42889.      Used if you use 'crtbeginS.o' and 'crtendS.o' in 'EXTRA-PARTS'.

  42890.      *Note Initialization::.

  42891. 

  42892. 'MULTILIB_OPTIONS'

  42893.      For some targets, invoking GCC in different ways produces objects

  42894.      that cannot be linked together.  For example, for some targets GCC

  42895.      produces both big and little endian code.  For these targets, you

  42896.      must arrange for multiple versions of 'libgcc.a' to be compiled,

  42897.      one for each set of incompatible options.  When GCC invokes the

  42898.      linker, it arranges to link in the right version of 'libgcc.a',

  42899.      based on the command line options used.

  42900. 

  42901.      The 'MULTILIB_OPTIONS' macro lists the set of options for which

  42902.      special versions of 'libgcc.a' must be built.  Write options that

  42903.      are mutually incompatible side by side, separated by a slash.

  42904.      Write options that may be used together separated by a space.  The

  42905.      build procedure will build all combinations of compatible options.

  42906. 

  42907.      For example, if you set 'MULTILIB_OPTIONS' to 'm68000/m68020

  42908.      msoft-float', 'Makefile' will build special versions of 'libgcc.a'

  42909.      using the following sets of options: '-m68000', '-m68020',

  42910.      '-msoft-float', '-m68000 -msoft-float', and '-m68020 -msoft-float'.

  42911. 

  42912. 'MULTILIB_DIRNAMES'

  42913.      If 'MULTILIB_OPTIONS' is used, this variable specifies the

  42914.      directory names that should be used to hold the various libraries.

  42915.      Write one element in 'MULTILIB_DIRNAMES' for each element in

  42916.      'MULTILIB_OPTIONS'.  If 'MULTILIB_DIRNAMES' is not used, the

  42917.      default value will be 'MULTILIB_OPTIONS', with all slashes treated

  42918.      as spaces.

  42919. 

  42920.      'MULTILIB_DIRNAMES' describes the multilib directories using GCC

  42921.      conventions and is applied to directories that are part of the GCC

  42922.      installation.  When multilib-enabled, the compiler will add a

  42923.      subdirectory of the form PREFIX/MULTILIB before each directory in

  42924.      the search path for libraries and crt files.

  42925. 

  42926.      For example, if 'MULTILIB_OPTIONS' is set to 'm68000/m68020

  42927.      msoft-float', then the default value of 'MULTILIB_DIRNAMES' is

  42928.      'm68000 m68020 msoft-float'.  You may specify a different value if

  42929.      you desire a different set of directory names.

  42930. 

  42931. 'MULTILIB_MATCHES'

  42932.      Sometimes the same option may be written in two different ways.  If

  42933.      an option is listed in 'MULTILIB_OPTIONS', GCC needs to know about

  42934.      any synonyms.  In that case, set 'MULTILIB_MATCHES' to a list of

  42935.      items of the form 'option=option' to describe all relevant

  42936.      synonyms.  For example, 'm68000=mc68000 m68020=mc68020'.

  42937. 

  42938. 'MULTILIB_EXCEPTIONS'

  42939.      Sometimes when there are multiple sets of 'MULTILIB_OPTIONS' being

  42940.      specified, there are combinations that should not be built.  In

  42941.      that case, set 'MULTILIB_EXCEPTIONS' to be all of the switch

  42942.      exceptions in shell case syntax that should not be built.

  42943. 

  42944.      For example the ARM processor cannot execute both hardware floating

  42945.      point instructions and the reduced size THUMB instructions at the

  42946.      same time, so there is no need to build libraries with both of

  42947.      these options enabled.  Therefore 'MULTILIB_EXCEPTIONS' is set to:

  42948.           *mthumb/*mhard-float*

  42949. 

  42950. 'MULTILIB_REQUIRED'

  42951.      Sometimes when there are only a few combinations are required, it

  42952.      would be a big effort to come up with a 'MULTILIB_EXCEPTIONS' list

  42953.      to cover all undesired ones.  In such a case, just listing all the

  42954.      required combinations in 'MULTILIB_REQUIRED' would be more

  42955.      straightforward.

  42956. 

  42957.      The way to specify the entries in 'MULTILIB_REQUIRED' is same with

  42958.      the way used for 'MULTILIB_EXCEPTIONS', only this time what are

  42959.      required will be specified.  Suppose there are multiple sets of

  42960.      'MULTILIB_OPTIONS' and only two combinations are required, one for

  42961.      ARMv7-M and one for ARMv7-R with hard floating-point ABI and FPU,

  42962.      the 'MULTILIB_REQUIRED' can be set to:

  42963.           MULTILIB_REQUIRED =  mthumb/march=armv7-m

  42964.           MULTILIB_REQUIRED += march=armv7-r/mfloat-abi=hard/mfpu=vfpv3-d16

  42965. 

  42966.      The 'MULTILIB_REQUIRED' can be used together with

  42967.      'MULTILIB_EXCEPTIONS'.  The option combinations generated from

  42968.      'MULTILIB_OPTIONS' will be filtered by 'MULTILIB_EXCEPTIONS' and

  42969.      then by 'MULTILIB_REQUIRED'.

  42970. 

  42971. 'MULTILIB_REUSE'

  42972.      Sometimes it is desirable to reuse one existing multilib for

  42973.      different sets of options.  Such kind of reuse can minimize the

  42974.      number of multilib variants.  And for some targets it is better to

  42975.      reuse an existing multilib than to fall back to default multilib

  42976.      when there is no corresponding multilib.  This can be done by

  42977.      adding reuse rules to 'MULTILIB_REUSE'.

  42978. 

  42979.      A reuse rule is comprised of two parts connected by equality sign.

  42980.      The left part is the option set used to build multilib and the

  42981.      right part is the option set that will reuse this multilib.  Both

  42982.      parts should only use options specified in 'MULTILIB_OPTIONS' and

  42983.      the equality signs found in options name should be replaced with

  42984.      periods.  An explicit period in the rule can be escaped by

  42985.      preceding it with a backslash.  The order of options in the left

  42986.      part matters and should be same with those specified in

  42987.      'MULTILIB_REQUIRED' or aligned with the order in

  42988.      'MULTILIB_OPTIONS'.  There is no such limitation for options in the

  42989.      right part as we don't build multilib from them.

  42990. 

  42991.      'MULTILIB_REUSE' is different from 'MULTILIB_MATCHES' in that it

  42992.      sets up relations between two option sets rather than two options.

  42993.      Here is an example to demo how we reuse libraries built in Thumb

  42994.      mode for applications built in ARM mode:

  42995.           MULTILIB_REUSE = mthumb/march.armv7-r=marm/march.armv7-r

  42996. 

  42997.      Before the advent of 'MULTILIB_REUSE', GCC select multilib by

  42998.      comparing command line options with options used to build multilib.

  42999.      The 'MULTILIB_REUSE' is complementary to that way.  Only when the

  43000.      original comparison matches nothing it will work to see if it is OK

  43001.      to reuse some existing multilib.

  43002. 

  43003. 'MULTILIB_EXTRA_OPTS'

  43004.      Sometimes it is desirable that when building multiple versions of

  43005.      'libgcc.a' certain options should always be passed on to the

  43006.      compiler.  In that case, set 'MULTILIB_EXTRA_OPTS' to be the list

  43007.      of options to be used for all builds.  If you set this, you should

  43008.      probably set 'CRTSTUFF_T_CFLAGS' to a dash followed by it.

  43009. 

  43010. 'MULTILIB_OSDIRNAMES'

  43011.      If 'MULTILIB_OPTIONS' is used, this variable specifies a list of

  43012.      subdirectory names, that are used to modify the search path

  43013.      depending on the chosen multilib.  Unlike 'MULTILIB_DIRNAMES',

  43014.      'MULTILIB_OSDIRNAMES' describes the multilib directories using

  43015.      operating systems conventions, and is applied to the directories

  43016.      such as 'lib' or those in the 'LIBRARY_PATH' environment variable.

  43017.      The format is either the same as of 'MULTILIB_DIRNAMES', or a set

  43018.      of mappings.  When it is the same as 'MULTILIB_DIRNAMES', it

  43019.      describes the multilib directories using operating system

  43020.      conventions, rather than GCC conventions.  When it is a set of

  43021.      mappings of the form GCCDIR=OSDIR, the left side gives the GCC

  43022.      convention and the right gives the equivalent OS defined location.

  43023.      If the OSDIR part begins with a '!', GCC will not search in the

  43024.      non-multilib directory and use exclusively the multilib directory.

  43025.      Otherwise, the compiler will examine the search path for libraries

  43026.      and crt files twice; the first time it will add MULTILIB to each

  43027.      directory in the search path, the second it will not.

  43028. 

  43029.      For configurations that support both multilib and multiarch,

  43030.      'MULTILIB_OSDIRNAMES' also encodes the multiarch name, thus

  43031.      subsuming 'MULTIARCH_DIRNAME'.  The multiarch name is appended to

  43032.      each directory name, separated by a colon (e.g.

  43033.      '../lib32:i386-linux-gnu').

  43034. 

  43035.      Each multiarch subdirectory will be searched before the

  43036.      corresponding OS multilib directory, for example

  43037.      '/lib/i386-linux-gnu' before '/lib/../lib32'.  The multiarch name

  43038.      will also be used to modify the system header search path, as

  43039.      explained for 'MULTIARCH_DIRNAME'.

  43040. 

  43041. 'MULTIARCH_DIRNAME'

  43042.      This variable specifies the multiarch name for configurations that

  43043.      are multiarch-enabled but not multilibbed configurations.

  43044. 

  43045.      The multiarch name is used to augment the search path for

  43046.      libraries, crt files and system header files with additional

  43047.      locations.  The compiler will add a multiarch subdirectory of the

  43048.      form PREFIX/MULTIARCH before each directory in the library and crt

  43049.      search path.  It will also add two directories

  43050.      'LOCAL_INCLUDE_DIR'/MULTIARCH and

  43051.      'NATIVE_SYSTEM_HEADER_DIR'/MULTIARCH) to the system header search

  43052.      path, respectively before 'LOCAL_INCLUDE_DIR' and

  43053.      'NATIVE_SYSTEM_HEADER_DIR'.

  43054. 

  43055.      'MULTIARCH_DIRNAME' is not used for configurations that support

  43056.      both multilib and multiarch.  In that case, multiarch names are

  43057.      encoded in 'MULTILIB_OSDIRNAMES' instead.

  43058. 

  43059.      More documentation about multiarch can be found at

  43060.      <https://wiki.debian.org/Multiarch>.

  43061. 

  43062. 'SPECS'

  43063.      Unfortunately, setting 'MULTILIB_EXTRA_OPTS' is not enough, since

  43064.      it does not affect the build of target libraries, at least not the

  43065.      build of the default multilib.  One possible work-around is to use

  43066.      'DRIVER_SELF_SPECS' to bring options from the 'specs' file as if

  43067.      they had been passed in the compiler driver command line.  However,

  43068.      you don't want to be adding these options after the toolchain is

  43069.      installed, so you can instead tweak the 'specs' file that will be

  43070.      used during the toolchain build, while you still install the

  43071.      original, built-in 'specs'.  The trick is to set 'SPECS' to some

  43072.      other filename (say 'specs.install'), that will then be created out

  43073.      of the built-in specs, and introduce a 'Makefile' rule to generate

  43074.      the 'specs' file that's going to be used at build time out of your

  43075.      'specs.install'.

  43076. 

  43077. 'T_CFLAGS'

  43078.      These are extra flags to pass to the C compiler.  They are used

  43079.      both when building GCC, and when compiling things with the

  43080.      just-built GCC.  This variable is deprecated and should not be

  43081.      used.

  43082. 

  43083. 

  43084. File: gccint.info,  Node: Host Fragment,  Prev: Target Fragment,  Up: Fragments

  43085. 

  43086. 20.2 Host Makefile Fragments

  43087. ============================

  43088. 

  43089. The use of 'x-HOST' fragments is discouraged.  You should only use it

  43090. for makefile dependencies.

  43091. 

  43092. 

  43093. File: gccint.info,  Node: Collect2,  Next: Header Dirs,  Prev: Fragments,  Up: Top

  43094. 

  43095. 21 'collect2'

  43096. *************

  43097. 

  43098. GCC uses a utility called 'collect2' on nearly all systems to arrange to

  43099. call various initialization functions at start time.

  43100. 

  43101.  The program 'collect2' works by linking the program once and looking

  43102. through the linker output file for symbols with particular names

  43103. indicating they are constructor functions.  If it finds any, it creates

  43104. a new temporary '.c' file containing a table of them, compiles it, and

  43105. links the program a second time including that file.

  43106. 

  43107.  The actual calls to the constructors are carried out by a subroutine

  43108. called '__main', which is called (automatically) at the beginning of the

  43109. body of 'main' (provided 'main' was compiled with GNU CC).  Calling

  43110. '__main' is necessary, even when compiling C code, to allow linking C

  43111. and C++ object code together.  (If you use '-nostdlib', you get an

  43112. unresolved reference to '__main', since it's defined in the standard GCC

  43113. library.  Include '-lgcc' at the end of your compiler command line to

  43114. resolve this reference.)

  43115. 

  43116.  The program 'collect2' is installed as 'ld' in the directory where the

  43117. passes of the compiler are installed.  When 'collect2' needs to find the

  43118. _real_ 'ld', it tries the following file names:

  43119. 

  43120.    * a hard coded linker file name, if GCC was configured with the

  43121.      '--with-ld' option.

  43122. 

  43123.    * 'real-ld' in the directories listed in the compiler's search

  43124.      directories.

  43125. 

  43126.    * 'real-ld' in the directories listed in the environment variable

  43127.      'PATH'.

  43128. 

  43129.    * The file specified in the 'REAL_LD_FILE_NAME' configuration macro,

  43130.      if specified.

  43131. 

  43132.    * 'ld' in the compiler's search directories, except that 'collect2'

  43133.      will not execute itself recursively.

  43134. 

  43135.    * 'ld' in 'PATH'.

  43136. 

  43137.  "The compiler's search directories" means all the directories where

  43138. 'gcc' searches for passes of the compiler.  This includes directories

  43139. that you specify with '-B'.

  43140. 

  43141.  Cross-compilers search a little differently:

  43142. 

  43143.    * 'real-ld' in the compiler's search directories.

  43144. 

  43145.    * 'TARGET-real-ld' in 'PATH'.

  43146. 

  43147.    * The file specified in the 'REAL_LD_FILE_NAME' configuration macro,

  43148.      if specified.

  43149. 

  43150.    * 'ld' in the compiler's search directories.

  43151. 

  43152.    * 'TARGET-ld' in 'PATH'.

  43153. 

  43154.  'collect2' explicitly avoids running 'ld' using the file name under

  43155. which 'collect2' itself was invoked.  In fact, it remembers up a list of

  43156. such names--in case one copy of 'collect2' finds another copy (or

  43157. version) of 'collect2' installed as 'ld' in a second place in the search

  43158. path.

  43159. 

  43160.  'collect2' searches for the utilities 'nm' and 'strip' using the same

  43161. algorithm as above for 'ld'.

  43162. 

  43163. 

  43164. File: gccint.info,  Node: Header Dirs,  Next: Type Information,  Prev: Collect2,  Up: Top

  43165. 

  43166. 22 Standard Header File Directories

  43167. ***********************************

  43168. 

  43169. 'GCC_INCLUDE_DIR' means the same thing for native and cross.  It is

  43170. where GCC stores its private include files, and also where GCC stores

  43171. the fixed include files.  A cross compiled GCC runs 'fixincludes' on the

  43172. header files in '$(tooldir)/include'.  (If the cross compilation header

  43173. files need to be fixed, they must be installed before GCC is built.  If

  43174. the cross compilation header files are already suitable for GCC, nothing

  43175. special need be done).

  43176. 

  43177.  'GPLUSPLUS_INCLUDE_DIR' means the same thing for native and cross.  It

  43178. is where 'g++' looks first for header files.  The C++ library installs

  43179. only target independent header files in that directory.

  43180. 

  43181.  'LOCAL_INCLUDE_DIR' is used only by native compilers.  GCC doesn't

  43182. install anything there.  It is normally '/usr/local/include'.  This is

  43183. where local additions to a packaged system should place header files.

  43184. 

  43185.  'CROSS_INCLUDE_DIR' is used only by cross compilers.  GCC doesn't

  43186. install anything there.

  43187. 

  43188.  'TOOL_INCLUDE_DIR' is used for both native and cross compilers.  It is

  43189. the place for other packages to install header files that GCC will use.

  43190. For a cross-compiler, this is the equivalent of '/usr/include'.  When

  43191. you build a cross-compiler, 'fixincludes' processes any header files in

  43192. this directory.

  43193. 

  43194. 

  43195. File: gccint.info,  Node: Type Information,  Next: Plugins,  Prev: Header Dirs,  Up: Top

  43196. 

  43197. 23 Memory Management and Type Information

  43198. *****************************************

  43199. 

  43200. GCC uses some fairly sophisticated memory management techniques, which

  43201. involve determining information about GCC's data structures from GCC's

  43202. source code and using this information to perform garbage collection and

  43203. implement precompiled headers.

  43204. 

  43205.  A full C++ parser would be too complicated for this task, so a limited

  43206. subset of C++ is interpreted and special markers are used to determine

  43207. what parts of the source to look at.  All 'struct', 'union' and

  43208. 'template' structure declarations that define data structures that are

  43209. allocated under control of the garbage collector must be marked.  All

  43210. global variables that hold pointers to garbage-collected memory must

  43211. also be marked.  Finally, all global variables that need to be saved and

  43212. restored by a precompiled header must be marked.  (The precompiled

  43213. header mechanism can only save static variables if they're scalar.

  43214. Complex data structures must be allocated in garbage-collected memory to

  43215. be saved in a precompiled header.)

  43216. 

  43217.  The full format of a marker is

  43218.      GTY (([OPTION] [(PARAM)], [OPTION] [(PARAM)] ...))

  43219. but in most cases no options are needed.  The outer double parentheses

  43220. are still necessary, though: 'GTY(())'.  Markers can appear:

  43221. 

  43222.    * In a structure definition, before the open brace;

  43223.    * In a global variable declaration, after the keyword 'static' or

  43224.      'extern'; and

  43225.    * In a structure field definition, before the name of the field.

  43226. 

  43227.  Here are some examples of marking simple data structures and globals.

  43228. 

  43229.      struct GTY(()) TAG

  43230.      {

  43231.        FIELDS...

  43232.      };

  43233. 

  43234.      typedef struct GTY(()) TAG

  43235.      {

  43236.        FIELDS...

  43237.      } *TYPENAME;

  43238. 

  43239.      static GTY(()) struct TAG *LIST;   /* points to GC memory */

  43240.      static GTY(()) int COUNTER;        /* save counter in a PCH */

  43241. 

  43242.  The parser understands simple typedefs such as 'typedef struct TAG

  43243. *NAME;' and 'typedef int NAME;'.  These don't need to be marked.

  43244. 

  43245.  Since 'gengtype''s understanding of C++ is limited, there are several

  43246. constructs and declarations that are not supported inside

  43247. classes/structures marked for automatic GC code generation.  The

  43248. following C++ constructs produce a 'gengtype' error on

  43249. structures/classes marked for automatic GC code generation:

  43250. 

  43251.    * Type definitions inside classes/structures are not supported.

  43252.    * Enumerations inside classes/structures are not supported.

  43253. 

  43254.  If you have a class or structure using any of the above constructs, you

  43255. need to mark that class as 'GTY ((user))' and provide your own marking

  43256. routines (see section *note User GC:: for details).

  43257. 

  43258.  It is always valid to include function definitions inside classes.

  43259. Those are always ignored by 'gengtype', as it only cares about data

  43260. members.

  43261. 

  43262. * Menu:

  43263. 

  43264. * GTY Options::         What goes inside a 'GTY(())'.

  43265. * Inheritance and GTY:: Adding GTY to a class hierarchy.

  43266. * User GC::		Adding user-provided GC marking routines.

  43267. * GGC Roots::           Making global variables GGC roots.

  43268. * Files::               How the generated files work.

  43269. * Invoking the garbage collector::   How to invoke the garbage collector.

  43270. * Troubleshooting::     When something does not work as expected.

  43271. 

  43272. 

  43273. File: gccint.info,  Node: GTY Options,  Next: Inheritance and GTY,  Up: Type Information

  43274. 

  43275. 23.1 The Inside of a 'GTY(())'

  43276. ==============================

  43277. 

  43278. Sometimes the C code is not enough to fully describe the type structure.

  43279. Extra information can be provided with 'GTY' options and additional

  43280. markers.  Some options take a parameter, which may be either a string or

  43281. a type name, depending on the parameter.  If an option takes no

  43282. parameter, it is acceptable either to omit the parameter entirely, or to

  43283. provide an empty string as a parameter.  For example, 'GTY ((skip))' and

  43284. 'GTY ((skip ("")))' are equivalent.

  43285. 

  43286.  When the parameter is a string, often it is a fragment of C code.  Four

  43287. special escapes may be used in these strings, to refer to pieces of the

  43288. data structure being marked:

  43289. 

  43290. '%h'

  43291.      The current structure.

  43292. '%1'

  43293.      The structure that immediately contains the current structure.

  43294. '%0'

  43295.      The outermost structure that contains the current structure.

  43296. '%a'

  43297.      A partial expression of the form '[i1][i2]...' that indexes the

  43298.      array item currently being marked.

  43299. 

  43300.  For instance, suppose that you have a structure of the form

  43301.      struct A {

  43302.        ...

  43303.      };

  43304.      struct B {

  43305.        struct A foo[12];

  43306.      };

  43307. and 'b' is a variable of type 'struct B'.  When marking 'b.foo[11]',

  43308. '%h' would expand to 'b.foo[11]', '%0' and '%1' would both expand to

  43309. 'b', and '%a' would expand to '[11]'.

  43310. 

  43311.  As in ordinary C, adjacent strings will be concatenated; this is

  43312. helpful when you have a complicated expression.

  43313.      GTY ((chain_next ("TREE_CODE (&%h.generic) == INTEGER_TYPE"

  43314.                        " ? TYPE_NEXT_VARIANT (&%h.generic)"

  43315.                        " : TREE_CHAIN (&%h.generic)")))

  43316. 

  43317.  The available options are:

  43318. 

  43319. 'length ("EXPRESSION")'

  43320. 

  43321.      There are two places the type machinery will need to be explicitly

  43322.      told the length of an array of non-atomic objects.  The first case

  43323.      is when a structure ends in a variable-length array, like this:

  43324.           struct GTY(()) rtvec_def {

  43325.             int num_elem;         /* number of elements */

  43326.             rtx GTY ((length ("%h.num_elem"))) elem[1];

  43327.           };

  43328. 

  43329.      In this case, the 'length' option is used to override the specified

  43330.      array length (which should usually be '1').  The parameter of the

  43331.      option is a fragment of C code that calculates the length.

  43332. 

  43333.      The second case is when a structure or a global variable contains a

  43334.      pointer to an array, like this:

  43335.           struct gimple_omp_for_iter * GTY((length ("%h.collapse"))) iter;

  43336.      In this case, 'iter' has been allocated by writing something like

  43337.             x->iter = ggc_alloc_cleared_vec_gimple_omp_for_iter (collapse);

  43338.      and the 'collapse' provides the length of the field.

  43339. 

  43340.      This second use of 'length' also works on global variables, like:

  43341.      static GTY((length("reg_known_value_size"))) rtx *reg_known_value;

  43342. 

  43343.      Note that the 'length' option is only meant for use with arrays of

  43344.      non-atomic objects, that is, objects that contain pointers pointing

  43345.      to other GTY-managed objects.  For other GC-allocated arrays and

  43346.      strings you should use 'atomic'.

  43347. 

  43348. 'skip'

  43349. 

  43350.      If 'skip' is applied to a field, the type machinery will ignore it.

  43351.      This is somewhat dangerous; the only safe use is in a union when

  43352.      one field really isn't ever used.

  43353. 

  43354. 'for_user'

  43355. 

  43356.      Use this to mark types that need to be marked by user gc routines,

  43357.      but are not refered to in a template argument.  So if you have some

  43358.      user gc type T1 and a non user gc type T2 you can give T2 the

  43359.      for_user option so that the marking functions for T1 can call non

  43360.      mangled functions to mark T2.

  43361. 

  43362. 'desc ("EXPRESSION")'

  43363. 'tag ("CONSTANT")'

  43364. 'default'

  43365. 

  43366.      The type machinery needs to be told which field of a 'union' is

  43367.      currently active.  This is done by giving each field a constant

  43368.      'tag' value, and then specifying a discriminator using 'desc'.  The

  43369.      value of the expression given by 'desc' is compared against each

  43370.      'tag' value, each of which should be different.  If no 'tag' is

  43371.      matched, the field marked with 'default' is used if there is one,

  43372.      otherwise no field in the union will be marked.

  43373. 

  43374.      In the 'desc' option, the "current structure" is the union that it

  43375.      discriminates.  Use '%1' to mean the structure containing it.

  43376.      There are no escapes available to the 'tag' option, since it is a

  43377.      constant.

  43378. 

  43379.      For example,

  43380.           struct GTY(()) tree_binding

  43381.           {

  43382.             struct tree_common common;

  43383.             union tree_binding_u {

  43384.               tree GTY ((tag ("0"))) scope;

  43385.               struct cp_binding_level * GTY ((tag ("1"))) level;

  43386.             } GTY ((desc ("BINDING_HAS_LEVEL_P ((tree)&%0)"))) xscope;

  43387.             tree value;

  43388.           };

  43389. 

  43390.      In this example, the value of BINDING_HAS_LEVEL_P when applied to a

  43391.      'struct tree_binding *' is presumed to be 0 or 1.  If 1, the type

  43392.      mechanism will treat the field 'level' as being present and if 0,

  43393.      will treat the field 'scope' as being present.

  43394. 

  43395.      The 'desc' and 'tag' options can also be used for inheritance to

  43396.      denote which subclass an instance is.  See *note Inheritance and

  43397.      GTY:: for more information.

  43398. 

  43399. 'cache'

  43400. 

  43401.      When the 'cache' option is applied to a global variable

  43402.      gt_clear_cache is called on that variable between the mark and

  43403.      sweep phases of garbage collection.  The gt_clear_cache function is

  43404.      free to mark blocks as used, or to clear pointers in the variable.

  43405. 

  43406. 'deletable'

  43407. 

  43408.      'deletable', when applied to a global variable, indicates that when

  43409.      garbage collection runs, there's no need to mark anything pointed

  43410.      to by this variable, it can just be set to 'NULL' instead.  This is

  43411.      used to keep a list of free structures around for re-use.

  43412. 

  43413. 'maybe_undef'

  43414. 

  43415.      When applied to a field, 'maybe_undef' indicates that it's OK if

  43416.      the structure that this fields points to is never defined, so long

  43417.      as this field is always 'NULL'.  This is used to avoid requiring

  43418.      backends to define certain optional structures.  It doesn't work

  43419.      with language frontends.

  43420. 

  43421. 'nested_ptr (TYPE, "TO EXPRESSION", "FROM EXPRESSION")'

  43422. 

  43423.      The type machinery expects all pointers to point to the start of an

  43424.      object.  Sometimes for abstraction purposes it's convenient to have

  43425.      a pointer which points inside an object.  So long as it's possible

  43426.      to convert the original object to and from the pointer, such

  43427.      pointers can still be used.  TYPE is the type of the original

  43428.      object, the TO EXPRESSION returns the pointer given the original

  43429.      object, and the FROM EXPRESSION returns the original object given

  43430.      the pointer.  The pointer will be available using the '%h' escape.

  43431. 

  43432. 'chain_next ("EXPRESSION")'

  43433. 'chain_prev ("EXPRESSION")'

  43434. 'chain_circular ("EXPRESSION")'

  43435. 

  43436.      It's helpful for the type machinery to know if objects are often

  43437.      chained together in long lists; this lets it generate code that

  43438.      uses less stack space by iterating along the list instead of

  43439.      recursing down it.  'chain_next' is an expression for the next item

  43440.      in the list, 'chain_prev' is an expression for the previous item.

  43441.      For singly linked lists, use only 'chain_next'; for doubly linked

  43442.      lists, use both.  The machinery requires that taking the next item

  43443.      of the previous item gives the original item.  'chain_circular' is

  43444.      similar to 'chain_next', but can be used for circular single linked

  43445.      lists.

  43446. 

  43447. 'reorder ("FUNCTION NAME")'

  43448. 

  43449.      Some data structures depend on the relative ordering of pointers.

  43450.      If the precompiled header machinery needs to change that ordering,

  43451.      it will call the function referenced by the 'reorder' option,

  43452.      before changing the pointers in the object that's pointed to by the

  43453.      field the option applies to.  The function must take four

  43454.      arguments, with the signature

  43455.      'void *, void *, gt_pointer_operator, void *'.  The first parameter

  43456.      is a pointer to the structure that contains the object being

  43457.      updated, or the object itself if there is no containing structure.

  43458.      The second parameter is a cookie that should be ignored.  The third

  43459.      parameter is a routine that, given a pointer, will update it to its

  43460.      correct new value.  The fourth parameter is a cookie that must be

  43461.      passed to the second parameter.

  43462. 

  43463.      PCH cannot handle data structures that depend on the absolute

  43464.      values of pointers.  'reorder' functions can be expensive.  When

  43465.      possible, it is better to depend on properties of the data, like an

  43466.      ID number or the hash of a string instead.

  43467. 

  43468. 'atomic'

  43469. 

  43470.      The 'atomic' option can only be used with pointers.  It informs the

  43471.      GC machinery that the memory that the pointer points to does not

  43472.      contain any pointers, and hence it should be treated by the GC and

  43473.      PCH machinery as an "atomic" block of memory that does not need to

  43474.      be examined when scanning memory for pointers.  In particular, the

  43475.      machinery will not scan that memory for pointers to mark them as

  43476.      reachable (when marking pointers for GC) or to relocate them (when

  43477.      writing a PCH file).

  43478. 

  43479.      The 'atomic' option differs from the 'skip' option.  'atomic' keeps

  43480.      the memory under Garbage Collection, but makes the GC ignore the

  43481.      contents of the memory.  'skip' is more drastic in that it causes

  43482.      the pointer and the memory to be completely ignored by the Garbage

  43483.      Collector.  So, memory marked as 'atomic' is automatically freed

  43484.      when no longer reachable, while memory marked as 'skip' is not.

  43485. 

  43486.      The 'atomic' option must be used with great care, because all sorts

  43487.      of problem can occur if used incorrectly, that is, if the memory

  43488.      the pointer points to does actually contain a pointer.

  43489. 

  43490.      Here is an example of how to use it:

  43491.           struct GTY(()) my_struct {

  43492.             int number_of_elements;

  43493.             unsigned int * GTY ((atomic)) elements;

  43494.           };

  43495.      In this case, 'elements' is a pointer under GC, and the memory it

  43496.      points to needs to be allocated using the Garbage Collector, and

  43497.      will be freed automatically by the Garbage Collector when it is no

  43498.      longer referenced.  But the memory that the pointer points to is an

  43499.      array of 'unsigned int' elements, and the GC must not try to scan

  43500.      it to find pointers to mark or relocate, which is why it is marked

  43501.      with the 'atomic' option.

  43502. 

  43503.      Note that, currently, global variables cannot be marked with

  43504.      'atomic'; only fields of a struct can.  This is a known limitation.

  43505.      It would be useful to be able to mark global pointers with 'atomic'

  43506.      to make the PCH machinery aware of them so that they are saved and

  43507.      restored correctly to PCH files.

  43508. 

  43509. 'special ("NAME")'

  43510. 

  43511.      The 'special' option is used to mark types that have to be dealt

  43512.      with by special case machinery.  The parameter is the name of the

  43513.      special case.  See 'gengtype.c' for further details.  Avoid adding

  43514.      new special cases unless there is no other alternative.

  43515. 

  43516. 'user'

  43517. 

  43518.      The 'user' option indicates that the code to mark structure fields

  43519.      is completely handled by user-provided routines.  See section *note

  43520.      User GC:: for details on what functions need to be provided.

  43521. 

  43522. 

  43523. File: gccint.info,  Node: Inheritance and GTY,  Next: User GC,  Prev: GTY Options,  Up: Type Information

  43524. 

  43525. 23.2 Support for inheritance

  43526. ============================

  43527. 

  43528. gengtype has some support for simple class hierarchies.  You can use

  43529. this to have gengtype autogenerate marking routines, provided:

  43530. 

  43531.    * There must be a concrete base class, with a discriminator

  43532.      expression that can be used to identify which subclass an instance

  43533.      is.

  43534.    * Only single inheritance is used.

  43535.    * None of the classes within the hierarchy are templates.

  43536. 

  43537.  If your class hierarchy does not fit in this pattern, you must use

  43538. *note User GC:: instead.

  43539. 

  43540.  The base class and its discriminator must be identified using the

  43541. "desc" option.  Each concrete subclass must use the "tag" option to

  43542. identify which value of the discriminator it corresponds to.

  43543. 

  43544.  Every class in the hierarchy must have a 'GTY(())' marker, as gengtype

  43545. will only attempt to parse classes that have such a marker (1).

  43546. 

  43547.      class GTY((desc("%h.kind"), tag("0"))) example_base

  43548.      {

  43549.      public:

  43550.          int kind;

  43551.          tree a;

  43552.      };

  43553. 

  43554.      class GTY((tag("1"))) some_subclass : public example_base

  43555.      {

  43556.      public:

  43557.          tree b;

  43558.      };

  43559. 

  43560.      class GTY((tag("2"))) some_other_subclass : public example_base

  43561.      {

  43562.      public:

  43563.          tree c;

  43564.      };

  43565. 

  43566.  The generated marking routines for the above will contain a "switch" on

  43567. "kind", visiting all appropriate fields.  For example, if kind is 2, it

  43568. will cast to "some_other_subclass" and visit fields a, b, and c.

  43569. 

  43570.    ---------- Footnotes ----------

  43571. 

  43572.    (1) Classes lacking such a marker will not be identified as being

  43573. part of the hierarchy, and so the marking routines will not handle them,

  43574. leading to a assertion failure within the marking routines due to an

  43575. unknown tag value (assuming that assertions are enabled).

  43576. 

  43577. 

  43578. File: gccint.info,  Node: User GC,  Next: GGC Roots,  Prev: Inheritance and GTY,  Up: Type Information

  43579. 

  43580. 23.3 Support for user-provided GC marking routines

  43581. ==================================================

  43582. 

  43583. The garbage collector supports types for which no automatic marking code

  43584. is generated.  For these types, the user is required to provide three

  43585. functions: one to act as a marker for garbage collection, and two

  43586. functions to act as marker and pointer walker for pre-compiled headers.

  43587. 

  43588.  Given a structure 'struct GTY((user)) my_struct', the following

  43589. functions should be defined to mark 'my_struct':

  43590. 

  43591.      void gt_ggc_mx (my_struct *p)

  43592.      {

  43593.        /* This marks field 'fld'.  */

  43594.        gt_ggc_mx (p->fld);

  43595.      }

  43596. 

  43597.      void gt_pch_nx (my_struct *p)

  43598.      {

  43599.        /* This marks field 'fld'.  */

  43600.        gt_pch_nx (tp->fld);

  43601.      }

  43602. 

  43603.      void gt_pch_nx (my_struct *p, gt_pointer_operator op, void *cookie)

  43604.      {

  43605.        /* For every field 'fld', call the given pointer operator.  */

  43606.        op (&(tp->fld), cookie);

  43607.      }

  43608. 

  43609.  In general, each marker 'M' should call 'M' for every pointer field in

  43610. the structure.  Fields that are not allocated in GC or are not pointers

  43611. must be ignored.

  43612. 

  43613.  For embedded lists (e.g., structures with a 'next' or 'prev' pointer),

  43614. the marker must follow the chain and mark every element in it.

  43615. 

  43616.  Note that the rules for the pointer walker 'gt_pch_nx (my_struct *,

  43617. gt_pointer_operator, void *)' are slightly different.  In this case, the

  43618. operation 'op' must be applied to the _address_ of every pointer field.

  43619. 

  43620. 23.3.1 User-provided marking routines for template types

  43621. --------------------------------------------------------

  43622. 

  43623. When a template type 'TP' is marked with 'GTY', all instances of that

  43624. type are considered user-provided types.  This means that the individual

  43625. instances of 'TP' do not need to be marked with 'GTY'.  The user needs

  43626. to provide template functions to mark all the fields of the type.

  43627. 

  43628.  The following code snippets represent all the functions that need to be

  43629. provided.  Note that type 'TP' may reference to more than one type.  In

  43630. these snippets, there is only one type 'T', but there could be more.

  43631. 

  43632.      template<typename T>

  43633.      void gt_ggc_mx (TP<T> *tp)

  43634.      {

  43635.        extern void gt_ggc_mx (T&);

  43636. 

  43637.        /* This marks field 'fld' of type 'T'.  */

  43638.        gt_ggc_mx (tp->fld);

  43639.      }

  43640. 

  43641.      template<typename T>

  43642.      void gt_pch_nx (TP<T> *tp)

  43643.      {

  43644.        extern void gt_pch_nx (T&);

  43645. 

  43646.        /* This marks field 'fld' of type 'T'.  */

  43647.        gt_pch_nx (tp->fld);

  43648.      }

  43649. 

  43650.      template<typename T>

  43651.      void gt_pch_nx (TP<T *> *tp, gt_pointer_operator op, void *cookie)

  43652.      {

  43653.        /* For every field 'fld' of 'tp' with type 'T *', call the given

  43654.           pointer operator.  */

  43655.        op (&(tp->fld), cookie);

  43656.      }

  43657. 

  43658.      template<typename T>

  43659.      void gt_pch_nx (TP<T> *tp, gt_pointer_operator, void *cookie)

  43660.      {

  43661.        extern void gt_pch_nx (T *, gt_pointer_operator, void *);

  43662. 

  43663.        /* For every field 'fld' of 'tp' with type 'T', call the pointer

  43664.           walker for all the fields of T.  */

  43665.        gt_pch_nx (&(tp->fld), op, cookie);

  43666.      }

  43667. 

  43668.  Support for user-defined types is currently limited.  The following

  43669. restrictions apply:

  43670. 

  43671.   1. Type 'TP' and all the argument types 'T' must be marked with 'GTY'.

  43672. 

  43673.   2. Type 'TP' can only have type names in its argument list.

  43674. 

  43675.   3. The pointer walker functions are different for 'TP<T>' and 'TP<T

  43676.      *>'.  In the case of 'TP<T>', references to 'T' must be handled by

  43677.      calling 'gt_pch_nx' (which will, in turn, walk all the pointers

  43678.      inside fields of 'T').  In the case of 'TP<T *>', references to 'T

  43679.      *' must be handled by calling the 'op' function on the address of

  43680.      the pointer (see the code snippets above).

  43681. 

  43682. 

  43683. File: gccint.info,  Node: GGC Roots,  Next: Files,  Prev: User GC,  Up: Type Information

  43684. 

  43685. 23.4 Marking Roots for the Garbage Collector

  43686. ============================================

  43687. 

  43688. In addition to keeping track of types, the type machinery also locates

  43689. the global variables ("roots") that the garbage collector starts at.

  43690. Roots must be declared using one of the following syntaxes:

  43691. 

  43692.    * 'extern GTY(([OPTIONS])) TYPE NAME;'

  43693.    * 'static GTY(([OPTIONS])) TYPE NAME;'

  43694. The syntax

  43695.    * 'GTY(([OPTIONS])) TYPE NAME;'

  43696. is _not_ accepted.  There should be an 'extern' declaration of such a

  43697. variable in a header somewhere--mark that, not the definition.  Or, if

  43698. the variable is only used in one file, make it 'static'.

  43699. 

  43700. 

  43701. File: gccint.info,  Node: Files,  Next: Invoking the garbage collector,  Prev: GGC Roots,  Up: Type Information

  43702. 

  43703. 23.5 Source Files Containing Type Information

  43704. =============================================

  43705. 

  43706. Whenever you add 'GTY' markers to a source file that previously had

  43707. none, or create a new source file containing 'GTY' markers, there are

  43708. three things you need to do:

  43709. 

  43710.   1. You need to add the file to the list of source files the type

  43711.      machinery scans.  There are four cases:

  43712. 

  43713.        a. For a back-end file, this is usually done automatically; if

  43714.           not, you should add it to 'target_gtfiles' in the appropriate

  43715.           port's entries in 'config.gcc'.

  43716. 

  43717.        b. For files shared by all front ends, add the filename to the

  43718.           'GTFILES' variable in 'Makefile.in'.

  43719. 

  43720.        c. For files that are part of one front end, add the filename to

  43721.           the 'gtfiles' variable defined in the appropriate

  43722.           'config-lang.in'.  Headers should appear before non-headers in

  43723.           this list.

  43724. 

  43725.        d. For files that are part of some but not all front ends, add

  43726.           the filename to the 'gtfiles' variable of _all_ the front ends

  43727.           that use it.

  43728. 

  43729.   2. If the file was a header file, you'll need to check that it's

  43730.      included in the right place to be visible to the generated files.

  43731.      For a back-end header file, this should be done automatically.  For

  43732.      a front-end header file, it needs to be included by the same file

  43733.      that includes 'gtype-LANG.h'.  For other header files, it needs to

  43734.      be included in 'gtype-desc.c', which is a generated file, so add it

  43735.      to 'ifiles' in 'open_base_file' in 'gengtype.c'.

  43736. 

  43737.      For source files that aren't header files, the machinery will

  43738.      generate a header file that should be included in the source file

  43739.      you just changed.  The file will be called 'gt-PATH.h' where PATH

  43740.      is the pathname relative to the 'gcc' directory with slashes

  43741.      replaced by -, so for example the header file to be included in

  43742.      'cp/parser.c' is called 'gt-cp-parser.c'.  The generated header

  43743.      file should be included after everything else in the source file.

  43744.      Don't forget to mention this file as a dependency in the

  43745.      'Makefile'!

  43746. 

  43747.  For language frontends, there is another file that needs to be included

  43748. somewhere.  It will be called 'gtype-LANG.h', where LANG is the name of

  43749. the subdirectory the language is contained in.

  43750. 

  43751.  Plugins can add additional root tables.  Run the 'gengtype' utility in

  43752. plugin mode as 'gengtype -P pluginout.h SOURCE-DIR FILE-LIST PLUGIN*.C'

  43753. with your plugin files PLUGIN*.C using 'GTY' to generate the PLUGINOUT.H

  43754. file.  The GCC build tree is needed to be present in that mode.

  43755. 

  43756. 

  43757. File: gccint.info,  Node: Invoking the garbage collector,  Next: Troubleshooting,  Prev: Files,  Up: Type Information

  43758. 

  43759. 23.6 How to invoke the garbage collector

  43760. ========================================

  43761. 

  43762. The GCC garbage collector GGC is only invoked explicitly.  In contrast

  43763. with many other garbage collectors, it is not implicitly invoked by

  43764. allocation routines when a lot of memory has been consumed.  So the only

  43765. way to have GGC reclaim storage is to call the 'ggc_collect' function

  43766. explicitly.  This call is an expensive operation, as it may have to scan

  43767. the entire heap.  Beware that local variables (on the GCC call stack)

  43768. are not followed by such an invocation (as many other garbage collectors

  43769. do): you should reference all your data from static or external 'GTY'-ed

  43770. variables, and it is advised to call 'ggc_collect' with a shallow call

  43771. stack.  The GGC is an exact mark and sweep garbage collector (so it does

  43772. not scan the call stack for pointers).  In practice GCC passes don't

  43773. often call 'ggc_collect' themselves, because it is called by the pass

  43774. manager between passes.

  43775. 

  43776.  At the time of the 'ggc_collect' call all pointers in the GC-marked

  43777. structures must be valid or 'NULL'.  In practice this means that there

  43778. should not be uninitialized pointer fields in the structures even if

  43779. your code never reads or writes those fields at a particular instance.

  43780. One way to ensure this is to use cleared versions of allocators unless

  43781. all the fields are initialized manually immediately after allocation.

  43782. 

  43783. 

  43784. File: gccint.info,  Node: Troubleshooting,  Prev: Invoking the garbage collector,  Up: Type Information

  43785. 

  43786. 23.7 Troubleshooting the garbage collector

  43787. ==========================================

  43788. 

  43789. With the current garbage collector implementation, most issues should

  43790. show up as GCC compilation errors.  Some of the most commonly

  43791. encountered issues are described below.

  43792. 

  43793.    * Gengtype does not produce allocators for a 'GTY'-marked type.

  43794.      Gengtype checks if there is at least one possible path from GC

  43795.      roots to at least one instance of each type before outputting

  43796.      allocators.  If there is no such path, the 'GTY' markers will be

  43797.      ignored and no allocators will be output.  Solve this by making

  43798.      sure that there exists at least one such path.  If creating it is

  43799.      unfeasible or raises a "code smell", consider if you really must

  43800.      use GC for allocating such type.

  43801. 

  43802.    * Link-time errors about undefined 'gt_ggc_r_foo_bar' and

  43803.      similarly-named symbols.  Check if your 'foo_bar' source file has

  43804.      '#include "gt-foo_bar.h"' as its very last line.

  43805. 

  43806. 

  43807. File: gccint.info,  Node: Plugins,  Next: LTO,  Prev: Type Information,  Up: Top

  43808. 

  43809. 24 Plugins

  43810. **********

  43811. 

  43812. GCC plugins are loadable modules that provide extra features to the

  43813. compiler.  Like GCC itself they can be distributed in source and binary

  43814. forms.

  43815. 

  43816.  GCC plugins provide developers with a rich subset of the GCC API to

  43817. allow them to extend GCC as they see fit.  Whether it is writing an

  43818. additional optimization pass, transforming code, or analyzing

  43819. information, plugins can be quite useful.

  43820. 

  43821. * Menu:

  43822. 

  43823. * Plugins loading::      How can we load plugins.

  43824. * Plugin API::           The APIs for plugins.

  43825. * Plugins pass::         How a plugin interact with the pass manager.

  43826. * Plugins GC::           How a plugin Interact with GCC Garbage Collector.

  43827. * Plugins description::  Giving information about a plugin itself.

  43828. * Plugins attr::         Registering custom attributes or pragmas.

  43829. * Plugins recording::    Recording information about pass execution.

  43830. * Plugins gate::         Controlling which passes are being run.

  43831. * Plugins tracking::     Keeping track of available passes.

  43832. * Plugins building::     How can we build a plugin.

  43833. 

  43834. 

  43835. File: gccint.info,  Node: Plugins loading,  Next: Plugin API,  Up: Plugins

  43836. 

  43837. 24.1 Loading Plugins

  43838. ====================

  43839. 

  43840. Plugins are supported on platforms that support '-ldl -rdynamic' as well

  43841. as Windows/MinGW. They are loaded by the compiler using 'dlopen' or

  43842. equivalent and invoked at pre-determined locations in the compilation

  43843. process.

  43844. 

  43845.  Plugins are loaded with

  43846. 

  43847.  '-fplugin=/path/to/NAME.EXT' '-fplugin-arg-NAME-KEY1[=VALUE1]'

  43848. 

  43849.  Where NAME is the plugin name and EXT is the platform-specific dynamic

  43850. library extension.  It should be 'dll' on Windows/MinGW, 'dylib' on

  43851. Darwin/Mac OS X, and 'so' on all other platforms.  The plugin arguments

  43852. are parsed by GCC and passed to respective plugins as key-value pairs.

  43853. Multiple plugins can be invoked by specifying multiple '-fplugin'

  43854. arguments.

  43855. 

  43856.  A plugin can be simply given by its short name (no dots or slashes).

  43857. When simply passing '-fplugin=NAME', the plugin is loaded from the

  43858. 'plugin' directory, so '-fplugin=NAME' is the same as '-fplugin=`gcc

  43859. -print-file-name=plugin`/NAME.EXT', using backquote shell syntax to

  43860. query the 'plugin' directory.

  43861. 

  43862. 

  43863. File: gccint.info,  Node: Plugin API,  Next: Plugins pass,  Prev: Plugins loading,  Up: Plugins

  43864. 

  43865. 24.2 Plugin API

  43866. ===============

  43867. 

  43868. Plugins are activated by the compiler at specific events as defined in

  43869. 'gcc-plugin.h'.  For each event of interest, the plugin should call

  43870. 'register_callback' specifying the name of the event and address of the

  43871. callback function that will handle that event.

  43872. 

  43873.  The header 'gcc-plugin.h' must be the first gcc header to be included.

  43874. 

  43875. 24.2.1 Plugin license check

  43876. ---------------------------

  43877. 

  43878. Every plugin should define the global symbol 'plugin_is_GPL_compatible'

  43879. to assert that it has been licensed under a GPL-compatible license.  If

  43880. this symbol does not exist, the compiler will emit a fatal error and

  43881. exit with the error message:

  43882. 

  43883.      fatal error: plugin NAME is not licensed under a GPL-compatible license

  43884.      NAME: undefined symbol: plugin_is_GPL_compatible

  43885.      compilation terminated

  43886. 

  43887.  The declared type of the symbol should be int, to match a forward

  43888. declaration in 'gcc-plugin.h' that suppresses C++ mangling.  It does not

  43889. need to be in any allocated section, though.  The compiler merely

  43890. asserts that the symbol exists in the global scope.  Something like this

  43891. is enough:

  43892. 

  43893.      int plugin_is_GPL_compatible;

  43894. 

  43895. 24.2.2 Plugin initialization

  43896. ----------------------------

  43897. 

  43898. Every plugin should export a function called 'plugin_init' that is

  43899. called right after the plugin is loaded.  This function is responsible

  43900. for registering all the callbacks required by the plugin and do any

  43901. other required initialization.

  43902. 

  43903.  This function is called from 'compile_file' right before invoking the

  43904. parser.  The arguments to 'plugin_init' are:

  43905. 

  43906.    * 'plugin_info': Plugin invocation information.

  43907.    * 'version': GCC version.

  43908. 

  43909.  The 'plugin_info' struct is defined as follows:

  43910. 

  43911.      struct plugin_name_args

  43912.      {

  43913.        char *base_name;              /* Short name of the plugin

  43914.                                         (filename without .so suffix). */

  43915.        const char *full_name;        /* Path to the plugin as specified with

  43916.                                         -fplugin=. */

  43917.        int argc;                     /* Number of arguments specified with

  43918.                                         -fplugin-arg-.... */

  43919.        struct plugin_argument *argv; /* Array of ARGC key-value pairs. */

  43920.        const char *version;          /* Version string provided by plugin. */

  43921.        const char *help;             /* Help string provided by plugin. */

  43922.      }

  43923. 

  43924.  If initialization fails, 'plugin_init' must return a non-zero value.

  43925. Otherwise, it should return 0.

  43926. 

  43927.  The version of the GCC compiler loading the plugin is described by the

  43928. following structure:

  43929. 

  43930.      struct plugin_gcc_version

  43931.      {

  43932.        const char *basever;

  43933.        const char *datestamp;

  43934.        const char *devphase;

  43935.        const char *revision;

  43936.        const char *configuration_arguments;

  43937.      };

  43938. 

  43939.  The function 'plugin_default_version_check' takes two pointers to such

  43940. structure and compare them field by field.  It can be used by the

  43941. plugin's 'plugin_init' function.

  43942. 

  43943.  The version of GCC used to compile the plugin can be found in the

  43944. symbol 'gcc_version' defined in the header 'plugin-version.h'.  The

  43945. recommended version check to perform looks like

  43946. 

  43947.      #include "plugin-version.h"

  43948.      ...

  43949. 

  43950.      int

  43951.      plugin_init (struct plugin_name_args *plugin_info,

  43952.                   struct plugin_gcc_version *version)

  43953.      {

  43954.        if (!plugin_default_version_check (version, &gcc_version))

  43955.          return 1;

  43956. 

  43957.      }

  43958. 

  43959.  but you can also check the individual fields if you want a less strict

  43960. check.

  43961. 

  43962. 24.2.3 Plugin callbacks

  43963. -----------------------

  43964. 

  43965. Callback functions have the following prototype:

  43966. 

  43967.      /* The prototype for a plugin callback function.

  43968.           gcc_data  - event-specific data provided by GCC

  43969.           user_data - plugin-specific data provided by the plug-in.  */

  43970.      typedef void (*plugin_callback_func)(void *gcc_data, void *user_data);

  43971. 

  43972.  Callbacks can be invoked at the following pre-determined events:

  43973. 

  43974.      enum plugin_event

  43975.      {

  43976.        PLUGIN_START_PARSE_FUNCTION,  /* Called before parsing the body of a function. */

  43977.        PLUGIN_FINISH_PARSE_FUNCTION, /* After finishing parsing a function. */

  43978.        PLUGIN_PASS_MANAGER_SETUP,    /* To hook into pass manager.  */

  43979.        PLUGIN_FINISH_TYPE,           /* After finishing parsing a type.  */

  43980.        PLUGIN_FINISH_DECL,           /* After finishing parsing a declaration. */

  43981.        PLUGIN_FINISH_UNIT,           /* Useful for summary processing.  */

  43982.        PLUGIN_PRE_GENERICIZE,        /* Allows to see low level AST in C and C++ frontends.  */

  43983.        PLUGIN_FINISH,                /* Called before GCC exits.  */

  43984.        PLUGIN_INFO,                  /* Information about the plugin. */

  43985.        PLUGIN_GGC_START,             /* Called at start of GCC Garbage Collection. */

  43986.        PLUGIN_GGC_MARKING,           /* Extend the GGC marking. */

  43987.        PLUGIN_GGC_END,               /* Called at end of GGC. */

  43988.        PLUGIN_REGISTER_GGC_ROOTS,    /* Register an extra GGC root table. */

  43989.        PLUGIN_ATTRIBUTES,            /* Called during attribute registration */

  43990.        PLUGIN_START_UNIT,            /* Called before processing a translation unit.  */

  43991.        PLUGIN_PRAGMAS,               /* Called during pragma registration. */

  43992.        /* Called before first pass from all_passes.  */

  43993.        PLUGIN_ALL_PASSES_START,

  43994.        /* Called after last pass from all_passes.  */

  43995.        PLUGIN_ALL_PASSES_END,

  43996.        /* Called before first ipa pass.  */

  43997.        PLUGIN_ALL_IPA_PASSES_START,

  43998.        /* Called after last ipa pass.  */

  43999.        PLUGIN_ALL_IPA_PASSES_END,

  44000.        /* Allows to override pass gate decision for current_pass.  */

  44001.        PLUGIN_OVERRIDE_GATE,

  44002.        /* Called before executing a pass.  */

  44003.        PLUGIN_PASS_EXECUTION,

  44004.        /* Called before executing subpasses of a GIMPLE_PASS in

  44005.           execute_ipa_pass_list.  */

  44006.        PLUGIN_EARLY_GIMPLE_PASSES_START,

  44007.        /* Called after executing subpasses of a GIMPLE_PASS in

  44008.           execute_ipa_pass_list.  */

  44009.        PLUGIN_EARLY_GIMPLE_PASSES_END,

  44010.        /* Called when a pass is first instantiated.  */

  44011.        PLUGIN_NEW_PASS,

  44012.      /* Called when a file is #include-d or given via the #line directive.

  44013.         This could happen many times.  The event data is the included file path,

  44014.         as a const char* pointer.  */

  44015.        PLUGIN_INCLUDE_FILE,

  44016. 

  44017.        PLUGIN_EVENT_FIRST_DYNAMIC    /* Dummy event used for indexing callback

  44018.                                         array.  */

  44019.      };

  44020. 

  44021.  In addition, plugins can also look up the enumerator of a named event,

  44022. and / or generate new events dynamically, by calling the function

  44023. 'get_named_event_id'.

  44024. 

  44025.  To register a callback, the plugin calls 'register_callback' with the

  44026. arguments:

  44027. 

  44028.    * 'char *name': Plugin name.

  44029.    * 'int event': The event code.

  44030.    * 'plugin_callback_func callback': The function that handles 'event'.

  44031.    * 'void *user_data': Pointer to plugin-specific data.

  44032. 

  44033.  For the PLUGIN_PASS_MANAGER_SETUP, PLUGIN_INFO, and

  44034. PLUGIN_REGISTER_GGC_ROOTS pseudo-events the 'callback' should be null,

  44035. and the 'user_data' is specific.

  44036. 

  44037.  When the PLUGIN_PRAGMAS event is triggered (with a null pointer as data

  44038. from GCC), plugins may register their own pragmas.  Notice that pragmas

  44039. are not available from 'lto1', so plugins used with '-flto' option to

  44040. GCC during link-time optimization cannot use pragmas and do not even see

  44041. functions like 'c_register_pragma' or 'pragma_lex'.

  44042. 

  44043.  The PLUGIN_INCLUDE_FILE event, with a 'const char*' file path as GCC

  44044. data, is triggered for processing of '#include' or '#line' directives.

  44045. 

  44046.  The PLUGIN_FINISH event is the last time that plugins can call GCC

  44047. functions, notably emit diagnostics with 'warning', 'error' etc.

  44048. 

  44049. 

  44050. File: gccint.info,  Node: Plugins pass,  Next: Plugins GC,  Prev: Plugin API,  Up: Plugins

  44051. 

  44052. 24.3 Interacting with the pass manager

  44053. ======================================

  44054. 

  44055. There needs to be a way to add/reorder/remove passes dynamically.  This

  44056. is useful for both analysis plugins (plugging in after a certain pass

  44057. such as CFG or an IPA pass) and optimization plugins.

  44058. 

  44059.  Basic support for inserting new passes or replacing existing passes is

  44060. provided.  A plugin registers a new pass with GCC by calling

  44061. 'register_callback' with the 'PLUGIN_PASS_MANAGER_SETUP' event and a

  44062. pointer to a 'struct register_pass_info' object defined as follows

  44063. 

  44064.      enum pass_positioning_ops

  44065.      {

  44066.        PASS_POS_INSERT_AFTER,  // Insert after the reference pass.

  44067.        PASS_POS_INSERT_BEFORE, // Insert before the reference pass.

  44068.        PASS_POS_REPLACE        // Replace the reference pass.

  44069.      };

  44070. 

  44071.      struct register_pass_info

  44072.      {

  44073.        struct opt_pass *pass;            /* New pass provided by the plugin.  */

  44074.        const char *reference_pass_name;  /* Name of the reference pass for hooking

  44075.                                             up the new pass.  */

  44076.        int ref_pass_instance_number;     /* Insert the pass at the specified

  44077.                                             instance number of the reference pass.  */

  44078.                                          /* Do it for every instance if it is 0.  */

  44079.        enum pass_positioning_ops pos_op; /* how to insert the new pass.  */

  44080.      };

  44081. 

  44082. 

  44083.      /* Sample plugin code that registers a new pass.  */

  44084.      int

  44085.      plugin_init (struct plugin_name_args *plugin_info,

  44086.                   struct plugin_gcc_version *version)

  44087.      {

  44088.        struct register_pass_info pass_info;

  44089. 

  44090.        ...

  44091. 

  44092.        /* Code to fill in the pass_info object with new pass information.  */

  44093. 

  44094.        ...

  44095. 

  44096.        /* Register the new pass.  */

  44097.        register_callback (plugin_info->base_name, PLUGIN_PASS_MANAGER_SETUP, NULL, &pass_info);

  44098. 

  44099.        ...

  44100.      }

  44101. 

  44102. 

  44103. File: gccint.info,  Node: Plugins GC,  Next: Plugins description,  Prev: Plugins pass,  Up: Plugins

  44104. 

  44105. 24.4 Interacting with the GCC Garbage Collector

  44106. ===============================================

  44107. 

  44108. Some plugins may want to be informed when GGC (the GCC Garbage

  44109. Collector) is running.  They can register callbacks for the

  44110. 'PLUGIN_GGC_START' and 'PLUGIN_GGC_END' events (for which the callback

  44111. is called with a null 'gcc_data') to be notified of the start or end of

  44112. the GCC garbage collection.

  44113. 

  44114.  Some plugins may need to have GGC mark additional data.  This can be

  44115. done by registering a callback (called with a null 'gcc_data') for the

  44116. 'PLUGIN_GGC_MARKING' event.  Such callbacks can call the 'ggc_set_mark'

  44117. routine, preferably through the 'ggc_mark' macro (and conversely, these

  44118. routines should usually not be used in plugins outside of the

  44119. 'PLUGIN_GGC_MARKING' event).  Plugins that wish to hold weak references

  44120. to gc data may also use this event to drop weak references when the

  44121. object is about to be collected.  The 'ggc_marked_p' function can be

  44122. used to tell if an object is marked, or is about to be collected.  The

  44123. 'gt_clear_cache' overloads which some types define may also be of use in

  44124. managing weak references.

  44125. 

  44126.  Some plugins may need to add extra GGC root tables, e.g. to handle

  44127. their own 'GTY'-ed data.  This can be done with the

  44128. 'PLUGIN_REGISTER_GGC_ROOTS' pseudo-event with a null callback and the

  44129. extra root table (of type 'struct ggc_root_tab*') as 'user_data'.

  44130. Running the 'gengtype -p SOURCE-DIR FILE-LIST PLUGIN*.C ...' utility

  44131. generates these extra root tables.

  44132. 

  44133.  You should understand the details of memory management inside GCC

  44134. before using 'PLUGIN_GGC_MARKING' or 'PLUGIN_REGISTER_GGC_ROOTS'.

  44135. 

  44136. 

  44137. File: gccint.info,  Node: Plugins description,  Next: Plugins attr,  Prev: Plugins GC,  Up: Plugins

  44138. 

  44139. 24.5 Giving information about a plugin

  44140. ======================================

  44141. 

  44142. A plugin should give some information to the user about itself.  This

  44143. uses the following structure:

  44144. 

  44145.      struct plugin_info

  44146.      {

  44147.        const char *version;

  44148.        const char *help;

  44149.      };

  44150. 

  44151.  Such a structure is passed as the 'user_data' by the plugin's init

  44152. routine using 'register_callback' with the 'PLUGIN_INFO' pseudo-event

  44153. and a null callback.

  44154. 

  44155. 

  44156. File: gccint.info,  Node: Plugins attr,  Next: Plugins recording,  Prev: Plugins description,  Up: Plugins

  44157. 

  44158. 24.6 Registering custom attributes or pragmas

  44159. =============================================

  44160. 

  44161. For analysis (or other) purposes it is useful to be able to add custom

  44162. attributes or pragmas.

  44163. 

  44164.  The 'PLUGIN_ATTRIBUTES' callback is called during attribute

  44165. registration.  Use the 'register_attribute' function to register custom

  44166. attributes.

  44167. 

  44168.      /* Attribute handler callback */

  44169.      static tree

  44170.      handle_user_attribute (tree *node, tree name, tree args,

  44171.                             int flags, bool *no_add_attrs)

  44172.      {

  44173.        return NULL_TREE;

  44174.      }

  44175. 

  44176.      /* Attribute definition */

  44177.      static struct attribute_spec user_attr =

  44178.        { "user", 1, 1, false,  false, false, false, handle_user_attribute, NULL };

  44179. 

  44180.      /* Plugin callback called during attribute registration.

  44181.      Registered with register_callback (plugin_name, PLUGIN_ATTRIBUTES, register_attributes, NULL)

  44182.      */

  44183.      static void

  44184.      register_attributes (void *event_data, void *data)

  44185.      {

  44186.        warning (0, G_("Callback to register attributes"));

  44187.        register_attribute (&user_attr);

  44188.      }

  44189. 

  44190. 

  44191.  The PLUGIN_PRAGMAS callback is called once during pragmas registration.

  44192. Use the 'c_register_pragma', 'c_register_pragma_with_data',

  44193. 'c_register_pragma_with_expansion',

  44194. 'c_register_pragma_with_expansion_and_data' functions to register custom

  44195. pragmas and their handlers (which often want to call 'pragma_lex') from

  44196. 'c-family/c-pragma.h'.

  44197. 

  44198.      /* Plugin callback called during pragmas registration. Registered with

  44199.           register_callback (plugin_name, PLUGIN_PRAGMAS,

  44200.                              register_my_pragma, NULL);

  44201.      */

  44202.      static void

  44203.      register_my_pragma (void *event_data, void *data)

  44204.      {

  44205.        warning (0, G_("Callback to register pragmas"));

  44206.        c_register_pragma ("GCCPLUGIN", "sayhello", handle_pragma_sayhello);

  44207.      }

  44208. 

  44209.  It is suggested to pass '"GCCPLUGIN"' (or a short name identifying your

  44210. plugin) as the "space" argument of your pragma.

  44211. 

  44212.  Pragmas registered with 'c_register_pragma_with_expansion' or

  44213. 'c_register_pragma_with_expansion_and_data' support preprocessor

  44214. expansions.  For example:

  44215. 

  44216.      #define NUMBER 10

  44217.      #pragma GCCPLUGIN foothreshold (NUMBER)

  44218. 

  44219. 

  44220. File: gccint.info,  Node: Plugins recording,  Next: Plugins gate,  Prev: Plugins attr,  Up: Plugins

  44221. 

  44222. 24.7 Recording information about pass execution

  44223. ===============================================

  44224. 

  44225. The event PLUGIN_PASS_EXECUTION passes the pointer to the executed pass

  44226. (the same as current_pass) as 'gcc_data' to the callback.  You can also

  44227. inspect cfun to find out about which function this pass is executed for.

  44228. Note that this event will only be invoked if the gate check (if

  44229. applicable, modified by PLUGIN_OVERRIDE_GATE) succeeds.  You can use

  44230. other hooks, like 'PLUGIN_ALL_PASSES_START', 'PLUGIN_ALL_PASSES_END',

  44231. 'PLUGIN_ALL_IPA_PASSES_START', 'PLUGIN_ALL_IPA_PASSES_END',

  44232. 'PLUGIN_EARLY_GIMPLE_PASSES_START', and/or

  44233. 'PLUGIN_EARLY_GIMPLE_PASSES_END' to manipulate global state in your

  44234. plugin(s) in order to get context for the pass execution.

  44235. 

  44236. 

  44237. File: gccint.info,  Node: Plugins gate,  Next: Plugins tracking,  Prev: Plugins recording,  Up: Plugins

  44238. 

  44239. 24.8 Controlling which passes are being run

  44240. ===========================================

  44241. 

  44242. After the original gate function for a pass is called, its result - the

  44243. gate status - is stored as an integer.  Then the event

  44244. 'PLUGIN_OVERRIDE_GATE' is invoked, with a pointer to the gate status in

  44245. the 'gcc_data' parameter to the callback function.  A nonzero value of

  44246. the gate status means that the pass is to be executed.  You can both

  44247. read and write the gate status via the passed pointer.

  44248. 

  44249. 

  44250. File: gccint.info,  Node: Plugins tracking,  Next: Plugins building,  Prev: Plugins gate,  Up: Plugins

  44251. 

  44252. 24.9 Keeping track of available passes

  44253. ======================================

  44254. 

  44255. When your plugin is loaded, you can inspect the various pass lists to

  44256. determine what passes are available.  However, other plugins might add

  44257. new passes.  Also, future changes to GCC might cause generic passes to

  44258. be added after plugin loading.  When a pass is first added to one of the

  44259. pass lists, the event 'PLUGIN_NEW_PASS' is invoked, with the callback

  44260. parameter 'gcc_data' pointing to the new pass.

  44261. 

  44262. 

  44263. File: gccint.info,  Node: Plugins building,  Prev: Plugins tracking,  Up: Plugins

  44264. 

  44265. 24.10 Building GCC plugins

  44266. ==========================

  44267. 

  44268. If plugins are enabled, GCC installs the headers needed to build a

  44269. plugin (somewhere in the installation tree, e.g. under '/usr/local').

  44270. In particular a 'plugin/include' directory is installed, containing all

  44271. the header files needed to build plugins.

  44272. 

  44273.  On most systems, you can query this 'plugin' directory by invoking 'gcc

  44274. -print-file-name=plugin' (replace if needed 'gcc' with the appropriate

  44275. program path).

  44276. 

  44277.  Inside plugins, this 'plugin' directory name can be queried by calling

  44278. 'default_plugin_dir_name ()'.

  44279. 

  44280.  Plugins may know, when they are compiled, the GCC version for which

  44281. 'plugin-version.h' is provided.  The constant macros

  44282. 'GCCPLUGIN_VERSION_MAJOR', 'GCCPLUGIN_VERSION_MINOR',

  44283. 'GCCPLUGIN_VERSION_PATCHLEVEL', 'GCCPLUGIN_VERSION' are integer numbers,

  44284. so a plugin could ensure it is built for GCC 4.7 with

  44285.      #if GCCPLUGIN_VERSION != 4007

  44286.      #error this GCC plugin is for GCC 4.7

  44287.      #endif

  44288. 

  44289.  The following GNU Makefile excerpt shows how to build a simple plugin:

  44290. 

  44291.      HOST_GCC=g++

  44292.      TARGET_GCC=gcc

  44293.      PLUGIN_SOURCE_FILES= plugin1.c plugin2.cc

  44294.      GCCPLUGINS_DIR:= $(shell $(TARGET_GCC) -print-file-name=plugin)

  44295.      CXXFLAGS+= -I$(GCCPLUGINS_DIR)/include -fPIC -fno-rtti -O2

  44296. 

  44297.      plugin.so: $(PLUGIN_SOURCE_FILES)

  44298.         $(HOST_GCC) -shared $(CXXFLAGS) $^ -o $@

  44299. 

  44300.  A single source file plugin may be built with 'g++ -I`gcc

  44301. -print-file-name=plugin`/include -fPIC -shared -fno-rtti -O2 plugin.c -o

  44302. plugin.so', using backquote shell syntax to query the 'plugin'

  44303. directory.

  44304. 

  44305.  Plugin support on Windows/MinGW has a number of limitations and

  44306. additional requirements.  When building a plugin on Windows we have to

  44307. link an import library for the corresponding backend executable, for

  44308. example, 'cc1.exe', 'cc1plus.exe', etc., in order to gain access to the

  44309. symbols provided by GCC. This means that on Windows a plugin is

  44310. language-specific, for example, for C, C++, etc.  If you wish to use

  44311. your plugin with multiple languages, then you will need to build

  44312. multiple plugin libraries and either instruct your users on how to load

  44313. the correct version or provide a compiler wrapper that does this

  44314. automatically.

  44315. 

  44316.  Additionally, on Windows the plugin library has to export the

  44317. 'plugin_is_GPL_compatible' and 'plugin_init' symbols.  If you do not

  44318. wish to modify the source code of your plugin, then you can use the

  44319. '-Wl,--export-all-symbols' option or provide a suitable DEF file.

  44320. Alternatively, you can export just these two symbols by decorating them

  44321. with '__declspec(dllexport)', for example:

  44322. 

  44323.      #ifdef _WIN32

  44324.      __declspec(dllexport)

  44325.      #endif

  44326.      int plugin_is_GPL_compatible;

  44327. 

  44328.      #ifdef _WIN32

  44329.      __declspec(dllexport)

  44330.      #endif

  44331.      int plugin_init (plugin_name_args *, plugin_gcc_version *)

  44332. 

  44333.  The import libraries are installed into the 'plugin' directory and

  44334. their names are derived by appending the '.a' extension to the backend

  44335. executable names, for example, 'cc1.exe.a', 'cc1plus.exe.a', etc.  The

  44336. following command line shows how to build the single source file plugin

  44337. on Windows to be used with the C++ compiler:

  44338. 

  44339.      g++ -I`gcc -print-file-name=plugin`/include -shared -Wl,--export-all-symbols \

  44340.      -o plugin.dll plugin.c `gcc -print-file-name=plugin`/cc1plus.exe.a

  44341. 

  44342.  When a plugin needs to use 'gengtype', be sure that both 'gengtype' and

  44343. 'gtype.state' have the same version as the GCC for which the plugin is

  44344. built.

  44345. 

  44346. 

  44347. File: gccint.info,  Node: LTO,  Next: Match and Simplify,  Prev: Plugins,  Up: Top

  44348. 

  44349. 25 Link Time Optimization

  44350. *************************

  44351. 

  44352. Link Time Optimization (LTO) gives GCC the capability of dumping its

  44353. internal representation (GIMPLE) to disk, so that all the different

  44354. compilation units that make up a single executable can be optimized as a

  44355. single module.  This expands the scope of inter-procedural optimizations

  44356. to encompass the whole program (or, rather, everything that is visible

  44357. at link time).

  44358. 

  44359. * Menu:

  44360. 

  44361. * LTO Overview::            Overview of LTO.

  44362. * LTO object file layout::  LTO file sections in ELF.

  44363. * IPA::                     Using summary information in IPA passes.

  44364. * WHOPR::                   Whole program assumptions,

  44365.                             linker plugin and symbol visibilities.

  44366. * Internal flags::          Internal flags controlling 'lto1'.

  44367. 

  44368. 

  44369. File: gccint.info,  Node: LTO Overview,  Next: LTO object file layout,  Up: LTO

  44370. 

  44371. 25.1 Design Overview

  44372. ====================

  44373. 

  44374. Link time optimization is implemented as a GCC front end for a bytecode

  44375. representation of GIMPLE that is emitted in special sections of '.o'

  44376. files.  Currently, LTO support is enabled in most ELF-based systems, as

  44377. well as darwin, cygwin and mingw systems.

  44378. 

  44379.  Since GIMPLE bytecode is saved alongside final object code, object

  44380. files generated with LTO support are larger than regular object files.

  44381. This "fat" object format makes it easy to integrate LTO into existing

  44382. build systems, as one can, for instance, produce archives of the files.

  44383. Additionally, one might be able to ship one set of fat objects which

  44384. could be used both for development and the production of optimized

  44385. builds.  A, perhaps surprising, side effect of this feature is that any

  44386. mistake in the toolchain leads to LTO information not being used (e.g.

  44387. an older 'libtool' calling 'ld' directly).  This is both an advantage,

  44388. as the system is more robust, and a disadvantage, as the user is not

  44389. informed that the optimization has been disabled.

  44390. 

  44391.  The current implementation only produces "fat" objects, effectively

  44392. doubling compilation time and increasing file sizes up to 5x the

  44393. original size.  This hides the problem that some tools, such as 'ar' and

  44394. 'nm', need to understand symbol tables of LTO sections.  These tools

  44395. were extended to use the plugin infrastructure, and with these problems

  44396. solved, GCC will also support "slim" objects consisting of the

  44397. intermediate code alone.

  44398. 

  44399.  At the highest level, LTO splits the compiler in two.  The first half

  44400. (the "writer") produces a streaming representation of all the internal

  44401. data structures needed to optimize and generate code.  This includes

  44402. declarations, types, the callgraph and the GIMPLE representation of

  44403. function bodies.

  44404. 

  44405.  When '-flto' is given during compilation of a source file, the pass

  44406. manager executes all the passes in 'all_lto_gen_passes'.  Currently,

  44407. this phase is composed of two IPA passes:

  44408. 

  44409.    * 'pass_ipa_lto_gimple_out' This pass executes the function

  44410.      'lto_output' in 'lto-streamer-out.c', which traverses the call

  44411.      graph encoding every reachable declaration, type and function.

  44412.      This generates a memory representation of all the file sections

  44413.      described below.

  44414. 

  44415.    * 'pass_ipa_lto_finish_out' This pass executes the function

  44416.      'produce_asm_for_decls' in 'lto-streamer-out.c', which takes the

  44417.      memory image built in the previous pass and encodes it in the

  44418.      corresponding ELF file sections.

  44419. 

  44420.  The second half of LTO support is the "reader".  This is implemented as

  44421. the GCC front end 'lto1' in 'lto/lto.c'.  When 'collect2' detects a link

  44422. set of '.o'/'.a' files with LTO information and the '-flto' is enabled,

  44423. it invokes 'lto1' which reads the set of files and aggregates them into

  44424. a single translation unit for optimization.  The main entry point for

  44425. the reader is 'lto/lto.c':'lto_main'.

  44426. 

  44427. 25.1.1 LTO modes of operation

  44428. -----------------------------

  44429. 

  44430. One of the main goals of the GCC link-time infrastructure was to allow

  44431. effective compilation of large programs.  For this reason GCC implements

  44432. two link-time compilation modes.

  44433. 

  44434.   1. _LTO mode_, in which the whole program is read into the compiler at

  44435.      link-time and optimized in a similar way as if it were a single

  44436.      source-level compilation unit.

  44437. 

  44438.   2. _WHOPR or partitioned mode_, designed to utilize multiple CPUs

  44439.      and/or a distributed compilation environment to quickly link large

  44440.      applications.  WHOPR stands for WHOle Program optimizeR (not to be

  44441.      confused with the semantics of '-fwhole-program').  It partitions

  44442.      the aggregated callgraph from many different '.o' files and

  44443.      distributes the compilation of the sub-graphs to different CPUs.

  44444. 

  44445.      Note that distributed compilation is not implemented yet, but since

  44446.      the parallelism is facilitated via generating a 'Makefile', it

  44447.      would be easy to implement.

  44448. 

  44449.  WHOPR splits LTO into three main stages:

  44450.   1. Local generation (LGEN) This stage executes in parallel.  Every

  44451.      file in the program is compiled into the intermediate language and

  44452.      packaged together with the local call-graph and summary

  44453.      information.  This stage is the same for both the LTO and WHOPR

  44454.      compilation mode.

  44455. 

  44456.   2. Whole Program Analysis (WPA) WPA is performed sequentially.  The

  44457.      global call-graph is generated, and a global analysis procedure

  44458.      makes transformation decisions.  The global call-graph is

  44459.      partitioned to facilitate parallel optimization during phase 3.

  44460.      The results of the WPA stage are stored into new object files which

  44461.      contain the partitions of program expressed in the intermediate

  44462.      language and the optimization decisions.

  44463. 

  44464.   3. Local transformations (LTRANS) This stage executes in parallel.

  44465.      All the decisions made during phase 2 are implemented locally in

  44466.      each partitioned object file, and the final object code is

  44467.      generated.  Optimizations which cannot be decided efficiently

  44468.      during the phase 2 may be performed on the local call-graph

  44469.      partitions.

  44470. 

  44471.  WHOPR can be seen as an extension of the usual LTO mode of compilation.

  44472. In LTO, WPA and LTRANS are executed within a single execution of the

  44473. compiler, after the whole program has been read into memory.

  44474. 

  44475.  When compiling in WHOPR mode, the callgraph is partitioned during the

  44476. WPA stage.  The whole program is split into a given number of partitions

  44477. of roughly the same size.  The compiler tries to minimize the number of

  44478. references which cross partition boundaries.  The main advantage of

  44479. WHOPR is to allow the parallel execution of LTRANS stages, which are the

  44480. most time-consuming part of the compilation process.  Additionally, it

  44481. avoids the need to load the whole program into memory.

  44482. 

  44483. 

  44484. File: gccint.info,  Node: LTO object file layout,  Next: IPA,  Prev: LTO Overview,  Up: LTO

  44485. 

  44486. 25.2 LTO file sections

  44487. ======================

  44488. 

  44489. LTO information is stored in several ELF sections inside object files.

  44490. Data structures and enum codes for sections are defined in

  44491. 'lto-streamer.h'.

  44492. 

  44493.  These sections are emitted from 'lto-streamer-out.c' and mapped in all

  44494. at once from 'lto/lto.c':'lto_file_read'.  The individual functions

  44495. dealing with the reading/writing of each section are described below.

  44496. 

  44497.    * Command line options ('.gnu.lto_.opts')

  44498. 

  44499.      This section contains the command line options used to generate the

  44500.      object files.  This is used at link time to determine the

  44501.      optimization level and other settings when they are not explicitly

  44502.      specified at the linker command line.

  44503. 

  44504.      Currently, GCC does not support combining LTO object files compiled

  44505.      with different set of the command line options into a single

  44506.      binary.  At link time, the options given on the command line and

  44507.      the options saved on all the files in a link-time set are applied

  44508.      globally.  No attempt is made at validating the combination of

  44509.      flags (other than the usual validation done by option processing).

  44510.      This is implemented in 'lto/lto.c':'lto_read_all_file_options'.

  44511. 

  44512.    * Symbol table ('.gnu.lto_.symtab')

  44513. 

  44514.      This table replaces the ELF symbol table for functions and

  44515.      variables represented in the LTO IL. Symbols used and exported by

  44516.      the optimized assembly code of "fat" objects might not match the

  44517.      ones used and exported by the intermediate code.  This table is

  44518.      necessary because the intermediate code is less optimized and thus

  44519.      requires a separate symbol table.

  44520. 

  44521.      Additionally, the binary code in the "fat" object will lack a call

  44522.      to a function, since the call was optimized out at compilation time

  44523.      after the intermediate language was streamed out.  In some special

  44524.      cases, the same optimization may not happen during link-time

  44525.      optimization.  This would lead to an undefined symbol if only one

  44526.      symbol table was used.

  44527. 

  44528.      The symbol table is emitted in

  44529.      'lto-streamer-out.c':'produce_symtab'.

  44530. 

  44531.    * Global declarations and types ('.gnu.lto_.decls')

  44532. 

  44533.      This section contains an intermediate language dump of all

  44534.      declarations and types required to represent the callgraph, static

  44535.      variables and top-level debug info.

  44536. 

  44537.      The contents of this section are emitted in

  44538.      'lto-streamer-out.c':'produce_asm_for_decls'.  Types and symbols

  44539.      are emitted in a topological order that preserves the sharing of

  44540.      pointers when the file is read back in

  44541.      ('lto.c':'read_cgraph_and_symbols').

  44542. 

  44543.    * The callgraph ('.gnu.lto_.cgraph')

  44544. 

  44545.      This section contains the basic data structure used by the GCC

  44546.      inter-procedural optimization infrastructure.  This section stores

  44547.      an annotated multi-graph which represents the functions and call

  44548.      sites as well as the variables, aliases and top-level 'asm'

  44549.      statements.

  44550. 

  44551.      This section is emitted in 'lto-streamer-out.c':'output_cgraph' and

  44552.      read in 'lto-cgraph.c':'input_cgraph'.

  44553. 

  44554.    * IPA references ('.gnu.lto_.refs')

  44555. 

  44556.      This section contains references between function and static

  44557.      variables.  It is emitted by 'lto-cgraph.c':'output_refs' and read

  44558.      by 'lto-cgraph.c':'input_refs'.

  44559. 

  44560.    * Function bodies ('.gnu.lto_.function_body.<name>')

  44561. 

  44562.      This section contains function bodies in the intermediate language

  44563.      representation.  Every function body is in a separate section to

  44564.      allow copying of the section independently to different object

  44565.      files or reading the function on demand.

  44566. 

  44567.      Functions are emitted in 'lto-streamer-out.c':'output_function' and

  44568.      read in 'lto-streamer-in.c':'input_function'.

  44569. 

  44570.    * Static variable initializers ('.gnu.lto_.vars')

  44571. 

  44572.      This section contains all the symbols in the global variable pool.

  44573.      It is emitted by 'lto-cgraph.c':'output_varpool' and read in

  44574.      'lto-cgraph.c':'input_cgraph'.

  44575. 

  44576.    * Summaries and optimization summaries used by IPA passes

  44577.      ('.gnu.lto_.<xxx>', where '<xxx>' is one of 'jmpfuncs', 'pureconst'

  44578.      or 'reference')

  44579. 

  44580.      These sections are used by IPA passes that need to emit summary

  44581.      information during LTO generation to be read and aggregated at link

  44582.      time.  Each pass is responsible for implementing two pass manager

  44583.      hooks: one for writing the summary and another for reading it in.

  44584.      The format of these sections is entirely up to each individual

  44585.      pass.  The only requirement is that the writer and reader hooks

  44586.      agree on the format.

  44587. 

  44588. 

  44589. File: gccint.info,  Node: IPA,  Next: WHOPR,  Prev: LTO object file layout,  Up: LTO

  44590. 

  44591. 25.3 Using summary information in IPA passes

  44592. ============================================

  44593. 

  44594. Programs are represented internally as a _callgraph_ (a multi-graph

  44595. where nodes are functions and edges are call sites) and a _varpool_ (a

  44596. list of static and external variables in the program).

  44597. 

  44598.  The inter-procedural optimization is organized as a sequence of

  44599. individual passes, which operate on the callgraph and the varpool.  To

  44600. make the implementation of WHOPR possible, every inter-procedural

  44601. optimization pass is split into several stages that are executed at

  44602. different times during WHOPR compilation:

  44603. 

  44604.    * LGEN time

  44605.        1. _Generate summary_ ('generate_summary' in 'struct

  44606.           ipa_opt_pass_d').  This stage analyzes every function body and

  44607.           variable initializer is examined and stores relevant

  44608.           information into a pass-specific data structure.

  44609. 

  44610.        2. _Write summary_ ('write_summary' in 'struct ipa_opt_pass_d').

  44611.           This stage writes all the pass-specific information generated

  44612.           by 'generate_summary'.  Summaries go into their own

  44613.           'LTO_section_*' sections that have to be declared in

  44614.           'lto-streamer.h':'enum lto_section_type'.  A new section is

  44615.           created by calling 'create_output_block' and data can be

  44616.           written using the 'lto_output_*' routines.

  44617. 

  44618.    * WPA time

  44619.        1. _Read summary_ ('read_summary' in 'struct ipa_opt_pass_d').

  44620.           This stage reads all the pass-specific information in exactly

  44621.           the same order that it was written by 'write_summary'.

  44622. 

  44623.        2. _Execute_ ('execute' in 'struct opt_pass').  This performs

  44624.           inter-procedural propagation.  This must be done without

  44625.           actual access to the individual function bodies or variable

  44626.           initializers.  Typically, this results in a transitive closure

  44627.           operation over the summary information of all the nodes in the

  44628.           callgraph.

  44629. 

  44630.        3. _Write optimization summary_ ('write_optimization_summary' in

  44631.           'struct ipa_opt_pass_d').  This writes the result of the

  44632.           inter-procedural propagation into the object file.  This can

  44633.           use the same data structures and helper routines used in

  44634.           'write_summary'.

  44635. 

  44636.    * LTRANS time

  44637.        1. _Read optimization summary_ ('read_optimization_summary' in

  44638.           'struct ipa_opt_pass_d').  The counterpart to

  44639.           'write_optimization_summary'.  This reads the interprocedural

  44640.           optimization decisions in exactly the same format emitted by

  44641.           'write_optimization_summary'.

  44642. 

  44643.        2. _Transform_ ('function_transform' and 'variable_transform' in

  44644.           'struct ipa_opt_pass_d').  The actual function bodies and

  44645.           variable initializers are updated based on the information

  44646.           passed down from the _Execute_ stage.

  44647. 

  44648.  The implementation of the inter-procedural passes are shared between

  44649. LTO, WHOPR and classic non-LTO compilation.

  44650. 

  44651.    * During the traditional file-by-file mode every pass executes its

  44652.      own _Generate summary_, _Execute_, and _Transform_ stages within

  44653.      the single execution context of the compiler.

  44654. 

  44655.    * In LTO compilation mode, every pass uses _Generate summary_ and

  44656.      _Write summary_ stages at compilation time, while the _Read

  44657.      summary_, _Execute_, and _Transform_ stages are executed at link

  44658.      time.

  44659. 

  44660.    * In WHOPR mode all stages are used.

  44661. 

  44662.  To simplify development, the GCC pass manager differentiates between

  44663. normal inter-procedural passes (*note Regular IPA passes::), small

  44664. inter-procedural passes (*note Small IPA passes::) and late

  44665. inter-procedural passes (*note Late IPA passes::).  A small or late IPA

  44666. pass ('SIMPLE_IPA_PASS') does everything at once and thus cannot be

  44667. executed during WPA in WHOPR mode.  It defines only the _Execute_ stage

  44668. and during this stage it accesses and modifies the function bodies.

  44669. Such passes are useful for optimization at LGEN or LTRANS time and are

  44670. used, for example, to implement early optimization before writing object

  44671. files.  The simple inter-procedural passes can also be used for easier

  44672. prototyping and development of a new inter-procedural pass.

  44673. 

  44674. 25.3.1 Virtual clones

  44675. ---------------------

  44676. 

  44677. One of the main challenges of introducing the WHOPR compilation mode was

  44678. addressing the interactions between optimization passes.  In LTO

  44679. compilation mode, the passes are executed in a sequence, each of which

  44680. consists of analysis (or _Generate summary_), propagation (or _Execute_)

  44681. and _Transform_ stages.  Once the work of one pass is finished, the next

  44682. pass sees the updated program representation and can execute.  This

  44683. makes the individual passes dependent on each other.

  44684. 

  44685.  In WHOPR mode all passes first execute their _Generate summary_ stage.

  44686. Then summary writing marks the end of the LGEN stage.  At WPA time, the

  44687. summaries are read back into memory and all passes run the _Execute_

  44688. stage.  Optimization summaries are streamed and sent to LTRANS, where

  44689. all the passes execute the _Transform_ stage.

  44690. 

  44691.  Most optimization passes split naturally into analysis, propagation and

  44692. transformation stages.  But some do not.  The main problem arises when

  44693. one pass performs changes and the following pass gets confused by seeing

  44694. different callgraphs between the _Transform_ stage and the _Generate

  44695. summary_ or _Execute_ stage.  This means that the passes are required to

  44696. communicate their decisions with each other.

  44697. 

  44698.  To facilitate this communication, the GCC callgraph infrastructure

  44699. implements _virtual clones_, a method of representing the changes

  44700. performed by the optimization passes in the callgraph without needing to

  44701. update function bodies.

  44702. 

  44703.  A _virtual clone_ in the callgraph is a function that has no associated

  44704. body, just a description of how to create its body based on a different

  44705. function (which itself may be a virtual clone).

  44706. 

  44707.  The description of function modifications includes adjustments to the

  44708. function's signature (which allows, for example, removing or adding

  44709. function arguments), substitutions to perform on the function body, and,

  44710. for inlined functions, a pointer to the function that it will be inlined

  44711. into.

  44712. 

  44713.  It is also possible to redirect any edge of the callgraph from a

  44714. function to its virtual clone.  This implies updating of the call site

  44715. to adjust for the new function signature.

  44716. 

  44717.  Most of the transformations performed by inter-procedural optimizations

  44718. can be represented via virtual clones.  For instance, a constant

  44719. propagation pass can produce a virtual clone of the function which

  44720. replaces one of its arguments by a constant.  The inliner can represent

  44721. its decisions by producing a clone of a function whose body will be

  44722. later integrated into a given function.

  44723. 

  44724.  Using _virtual clones_, the program can be easily updated during the

  44725. _Execute_ stage, solving most of pass interactions problems that would

  44726. otherwise occur during _Transform_.

  44727. 

  44728.  Virtual clones are later materialized in the LTRANS stage and turned

  44729. into real functions.  Passes executed after the virtual clone were

  44730. introduced also perform their _Transform_ stage on new functions, so for

  44731. a pass there is no significant difference between operating on a real

  44732. function or a virtual clone introduced before its _Execute_ stage.

  44733. 

  44734.  Optimization passes then work on virtual clones introduced before their

  44735. _Execute_ stage as if they were real functions.  The only difference is

  44736. that clones are not visible during the _Generate Summary_ stage.

  44737. 

  44738.  To keep function summaries updated, the callgraph interface allows an

  44739. optimizer to register a callback that is called every time a new clone

  44740. is introduced as well as when the actual function or variable is

  44741. generated or when a function or variable is removed.  These hooks are

  44742. registered in the _Generate summary_ stage and allow the pass to keep

  44743. its information intact until the _Execute_ stage.  The same hooks can

  44744. also be registered during the _Execute_ stage to keep the optimization

  44745. summaries updated for the _Transform_ stage.

  44746. 

  44747. 25.3.2 IPA references

  44748. ---------------------

  44749. 

  44750. GCC represents IPA references in the callgraph.  For a function or

  44751. variable 'A', the _IPA reference_ is a list of all locations where the

  44752. address of 'A' is taken and, when 'A' is a variable, a list of all

  44753. direct stores and reads to/from 'A'.  References represent an oriented

  44754. multi-graph on the union of nodes of the callgraph and the varpool.  See

  44755. 'ipa-reference.c':'ipa_reference_write_optimization_summary' and

  44756. 'ipa-reference.c':'ipa_reference_read_optimization_summary' for details.

  44757. 

  44758. 25.3.3 Jump functions

  44759. ---------------------

  44760. 

  44761. Suppose that an optimization pass sees a function 'A' and it knows the

  44762. values of (some of) its arguments.  The _jump function_ describes the

  44763. value of a parameter of a given function call in function 'A' based on

  44764. this knowledge.

  44765. 

  44766.  Jump functions are used by several optimizations, such as the

  44767. inter-procedural constant propagation pass and the devirtualization

  44768. pass.  The inliner also uses jump functions to perform inlining of

  44769. callbacks.

  44770. 

  44771. 

  44772. File: gccint.info,  Node: WHOPR,  Next: Internal flags,  Prev: IPA,  Up: LTO

  44773. 

  44774. 25.4 Whole program assumptions, linker plugin and symbol visibilities

  44775. =====================================================================

  44776. 

  44777. Link-time optimization gives relatively minor benefits when used alone.

  44778. The problem is that propagation of inter-procedural information does not

  44779. work well across functions and variables that are called or referenced

  44780. by other compilation units (such as from a dynamically linked library).

  44781. We say that such functions and variables are _externally visible_.

  44782. 

  44783.  To make the situation even more difficult, many applications organize

  44784. themselves as a set of shared libraries, and the default ELF visibility

  44785. rules allow one to overwrite any externally visible symbol with a

  44786. different symbol at runtime.  This basically disables any optimizations

  44787. across such functions and variables, because the compiler cannot be sure

  44788. that the function body it is seeing is the same function body that will

  44789. be used at runtime.  Any function or variable not declared 'static' in

  44790. the sources degrades the quality of inter-procedural optimization.

  44791. 

  44792.  To avoid this problem the compiler must assume that it sees the whole

  44793. program when doing link-time optimization.  Strictly speaking, the whole

  44794. program is rarely visible even at link-time.  Standard system libraries

  44795. are usually linked dynamically or not provided with the link-time

  44796. information.  In GCC, the whole program option ('-fwhole-program')

  44797. asserts that every function and variable defined in the current

  44798. compilation unit is static, except for function 'main' (note: at link

  44799. time, the current unit is the union of all objects compiled with LTO).

  44800. Since some functions and variables need to be referenced externally, for

  44801. example by another DSO or from an assembler file, GCC also provides the

  44802. function and variable attribute 'externally_visible' which can be used

  44803. to disable the effect of '-fwhole-program' on a specific symbol.

  44804. 

  44805.  The whole program mode assumptions are slightly more complex in C++,

  44806. where inline functions in headers are put into _COMDAT_ sections.

  44807. COMDAT function and variables can be defined by multiple object files

  44808. and their bodies are unified at link-time and dynamic link-time.  COMDAT

  44809. functions are changed to local only when their address is not taken and

  44810. thus un-sharing them with a library is not harmful.  COMDAT variables

  44811. always remain externally visible, however for readonly variables it is

  44812. assumed that their initializers cannot be overwritten by a different

  44813. value.

  44814. 

  44815.  GCC provides the function and variable attribute 'visibility' that can

  44816. be used to specify the visibility of externally visible symbols (or

  44817. alternatively an '-fdefault-visibility' command line option).  ELF

  44818. defines the 'default', 'protected', 'hidden' and 'internal'

  44819. visibilities.

  44820. 

  44821.  The most commonly used is visibility is 'hidden'.  It specifies that

  44822. the symbol cannot be referenced from outside of the current shared

  44823. library.  Unfortunately, this information cannot be used directly by the

  44824. link-time optimization in the compiler since the whole shared library

  44825. also might contain non-LTO objects and those are not visible to the

  44826. compiler.

  44827. 

  44828.  GCC solves this problem using linker plugins.  A _linker plugin_ is an

  44829. interface to the linker that allows an external program to claim the

  44830. ownership of a given object file.  The linker then performs the linking

  44831. procedure by querying the plugin about the symbol table of the claimed

  44832. objects and once the linking decisions are complete, the plugin is

  44833. allowed to provide the final object file before the actual linking is

  44834. made.  The linker plugin obtains the symbol resolution information which

  44835. specifies which symbols provided by the claimed objects are bound from

  44836. the rest of a binary being linked.

  44837. 

  44838.  GCC is designed to be independent of the rest of the toolchain and aims

  44839. to support linkers without plugin support.  For this reason it does not

  44840. use the linker plugin by default.  Instead, the object files are

  44841. examined by 'collect2' before being passed to the linker and objects

  44842. found to have LTO sections are passed to 'lto1' first.  This mode does

  44843. not work for library archives.  The decision on what object files from

  44844. the archive are needed depends on the actual linking and thus GCC would

  44845. have to implement the linker itself.  The resolution information is

  44846. missing too and thus GCC needs to make an educated guess based on

  44847. '-fwhole-program'.  Without the linker plugin GCC also assumes that

  44848. symbols are declared 'hidden' and not referred by non-LTO code by

  44849. default.

  44850. 

  44851. 

  44852. File: gccint.info,  Node: Internal flags,  Prev: WHOPR,  Up: LTO

  44853. 

  44854. 25.5 Internal flags controlling 'lto1'

  44855. ======================================

  44856. 

  44857. The following flags are passed into 'lto1' and are not meant to be used

  44858. directly from the command line.

  44859. 

  44860.    * -fwpa This option runs the serial part of the link-time optimizer

  44861.      performing the inter-procedural propagation (WPA mode).  The

  44862.      compiler reads in summary information from all inputs and performs

  44863.      an analysis based on summary information only.  It generates object

  44864.      files for subsequent runs of the link-time optimizer where

  44865.      individual object files are optimized using both summary

  44866.      information from the WPA mode and the actual function bodies.  It

  44867.      then drives the LTRANS phase.

  44868. 

  44869.    * -fltrans This option runs the link-time optimizer in the

  44870.      local-transformation (LTRANS) mode, which reads in output from a

  44871.      previous run of the LTO in WPA mode.  In the LTRANS mode, LTO

  44872.      optimizes an object and produces the final assembly.

  44873. 

  44874.    * -fltrans-output-list=FILE This option specifies a file to which the

  44875.      names of LTRANS output files are written.  This option is only

  44876.      meaningful in conjunction with '-fwpa'.

  44877. 

  44878.    * -fresolution=FILE This option specifies the linker resolution file.

  44879.      This option is only meaningful in conjunction with '-fwpa' and as

  44880.      option to pass through to the LTO linker plugin.

  44881. 

  44882. 

  44883. File: gccint.info,  Node: Match and Simplify,  Next: Static Analyzer,  Prev: LTO,  Up: Top

  44884. 

  44885. 26 Match and Simplify

  44886. *********************

  44887. 

  44888. The GIMPLE and GENERIC pattern matching project match-and-simplify tries

  44889. to address several issues.

  44890. 

  44891.   1. unify expression simplifications currently spread and duplicated

  44892.      over separate files like fold-const.c, gimple-fold.c and builtins.c

  44893.   2. allow for a cheap way to implement building and simplifying

  44894.      non-trivial GIMPLE expressions, avoiding the need to go through

  44895.      building and simplifying GENERIC via fold_buildN and then

  44896.      gimplifying via force_gimple_operand

  44897. 

  44898.  To address these the project introduces a simple domain specific

  44899. language to write expression simplifications from which code targeting

  44900. GIMPLE and GENERIC is auto-generated.  The GENERIC variant follows the

  44901. fold_buildN API while for the GIMPLE variant and to address 2) new APIs

  44902. are introduced.

  44903. 

  44904. * Menu:

  44905. 

  44906. * GIMPLE API::

  44907. * The Language::

  44908. 

  44909. 

  44910. File: gccint.info,  Node: GIMPLE API,  Next: The Language,  Up: Match and Simplify

  44911. 

  44912. 26.1 GIMPLE API

  44913. ===============

  44914. 

  44915.  -- GIMPLE function: tree gimple_simplify (enum tree_code, tree, tree,

  44916.           gimple_seq *, tree (*)(tree))

  44917.  -- GIMPLE function: tree gimple_simplify (enum tree_code, tree, tree,

  44918.           tree, gimple_seq *, tree (*)(tree))

  44919.  -- GIMPLE function: tree gimple_simplify (enum tree_code, tree, tree,

  44920.           tree, tree, gimple_seq *, tree (*)(tree))

  44921.  -- GIMPLE function: tree gimple_simplify (enum built_in_function, tree,

  44922.           tree, gimple_seq *, tree (*)(tree))

  44923.  -- GIMPLE function: tree gimple_simplify (enum built_in_function, tree,

  44924.           tree, tree, gimple_seq *, tree (*)(tree))

  44925.  -- GIMPLE function: tree gimple_simplify (enum built_in_function, tree,

  44926.           tree, tree, tree, gimple_seq *, tree (*)(tree))

  44927.      The main GIMPLE API entry to the expression simplifications

  44928.      mimicing that of the GENERIC fold_{unary,binary,ternary} functions.

  44929. 

  44930.  thus providing n-ary overloads for operation or function.  The

  44931. additional arguments are a gimple_seq where built statements are

  44932. inserted on (if 'NULL' then simplifications requiring new statements are

  44933. not performed) and a valueization hook that can be used to tie

  44934. simplifications to a SSA lattice.

  44935. 

  44936.  In addition to those APIs 'fold_stmt' is overloaded with a valueization

  44937. hook:

  44938. 

  44939.  -- bool: fold_stmt (gimple_stmt_iterator *, tree (*)(tree));

  44940. 

  44941.  Ontop of these a 'fold_buildN'-like API for GIMPLE is introduced:

  44942. 

  44943.  -- GIMPLE function: tree gimple_build (gimple_seq *, location_t, enum

  44944.           tree_code, tree, tree, tree (*valueize) (tree) = NULL);

  44945.  -- GIMPLE function: tree gimple_build (gimple_seq *, location_t, enum

  44946.           tree_code, tree, tree, tree, tree (*valueize) (tree) = NULL);

  44947.  -- GIMPLE function: tree gimple_build (gimple_seq *, location_t, enum

  44948.           tree_code, tree, tree, tree, tree, tree (*valueize) (tree) =

  44949.           NULL);

  44950.  -- GIMPLE function: tree gimple_build (gimple_seq *, location_t, enum

  44951.           built_in_function, tree, tree, tree (*valueize) (tree) =

  44952.           NULL);

  44953.  -- GIMPLE function: tree gimple_build (gimple_seq *, location_t, enum

  44954.           built_in_function, tree, tree, tree, tree (*valueize) (tree) =

  44955.           NULL);

  44956.  -- GIMPLE function: tree gimple_build (gimple_seq *, location_t, enum

  44957.           built_in_function, tree, tree, tree, tree, tree (*valueize)

  44958.           (tree) = NULL);

  44959.  -- GIMPLE function: tree gimple_convert (gimple_seq *, location_t,

  44960.           tree, tree);

  44961. 

  44962.  which is supposed to replace 'force_gimple_operand (fold_buildN (...),

  44963. ...)' and calls to 'fold_convert'.  Overloads without the 'location_t'

  44964. argument exist.  Built statements are inserted on the provided sequence

  44965. and simplification is performed using the optional valueization hook.

  44966. 

  44967. 

  44968. File: gccint.info,  Node: The Language,  Prev: GIMPLE API,  Up: Match and Simplify

  44969. 

  44970. 26.2 The Language

  44971. =================

  44972. 

  44973. The language to write expression simplifications in resembles other

  44974. domain-specific languages GCC uses.  Thus it is lispy.  Lets start with

  44975. an example from the match.pd file:

  44976. 

  44977.      (simplify

  44978.        (bit_and @0 integer_all_onesp)

  44979.        @0)

  44980. 

  44981.  This example contains all required parts of an expression

  44982. simplification.  A simplification is wrapped inside a '(simplify ...)'

  44983. expression.  That contains at least two operands - an expression that is

  44984. matched with the GIMPLE or GENERIC IL and a replacement expression that

  44985. is returned if the match was successful.

  44986. 

  44987.  Expressions have an operator ID, 'bit_and' in this case.  Expressions

  44988. can be lower-case tree codes with '_expr' stripped off or builtin

  44989. function code names in all-caps, like 'BUILT_IN_SQRT'.

  44990. 

  44991.  '@n' denotes a so-called capture.  It captures the operand and lets you

  44992. refer to it in other places of the match-and-simplify.  In the above

  44993. example it is refered to in the replacement expression.  Captures are

  44994. '@' followed by a number or an identifier.

  44995. 

  44996.      (simplify

  44997.        (bit_xor @0 @0)

  44998.        { build_zero_cst (type); })

  44999. 

  45000.  In this example '@0' is mentioned twice which constrains the matched

  45001. expression to have two equal operands.  Usually matches are constraint

  45002. to equal types.  If operands may be constants and conversions are

  45003. involved matching by value might be preferred in which case use '@@0' to

  45004. denote a by value match and the specific operand you want to refer to in

  45005. the result part.  This example also introduces operands written in C

  45006. code.  These can be used in the expression replacements and are supposed

  45007. to evaluate to a tree node which has to be a valid GIMPLE operand (so

  45008. you cannot generate expressions in C code).

  45009. 

  45010.      (simplify

  45011.        (trunc_mod integer_zerop@0 @1)

  45012.        (if (!integer_zerop (@1))

  45013.         @0))

  45014. 

  45015.  Here '@0' captures the first operand of the trunc_mod expression which

  45016. is also predicated with 'integer_zerop'.  Expression operands may be

  45017. either expressions, predicates or captures.  Captures can be

  45018. unconstrained or capture expresions or predicates.

  45019. 

  45020.  This example introduces an optional operand of simplify, the

  45021. if-expression.  This condition is evaluated after the expression matched

  45022. in the IL and is required to evaluate to true to enable the replacement

  45023. expression in the second operand position.  The expression operand of

  45024. the 'if' is a standard C expression which may contain references to

  45025. captures.  The 'if' has an optional third operand which may contain the

  45026. replacement expression that is enabled when the condition evaluates to

  45027. false.

  45028. 

  45029.  A 'if' expression can be used to specify a common condition for

  45030. multiple simplify patterns, avoiding the need to repeat that multiple

  45031. times:

  45032. 

  45033.      (if (!TYPE_SATURATING (type)

  45034.           && !FLOAT_TYPE_P (type) && !FIXED_POINT_TYPE_P (type))

  45035.        (simplify

  45036.          (minus (plus @0 @1) @0)

  45037.          @1)

  45038.        (simplify

  45039.          (minus (minus @0 @1) @0)

  45040.          (negate @1)))

  45041. 

  45042.  Note that 'if's in outer position do not have the optional else clause

  45043. but instead have multiple then clauses.

  45044. 

  45045.  Ifs can be nested.

  45046. 

  45047.  There exists a 'switch' expression which can be used to chain

  45048. conditions avoiding nesting 'if's too much:

  45049. 

  45050.      (simplify

  45051.       (simple_comparison @0 REAL_CST@1)

  45052.       (switch

  45053.        /* a CMP (-0) -> a CMP 0  */

  45054.        (if (REAL_VALUE_MINUS_ZERO (TREE_REAL_CST (@1)))

  45055.         (cmp @0 { build_real (TREE_TYPE (@1), dconst0); }))

  45056.        /* x != NaN is always true, other ops are always false.  */

  45057.        (if (REAL_VALUE_ISNAN (TREE_REAL_CST (@1))

  45058.             && ! HONOR_SNANS (@1))

  45059.         { constant_boolean_node (cmp == NE_EXPR, type); })))

  45060. 

  45061.  Is equal to

  45062. 

  45063.      (simplify

  45064.       (simple_comparison @0 REAL_CST@1)

  45065.       (switch

  45066.        /* a CMP (-0) -> a CMP 0  */

  45067.        (if (REAL_VALUE_MINUS_ZERO (TREE_REAL_CST (@1)))

  45068.         (cmp @0 { build_real (TREE_TYPE (@1), dconst0); })

  45069.         /* x != NaN is always true, other ops are always false.  */

  45070.         (if (REAL_VALUE_ISNAN (TREE_REAL_CST (@1))

  45071.              && ! HONOR_SNANS (@1))

  45072.          { constant_boolean_node (cmp == NE_EXPR, type); }))))

  45073. 

  45074.  which has the second 'if' in the else operand of the first.  The

  45075. 'switch' expression takes 'if' expressions as operands (which may not

  45076. have else clauses) and as a last operand a replacement expression which

  45077. should be enabled by default if no other condition evaluated to true.

  45078. 

  45079.  Captures can also be used for capturing results of sub-expressions.

  45080. 

  45081.      #if GIMPLE

  45082.      (simplify

  45083.        (pointer_plus (addr@2 @0) INTEGER_CST_P@1)

  45084.        (if (is_gimple_min_invariant (@2)))

  45085.        {

  45086.          poly_int64 off;

  45087.          tree base = get_addr_base_and_unit_offset (@0, &off);

  45088.          off += tree_to_uhwi (@1);

  45089.          /* Now with that we should be able to simply write

  45090.             (addr (mem_ref (addr @base) (plus @off @1)))  */

  45091.          build1 (ADDR_EXPR, type,

  45092.                  build2 (MEM_REF, TREE_TYPE (TREE_TYPE (@2)),

  45093.                          build_fold_addr_expr (base),

  45094.                          build_int_cst (ptr_type_node, off)));

  45095.        })

  45096.      #endif

  45097. 

  45098.  In the above example, '@2' captures the result of the expression '(addr

  45099. @0)'.  For outermost expression only its type can be captured, and the

  45100. keyword 'type' is reserved for this purpose.  The above example also

  45101. gives a way to conditionalize patterns to only apply to 'GIMPLE' or

  45102. 'GENERIC' by means of using the pre-defined preprocessor macros 'GIMPLE'

  45103. and 'GENERIC' and using preprocessor directives.

  45104. 

  45105.      (simplify

  45106.        (bit_and:c integral_op_p@0 (bit_ior:c (bit_not @0) @1))

  45107.        (bit_and @1 @0))

  45108. 

  45109.  Here we introduce flags on match expressions.  The flag used above,

  45110. 'c', denotes that the expression should be also matched commutated.

  45111. Thus the above match expression is really the following four match

  45112. expressions:

  45113. 

  45114.        (bit_and integral_op_p@0 (bit_ior (bit_not @0) @1))

  45115.        (bit_and (bit_ior (bit_not @0) @1) integral_op_p@0)

  45116.        (bit_and integral_op_p@0 (bit_ior @1 (bit_not @0)))

  45117.        (bit_and (bit_ior @1 (bit_not @0)) integral_op_p@0)

  45118. 

  45119.  Usual canonicalizations you know from GENERIC expressions are applied

  45120. before matching, so for example constant operands always come second in

  45121. commutative expressions.

  45122. 

  45123.  The second supported flag is 's' which tells the code generator to fail

  45124. the pattern if the expression marked with 's' does have more than one

  45125. use and the simplification results in an expression with more than one

  45126. operator.  For example in

  45127. 

  45128.      (simplify

  45129.        (pointer_plus (pointer_plus:s @0 @1) @3)

  45130.        (pointer_plus @0 (plus @1 @3)))

  45131. 

  45132.  this avoids the association if '(pointer_plus @0 @1)' is used outside

  45133. of the matched expression and thus it would stay live and not trivially

  45134. removed by dead code elimination.  Now consider '((x + 3) + -3)' with

  45135. the temporary holding '(x + 3)' used elsewhere.  This simplifies down to

  45136. 'x' which is desirable and thus flagging with 's' does not prevent the

  45137. transform.  Now consider '((x + 3) + 1)' which simplifies to '(x + 4)'.

  45138. Despite being flagged with 's' the simplification will be performed.

  45139. The simplification of '((x + a) + 1)' to '(x + (a + 1))' will not

  45140. performed in this case though.

  45141. 

  45142.  More features exist to avoid too much repetition.

  45143. 

  45144.      (for op (plus pointer_plus minus bit_ior bit_xor)

  45145.        (simplify

  45146.          (op @0 integer_zerop)

  45147.          @0))

  45148. 

  45149.  A 'for' expression can be used to repeat a pattern for each operator

  45150. specified, substituting 'op'.  'for' can be nested and a 'for' can have

  45151. multiple operators to iterate.

  45152. 

  45153.      (for opa (plus minus)

  45154.           opb (minus plus)

  45155.        (for opc (plus minus)

  45156.          (simplify...

  45157. 

  45158.  In this example the pattern will be repeated four times with 'opa, opb,

  45159. opc' being 'plus, minus, plus'; 'plus, minus, minus'; 'minus, plus,

  45160. plus'; 'minus, plus, minus'.

  45161. 

  45162.  To avoid repeating operator lists in 'for' you can name them via

  45163. 

  45164.      (define_operator_list pmm plus minus mult)

  45165. 

  45166.  and use them in 'for' operator lists where they get expanded.

  45167. 

  45168.      (for opa (pmm trunc_div)

  45169.       (simplify...

  45170. 

  45171.  So this example iterates over 'plus', 'minus', 'mult' and 'trunc_div'.

  45172. 

  45173.  Using operator lists can also remove the need to explicitely write a

  45174. 'for'.  All operator list uses that appear in a 'simplify' or 'match'

  45175. pattern in operator positions will implicitely be added to a new 'for'.

  45176. For example

  45177. 

  45178.      (define_operator_list SQRT BUILT_IN_SQRTF BUILT_IN_SQRT BUILT_IN_SQRTL)

  45179.      (define_operator_list POW BUILT_IN_POWF BUILT_IN_POW BUILT_IN_POWL)

  45180.      (simplify

  45181.       (SQRT (POW @0 @1))

  45182.       (POW (abs @0) (mult @1 { built_real (TREE_TYPE (@1), dconsthalf); })))

  45183. 

  45184.  is the same as

  45185. 

  45186.      (for SQRT (BUILT_IN_SQRTF BUILT_IN_SQRT BUILT_IN_SQRTL)

  45187.           POW (BUILT_IN_POWF BUILT_IN_POW BUILT_IN_POWL)

  45188.       (simplify

  45189.        (SQRT (POW @0 @1))

  45190.        (POW (abs @0) (mult @1 { built_real (TREE_TYPE (@1), dconsthalf); }))))

  45191. 

  45192.  'for's and operator lists can include the special identifier 'null'

  45193. that matches nothing and can never be generated.  This can be used to

  45194. pad an operator list so that it has a standard form, even if there isn't

  45195. a suitable operator for every form.

  45196. 

  45197.  Another building block are 'with' expressions in the result expression

  45198. which nest the generated code in a new C block followed by its argument:

  45199. 

  45200.      (simplify

  45201.       (convert (mult @0 @1))

  45202.       (with { tree utype = unsigned_type_for (type); }

  45203.        (convert (mult (convert:utype @0) (convert:utype @1)))))

  45204. 

  45205.  This allows code nested in the 'with' to refer to the declared

  45206. variables.  In the above case we use the feature to specify the type of

  45207. a generated expression with the ':type' syntax where 'type' needs to be

  45208. an identifier that refers to the desired type.  Usually the types of the

  45209. generated result expressions are determined from the context, but

  45210. sometimes like in the above case it is required that you specify them

  45211. explicitely.

  45212. 

  45213.  As intermediate conversions are often optional there is a way to avoid

  45214. the need to repeat patterns both with and without such conversions.

  45215. Namely you can mark a conversion as being optional with a '?':

  45216. 

  45217.      (simplify

  45218.       (eq (convert@0 @1) (convert? @2))

  45219.       (eq @1 (convert @2)))

  45220. 

  45221.  which will match both '(eq (convert @1) (convert @2))' and '(eq

  45222. (convert @1) @2)'.  The optional converts are supposed to be all either

  45223. present or not, thus '(eq (convert? @1) (convert? @2))' will result in

  45224. two patterns only.  If you want to match all four combinations you have

  45225. access to two additional conditional converts as in '(eq (convert1? @1)

  45226. (convert2? @2))'.

  45227. 

  45228.  The support for '?' marking extends to all unary operations including

  45229. predicates you declare yourself with 'match'.

  45230. 

  45231.  Predicates available from the GCC middle-end need to be made available

  45232. explicitely via 'define_predicates':

  45233. 

  45234.      (define_predicates

  45235.       integer_onep integer_zerop integer_all_onesp)

  45236. 

  45237.  You can also define predicates using the pattern matching language and

  45238. the 'match' form:

  45239. 

  45240.      (match negate_expr_p

  45241.       INTEGER_CST

  45242.       (if (TYPE_OVERFLOW_WRAPS (type)

  45243.            || may_negate_without_overflow_p (t))))

  45244.      (match negate_expr_p

  45245.       (negate @0))

  45246. 

  45247.  This shows that for 'match' expressions there is 't' available which

  45248. captures the outermost expression (something not possible in the

  45249. 'simplify' context).  As you can see 'match' has an identifier as first

  45250. operand which is how you refer to the predicate in patterns.  Multiple

  45251. 'match' for the same identifier add additional cases where the predicate

  45252. matches.

  45253. 

  45254.  Predicates can also match an expression in which case you need to

  45255. provide a template specifying the identifier and where to get its

  45256. operands from:

  45257. 

  45258.      (match (logical_inverted_value @0)

  45259.       (eq @0 integer_zerop))

  45260.      (match (logical_inverted_value @0)

  45261.       (bit_not truth_valued_p@0))

  45262. 

  45263.  You can use the above predicate like

  45264. 

  45265.      (simplify

  45266.       (bit_and @0 (logical_inverted_value @0))

  45267.       { build_zero_cst (type); })

  45268. 

  45269.  Which will match a bitwise and of an operand with its logical inverted

  45270. value.

  45271. 

  45272. 

  45273. File: gccint.info,  Node: Static Analyzer,  Next: User Experience Guidelines,  Prev: Match and Simplify,  Up: Top

  45274. 

  45275. 27 Static Analyzer

  45276. ******************

  45277. 

  45278. * Menu:

  45279. 

  45280. * Analyzer Internals::       Analyzer Internals

  45281. * Debugging the Analyzer::   Useful debugging tips

  45282. 

  45283. 

  45284. File: gccint.info,  Node: Analyzer Internals,  Next: Debugging the Analyzer,  Up: Static Analyzer

  45285. 

  45286. 27.1 Analyzer Internals

  45287. =======================

  45288. 

  45289. 27.1.1 Overview

  45290. ---------------

  45291. 

  45292. The analyzer implementation works on the gimple-SSA representation.  (I

  45293. chose this in the hopes of making it easy to work with LTO to do

  45294. whole-program analysis).

  45295. 

  45296.  The implementation is read-only: it doesn't attempt to change anything,

  45297. just emit warnings.

  45298. 

  45299.  The gimple representation can be seen using '-fdump-ipa-analyzer'.

  45300. 

  45301.  First, we build a 'supergraph' which combines the callgraph and all of

  45302. the CFGs into a single directed graph, with both interprocedural and

  45303. intraprocedural edges.  The nodes and edges in the supergraph are called

  45304. "supernodes" and "superedges", and often referred to in code as 'snodes'

  45305. and 'sedges'.  Basic blocks in the CFGs are split at interprocedural

  45306. calls, so there can be more than one supernode per basic block.  Most

  45307. statements will be in just one supernode, but a call statement can

  45308. appear in two supernodes: at the end of one for the call, and again at

  45309. the start of another for the return.

  45310. 

  45311.  The supergraph can be seen using '-fdump-analyzer-supergraph'.

  45312. 

  45313.  We then build an 'analysis_plan' which walks the callgraph to determine

  45314. which calls might be suitable for being summarized (rather than fully

  45315. explored) and thus in what order to explore the functions.

  45316. 

  45317.  Next is the heart of the analyzer: we use a worklist to explore state

  45318. within the supergraph, building an "exploded graph".  Nodes in the

  45319. exploded graph correspond to <point, state> pairs, as in "Precise

  45320. Interprocedural Dataflow Analysis via Graph Reachability" (Thomas Reps,

  45321. Susan Horwitz and Mooly Sagiv).

  45322. 

  45323.  We reuse nodes for <point, state> pairs we've already seen, and avoid

  45324. tracking state too closely, so that (hopefully) we rapidly converge on a

  45325. final exploded graph, and terminate the analysis.  We also bail out if

  45326. the number of exploded <end-of-basic-block, state> nodes gets larger

  45327. than a particular multiple of the total number of basic blocks (to

  45328. ensure termination in the face of pathological state-explosion cases, or

  45329. bugs).  We also stop exploring a point once we hit a limit of states for

  45330. that point.

  45331. 

  45332.  We can identify problems directly when processing a <point, state>

  45333. instance.  For example, if we're finding the successors of

  45334. 

  45335.         <point: before-stmt: "free (ptr);",

  45336.          state: {"ptr": freed}>

  45337. 

  45338.  then we can detect a double-free of "ptr".  We can then emit a path to

  45339. reach the problem by finding the simplest route through the graph.

  45340. 

  45341.  Program points in the analysis are much more fine-grained than in the

  45342. CFG and supergraph, with points (and thus potentially exploded nodes)

  45343. for various events, including before individual statements.  By default

  45344. the exploded graph merges multiple consecutive statements in a supernode

  45345. into one exploded edge to minimize the size of the exploded graph.  This

  45346. can be suppressed via '-fanalyzer-fine-grained'.  The fine-grained

  45347. approach seems to make things simpler and more debuggable that other

  45348. approaches I tried, in that each point is responsible for one thing.

  45349. 

  45350.  Program points in the analysis also have a "call string" identifying

  45351. the stack of callsites below them, so that paths in the exploded graph

  45352. correspond to interprocedurally valid paths: we always return to the

  45353. correct call site, propagating state information accordingly.  We avoid

  45354. infinite recursion by stopping the analysis if a callsite appears more

  45355. than 'analyzer-max-recursion-depth' in a callstring (defaulting to 2).

  45356. 

  45357. 27.1.2 Graphs

  45358. -------------

  45359. 

  45360. Nodes and edges in the exploded graph are called "exploded nodes" and

  45361. "exploded edges" and often referred to in the code as 'enodes' and

  45362. 'eedges' (especially when distinguishing them from the 'snodes' and

  45363. 'sedges' in the supergraph).

  45364. 

  45365.  Each graph numbers its nodes, giving unique identifiers - supernodes

  45366. are referred to throughout dumps in the form 'SN': INDEX' and exploded

  45367. nodes in the form 'EN: INDEX' (e.g.  'SN: 2' and 'EN:29').

  45368. 

  45369.  The supergraph can be seen using '-fdump-analyzer-supergraph-graph'.

  45370. 

  45371.  The exploded graph can be seen using '-fdump-analyzer-exploded-graph'

  45372. and other dump options.  Exploded nodes are color-coded in the .dot

  45373. output based on state-machine states to make it easier to see state

  45374. changes at a glance.

  45375. 

  45376. 27.1.3 State Tracking

  45377. ---------------------

  45378. 

  45379. There's a tension between:

  45380.    * precision of analysis in the straight-line case, vs

  45381.    * exponential blow-up in the face of control flow.

  45382. 

  45383.  For example, in general, given this CFG:

  45384. 

  45385.            A

  45386.           / \

  45387.          B   C

  45388.           \ /

  45389.            D

  45390.           / \

  45391.          E   F

  45392.           \ /

  45393.            G

  45394. 

  45395.  we want to avoid differences in state-tracking in B and C from leading

  45396. to blow-up.  If we don't prevent state blowup, we end up with

  45397. exponential growth of the exploded graph like this:

  45398. 

  45399. 

  45400.                 1:A

  45401.                /   \

  45402.               /     \

  45403.              /       \

  45404.            2:B       3:C

  45405.             |         |

  45406.            4:D       5:D        (2 exploded nodes for D)

  45407.           /   \     /   \

  45408.         6:E   7:F 8:E   9:F

  45409.          |     |   |     |

  45410.         10:G 11:G 12:G  13:G    (4 exploded nodes for G)

  45411. 

  45412. 

  45413.  Similar issues arise with loops.

  45414. 

  45415.  To prevent this, we follow various approaches:

  45416. 

  45417.   a. state pruning: which tries to discard state that won't be relevant

  45418.      later on withing the function.  This can be disabled via

  45419.      '-fno-analyzer-state-purge'.

  45420. 

  45421.   b. state merging.  We can try to find the commonality between two

  45422.      program_state instances to make a third, simpler program_state.  We

  45423.      have two strategies here:

  45424. 

  45425.        1. the worklist keeps new nodes for the same program_point

  45426.           together, and tries to merge them before processing, and thus

  45427.           before they have successors.  Hence, in the above, the two

  45428.           nodes for D (4 and 5) reach the front of the worklist

  45429.           together, and we create a node for D with the merger of the

  45430.           incoming states.

  45431. 

  45432.        2. try merging with the state of existing enodes for the

  45433.           program_point (which may have already been explored).  There

  45434.           will be duplication, but only one set of duplication;

  45435.           subsequent duplicates are more likely to hit the cache.  In

  45436.           particular, (hopefully) all merger chains are finite, and so

  45437.           we guarantee termination.  This is intended to help with

  45438.           loops: we ought to explore the first iteration, and then have

  45439.           a "subsequent iterations" exploration, which uses a state

  45440.           merged from that of the first, to be more abstract.

  45441. 

  45442.      We avoid merging pairs of states that have state-machine

  45443.      differences, as these are the kinds of differences that are likely

  45444.      to be most interesting.  So, for example, given:

  45445. 

  45446.                 if (condition)

  45447.                   ptr = malloc (size);

  45448.                 else

  45449.                   ptr = local_buf;

  45450. 

  45451.                 .... do things with 'ptr'

  45452. 

  45453.                 if (condition)

  45454.                   free (ptr);

  45455. 

  45456.                 ...etc

  45457. 

  45458.      then we end up with an exploded graph that looks like this:

  45459. 

  45460. 

  45461.                              if (condition)

  45462.                                / T      \ F

  45463.                       ---------          ----------

  45464.                      /                             \

  45465.                 ptr = malloc (size)             ptr = local_buf

  45466.                     |                               |

  45467.                 copy of                         copy of

  45468.                   "do things with 'ptr'"          "do things with 'ptr'"

  45469.                 with ptr: heap-allocated        with ptr: stack-allocated

  45470.                     |                               |

  45471.                 if (condition)                  if (condition)

  45472.                     | known to be T                 | known to be F

  45473.                 free (ptr);                         |

  45474.                      \                             /

  45475.                       -----------------------------

  45476.                                    | ('ptr' is pruned, so states can be merged)

  45477.                                   etc

  45478. 

  45479. 

  45480.      where some duplication has occurred, but only for the places where

  45481.      the the different paths are worth exploringly separately.

  45482. 

  45483.      Merging can be disabled via '-fno-analyzer-state-merge'.

  45484. 

  45485. 27.1.4 Region Model

  45486. -------------------

  45487. 

  45488. Part of the state stored at a 'exploded_node' is a 'region_model'.  This

  45489. is an implementation of the region-based ternary model described in "A

  45490. Memory Model for Static Analysis of C Programs"

  45491. (http://lcs.ios.ac.cn/~xuzb/canalyze/memmodel.pdf) (Zhongxing Xu, Ted

  45492. Kremenek, and Jian Zhang).

  45493. 

  45494.  A 'region_model' encapsulates a representation of the state of memory,

  45495. with a tree of 'region' instances, along with their associated values.

  45496. The representation is graph-like because values can be pointers to

  45497. regions.  It also stores a constraint_manager, capturing relationships

  45498. between the values.

  45499. 

  45500.  Because each node in the 'exploded_graph' has a 'region_model', and

  45501. each of the latter is graph-like, the 'exploded_graph' is in some ways a

  45502. graph of graphs.

  45503. 

  45504.  Here's an example of printing a 'region_model', showing the ASCII-art

  45505. used to visualize the region hierarchy (colorized when printing to

  45506. stderr):

  45507. 

  45508.      (gdb) call debug (*this)

  45509.      r0: {kind: 'root', parent: null, sval: null}

  45510.      |-stack: r1: {kind: 'stack', parent: r0, sval: sv1}

  45511.      |  |: sval: sv1: {poisoned: uninit}

  45512.      |  |-frame for 'test': r2: {kind: 'frame', parent: r1, sval: null, map: {'ptr_3': r3}, function: 'test', depth: 0}

  45513.      |  |  `-'ptr_3': r3: {kind: 'map', parent: r2, sval: sv3, type: 'void *', map: {}}

  45514.      |  |    |: sval: sv3: {type: 'void *', unknown}

  45515.      |  |    |: type: 'void *'

  45516.      |  `-frame for 'calls_malloc': r4: {kind: 'frame', parent: r1, sval: null, map: {'result_3': r7, '_4': r8, '<anonymous>': r5}, function: 'calls_malloc', depth: 1}

  45517.      |    |-'<anonymous>': r5: {kind: 'map', parent: r4, sval: sv4, type: 'void *', map: {}}

  45518.      |    |  |: sval: sv4: {type: 'void *', &r6}

  45519.      |    |  |: type: 'void *'

  45520.      |    |-'result_3': r7: {kind: 'map', parent: r4, sval: sv4, type: 'void *', map: {}}

  45521.      |    |  |: sval: sv4: {type: 'void *', &r6}

  45522.      |    |  |: type: 'void *'

  45523.      |    `-'_4': r8: {kind: 'map', parent: r4, sval: sv4, type: 'void *', map: {}}

  45524.      |      |: sval: sv4: {type: 'void *', &r6}

  45525.      |      |: type: 'void *'

  45526.      `-heap: r9: {kind: 'heap', parent: r0, sval: sv2}

  45527.        |: sval: sv2: {poisoned: uninit}

  45528.        `-r6: {kind: 'symbolic', parent: r9, sval: null, map: {}}

  45529.      svalues:

  45530.        sv0: {type: 'size_t', '1024'}

  45531.        sv1: {poisoned: uninit}

  45532.        sv2: {poisoned: uninit}

  45533.        sv3: {type: 'void *', unknown}

  45534.        sv4: {type: 'void *', &r6}

  45535.      constraint manager:

  45536.        equiv classes:

  45537.          ec0: {sv0 == '1024'}

  45538.          ec1: {sv4}

  45539.        constraints:

  45540. 

  45541.  This is the state at the point of returning from 'calls_malloc' back to

  45542. 'test' in the following:

  45543. 

  45544.      void *

  45545.      calls_malloc (void)

  45546.      {

  45547.        void *result = malloc (1024);

  45548.        return result;

  45549.      }

  45550. 

  45551.      void test (void)

  45552.      {

  45553.        void *ptr = calls_malloc ();

  45554.        /* etc.  */

  45555.      }

  45556. 

  45557.  The "root" region ("r0") has a "stack" child ("r1"), with two children:

  45558. a frame for 'test' ("r2"), and a frame for 'calls_malloc' ("r4").  These

  45559. frame regions have child regions for storing their local variables.  For

  45560. example, the return region and that of various other regions within the

  45561. "calls_malloc" frame all have value "sv4", a pointer to a heap-allocated

  45562. region "r6".  Within the parent frame, 'ptr_3' has value "sv3", an

  45563. unknown 'void *'.

  45564. 

  45565. 27.1.5 Analyzer Paths

  45566. ---------------------

  45567. 

  45568. We need to explain to the user what the problem is, and to persuade them

  45569. that there really is a problem.  Hence having a 'diagnostic_path' isn't

  45570. just an incidental detail of the analyzer; it's required.

  45571. 

  45572.  Paths ought to be:

  45573.    * interprocedurally-valid

  45574.    * feasible

  45575. 

  45576.  Without state-merging, all paths in the exploded graph are feasible (in

  45577. terms of constraints being satisified).  With state-merging, paths in

  45578. the exploded graph can be infeasible.

  45579. 

  45580.  We collate warnings and only emit them for the simplest path e.g.  for

  45581. a bug in a utility function, with lots of routes to calling it, we only

  45582. emit the simplest path (which could be intraprocedural, if it can be

  45583. reproduced without a caller).  We apply a check that each duplicate

  45584. warning's shortest path is feasible, rejecting any warnings for which

  45585. the shortest path is infeasible (which could lead to false negatives).

  45586. 

  45587.  We use the shortest feasible 'exploded_path' through the

  45588. 'exploded_graph' (a list of 'exploded_edge *') to build a

  45589. 'diagnostic_path' (a list of events for the diagnostic subsystem) -

  45590. specifically a 'checker_path'.

  45591. 

  45592.  Having built the 'checker_path', we prune it to try to eliminate events

  45593. that aren't relevant, to minimize how much the user has to read.

  45594. 

  45595.  After pruning, we notify each event in the path of its ID and record

  45596. the IDs of interesting events, allowing for events to refer to other

  45597. events in their descriptions.  The 'pending_diagnostic' class has

  45598. various vfuncs to support emitting more precise descriptions, so that

  45599. e.g.

  45600. 

  45601.    * a deref-of-unchecked-malloc diagnostic might use:

  45602.             returning possibly-NULL pointer to 'make_obj' from 'allocator'

  45603.      for a 'return_event' to make it clearer how the unchecked value

  45604.      moves from callee back to caller

  45605.    * a double-free diagnostic might use:

  45606.             second 'free' here; first 'free' was at (3)

  45607.      and a use-after-free might use

  45608.             use after 'free' here; memory was freed at (2)

  45609. 

  45610.  At this point we can emit the diagnostic.

  45611. 

  45612. 27.1.6 Limitations

  45613. ------------------

  45614. 

  45615.    * Only for C so far

  45616.    * The implementation of call summaries is currently very simplistic.

  45617.    * Lack of function pointer analysis

  45618.    * The constraint-handling code assumes reflexivity in some places

  45619.      (that values are equal to themselves), which is not the case for

  45620.      NaN. As a simple workaround, constraints on floating-point values

  45621.      are currently ignored.

  45622.    * The region model code creates lots of little mutable objects at

  45623.      each 'region_model' (and thus per 'exploded_node') rather than

  45624.      sharing immutable objects and having the mutable state in the

  45625.      'program_state' or 'region_model'.  The latter approach might be

  45626.      more efficient, and might avoid dealing with IDs rather than

  45627.      pointers (which requires us to impose an ordering to get meaningful

  45628.      equality).

  45629.    * The region model code doesn't yet support 'memcpy'.  At the

  45630.      gimple-ssa level these have been optimized to statements like this:

  45631.           _10 = MEM <long unsigned int> [(char * {ref-all})&c]

  45632.           MEM <long unsigned int> [(char * {ref-all})&d] = _10;

  45633.      Perhaps they could be supported via a new 'compound_svalue' type.

  45634.    * There are various other limitations in the region model (grep for

  45635.      TODO/xfail in the testsuite).

  45636.    * The constraint_manager's implementation of transitivity is

  45637.      currently too expensive to enable by default and so must be

  45638.      manually enabled via '-fanalyzer-transitivity').

  45639.    * The checkers are currently hardcoded and don't allow for user

  45640.      extensibility (e.g.  adding allocate/release pairs).

  45641.    * Although the analyzer's test suite has a proof-of-concept test case

  45642.      for LTO, LTO support hasn't had extensive testing.  There are

  45643.      various lang-specific things in the analyzer that assume C rather

  45644.      than LTO. For example, SSA names are printed to the user in "raw"

  45645.      form, rather than printing the underlying variable name.

  45646. 

  45647.  Some ideas for other checkers

  45648.    * File-descriptor-based APIs

  45649.    * Linux kernel internal APIs

  45650.    * Signal handling

  45651. 

  45652. 

  45653. File: gccint.info,  Node: Debugging the Analyzer,  Prev: Analyzer Internals,  Up: Static Analyzer

  45654. 

  45655. 27.2 Debugging the Analyzer

  45656. ===========================

  45657. 

  45658. 27.2.1 Special Functions for Debugging the Analyzer

  45659. ---------------------------------------------------

  45660. 

  45661. The analyzer recognizes various special functions by name, for use in

  45662. debugging the analyzer.  Declarations can be seen in the testsuite in

  45663. 'analyzer-decls.h'.  None of these functions are actually implemented.

  45664. 

  45665.  Add:

  45666.        __analyzer_break ();

  45667.  to the source being analyzed to trigger a breakpoint in the analyzer

  45668. when that source is reached.  By putting a series of these in the

  45669. source, it's much easier to effectively step through the program state

  45670. as it's analyzed.

  45671. 

  45672.      __analyzer_dump ();

  45673. 

  45674.  will dump the copious information about the analyzer's state each time

  45675. it reaches the call in its traversal of the source.

  45676. 

  45677.      __analyzer_dump_path ();

  45678. 

  45679.  will emit a placeholder "note" diagnostic with a path to that call

  45680. site, if the analyzer finds a feasible path to it.

  45681. 

  45682.  The builtin '__analyzer_dump_exploded_nodes' will emit a warning after

  45683. analysis containing information on all of the exploded nodes at that

  45684. program point:

  45685. 

  45686.        __analyzer_dump_exploded_nodes (0);

  45687. 

  45688.  will output the number of "processed" nodes, and the IDs of both

  45689. "processed" and "merger" nodes, such as:

  45690. 

  45691.      warning: 2 processed enodes: [EN: 56, EN: 58] merger(s): [EN: 54-55, EN: 57, EN: 59]

  45692. 

  45693.  With a non-zero argument

  45694. 

  45695.        __analyzer_dump_exploded_nodes (1);

  45696. 

  45697.  it will also dump all of the states within the "processed" nodes.

  45698. 

  45699.         __analyzer_dump_region_model ();

  45700.  will dump the region_model's state to stderr.

  45701. 

  45702.      __analyzer_eval (expr);

  45703.  will emit a warning with text "TRUE", FALSE" or "UNKNOWN" based on the

  45704. truthfulness of the argument.  This is useful for writing DejaGnu tests.

  45705. 

  45706. 27.2.2 Other Debugging Techniques

  45707. ---------------------------------

  45708. 

  45709. One approach when tracking down where a particular bogus state is

  45710. introduced into the 'exploded_graph' is to add custom code to

  45711. 'region_model::validate'.

  45712. 

  45713.  For example, this custom code (added to 'region_model::validate')

  45714. breaks with an assertion failure when a variable called 'ptr' acquires a

  45715. value that's unknown, using 'region_model::get_value_by_name' to locate

  45716. the variable

  45717. 

  45718.          /* Find a variable matching "ptr".  */

  45719.          svalue_id sid = get_value_by_name ("ptr");

  45720.          if (!sid.null_p ())

  45721.            {

  45722.      	svalue *sval = get_svalue (sid);

  45723.      	gcc_assert (sval->get_kind () != SK_UNKNOWN);

  45724.            }

  45725. 

  45726.  making it easier to investigate further in a debugger when this occurs.

  45727. 

  45728. 

  45729. File: gccint.info,  Node: User Experience Guidelines,  Next: Funding,  Prev: Static Analyzer,  Up: Top

  45730. 

  45731. 28 User Experience Guidelines

  45732. *****************************

  45733. 

  45734. To borrow a slogan from Elm

  45735. (https://elm-lang.org/news/compilers-as-assistants),

  45736. 

  45737.      *Compilers should be assistants, not adversaries.*  A compiler

  45738.      should not just detect bugs, it should then help you understand why

  45739.      there is a bug.  It should not berate you in a robot voice, it

  45740.      should give you specific hints that help you write better code.

  45741.      Ultimately, a compiler should make programming faster and more fun!

  45742.                           -- _Evan Czaplicki_

  45743. 

  45744.  This chapter provides guidelines on how to implement diagnostics and

  45745. command-line options in ways that we hope achieve the above ideal.

  45746. 

  45747. * Menu:

  45748. 

  45749. * Guidelines for Diagnostics::       How to implement diagnostics.

  45750. * Guidelines for Options::           Guidelines for command-line options.

  45751. 

  45752. 

  45753. File: gccint.info,  Node: Guidelines for Diagnostics,  Next: Guidelines for Options,  Up: User Experience Guidelines

  45754. 

  45755. 28.1 Guidelines for Diagnostics

  45756. ===============================

  45757. 

  45758. 28.1.1 Talk in terms of the user's code

  45759. ---------------------------------------

  45760. 

  45761. Diagnostics should be worded in terms of the user's source code, and the

  45762. source language, rather than GCC's own implementation details.

  45763. 

  45764. 28.1.2 Diagnostics are actionable

  45765. ---------------------------------

  45766. 

  45767. A good diagnostic is "actionable": it should assist the user in taking

  45768. action.

  45769. 

  45770.  Consider what an end user will want to do when encountering a

  45771. diagnostic.

  45772. 

  45773.  Given an error, an end user will think: "How do I fix this?"

  45774. 

  45775.  Given a warning, an end user will think:

  45776. 

  45777.    * "Is this a real problem?"

  45778.    * "Do I care?"

  45779.    * if they decide it's genuine: "How do I fix this?"

  45780. 

  45781.  A good diagnostic provides pertinent information to allow the user to

  45782. easily answer the above questions.

  45783. 

  45784. 28.1.3 The user's attention is important

  45785. ----------------------------------------

  45786. 

  45787. A perfect compiler would issue a warning on every aspect of the user's

  45788. source code that ought to be fixed, and issue no other warnings.

  45789. Naturally, this ideal is impossible to achieve.

  45790. 

  45791.  Warnings should have a good "signal-to-noise ratio": we should have few

  45792. "false positives" (falsely issuing a warning when no warning is

  45793. warranted) and few "false negatives" (failing to issue a warning when

  45794. one _is_ justified).

  45795. 

  45796.  Note that a false positive can mean, in practice, a warning that the

  45797. user doesn't agree with.  Ideally a diagnostic should contain enough

  45798. information to allow the user to make an informed choice about whether

  45799. they should care (and how to fix it), but a balance must be drawn

  45800. against overloading the user with irrelevant data.

  45801. 

  45802. 28.1.4 Precision of Wording

  45803. ---------------------------

  45804. 

  45805. Provide the user with details that allow them to identify what the

  45806. problem is.  For example, the vaguely-worded message:

  45807. 

  45808.      demo.c:1:1: warning: 'noinline' attribute ignored [-Wattributes]

  45809.          1 | int foo __attribute__((noinline));

  45810.            | ^~~

  45811. 

  45812. doesn't tell the user why the attribute was ignored, or what kind of

  45813. entity the compiler thought the attribute was being applied to (the

  45814. source location for the diagnostic is also poor; *note discussion of

  45815. 'input_location': input_location_example.).  A better message would be:

  45816. 

  45817.      demo.c:1:24: warning: attribute 'noinline' on variable 'foo' was

  45818.         ignored [-Wattributes]

  45819.          1 | int foo __attribute__((noinline));

  45820.            |     ~~~ ~~~~~~~~~~~~~~~^~~~~~~~~

  45821.      demo.c:1:24: note: attribute 'noinline' is only applicable to functions

  45822. 

  45823. which spells out the missing information (and fixes the location

  45824. information, as discussed below).

  45825. 

  45826.  The above example uses a note to avoid a combinatorial explosion of

  45827. possible messages.

  45828. 

  45829. 28.1.5 Try the diagnostic on real-world code

  45830. --------------------------------------------

  45831. 

  45832. It's worth testing a new warning on many instances of real-world code,

  45833. written by different people, and seeing what it complains about, and

  45834. what it doesn't complain about.

  45835. 

  45836.  This may suggest heuristics that silence common false positives.

  45837. 

  45838.  It may also suggest ways to improve the precision of the message.

  45839. 

  45840. 28.1.6 Make mismatches clear

  45841. ----------------------------

  45842. 

  45843. Many diagnostics relate to a mismatch between two different places in

  45844. the user's source code.  Examples include:

  45845.    * a type mismatch, where the type at a usage site does not match the

  45846.      type at a declaration

  45847. 

  45848.    * the argument count at a call site does not match the parameter

  45849.      count at the declaration

  45850. 

  45851.    * something is erroneously duplicated (e.g. an error, due to breaking

  45852.      a uniqueness requirement, or a warning, if it's suggestive of a

  45853.      bug)

  45854. 

  45855.    * an "opened" syntactic construct (such as an open-parenthesis) is

  45856.      not closed

  45857. 

  45858.  In each case, the diagnostic should indicate *both* pertinent locations

  45859. (so that the user can easily see the problem and how to fix it).

  45860. 

  45861.  The standard way to do this is with a note (via 'inform').  For

  45862. example:

  45863. 

  45864.        auto_diagnostic_group d;

  45865.        if (warning_at (loc, OPT_Wduplicated_cond,

  45866.                        "duplicated %<if%> condition"))

  45867.          inform (EXPR_LOCATION (t), "previously used here");

  45868. 

  45869. which leads to:

  45870. 

  45871.      demo.c: In function 'test':

  45872.      demo.c:5:17: warning: duplicated 'if' condition [-Wduplicated-cond]

  45873.          5 |   else if (flag > 3)

  45874.            |            ~~~~~^~~

  45875.      demo.c:3:12: note: previously used here

  45876.          3 |   if (flag > 3)

  45877.            |       ~~~~~^~~

  45878. 

  45879. The 'inform' call should be guarded by the return value from the

  45880. 'warning_at' call so that the note isn't emitted when the warning is

  45881. suppressed.

  45882. 

  45883.  For cases involving punctuation where the locations might be near each

  45884. other, they can be conditionally consolidated via

  45885. 'gcc_rich_location::add_location_if_nearby':

  45886. 

  45887.          auto_diagnostic_group d;

  45888.          gcc_rich_location richloc (primary_loc);

  45889.          bool added secondary = richloc.add_location_if_nearby (secondary_loc);

  45890.          error_at (&richloc, "main message");

  45891.          if (!added secondary)

  45892.            inform (secondary_loc, "message for secondary");

  45893. 

  45894. This will emit either one diagnostic with two locations:

  45895.        demo.c:42:10: error: main message

  45896.          (foo)

  45897.          ~   ^

  45898. 

  45899. or two diagnostics:

  45900. 

  45901.        demo.c:42:4: error: main message

  45902.          foo)

  45903.             ^

  45904.        demo.c:40:2: note: message for secondary

  45905.          (

  45906.          ^

  45907. 

  45908. 28.1.7 Location Information

  45909. ---------------------------

  45910. 

  45911. GCC's 'location_t' type can support both ordinary locations, and

  45912. locations relating to a macro expansion.

  45913. 

  45914.  As of GCC 6, ordinary locations changed from supporting just a point in

  45915. the user's source code to supporting three points: the "caret" location,

  45916. plus a start and a finish:

  45917. 

  45918.            a = foo && bar;

  45919.                ~~~~^~~~~~

  45920.                |   |    |

  45921.                |   |    finish

  45922.                |   caret

  45923.                start

  45924. 

  45925.  Tokens coming out of libcpp have locations of the form 'caret ==

  45926. start', such as for 'foo' here:

  45927. 

  45928.            a = foo && bar;

  45929.                ^~~

  45930.                | |

  45931.                | finish

  45932.                caret == start

  45933. 

  45934.  Compound expressions should be reported using the location of the

  45935. expression as a whole, rather than just of one token within it.

  45936. 

  45937.  For example, in '-Wformat', rather than underlining just the first

  45938. token of a bad argument:

  45939. 

  45940.         printf("hello %i %s", (long)0, "world");

  45941.                       ~^      ~

  45942.                       %li

  45943. 

  45944. the whole of the expression should be underlined, so that the user can

  45945. easily identify what is being referred to:

  45946. 

  45947.         printf("hello %i %s", (long)0, "world");

  45948.                       ~^      ~~~~~~~

  45949.                       %li

  45950. 

  45951.  Avoid using the 'input_location' global, and the diagnostic functions

  45952. that implicitly use it--use 'error_at' and 'warning_at' rather than

  45953. 'error' and 'warning', and provide the most appropriate 'location_t'

  45954. value available at that phase of the compilation.  It's possible to

  45955. supply secondary 'location_t' values via 'rich_location'.

  45956. 

  45957. For example, in the example of imprecise wording above, generating the

  45958. diagnostic using 'warning':

  45959. 

  45960.        // BAD: implicitly uses input_location

  45961.        warning (OPT_Wattributes, "%qE attribute ignored", name);

  45962. 

  45963. leads to:

  45964. 

  45965.      // BAD: uses input_location

  45966.      demo.c:1:1: warning: 'noinline' attribute ignored [-Wattributes]

  45967.          1 | int foo __attribute__((noinline));

  45968.            | ^~~

  45969. 

  45970. which thus happened to use the location of the 'int' token, rather than

  45971. that of the attribute.  Using 'warning_at' with the location of the

  45972. attribute, providing the location of the declaration in question as a

  45973. secondary location, and adding a note:

  45974. 

  45975.        auto_diagnostic_group d;

  45976.        gcc_rich_location richloc (attrib_loc);

  45977.        richloc.add_range (decl_loc);

  45978.        if (warning_at (OPT_Wattributes, &richloc,

  45979.                        "attribute %qE on variable %qE was ignored", name))

  45980.          inform (attrib_loc, "attribute %qE is only applicable to functions");

  45981. 

  45982. would lead to:

  45983. 

  45984.      // OK: use location of attribute, with a secondary location

  45985.      demo.c:1:24: warning: attribute 'noinline' on variable 'foo' was

  45986.         ignored [-Wattributes]

  45987.          1 | int foo __attribute__((noinline));

  45988.            |     ~~~ ~~~~~~~~~~~~~~~^~~~~~~~~

  45989.      demo.c:1:24: note: attribute 'noinline' is only applicable to functions

  45990. 

  45991. 28.1.8 Coding Conventions

  45992. -------------------------

  45993. 

  45994. See the diagnostics section

  45995. (https://gcc.gnu.org/codingconventions.html#Diagnostics) of the GCC

  45996. coding conventions.

  45997. 

  45998.  In the C++ front end, when comparing two types in a message, use '%H'

  45999. and '%I' rather than '%T', as this allows the diagnostics subsystem to

  46000. highlight differences between template-based types.  For example, rather

  46001. than using '%qT':

  46002. 

  46003.        // BAD: a pair of %qT used in C++ front end for type comparison

  46004.        error_at (loc, "could not convert %qE from %qT to %qT", expr,

  46005.                  TREE_TYPE (expr), type);

  46006. 

  46007. which could lead to:

  46008. 

  46009.      error: could not convert 'map<int, double>()' from 'map<int,double>'

  46010.         to 'map<int,int>'

  46011. 

  46012. using '%H' and '%I' (via '%qH' and '%qI'):

  46013. 

  46014.        // OK: compare types in C++ front end via %qH and %qI

  46015.        error_at (loc, "could not convert %qE from %qH to %qI", expr,

  46016.                  TREE_TYPE (expr), type);

  46017. 

  46018. allows the above output to be simplified to:

  46019. 

  46020.      error: could not convert 'map<int, double>()' from 'map<[...],double>'

  46021.         to 'map<[...],int>'

  46022. 

  46023. where the 'double' and 'int' are colorized to highlight them.

  46024. 

  46025. 28.1.9 Group logically-related diagnostics

  46026. ------------------------------------------

  46027. 

  46028. Use 'auto_diagnostic_group' when issuing multiple related diagnostics

  46029. (seen in various examples on this page).  This informs the diagnostic

  46030. subsystem that all diagnostics issued within the lifetime of the

  46031. 'auto_diagnostic_group' are related.  For example,

  46032. '-fdiagnostics-format=json' will treat the first diagnostic emitted

  46033. within the group as a top-level diagnostic, and all subsequent

  46034. diagnostics within the group as its children.

  46035. 

  46036. 28.1.10 Quoting

  46037. ---------------

  46038. 

  46039. Text should be quoted by either using the 'q' modifier in a directive

  46040. such as '%qE', or by enclosing the quoted text in a pair of '%<' and

  46041. '%>' directives, and never by using explicit quote characters.  The

  46042. directives handle the appropriate quote characters for each language and

  46043. apply the correct color or highlighting.

  46044. 

  46045.  The following elements should be quoted in GCC diagnostics:

  46046. 

  46047.    * Language keywords.

  46048.    * Tokens.

  46049.    * Boolean, numerical, character, and string constants that appear in

  46050.      the source code.

  46051.    * Identifiers, including function, macro, type, and variable names.

  46052. 

  46053.  Other elements such as numbers that do not refer to numeric constants

  46054. that appear in the source code should not be quoted.  For example, in

  46055. the message:

  46056. 

  46057.      argument %d of %qE must be a pointer type

  46058. 

  46059. since the argument number does not refer to a numerical constant in the

  46060. source code it should not be quoted.

  46061. 

  46062. 28.1.11 Spelling and Terminology

  46063. --------------------------------

  46064. 

  46065. See the terminology and markup

  46066. (https://gcc.gnu.org/codingconventions.html#Spelling Spelling) section

  46067. of the GCC coding conventions.

  46068. 

  46069. 28.1.12 Fix-it hints

  46070. --------------------

  46071. 

  46072. GCC's diagnostic subsystem can emit "fix-it hints": small suggested

  46073. edits to the user's source code.

  46074. 

  46075.  They are printed by default underneath the code in question.  They can

  46076. also be viewed via '-fdiagnostics-generate-patch' and

  46077. '-fdiagnostics-parseable-fixits'.  With the latter, an IDE ought to be

  46078. able to offer to automatically apply the suggested fix.

  46079. 

  46080.  Fix-it hints contain code fragments, and thus they should not be marked

  46081. for translation.

  46082. 

  46083.  Fix-it hints can be added to a diagnostic by using a 'rich_location'

  46084. rather than a 'location_t' - the fix-it hints are added to the

  46085. 'rich_location' using one of the various 'add_fixit' member functions of

  46086. 'rich_location'.  They are documented with 'rich_location' in

  46087. 'libcpp/line-map.h'.  It's easiest to use the 'gcc_rich_location'

  46088. subclass of 'rich_location' found in 'gcc-rich-location.h', as this

  46089. implicitly supplies the 'line_table' variable.

  46090. 

  46091.  For example:

  46092. 

  46093.         if (const char *suggestion = hint.suggestion ())

  46094.           {

  46095.             gcc_rich_location richloc (location);

  46096.             richloc.add_fixit_replace (suggestion);

  46097.             error_at (&richloc,

  46098.                       "%qE does not name a type; did you mean %qs?",

  46099.                       id, suggestion);

  46100.           }

  46101. 

  46102. which can lead to:

  46103. 

  46104.      spellcheck-typenames.C:73:1: error: 'singed' does not name a type; did

  46105.         you mean 'signed'?

  46106.         73 | singed char ch;

  46107.            | ^~~~~~

  46108.            | signed

  46109. 

  46110.  Non-trivial edits can be built up by adding multiple fix-it hints to

  46111. one 'rich_location'.  It's best to express the edits in terms of the

  46112. locations of individual tokens.  Various handy functions for adding

  46113. fix-it hints for idiomatic C and C++ can be seen in

  46114. 'gcc-rich-location.h'.

  46115. 

  46116. 28.1.12.1 Fix-it hints should work

  46117. ..................................

  46118. 

  46119. When implementing a fix-it hint, please verify that the suggested edit

  46120. leads to fixed, compilable code.  (Unfortunately, this currently must be

  46121. done by hand using '-fdiagnostics-generate-patch'.  It would be good to

  46122. have an automated way of verifying that fix-it hints actually fix the

  46123. code).

  46124. 

  46125.  For example, a "gotcha" here is to forget to add a space when adding a

  46126. missing reserved word.  Consider a C++ fix-it hint that adds 'typename'

  46127. in front of a template declaration.  A naive way to implement this might

  46128. be:

  46129. 

  46130.      gcc_rich_location richloc (loc);

  46131.      // BAD: insertion is missing a trailing space

  46132.      richloc.add_fixit_insert_before ("typename");

  46133.      error_at (&richloc, "need %<typename%> before %<%T::%E%> because "

  46134.                           "%qT is a dependent scope",

  46135.                           parser->scope, id, parser->scope);

  46136. 

  46137. When applied to the code, this might lead to:

  46138. 

  46139.      T::type x;

  46140. 

  46141. being "corrected" to:

  46142. 

  46143.      typenameT::type x;

  46144. 

  46145. In this case, the correct thing to do is to add a trailing space after

  46146. 'typename':

  46147. 

  46148.      gcc_rich_location richloc (loc);

  46149.      // OK: note that here we have a trailing space

  46150.      richloc.add_fixit_insert_before ("typename ");

  46151.      error_at (&richloc, "need %<typename%> before %<%T::%E%> because "

  46152.                           "%qT is a dependent scope",

  46153.                           parser->scope, id, parser->scope);

  46154. 

  46155. leading to this corrected code:

  46156. 

  46157.      typename T::type x;

  46158. 

  46159. 28.1.12.2 Express deletion in terms of deletion, not replacement

  46160. ................................................................

  46161. 

  46162. It's best to express deletion suggestions in terms of deletion fix-it

  46163. hints, rather than replacement fix-it hints.  For example, consider

  46164. this:

  46165. 

  46166.          auto_diagnostic_group d;

  46167.          gcc_rich_location richloc (location_of (retval));

  46168.          tree name = DECL_NAME (arg);

  46169.          richloc.add_fixit_replace (IDENTIFIER_POINTER (name));

  46170.          warning_at (&richloc, OPT_Wredundant_move,

  46171.                      "redundant move in return statement");

  46172. 

  46173. which is intended to e.g. replace a 'std::move' with the underlying

  46174. value:

  46175. 

  46176.         return std::move (retval);

  46177.                ~~~~~~~~~~^~~~~~~~

  46178.                retval

  46179. 

  46180. where the change has been expressed as replacement, replacing with the

  46181. name of the declaration.  This works for simple cases, but consider this

  46182. case:

  46183. 

  46184.      #ifdef SOME_CONFIG_FLAG

  46185.      # define CONFIGURY_GLOBAL global_a

  46186.      #else

  46187.      # define CONFIGURY_GLOBAL global_b

  46188.      #endif

  46189. 

  46190.      int fn ()

  46191.      {

  46192.        return std::move (CONFIGURY_GLOBAL /* some comment */);

  46193.      }

  46194. 

  46195. The above implementation erroneously strips out the macro and the

  46196. comment in the fix-it hint:

  46197. 

  46198.         return std::move (CONFIGURY_GLOBAL /* some comment */);

  46199.                ~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

  46200.                global_a

  46201. 

  46202. and thus this resulting code:

  46203. 

  46204.         return global_a;

  46205. 

  46206. It's better to do deletions in terms of deletions; deleting the

  46207. 'std::move (' and the trailing close-paren, leading to this:

  46208. 

  46209.         return std::move (CONFIGURY_GLOBAL /* some comment */);

  46210.                ~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

  46211.                CONFIGURY_GLOBAL /* some comment */

  46212. 

  46213. and thus this result:

  46214. 

  46215.         return CONFIGURY_GLOBAL /* some comment */;

  46216. 

  46217. Unfortunately, the pertinent 'location_t' values are not always

  46218. available.

  46219. 

  46220. 28.1.12.3 Multiple suggestions

  46221. ..............................

  46222. 

  46223. In the rare cases where you need to suggest more than one mutually

  46224. exclusive solution to a problem, this can be done by emitting multiple

  46225. notes and calling 'rich_location::fixits_cannot_be_auto_applied' on each

  46226. note's 'rich_location'.  If this is called, then the fix-it hints in the

  46227. 'rich_location' will be printed, but will not be added to generated

  46228. patches.

  46229. 

  46230. 

  46231. File: gccint.info,  Node: Guidelines for Options,  Prev: Guidelines for Diagnostics,  Up: User Experience Guidelines

  46232. 

  46233. 28.2 Guidelines for Options

  46234. ===========================

  46235. 

  46236. 

  46237. File: gccint.info,  Node: Funding,  Next: GNU Project,  Prev: User Experience Guidelines,  Up: Top

  46238. 

  46239. Funding Free Software

  46240. *********************

  46241. 

  46242. If you want to have more free software a few years from now, it makes

  46243. sense for you to help encourage people to contribute funds for its

  46244. development.  The most effective approach known is to encourage

  46245. commercial redistributors to donate.

  46246. 

  46247.  Users of free software systems can boost the pace of development by

  46248. encouraging for-a-fee distributors to donate part of their selling price

  46249. to free software developers--the Free Software Foundation, and others.

  46250. 

  46251.  The way to convince distributors to do this is to demand it and expect

  46252. it from them.  So when you compare distributors, judge them partly by

  46253. how much they give to free software development.  Show distributors they

  46254. must compete to be the one who gives the most.

  46255. 

  46256.  To make this approach work, you must insist on numbers that you can

  46257. compare, such as, "We will donate ten dollars to the Frobnitz project

  46258. for each disk sold."  Don't be satisfied with a vague promise, such as

  46259. "A portion of the profits are donated," since it doesn't give a basis

  46260. for comparison.

  46261. 

  46262.  Even a precise fraction "of the profits from this disk" is not very

  46263. meaningful, since creative accounting and unrelated business decisions

  46264. can greatly alter what fraction of the sales price counts as profit.  If

  46265. the price you pay is $50, ten percent of the profit is probably less

  46266. than a dollar; it might be a few cents, or nothing at all.

  46267. 

  46268.  Some redistributors do development work themselves.  This is useful

  46269. too; but to keep everyone honest, you need to inquire how much they do,

  46270. and what kind.  Some kinds of development make much more long-term

  46271. difference than others.  For example, maintaining a separate version of

  46272. a program contributes very little; maintaining the standard version of a

  46273. program for the whole community contributes much.  Easy new ports

  46274. contribute little, since someone else would surely do them; difficult

  46275. ports such as adding a new CPU to the GNU Compiler Collection contribute

  46276. more; major new features or packages contribute the most.

  46277. 

  46278.  By establishing the idea that supporting further development is "the

  46279. proper thing to do" when distributing free software for a fee, we can

  46280. assure a steady flow of resources into making more free software.

  46281. 

  46282.      Copyright (C) 1994 Free Software Foundation, Inc.

  46283.      Verbatim copying and redistribution of this section is permitted

  46284.      without royalty; alteration is not permitted.

  46285. 

  46286. 

  46287. File: gccint.info,  Node: GNU Project,  Next: Copying,  Prev: Funding,  Up: Top

  46288. 

  46289. The GNU Project and GNU/Linux

  46290. *****************************

  46291. 

  46292. The GNU Project was launched in 1984 to develop a complete Unix-like

  46293. operating system which is free software: the GNU system.  (GNU is a

  46294. recursive acronym for "GNU's Not Unix"; it is pronounced "guh-NEW".)

  46295. Variants of the GNU operating system, which use the kernel Linux, are

  46296. now widely used; though these systems are often referred to as "Linux",

  46297. they are more accurately called GNU/Linux systems.

  46298. 

  46299.  For more information, see:

  46300.      <http://www.gnu.org/>

  46301.      <http://www.gnu.org/gnu/linux-and-gnu.html>

  46302. 

  46303. 

  46304. File: gccint.info,  Node: Copying,  Next: GNU Free Documentation License,  Prev: GNU Project,  Up: Top

  46305. 

  46306. GNU General Public License

  46307. **************************

  46308. 

  46309.                         Version 3, 29 June 2007

  46310. 

  46311.      Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>

  46312. 

  46313.      Everyone is permitted to copy and distribute verbatim copies of this

  46314.      license document, but changing it is not allowed.

  46315. 

  46316. Preamble

  46317. ========

  46318. 

  46319. The GNU General Public License is a free, copyleft license for software

  46320. and other kinds of works.

  46321. 

  46322.  The licenses for most software and other practical works are designed

  46323. to take away your freedom to share and change the works.  By contrast,

  46324. the GNU General Public License is intended to guarantee your freedom to

  46325. share and change all versions of a program-to make sure it remains free

  46326. software for all its users.  We, the Free Software Foundation, use the

  46327. GNU General Public License for most of our software; it applies also to

  46328. any other work released this way by its authors.  You can apply it to

  46329. your programs, too.

  46330. 

  46331.  When we speak of free software, we are referring to freedom, not price.

  46332. Our General Public Licenses are designed to make sure that you have the

  46333. freedom to distribute copies of free software (and charge for them if

  46334. you wish), that you receive source code or can get it if you want it,

  46335. that you can change the software or use pieces of it in new free

  46336. programs, and that you know you can do these things.

  46337. 

  46338.  To protect your rights, we need to prevent others from denying you

  46339. these rights or asking you to surrender the rights.  Therefore, you have

  46340. certain responsibilities if you distribute copies of the software, or if

  46341. you modify it: responsibilities to respect the freedom of others.

  46342. 

  46343.  For example, if you distribute copies of such a program, whether gratis

  46344. or for a fee, you must pass on to the recipients the same freedoms that

  46345. you received.  You must make sure that they, too, receive or can get the

  46346. source code.  And you must show them these terms so they know their

  46347. rights.

  46348. 

  46349.  Developers that use the GNU GPL protect your rights with two steps: (1)

  46350. assert copyright on the software, and (2) offer you this License giving

  46351. you legal permission to copy, distribute and/or modify it.

  46352. 

  46353.  For the developers' and authors' protection, the GPL clearly explains

  46354. that there is no warranty for this free software.  For both users' and

  46355. authors' sake, the GPL requires that modified versions be marked as

  46356. changed, so that their problems will not be attributed erroneously to

  46357. authors of previous versions.

  46358. 

  46359.  Some devices are designed to deny users access to install or run

  46360. modified versions of the software inside them, although the manufacturer

  46361. can do so.  This is fundamentally incompatible with the aim of

  46362. protecting users' freedom to change the software.  The systematic

  46363. pattern of such abuse occurs in the area of products for individuals to

  46364. use, which is precisely where it is most unacceptable.  Therefore, we

  46365. have designed this version of the GPL to prohibit the practice for those

  46366. products.  If such problems arise substantially in other domains, we

  46367. stand ready to extend this provision to those domains in future versions

  46368. of the GPL, as needed to protect the freedom of users.

  46369. 

  46370.  Finally, every program is threatened constantly by software patents.

  46371. States should not allow patents to restrict development and use of

  46372. software on general-purpose computers, but in those that do, we wish to

  46373. avoid the special danger that patents applied to a free program could

  46374. make it effectively proprietary.  To prevent this, the GPL assures that

  46375. patents cannot be used to render the program non-free.

  46376. 

  46377.  The precise terms and conditions for copying, distribution and

  46378. modification follow.

  46379. 

  46380. TERMS AND CONDITIONS

  46381. ====================

  46382. 

  46383.   0. Definitions.

  46384. 

  46385.      "This License" refers to version 3 of the GNU General Public

  46386.      License.

  46387. 

  46388.      "Copyright" also means copyright-like laws that apply to other

  46389.      kinds of works, such as semiconductor masks.

  46390. 

  46391.      "The Program" refers to any copyrightable work licensed under this

  46392.      License.  Each licensee is addressed as "you".  "Licensees" and

  46393.      "recipients" may be individuals or organizations.

  46394. 

  46395.      To "modify" a work means to copy from or adapt all or part of the

  46396.      work in a fashion requiring copyright permission, other than the

  46397.      making of an exact copy.  The resulting work is called a "modified

  46398.      version" of the earlier work or a work "based on" the earlier work.

  46399. 

  46400.      A "covered work" means either the unmodified Program or a work

  46401.      based on the Program.

  46402. 

  46403.      To "propagate" a work means to do anything with it that, without

  46404.      permission, would make you directly or secondarily liable for

  46405.      infringement under applicable copyright law, except executing it on

  46406.      a computer or modifying a private copy.  Propagation includes

  46407.      copying, distribution (with or without modification), making

  46408.      available to the public, and in some countries other activities as

  46409.      well.

  46410. 

  46411.      To "convey" a work means any kind of propagation that enables other

  46412.      parties to make or receive copies.  Mere interaction with a user

  46413.      through a computer network, with no transfer of a copy, is not

  46414.      conveying.

  46415. 

  46416.      An interactive user interface displays "Appropriate Legal Notices"

  46417.      to the extent that it includes a convenient and prominently visible

  46418.      feature that (1) displays an appropriate copyright notice, and (2)

  46419.      tells the user that there is no warranty for the work (except to

  46420.      the extent that warranties are provided), that licensees may convey

  46421.      the work under this License, and how to view a copy of this

  46422.      License.  If the interface presents a list of user commands or

  46423.      options, such as a menu, a prominent item in the list meets this

  46424.      criterion.

  46425. 

  46426.   1. Source Code.

  46427. 

  46428.      The "source code" for a work means the preferred form of the work

  46429.      for making modifications to it.  "Object code" means any non-source

  46430.      form of a work.

  46431. 

  46432.      A "Standard Interface" means an interface that either is an

  46433.      official standard defined by a recognized standards body, or, in

  46434.      the case of interfaces specified for a particular programming

  46435.      language, one that is widely used among developers working in that

  46436.      language.

  46437. 

  46438.      The "System Libraries" of an executable work include anything,

  46439.      other than the work as a whole, that (a) is included in the normal

  46440.      form of packaging a Major Component, but which is not part of that

  46441.      Major Component, and (b) serves only to enable use of the work with

  46442.      that Major Component, or to implement a Standard Interface for

  46443.      which an implementation is available to the public in source code

  46444.      form.  A "Major Component", in this context, means a major

  46445.      essential component (kernel, window system, and so on) of the

  46446.      specific operating system (if any) on which the executable work

  46447.      runs, or a compiler used to produce the work, or an object code

  46448.      interpreter used to run it.

  46449. 

  46450.      The "Corresponding Source" for a work in object code form means all

  46451.      the source code needed to generate, install, and (for an executable

  46452.      work) run the object code and to modify the work, including scripts

  46453.      to control those activities.  However, it does not include the

  46454.      work's System Libraries, or general-purpose tools or generally

  46455.      available free programs which are used unmodified in performing

  46456.      those activities but which are not part of the work.  For example,

  46457.      Corresponding Source includes interface definition files associated

  46458.      with source files for the work, and the source code for shared

  46459.      libraries and dynamically linked subprograms that the work is

  46460.      specifically designed to require, such as by intimate data

  46461.      communication or control flow between those subprograms and other

  46462.      parts of the work.

  46463. 

  46464.      The Corresponding Source need not include anything that users can

  46465.      regenerate automatically from other parts of the Corresponding

  46466.      Source.

  46467. 

  46468.      The Corresponding Source for a work in source code form is that

  46469.      same work.

  46470. 

  46471.   2. Basic Permissions.

  46472. 

  46473.      All rights granted under this License are granted for the term of

  46474.      copyright on the Program, and are irrevocable provided the stated

  46475.      conditions are met.  This License explicitly affirms your unlimited

  46476.      permission to run the unmodified Program.  The output from running

  46477.      a covered work is covered by this License only if the output, given

  46478.      its content, constitutes a covered work.  This License acknowledges

  46479.      your rights of fair use or other equivalent, as provided by

  46480.      copyright law.

  46481. 

  46482.      You may make, run and propagate covered works that you do not

  46483.      convey, without conditions so long as your license otherwise

  46484.      remains in force.  You may convey covered works to others for the

  46485.      sole purpose of having them make modifications exclusively for you,

  46486.      or provide you with facilities for running those works, provided

  46487.      that you comply with the terms of this License in conveying all

  46488.      material for which you do not control copyright.  Those thus making

  46489.      or running the covered works for you must do so exclusively on your

  46490.      behalf, under your direction and control, on terms that prohibit

  46491.      them from making any copies of your copyrighted material outside

  46492.      their relationship with you.

  46493. 

  46494.      Conveying under any other circumstances is permitted solely under

  46495.      the conditions stated below.  Sublicensing is not allowed; section

  46496.      10 makes it unnecessary.

  46497. 

  46498.   3. Protecting Users' Legal Rights From Anti-Circumvention Law.

  46499. 

  46500.      No covered work shall be deemed part of an effective technological

  46501.      measure under any applicable law fulfilling obligations under

  46502.      article 11 of the WIPO copyright treaty adopted on 20 December

  46503.      1996, or similar laws prohibiting or restricting circumvention of

  46504.      such measures.

  46505. 

  46506.      When you convey a covered work, you waive any legal power to forbid

  46507.      circumvention of technological measures to the extent such

  46508.      circumvention is effected by exercising rights under this License

  46509.      with respect to the covered work, and you disclaim any intention to

  46510.      limit operation or modification of the work as a means of

  46511.      enforcing, against the work's users, your or third parties' legal

  46512.      rights to forbid circumvention of technological measures.

  46513. 

  46514.   4. Conveying Verbatim Copies.

  46515. 

  46516.      You may convey verbatim copies of the Program's source code as you

  46517.      receive it, in any medium, provided that you conspicuously and

  46518.      appropriately publish on each copy an appropriate copyright notice;

  46519.      keep intact all notices stating that this License and any

  46520.      non-permissive terms added in accord with section 7 apply to the

  46521.      code; keep intact all notices of the absence of any warranty; and

  46522.      give all recipients a copy of this License along with the Program.

  46523. 

  46524.      You may charge any price or no price for each copy that you convey,

  46525.      and you may offer support or warranty protection for a fee.

  46526. 

  46527.   5. Conveying Modified Source Versions.

  46528. 

  46529.      You may convey a work based on the Program, or the modifications to

  46530.      produce it from the Program, in the form of source code under the

  46531.      terms of section 4, provided that you also meet all of these

  46532.      conditions:

  46533. 

  46534.        a. The work must carry prominent notices stating that you

  46535.           modified it, and giving a relevant date.

  46536. 

  46537.        b. The work must carry prominent notices stating that it is

  46538.           released under this License and any conditions added under

  46539.           section 7.  This requirement modifies the requirement in

  46540.           section 4 to "keep intact all notices".

  46541. 

  46542.        c. You must license the entire work, as a whole, under this

  46543.           License to anyone who comes into possession of a copy.  This

  46544.           License will therefore apply, along with any applicable

  46545.           section 7 additional terms, to the whole of the work, and all

  46546.           its parts, regardless of how they are packaged.  This License

  46547.           gives no permission to license the work in any other way, but

  46548.           it does not invalidate such permission if you have separately

  46549.           received it.

  46550. 

  46551.        d. If the work has interactive user interfaces, each must display

  46552.           Appropriate Legal Notices; however, if the Program has

  46553.           interactive interfaces that do not display Appropriate Legal

  46554.           Notices, your work need not make them do so.

  46555. 

  46556.      A compilation of a covered work with other separate and independent

  46557.      works, which are not by their nature extensions of the covered

  46558.      work, and which are not combined with it such as to form a larger

  46559.      program, in or on a volume of a storage or distribution medium, is

  46560.      called an "aggregate" if the compilation and its resulting

  46561.      copyright are not used to limit the access or legal rights of the

  46562.      compilation's users beyond what the individual works permit.

  46563.      Inclusion of a covered work in an aggregate does not cause this

  46564.      License to apply to the other parts of the aggregate.

  46565. 

  46566.   6. Conveying Non-Source Forms.

  46567. 

  46568.      You may convey a covered work in object code form under the terms

  46569.      of sections 4 and 5, provided that you also convey the

  46570.      machine-readable Corresponding Source under the terms of this

  46571.      License, in one of these ways:

  46572. 

  46573.        a. Convey the object code in, or embodied in, a physical product

  46574.           (including a physical distribution medium), accompanied by the

  46575.           Corresponding Source fixed on a durable physical medium

  46576.           customarily used for software interchange.

  46577. 

  46578.        b. Convey the object code in, or embodied in, a physical product

  46579.           (including a physical distribution medium), accompanied by a

  46580.           written offer, valid for at least three years and valid for as

  46581.           long as you offer spare parts or customer support for that

  46582.           product model, to give anyone who possesses the object code

  46583.           either (1) a copy of the Corresponding Source for all the

  46584.           software in the product that is covered by this License, on a

  46585.           durable physical medium customarily used for software

  46586.           interchange, for a price no more than your reasonable cost of

  46587.           physically performing this conveying of source, or (2) access

  46588.           to copy the Corresponding Source from a network server at no

  46589.           charge.

  46590. 

  46591.        c. Convey individual copies of the object code with a copy of the

  46592.           written offer to provide the Corresponding Source.  This

  46593.           alternative is allowed only occasionally and noncommercially,

  46594.           and only if you received the object code with such an offer,

  46595.           in accord with subsection 6b.

  46596. 

  46597.        d. Convey the object code by offering access from a designated

  46598.           place (gratis or for a charge), and offer equivalent access to

  46599.           the Corresponding Source in the same way through the same

  46600.           place at no further charge.  You need not require recipients

  46601.           to copy the Corresponding Source along with the object code.

  46602.           If the place to copy the object code is a network server, the

  46603.           Corresponding Source may be on a different server (operated by

  46604.           you or a third party) that supports equivalent copying

  46605.           facilities, provided you maintain clear directions next to the

  46606.           object code saying where to find the Corresponding Source.

  46607.           Regardless of what server hosts the Corresponding Source, you

  46608.           remain obligated to ensure that it is available for as long as

  46609.           needed to satisfy these requirements.

  46610. 

  46611.        e. Convey the object code using peer-to-peer transmission,

  46612.           provided you inform other peers where the object code and

  46613.           Corresponding Source of the work are being offered to the

  46614.           general public at no charge under subsection 6d.

  46615. 

  46616.      A separable portion of the object code, whose source code is

  46617.      excluded from the Corresponding Source as a System Library, need

  46618.      not be included in conveying the object code work.

  46619. 

  46620.      A "User Product" is either (1) a "consumer product", which means

  46621.      any tangible personal property which is normally used for personal,

  46622.      family, or household purposes, or (2) anything designed or sold for

  46623.      incorporation into a dwelling.  In determining whether a product is

  46624.      a consumer product, doubtful cases shall be resolved in favor of

  46625.      coverage.  For a particular product received by a particular user,

  46626.      "normally used" refers to a typical or common use of that class of

  46627.      product, regardless of the status of the particular user or of the

  46628.      way in which the particular user actually uses, or expects or is

  46629.      expected to use, the product.  A product is a consumer product

  46630.      regardless of whether the product has substantial commercial,

  46631.      industrial or non-consumer uses, unless such uses represent the

  46632.      only significant mode of use of the product.

  46633. 

  46634.      "Installation Information" for a User Product means any methods,

  46635.      procedures, authorization keys, or other information required to

  46636.      install and execute modified versions of a covered work in that

  46637.      User Product from a modified version of its Corresponding Source.

  46638.      The information must suffice to ensure that the continued

  46639.      functioning of the modified object code is in no case prevented or

  46640.      interfered with solely because modification has been made.

  46641. 

  46642.      If you convey an object code work under this section in, or with,

  46643.      or specifically for use in, a User Product, and the conveying

  46644.      occurs as part of a transaction in which the right of possession

  46645.      and use of the User Product is transferred to the recipient in

  46646.      perpetuity or for a fixed term (regardless of how the transaction

  46647.      is characterized), the Corresponding Source conveyed under this

  46648.      section must be accompanied by the Installation Information.  But

  46649.      this requirement does not apply if neither you nor any third party

  46650.      retains the ability to install modified object code on the User

  46651.      Product (for example, the work has been installed in ROM).

  46652. 

  46653.      The requirement to provide Installation Information does not

  46654.      include a requirement to continue to provide support service,

  46655.      warranty, or updates for a work that has been modified or installed

  46656.      by the recipient, or for the User Product in which it has been

  46657.      modified or installed.  Access to a network may be denied when the

  46658.      modification itself materially and adversely affects the operation

  46659.      of the network or violates the rules and protocols for

  46660.      communication across the network.

  46661. 

  46662.      Corresponding Source conveyed, and Installation Information

  46663.      provided, in accord with this section must be in a format that is

  46664.      publicly documented (and with an implementation available to the

  46665.      public in source code form), and must require no special password

  46666.      or key for unpacking, reading or copying.

  46667. 

  46668.   7. Additional Terms.

  46669. 

  46670.      "Additional permissions" are terms that supplement the terms of

  46671.      this License by making exceptions from one or more of its

  46672.      conditions.  Additional permissions that are applicable to the

  46673.      entire Program shall be treated as though they were included in

  46674.      this License, to the extent that they are valid under applicable

  46675.      law.  If additional permissions apply only to part of the Program,

  46676.      that part may be used separately under those permissions, but the

  46677.      entire Program remains governed by this License without regard to

  46678.      the additional permissions.

  46679. 

  46680.      When you convey a copy of a covered work, you may at your option

  46681.      remove any additional permissions from that copy, or from any part

  46682.      of it.  (Additional permissions may be written to require their own

  46683.      removal in certain cases when you modify the work.)  You may place

  46684.      additional permissions on material, added by you to a covered work,

  46685.      for which you have or can give appropriate copyright permission.

  46686. 

  46687.      Notwithstanding any other provision of this License, for material

  46688.      you add to a covered work, you may (if authorized by the copyright

  46689.      holders of that material) supplement the terms of this License with

  46690.      terms:

  46691. 

  46692.        a. Disclaiming warranty or limiting liability differently from

  46693.           the terms of sections 15 and 16 of this License; or

  46694. 

  46695.        b. Requiring preservation of specified reasonable legal notices

  46696.           or author attributions in that material or in the Appropriate

  46697.           Legal Notices displayed by works containing it; or

  46698. 

  46699.        c. Prohibiting misrepresentation of the origin of that material,

  46700.           or requiring that modified versions of such material be marked

  46701.           in reasonable ways as different from the original version; or

  46702. 

  46703.        d. Limiting the use for publicity purposes of names of licensors

  46704.           or authors of the material; or

  46705. 

  46706.        e. Declining to grant rights under trademark law for use of some

  46707.           trade names, trademarks, or service marks; or

  46708. 

  46709.        f. Requiring indemnification of licensors and authors of that

  46710.           material by anyone who conveys the material (or modified

  46711.           versions of it) with contractual assumptions of liability to

  46712.           the recipient, for any liability that these contractual

  46713.           assumptions directly impose on those licensors and authors.

  46714. 

  46715.      All other non-permissive additional terms are considered "further

  46716.      restrictions" within the meaning of section 10.  If the Program as

  46717.      you received it, or any part of it, contains a notice stating that

  46718.      it is governed by this License along with a term that is a further

  46719.      restriction, you may remove that term.  If a license document

  46720.      contains a further restriction but permits relicensing or conveying

  46721.      under this License, you may add to a covered work material governed

  46722.      by the terms of that license document, provided that the further

  46723.      restriction does not survive such relicensing or conveying.

  46724. 

  46725.      If you add terms to a covered work in accord with this section, you

  46726.      must place, in the relevant source files, a statement of the

  46727.      additional terms that apply to those files, or a notice indicating

  46728.      where to find the applicable terms.

  46729. 

  46730.      Additional terms, permissive or non-permissive, may be stated in

  46731.      the form of a separately written license, or stated as exceptions;

  46732.      the above requirements apply either way.

  46733. 

  46734.   8. Termination.

  46735. 

  46736.      You may not propagate or modify a covered work except as expressly

  46737.      provided under this License.  Any attempt otherwise to propagate or

  46738.      modify it is void, and will automatically terminate your rights

  46739.      under this License (including any patent licenses granted under the

  46740.      third paragraph of section 11).

  46741. 

  46742.      However, if you cease all violation of this License, then your

  46743.      license from a particular copyright holder is reinstated (a)

  46744.      provisionally, unless and until the copyright holder explicitly and

  46745.      finally terminates your license, and (b) permanently, if the

  46746.      copyright holder fails to notify you of the violation by some

  46747.      reasonable means prior to 60 days after the cessation.

  46748. 

  46749.      Moreover, your license from a particular copyright holder is

  46750.      reinstated permanently if the copyright holder notifies you of the

  46751.      violation by some reasonable means, this is the first time you have

  46752.      received notice of violation of this License (for any work) from

  46753.      that copyright holder, and you cure the violation prior to 30 days

  46754.      after your receipt of the notice.

  46755. 

  46756.      Termination of your rights under this section does not terminate

  46757.      the licenses of parties who have received copies or rights from you

  46758.      under this License.  If your rights have been terminated and not

  46759.      permanently reinstated, you do not qualify to receive new licenses

  46760.      for the same material under section 10.

  46761. 

  46762.   9. Acceptance Not Required for Having Copies.

  46763. 

  46764.      You are not required to accept this License in order to receive or

  46765.      run a copy of the Program.  Ancillary propagation of a covered work

  46766.      occurring solely as a consequence of using peer-to-peer

  46767.      transmission to receive a copy likewise does not require

  46768.      acceptance.  However, nothing other than this License grants you

  46769.      permission to propagate or modify any covered work.  These actions

  46770.      infringe copyright if you do not accept this License.  Therefore,

  46771.      by modifying or propagating a covered work, you indicate your

  46772.      acceptance of this License to do so.

  46773. 

  46774.   10. Automatic Licensing of Downstream Recipients.

  46775. 

  46776.      Each time you convey a covered work, the recipient automatically

  46777.      receives a license from the original licensors, to run, modify and

  46778.      propagate that work, subject to this License.  You are not

  46779.      responsible for enforcing compliance by third parties with this

  46780.      License.

  46781. 

  46782.      An "entity transaction" is a transaction transferring control of an

  46783.      organization, or substantially all assets of one, or subdividing an

  46784.      organization, or merging organizations.  If propagation of a

  46785.      covered work results from an entity transaction, each party to that

  46786.      transaction who receives a copy of the work also receives whatever

  46787.      licenses to the work the party's predecessor in interest had or

  46788.      could give under the previous paragraph, plus a right to possession

  46789.      of the Corresponding Source of the work from the predecessor in

  46790.      interest, if the predecessor has it or can get it with reasonable

  46791.      efforts.

  46792. 

  46793.      You may not impose any further restrictions on the exercise of the

  46794.      rights granted or affirmed under this License.  For example, you

  46795.      may not impose a license fee, royalty, or other charge for exercise

  46796.      of rights granted under this License, and you may not initiate

  46797.      litigation (including a cross-claim or counterclaim in a lawsuit)

  46798.      alleging that any patent claim is infringed by making, using,

  46799.      selling, offering for sale, or importing the Program or any portion

  46800.      of it.

  46801. 

  46802.   11. Patents.

  46803. 

  46804.      A "contributor" is a copyright holder who authorizes use under this

  46805.      License of the Program or a work on which the Program is based.

  46806.      The work thus licensed is called the contributor's "contributor

  46807.      version".

  46808. 

  46809.      A contributor's "essential patent claims" are all patent claims

  46810.      owned or controlled by the contributor, whether already acquired or

  46811.      hereafter acquired, that would be infringed by some manner,

  46812.      permitted by this License, of making, using, or selling its

  46813.      contributor version, but do not include claims that would be

  46814.      infringed only as a consequence of further modification of the

  46815.      contributor version.  For purposes of this definition, "control"

  46816.      includes the right to grant patent sublicenses in a manner

  46817.      consistent with the requirements of this License.

  46818. 

  46819.      Each contributor grants you a non-exclusive, worldwide,

  46820.      royalty-free patent license under the contributor's essential

  46821.      patent claims, to make, use, sell, offer for sale, import and

  46822.      otherwise run, modify and propagate the contents of its contributor

  46823.      version.

  46824. 

  46825.      In the following three paragraphs, a "patent license" is any

  46826.      express agreement or commitment, however denominated, not to

  46827.      enforce a patent (such as an express permission to practice a

  46828.      patent or covenant not to sue for patent infringement).  To "grant"

  46829.      such a patent license to a party means to make such an agreement or

  46830.      commitment not to enforce a patent against the party.

  46831. 

  46832.      If you convey a covered work, knowingly relying on a patent

  46833.      license, and the Corresponding Source of the work is not available

  46834.      for anyone to copy, free of charge and under the terms of this

  46835.      License, through a publicly available network server or other

  46836.      readily accessible means, then you must either (1) cause the

  46837.      Corresponding Source to be so available, or (2) arrange to deprive

  46838.      yourself of the benefit of the patent license for this particular

  46839.      work, or (3) arrange, in a manner consistent with the requirements

  46840.      of this License, to extend the patent license to downstream

  46841.      recipients.  "Knowingly relying" means you have actual knowledge

  46842.      that, but for the patent license, your conveying the covered work

  46843.      in a country, or your recipient's use of the covered work in a

  46844.      country, would infringe one or more identifiable patents in that

  46845.      country that you have reason to believe are valid.

  46846. 

  46847.      If, pursuant to or in connection with a single transaction or

  46848.      arrangement, you convey, or propagate by procuring conveyance of, a

  46849.      covered work, and grant a patent license to some of the parties

  46850.      receiving the covered work authorizing them to use, propagate,

  46851.      modify or convey a specific copy of the covered work, then the

  46852.      patent license you grant is automatically extended to all

  46853.      recipients of the covered work and works based on it.

  46854. 

  46855.      A patent license is "discriminatory" if it does not include within

  46856.      the scope of its coverage, prohibits the exercise of, or is

  46857.      conditioned on the non-exercise of one or more of the rights that

  46858.      are specifically granted under this License.  You may not convey a

  46859.      covered work if you are a party to an arrangement with a third

  46860.      party that is in the business of distributing software, under which

  46861.      you make payment to the third party based on the extent of your

  46862.      activity of conveying the work, and under which the third party

  46863.      grants, to any of the parties who would receive the covered work

  46864.      from you, a discriminatory patent license (a) in connection with

  46865.      copies of the covered work conveyed by you (or copies made from

  46866.      those copies), or (b) primarily for and in connection with specific

  46867.      products or compilations that contain the covered work, unless you

  46868.      entered into that arrangement, or that patent license was granted,

  46869.      prior to 28 March 2007.

  46870. 

  46871.      Nothing in this License shall be construed as excluding or limiting

  46872.      any implied license or other defenses to infringement that may

  46873.      otherwise be available to you under applicable patent law.

  46874. 

  46875.   12. No Surrender of Others' Freedom.

  46876. 

  46877.      If conditions are imposed on you (whether by court order, agreement

  46878.      or otherwise) that contradict the conditions of this License, they

  46879.      do not excuse you from the conditions of this License.  If you

  46880.      cannot convey a covered work so as to satisfy simultaneously your

  46881.      obligations under this License and any other pertinent obligations,

  46882.      then as a consequence you may not convey it at all.  For example,

  46883.      if you agree to terms that obligate you to collect a royalty for

  46884.      further conveying from those to whom you convey the Program, the

  46885.      only way you could satisfy both those terms and this License would

  46886.      be to refrain entirely from conveying the Program.

  46887. 

  46888.   13. Use with the GNU Affero General Public License.

  46889. 

  46890.      Notwithstanding any other provision of this License, you have

  46891.      permission to link or combine any covered work with a work licensed

  46892.      under version 3 of the GNU Affero General Public License into a

  46893.      single combined work, and to convey the resulting work.  The terms

  46894.      of this License will continue to apply to the part which is the

  46895.      covered work, but the special requirements of the GNU Affero

  46896.      General Public License, section 13, concerning interaction through

  46897.      a network will apply to the combination as such.

  46898. 

  46899.   14. Revised Versions of this License.

  46900. 

  46901.      The Free Software Foundation may publish revised and/or new

  46902.      versions of the GNU General Public License from time to time.  Such

  46903.      new versions will be similar in spirit to the present version, but

  46904.      may differ in detail to address new problems or concerns.

  46905. 

  46906.      Each version is given a distinguishing version number.  If the

  46907.      Program specifies that a certain numbered version of the GNU

  46908.      General Public License "or any later version" applies to it, you

  46909.      have the option of following the terms and conditions either of

  46910.      that numbered version or of any later version published by the Free

  46911.      Software Foundation.  If the Program does not specify a version

  46912.      number of the GNU General Public License, you may choose any

  46913.      version ever published by the Free Software Foundation.

  46914. 

  46915.      If the Program specifies that a proxy can decide which future

  46916.      versions of the GNU General Public License can be used, that

  46917.      proxy's public statement of acceptance of a version permanently

  46918.      authorizes you to choose that version for the Program.

  46919. 

  46920.      Later license versions may give you additional or different

  46921.      permissions.  However, no additional obligations are imposed on any

  46922.      author or copyright holder as a result of your choosing to follow a

  46923.      later version.

  46924. 

  46925.   15. Disclaimer of Warranty.

  46926. 

  46927.      THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY

  46928.      APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE

  46929.      COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS"

  46930.      WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED,

  46931.      INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF

  46932.      MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE

  46933.      RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU.

  46934.      SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL

  46935.      NECESSARY SERVICING, REPAIR OR CORRECTION.

  46936. 

  46937.   16. Limitation of Liability.

  46938. 

  46939.      IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN

  46940.      WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES

  46941.      AND/OR CONVEYS THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR

  46942.      DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR

  46943.      CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE

  46944.      THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA

  46945.      BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD

  46946.      PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER

  46947.      PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF

  46948.      THE POSSIBILITY OF SUCH DAMAGES.

  46949. 

  46950.   17. Interpretation of Sections 15 and 16.

  46951. 

  46952.      If the disclaimer of warranty and limitation of liability provided

  46953.      above cannot be given local legal effect according to their terms,

  46954.      reviewing courts shall apply local law that most closely

  46955.      approximates an absolute waiver of all civil liability in

  46956.      connection with the Program, unless a warranty or assumption of

  46957.      liability accompanies a copy of the Program in return for a fee.

  46958. 

  46959. END OF TERMS AND CONDITIONS

  46960. ===========================

  46961. 

  46962. How to Apply These Terms to Your New Programs

  46963. =============================================

  46964. 

  46965. If you develop a new program, and you want it to be of the greatest

  46966. possible use to the public, the best way to achieve this is to make it

  46967. free software which everyone can redistribute and change under these

  46968. terms.

  46969. 

  46970.  To do so, attach the following notices to the program.  It is safest to

  46971. attach them to the start of each source file to most effectively state

  46972. the exclusion of warranty; and each file should have at least the

  46973. "copyright" line and a pointer to where the full notice is found.

  46974. 

  46975.      ONE LINE TO GIVE THE PROGRAM'S NAME AND A BRIEF IDEA OF WHAT IT DOES.

  46976.      Copyright (C) YEAR NAME OF AUTHOR

  46977. 

  46978.      This program is free software: you can redistribute it and/or modify

  46979.      it under the terms of the GNU General Public License as published by

  46980.      the Free Software Foundation, either version 3 of the License, or (at

  46981.      your option) any later version.

  46982. 

  46983.      This program is distributed in the hope that it will be useful, but

  46984.      WITHOUT ANY WARRANTY; without even the implied warranty of

  46985.      MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  46986.      General Public License for more details.

  46987. 

  46988.      You should have received a copy of the GNU General Public License

  46989.      along with this program.  If not, see <http://www.gnu.org/licenses/>.

  46990. 

  46991.  Also add information on how to contact you by electronic and paper

  46992. mail.

  46993. 

  46994.  If the program does terminal interaction, make it output a short notice

  46995. like this when it starts in an interactive mode:

  46996. 

  46997.      PROGRAM Copyright (C) YEAR NAME OF AUTHOR

  46998.      This program comes with ABSOLUTELY NO WARRANTY; for details type 'show w'.

  46999.      This is free software, and you are welcome to redistribute it

  47000.      under certain conditions; type 'show c' for details.

  47001. 

  47002.  The hypothetical commands 'show w' and 'show c' should show the

  47003. appropriate parts of the General Public License.  Of course, your

  47004. program's commands might be different; for a GUI interface, you would

  47005. use an "about box".

  47006. 

  47007.  You should also get your employer (if you work as a programmer) or

  47008. school, if any, to sign a "copyright disclaimer" for the program, if

  47009. necessary.  For more information on this, and how to apply and follow

  47010. the GNU GPL, see <http://www.gnu.org/licenses/>.

  47011. 

  47012.  The GNU General Public License does not permit incorporating your

  47013. program into proprietary programs.  If your program is a subroutine

  47014. library, you may consider it more useful to permit linking proprietary

  47015. applications with the library.  If this is what you want to do, use the

  47016. GNU Lesser General Public License instead of this License.  But first,

  47017. please read <https://www.gnu.org/licenses/why-not-lgpl.html>.

  47018. 

  47019. 

  47020. File: gccint.info,  Node: GNU Free Documentation License,  Next: Contributors,  Prev: Copying,  Up: Top

  47021. 

  47022. GNU Free Documentation License

  47023. ******************************

  47024. 

  47025.                      Version 1.3, 3 November 2008

  47026. 

  47027.      Copyright (C) 2000, 2001, 2002, 2007, 2008 Free Software Foundation, Inc.

  47028.      <http://fsf.org/>

  47029. 

  47030.      Everyone is permitted to copy and distribute verbatim copies

  47031.      of this license document, but changing it is not allowed.

  47032. 

  47033.   0. PREAMBLE

  47034. 

  47035.      The purpose of this License is to make a manual, textbook, or other

  47036.      functional and useful document "free" in the sense of freedom: to

  47037.      assure everyone the effective freedom to copy and redistribute it,

  47038.      with or without modifying it, either commercially or

  47039.      noncommercially.  Secondarily, this License preserves for the

  47040.      author and publisher a way to get credit for their work, while not

  47041.      being considered responsible for modifications made by others.

  47042. 

  47043.      This License is a kind of "copyleft", which means that derivative

  47044.      works of the document must themselves be free in the same sense.

  47045.      It complements the GNU General Public License, which is a copyleft

  47046.      license designed for free software.

  47047. 

  47048.      We have designed this License in order to use it for manuals for

  47049.      free software, because free software needs free documentation: a

  47050.      free program should come with manuals providing the same freedoms

  47051.      that the software does.  But this License is not limited to

  47052.      software manuals; it can be used for any textual work, regardless

  47053.      of subject matter or whether it is published as a printed book.  We

  47054.      recommend this License principally for works whose purpose is

  47055.      instruction or reference.

  47056. 

  47057.   1. APPLICABILITY AND DEFINITIONS

  47058. 

  47059.      This License applies to any manual or other work, in any medium,

  47060.      that contains a notice placed by the copyright holder saying it can

  47061.      be distributed under the terms of this License.  Such a notice

  47062.      grants a world-wide, royalty-free license, unlimited in duration,

  47063.      to use that work under the conditions stated herein.  The

  47064.      "Document", below, refers to any such manual or work.  Any member

  47065.      of the public is a licensee, and is addressed as "you".  You accept

  47066.      the license if you copy, modify or distribute the work in a way

  47067.      requiring permission under copyright law.

  47068. 

  47069.      A "Modified Version" of the Document means any work containing the

  47070.      Document or a portion of it, either copied verbatim, or with

  47071.      modifications and/or translated into another language.

  47072. 

  47073.      A "Secondary Section" is a named appendix or a front-matter section

  47074.      of the Document that deals exclusively with the relationship of the

  47075.      publishers or authors of the Document to the Document's overall

  47076.      subject (or to related matters) and contains nothing that could

  47077.      fall directly within that overall subject.  (Thus, if the Document

  47078.      is in part a textbook of mathematics, a Secondary Section may not

  47079.      explain any mathematics.)  The relationship could be a matter of

  47080.      historical connection with the subject or with related matters, or

  47081.      of legal, commercial, philosophical, ethical or political position

  47082.      regarding them.

  47083. 

  47084.      The "Invariant Sections" are certain Secondary Sections whose

  47085.      titles are designated, as being those of Invariant Sections, in the

  47086.      notice that says that the Document is released under this License.

  47087.      If a section does not fit the above definition of Secondary then it

  47088.      is not allowed to be designated as Invariant.  The Document may

  47089.      contain zero Invariant Sections.  If the Document does not identify

  47090.      any Invariant Sections then there are none.

  47091. 

  47092.      The "Cover Texts" are certain short passages of text that are

  47093.      listed, as Front-Cover Texts or Back-Cover Texts, in the notice

  47094.      that says that the Document is released under this License.  A

  47095.      Front-Cover Text may be at most 5 words, and a Back-Cover Text may

  47096.      be at most 25 words.

  47097. 

  47098.      A "Transparent" copy of the Document means a machine-readable copy,

  47099.      represented in a format whose specification is available to the

  47100.      general public, that is suitable for revising the document

  47101.      straightforwardly with generic text editors or (for images composed

  47102.      of pixels) generic paint programs or (for drawings) some widely

  47103.      available drawing editor, and that is suitable for input to text

  47104.      formatters or for automatic translation to a variety of formats

  47105.      suitable for input to text formatters.  A copy made in an otherwise

  47106.      Transparent file format whose markup, or absence of markup, has

  47107.      been arranged to thwart or discourage subsequent modification by

  47108.      readers is not Transparent.  An image format is not Transparent if

  47109.      used for any substantial amount of text.  A copy that is not

  47110.      "Transparent" is called "Opaque".

  47111. 

  47112.      Examples of suitable formats for Transparent copies include plain

  47113.      ASCII without markup, Texinfo input format, LaTeX input format,

  47114.      SGML or XML using a publicly available DTD, and standard-conforming

  47115.      simple HTML, PostScript or PDF designed for human modification.

  47116.      Examples of transparent image formats include PNG, XCF and JPG.

  47117.      Opaque formats include proprietary formats that can be read and

  47118.      edited only by proprietary word processors, SGML or XML for which

  47119.      the DTD and/or processing tools are not generally available, and

  47120.      the machine-generated HTML, PostScript or PDF produced by some word

  47121.      processors for output purposes only.

  47122. 

  47123.      The "Title Page" means, for a printed book, the title page itself,

  47124.      plus such following pages as are needed to hold, legibly, the

  47125.      material this License requires to appear in the title page.  For

  47126.      works in formats which do not have any title page as such, "Title

  47127.      Page" means the text near the most prominent appearance of the

  47128.      work's title, preceding the beginning of the body of the text.

  47129. 

  47130.      The "publisher" means any person or entity that distributes copies

  47131.      of the Document to the public.

  47132. 

  47133.      A section "Entitled XYZ" means a named subunit of the Document

  47134.      whose title either is precisely XYZ or contains XYZ in parentheses

  47135.      following text that translates XYZ in another language.  (Here XYZ

  47136.      stands for a specific section name mentioned below, such as

  47137.      "Acknowledgements", "Dedications", "Endorsements", or "History".)

  47138.      To "Preserve the Title" of such a section when you modify the

  47139.      Document means that it remains a section "Entitled XYZ" according

  47140.      to this definition.

  47141. 

  47142.      The Document may include Warranty Disclaimers next to the notice

  47143.      which states that this License applies to the Document.  These

  47144.      Warranty Disclaimers are considered to be included by reference in

  47145.      this License, but only as regards disclaiming warranties: any other

  47146.      implication that these Warranty Disclaimers may have is void and

  47147.      has no effect on the meaning of this License.

  47148. 

  47149.   2. VERBATIM COPYING

  47150. 

  47151.      You may copy and distribute the Document in any medium, either

  47152.      commercially or noncommercially, provided that this License, the

  47153.      copyright notices, and the license notice saying this License

  47154.      applies to the Document are reproduced in all copies, and that you

  47155.      add no other conditions whatsoever to those of this License.  You

  47156.      may not use technical measures to obstruct or control the reading

  47157.      or further copying of the copies you make or distribute.  However,

  47158.      you may accept compensation in exchange for copies.  If you

  47159.      distribute a large enough number of copies you must also follow the

  47160.      conditions in section 3.

  47161. 

  47162.      You may also lend copies, under the same conditions stated above,

  47163.      and you may publicly display copies.

  47164. 

  47165.   3. COPYING IN QUANTITY

  47166. 

  47167.      If you publish printed copies (or copies in media that commonly

  47168.      have printed covers) of the Document, numbering more than 100, and

  47169.      the Document's license notice requires Cover Texts, you must

  47170.      enclose the copies in covers that carry, clearly and legibly, all

  47171.      these Cover Texts: Front-Cover Texts on the front cover, and

  47172.      Back-Cover Texts on the back cover.  Both covers must also clearly

  47173.      and legibly identify you as the publisher of these copies.  The

  47174.      front cover must present the full title with all words of the title

  47175.      equally prominent and visible.  You may add other material on the

  47176.      covers in addition.  Copying with changes limited to the covers, as

  47177.      long as they preserve the title of the Document and satisfy these

  47178.      conditions, can be treated as verbatim copying in other respects.

  47179. 

  47180.      If the required texts for either cover are too voluminous to fit

  47181.      legibly, you should put the first ones listed (as many as fit

  47182.      reasonably) on the actual cover, and continue the rest onto

  47183.      adjacent pages.

  47184. 

  47185.      If you publish or distribute Opaque copies of the Document

  47186.      numbering more than 100, you must either include a machine-readable

  47187.      Transparent copy along with each Opaque copy, or state in or with

  47188.      each Opaque copy a computer-network location from which the general

  47189.      network-using public has access to download using public-standard

  47190.      network protocols a complete Transparent copy of the Document, free

  47191.      of added material.  If you use the latter option, you must take

  47192.      reasonably prudent steps, when you begin distribution of Opaque

  47193.      copies in quantity, to ensure that this Transparent copy will

  47194.      remain thus accessible at the stated location until at least one

  47195.      year after the last time you distribute an Opaque copy (directly or

  47196.      through your agents or retailers) of that edition to the public.

  47197. 

  47198.      It is requested, but not required, that you contact the authors of

  47199.      the Document well before redistributing any large number of copies,

  47200.      to give them a chance to provide you with an updated version of the

  47201.      Document.

  47202. 

  47203.   4. MODIFICATIONS

  47204. 

  47205.      You may copy and distribute a Modified Version of the Document

  47206.      under the conditions of sections 2 and 3 above, provided that you

  47207.      release the Modified Version under precisely this License, with the

  47208.      Modified Version filling the role of the Document, thus licensing

  47209.      distribution and modification of the Modified Version to whoever

  47210.      possesses a copy of it.  In addition, you must do these things in

  47211.      the Modified Version:

  47212. 

  47213.        A. Use in the Title Page (and on the covers, if any) a title

  47214.           distinct from that of the Document, and from those of previous

  47215.           versions (which should, if there were any, be listed in the

  47216.           History section of the Document).  You may use the same title

  47217.           as a previous version if the original publisher of that

  47218.           version gives permission.

  47219. 

  47220.        B. List on the Title Page, as authors, one or more persons or

  47221.           entities responsible for authorship of the modifications in

  47222.           the Modified Version, together with at least five of the

  47223.           principal authors of the Document (all of its principal

  47224.           authors, if it has fewer than five), unless they release you

  47225.           from this requirement.

  47226. 

  47227.        C. State on the Title page the name of the publisher of the

  47228.           Modified Version, as the publisher.

  47229. 

  47230.        D. Preserve all the copyright notices of the Document.

  47231. 

  47232.        E. Add an appropriate copyright notice for your modifications

  47233.           adjacent to the other copyright notices.

  47234. 

  47235.        F. Include, immediately after the copyright notices, a license

  47236.           notice giving the public permission to use the Modified

  47237.           Version under the terms of this License, in the form shown in

  47238.           the Addendum below.

  47239. 

  47240.        G. Preserve in that license notice the full lists of Invariant

  47241.           Sections and required Cover Texts given in the Document's

  47242.           license notice.

  47243. 

  47244.        H. Include an unaltered copy of this License.

  47245. 

  47246.        I. Preserve the section Entitled "History", Preserve its Title,

  47247.           and add to it an item stating at least the title, year, new

  47248.           authors, and publisher of the Modified Version as given on the

  47249.           Title Page.  If there is no section Entitled "History" in the

  47250.           Document, create one stating the title, year, authors, and

  47251.           publisher of the Document as given on its Title Page, then add

  47252.           an item describing the Modified Version as stated in the

  47253.           previous sentence.

  47254. 

  47255.        J. Preserve the network location, if any, given in the Document

  47256.           for public access to a Transparent copy of the Document, and

  47257.           likewise the network locations given in the Document for

  47258.           previous versions it was based on.  These may be placed in the

  47259.           "History" section.  You may omit a network location for a work

  47260.           that was published at least four years before the Document

  47261.           itself, or if the original publisher of the version it refers

  47262.           to gives permission.

  47263. 

  47264.        K. For any section Entitled "Acknowledgements" or "Dedications",

  47265.           Preserve the Title of the section, and preserve in the section

  47266.           all the substance and tone of each of the contributor

  47267.           acknowledgements and/or dedications given therein.

  47268. 

  47269.        L. Preserve all the Invariant Sections of the Document, unaltered

  47270.           in their text and in their titles.  Section numbers or the

  47271.           equivalent are not considered part of the section titles.

  47272. 

  47273.        M. Delete any section Entitled "Endorsements".  Such a section

  47274.           may not be included in the Modified Version.

  47275. 

  47276.        N. Do not retitle any existing section to be Entitled

  47277.           "Endorsements" or to conflict in title with any Invariant

  47278.           Section.

  47279. 

  47280.        O. Preserve any Warranty Disclaimers.

  47281. 

  47282.      If the Modified Version includes new front-matter sections or

  47283.      appendices that qualify as Secondary Sections and contain no

  47284.      material copied from the Document, you may at your option designate

  47285.      some or all of these sections as invariant.  To do this, add their

  47286.      titles to the list of Invariant Sections in the Modified Version's

  47287.      license notice.  These titles must be distinct from any other

  47288.      section titles.

  47289. 

  47290.      You may add a section Entitled "Endorsements", provided it contains

  47291.      nothing but endorsements of your Modified Version by various

  47292.      parties--for example, statements of peer review or that the text

  47293.      has been approved by an organization as the authoritative

  47294.      definition of a standard.

  47295. 

  47296.      You may add a passage of up to five words as a Front-Cover Text,

  47297.      and a passage of up to 25 words as a Back-Cover Text, to the end of

  47298.      the list of Cover Texts in the Modified Version.  Only one passage

  47299.      of Front-Cover Text and one of Back-Cover Text may be added by (or

  47300.      through arrangements made by) any one entity.  If the Document

  47301.      already includes a cover text for the same cover, previously added

  47302.      by you or by arrangement made by the same entity you are acting on

  47303.      behalf of, you may not add another; but you may replace the old

  47304.      one, on explicit permission from the previous publisher that added

  47305.      the old one.

  47306. 

  47307.      The author(s) and publisher(s) of the Document do not by this

  47308.      License give permission to use their names for publicity for or to

  47309.      assert or imply endorsement of any Modified Version.

  47310. 

  47311.   5. COMBINING DOCUMENTS

  47312. 

  47313.      You may combine the Document with other documents released under

  47314.      this License, under the terms defined in section 4 above for

  47315.      modified versions, provided that you include in the combination all

  47316.      of the Invariant Sections of all of the original documents,

  47317.      unmodified, and list them all as Invariant Sections of your

  47318.      combined work in its license notice, and that you preserve all

  47319.      their Warranty Disclaimers.

  47320. 

  47321.      The combined work need only contain one copy of this License, and

  47322.      multiple identical Invariant Sections may be replaced with a single

  47323.      copy.  If there are multiple Invariant Sections with the same name

  47324.      but different contents, make the title of each such section unique

  47325.      by adding at the end of it, in parentheses, the name of the

  47326.      original author or publisher of that section if known, or else a

  47327.      unique number.  Make the same adjustment to the section titles in

  47328.      the list of Invariant Sections in the license notice of the

  47329.      combined work.

  47330. 

  47331.      In the combination, you must combine any sections Entitled

  47332.      "History" in the various original documents, forming one section

  47333.      Entitled "History"; likewise combine any sections Entitled

  47334.      "Acknowledgements", and any sections Entitled "Dedications".  You

  47335.      must delete all sections Entitled "Endorsements."

  47336. 

  47337.   6. COLLECTIONS OF DOCUMENTS

  47338. 

  47339.      You may make a collection consisting of the Document and other

  47340.      documents released under this License, and replace the individual

  47341.      copies of this License in the various documents with a single copy

  47342.      that is included in the collection, provided that you follow the

  47343.      rules of this License for verbatim copying of each of the documents

  47344.      in all other respects.

  47345. 

  47346.      You may extract a single document from such a collection, and

  47347.      distribute it individually under this License, provided you insert

  47348.      a copy of this License into the extracted document, and follow this

  47349.      License in all other respects regarding verbatim copying of that

  47350.      document.

  47351. 

  47352.   7. AGGREGATION WITH INDEPENDENT WORKS

  47353. 

  47354.      A compilation of the Document or its derivatives with other

  47355.      separate and independent documents or works, in or on a volume of a

  47356.      storage or distribution medium, is called an "aggregate" if the

  47357.      copyright resulting from the compilation is not used to limit the

  47358.      legal rights of the compilation's users beyond what the individual

  47359.      works permit.  When the Document is included in an aggregate, this

  47360.      License does not apply to the other works in the aggregate which

  47361.      are not themselves derivative works of the Document.

  47362. 

  47363.      If the Cover Text requirement of section 3 is applicable to these

  47364.      copies of the Document, then if the Document is less than one half

  47365.      of the entire aggregate, the Document's Cover Texts may be placed

  47366.      on covers that bracket the Document within the aggregate, or the

  47367.      electronic equivalent of covers if the Document is in electronic

  47368.      form.  Otherwise they must appear on printed covers that bracket

  47369.      the whole aggregate.

  47370. 

  47371.   8. TRANSLATION

  47372. 

  47373.      Translation is considered a kind of modification, so you may

  47374.      distribute translations of the Document under the terms of section

  47375.      4.  Replacing Invariant Sections with translations requires special

  47376.      permission from their copyright holders, but you may include

  47377.      translations of some or all Invariant Sections in addition to the

  47378.      original versions of these Invariant Sections.  You may include a

  47379.      translation of this License, and all the license notices in the

  47380.      Document, and any Warranty Disclaimers, provided that you also

  47381.      include the original English version of this License and the

  47382.      original versions of those notices and disclaimers.  In case of a

  47383.      disagreement between the translation and the original version of

  47384.      this License or a notice or disclaimer, the original version will

  47385.      prevail.

  47386. 

  47387.      If a section in the Document is Entitled "Acknowledgements",

  47388.      "Dedications", or "History", the requirement (section 4) to

  47389.      Preserve its Title (section 1) will typically require changing the

  47390.      actual title.

  47391. 

  47392.   9. TERMINATION

  47393. 

  47394.      You may not copy, modify, sublicense, or distribute the Document

  47395.      except as expressly provided under this License.  Any attempt

  47396.      otherwise to copy, modify, sublicense, or distribute it is void,

  47397.      and will automatically terminate your rights under this License.

  47398. 

  47399.      However, if you cease all violation of this License, then your

  47400.      license from a particular copyright holder is reinstated (a)

  47401.      provisionally, unless and until the copyright holder explicitly and

  47402.      finally terminates your license, and (b) permanently, if the

  47403.      copyright holder fails to notify you of the violation by some

  47404.      reasonable means prior to 60 days after the cessation.

  47405. 

  47406.      Moreover, your license from a particular copyright holder is

  47407.      reinstated permanently if the copyright holder notifies you of the

  47408.      violation by some reasonable means, this is the first time you have

  47409.      received notice of violation of this License (for any work) from

  47410.      that copyright holder, and you cure the violation prior to 30 days

  47411.      after your receipt of the notice.

  47412. 

  47413.      Termination of your rights under this section does not terminate

  47414.      the licenses of parties who have received copies or rights from you

  47415.      under this License.  If your rights have been terminated and not

  47416.      permanently reinstated, receipt of a copy of some or all of the

  47417.      same material does not give you any rights to use it.

  47418. 

  47419.   10. FUTURE REVISIONS OF THIS LICENSE

  47420. 

  47421.      The Free Software Foundation may publish new, revised versions of

  47422.      the GNU Free Documentation License from time to time.  Such new

  47423.      versions will be similar in spirit to the present version, but may

  47424.      differ in detail to address new problems or concerns.  See

  47425.      <http://www.gnu.org/copyleft/>.

  47426. 

  47427.      Each version of the License is given a distinguishing version

  47428.      number.  If the Document specifies that a particular numbered

  47429.      version of this License "or any later version" applies to it, you

  47430.      have the option of following the terms and conditions either of

  47431.      that specified version or of any later version that has been

  47432.      published (not as a draft) by the Free Software Foundation.  If the

  47433.      Document does not specify a version number of this License, you may

  47434.      choose any version ever published (not as a draft) by the Free

  47435.      Software Foundation.  If the Document specifies that a proxy can

  47436.      decide which future versions of this License can be used, that

  47437.      proxy's public statement of acceptance of a version permanently

  47438.      authorizes you to choose that version for the Document.

  47439. 

  47440.   11. RELICENSING

  47441. 

  47442.      "Massive Multiauthor Collaboration Site" (or "MMC Site") means any

  47443.      World Wide Web server that publishes copyrightable works and also

  47444.      provides prominent facilities for anybody to edit those works.  A

  47445.      public wiki that anybody can edit is an example of such a server.

  47446.      A "Massive Multiauthor Collaboration" (or "MMC") contained in the

  47447.      site means any set of copyrightable works thus published on the MMC

  47448.      site.

  47449. 

  47450.      "CC-BY-SA" means the Creative Commons Attribution-Share Alike 3.0

  47451.      license published by Creative Commons Corporation, a not-for-profit

  47452.      corporation with a principal place of business in San Francisco,

  47453.      California, as well as future copyleft versions of that license

  47454.      published by that same organization.

  47455. 

  47456.      "Incorporate" means to publish or republish a Document, in whole or

  47457.      in part, as part of another Document.

  47458. 

  47459.      An MMC is "eligible for relicensing" if it is licensed under this

  47460.      License, and if all works that were first published under this

  47461.      License somewhere other than this MMC, and subsequently

  47462.      incorporated in whole or in part into the MMC, (1) had no cover

  47463.      texts or invariant sections, and (2) were thus incorporated prior

  47464.      to November 1, 2008.

  47465. 

  47466.      The operator of an MMC Site may republish an MMC contained in the

  47467.      site under CC-BY-SA on the same site at any time before August 1,

  47468.      2009, provided the MMC is eligible for relicensing.

  47469. 

  47470. ADDENDUM: How to use this License for your documents

  47471. ====================================================

  47472. 

  47473. To use this License in a document you have written, include a copy of

  47474. the License in the document and put the following copyright and license

  47475. notices just after the title page:

  47476. 

  47477.        Copyright (C)  YEAR  YOUR NAME.

  47478.        Permission is granted to copy, distribute and/or modify this document

  47479.        under the terms of the GNU Free Documentation License, Version 1.3

  47480.        or any later version published by the Free Software Foundation;

  47481.        with no Invariant Sections, no Front-Cover Texts, and no Back-Cover

  47482.        Texts.  A copy of the license is included in the section entitled ``GNU

  47483.        Free Documentation License''.

  47484. 

  47485.  If you have Invariant Sections, Front-Cover Texts and Back-Cover Texts,

  47486. replace the "with...Texts."  line with this:

  47487. 

  47488.          with the Invariant Sections being LIST THEIR TITLES, with

  47489.          the Front-Cover Texts being LIST, and with the Back-Cover Texts

  47490.          being LIST.

  47491. 

  47492.  If you have Invariant Sections without Cover Texts, or some other

  47493. combination of the three, merge those two alternatives to suit the

  47494. situation.

  47495. 

  47496.  If your document contains nontrivial examples of program code, we

  47497. recommend releasing these examples in parallel under your choice of free

  47498. software license, such as the GNU General Public License, to permit

  47499. their use in free software.

  47500. 

  47501. 

  47502. File: gccint.info,  Node: Contributors,  Next: Option Index,  Prev: GNU Free Documentation License,  Up: Top

  47503. 

  47504. Contributors to GCC

  47505. *******************

  47506. 

  47507. The GCC project would like to thank its many contributors.  Without them

  47508. the project would not have been nearly as successful as it has been.

  47509. Any omissions in this list are accidental.  Feel free to contact

  47510. <law@redhat.com> or <gerald@pfeifer.com> if you have been left out or

  47511. some of your contributions are not listed.  Please keep this list in

  47512. alphabetical order.

  47513. 

  47514.    * Analog Devices helped implement the support for complex data types

  47515.      and iterators.

  47516. 

  47517.    * John David Anglin for threading-related fixes and improvements to

  47518.      libstdc++-v3, and the HP-UX port.

  47519. 

  47520.    * James van Artsdalen wrote the code that makes efficient use of the

  47521.      Intel 80387 register stack.

  47522. 

  47523.    * Abramo and Roberto Bagnara for the SysV68 Motorola 3300 Delta

  47524.      Series port.

  47525. 

  47526.    * Alasdair Baird for various bug fixes.

  47527. 

  47528.    * Giovanni Bajo for analyzing lots of complicated C++ problem

  47529.      reports.

  47530. 

  47531.    * Peter Barada for his work to improve code generation for new

  47532.      ColdFire cores.

  47533. 

  47534.    * Gerald Baumgartner added the signature extension to the C++ front

  47535.      end.

  47536. 

  47537.    * Godmar Back for his Java improvements and encouragement.

  47538. 

  47539.    * Scott Bambrough for help porting the Java compiler.

  47540. 

  47541.    * Wolfgang Bangerth for processing tons of bug reports.

  47542. 

  47543.    * Jon Beniston for his Microsoft Windows port of Java and port to

  47544.      Lattice Mico32.

  47545. 

  47546.    * Daniel Berlin for better DWARF 2 support, faster/better

  47547.      optimizations, improved alias analysis, plus migrating GCC to

  47548.      Bugzilla.

  47549. 

  47550.    * Geoff Berry for his Java object serialization work and various

  47551.      patches.

  47552. 

  47553.    * David Binderman tests weekly snapshots of GCC trunk against Fedora

  47554.      Rawhide for several architectures.

  47555. 

  47556.    * Laurynas Biveinis for memory management work and DJGPP port fixes.

  47557. 

  47558.    * Uros Bizjak for the implementation of x87 math built-in functions

  47559.      and for various middle end and i386 back end improvements and bug

  47560.      fixes.

  47561. 

  47562.    * Eric Blake for helping to make GCJ and libgcj conform to the

  47563.      specifications.

  47564. 

  47565.    * Janne Blomqvist for contributions to GNU Fortran.

  47566. 

  47567.    * Hans-J. Boehm for his garbage collector, IA-64 libffi port, and

  47568.      other Java work.

  47569. 

  47570.    * Segher Boessenkool for helping maintain the PowerPC port and the

  47571.      instruction combiner plus various contributions to the middle end.

  47572. 

  47573.    * Neil Booth for work on cpplib, lang hooks, debug hooks and other

  47574.      miscellaneous clean-ups.

  47575. 

  47576.    * Steven Bosscher for integrating the GNU Fortran front end into GCC

  47577.      and for contributing to the tree-ssa branch.

  47578. 

  47579.    * Eric Botcazou for fixing middle- and backend bugs left and right.

  47580. 

  47581.    * Per Bothner for his direction via the steering committee and

  47582.      various improvements to the infrastructure for supporting new

  47583.      languages.  Chill front end implementation.  Initial

  47584.      implementations of cpplib, fix-header, config.guess, libio, and

  47585.      past C++ library (libg++) maintainer.  Dreaming up, designing and

  47586.      implementing much of GCJ.

  47587. 

  47588.    * Devon Bowen helped port GCC to the Tahoe.

  47589. 

  47590.    * Don Bowman for mips-vxworks contributions.

  47591. 

  47592.    * James Bowman for the FT32 port.

  47593. 

  47594.    * Dave Brolley for work on cpplib and Chill.

  47595. 

  47596.    * Paul Brook for work on the ARM architecture and maintaining GNU

  47597.      Fortran.

  47598. 

  47599.    * Robert Brown implemented the support for Encore 32000 systems.

  47600. 

  47601.    * Christian Bruel for improvements to local store elimination.

  47602. 

  47603.    * Herman A.J. ten Brugge for various fixes.

  47604. 

  47605.    * Joerg Brunsmann for Java compiler hacking and help with the GCJ

  47606.      FAQ.

  47607. 

  47608.    * Joe Buck for his direction via the steering committee from its

  47609.      creation to 2013.

  47610. 

  47611.    * Iain Buclaw for the D frontend.

  47612. 

  47613.    * Craig Burley for leadership of the G77 Fortran effort.

  47614. 

  47615.    * Tobias Burnus for contributions to GNU Fortran.

  47616. 

  47617.    * Stephan Buys for contributing Doxygen notes for libstdc++.

  47618. 

  47619.    * Paolo Carlini for libstdc++ work: lots of efficiency improvements

  47620.      to the C++ strings, streambufs and formatted I/O, hard detective

  47621.      work on the frustrating localization issues, and keeping up with

  47622.      the problem reports.

  47623. 

  47624.    * John Carr for his alias work, SPARC hacking, infrastructure

  47625.      improvements, previous contributions to the steering committee,

  47626.      loop optimizations, etc.

  47627. 

  47628.    * Stephane Carrez for 68HC11 and 68HC12 ports.

  47629. 

  47630.    * Steve Chamberlain for support for the Renesas SH and H8 processors

  47631.      and the PicoJava processor, and for GCJ config fixes.

  47632. 

  47633.    * Glenn Chambers for help with the GCJ FAQ.

  47634. 

  47635.    * John-Marc Chandonia for various libgcj patches.

  47636. 

  47637.    * Denis Chertykov for contributing and maintaining the AVR port, the

  47638.      first GCC port for an 8-bit architecture.

  47639. 

  47640.    * Kito Cheng for his work on the RISC-V port, including bringing up

  47641.      the test suite and maintenance.

  47642. 

  47643.    * Scott Christley for his Objective-C contributions.

  47644. 

  47645.    * Eric Christopher for his Java porting help and clean-ups.

  47646. 

  47647.    * Branko Cibej for more warning contributions.

  47648. 

  47649.    * The GNU Classpath project for all of their merged runtime code.

  47650. 

  47651.    * Nick Clifton for arm, mcore, fr30, v850, m32r, msp430 rx work,

  47652.      '--help', and other random hacking.

  47653. 

  47654.    * Michael Cook for libstdc++ cleanup patches to reduce warnings.

  47655. 

  47656.    * R. Kelley Cook for making GCC buildable from a read-only directory

  47657.      as well as other miscellaneous build process and documentation

  47658.      clean-ups.

  47659. 

  47660.    * Ralf Corsepius for SH testing and minor bug fixing.

  47661. 

  47662.    * Franc,ois-Xavier Coudert for contributions to GNU Fortran.

  47663. 

  47664.    * Stan Cox for care and feeding of the x86 port and lots of behind

  47665.      the scenes hacking.

  47666. 

  47667.    * Alex Crain provided changes for the 3b1.

  47668. 

  47669.    * Ian Dall for major improvements to the NS32k port.

  47670. 

  47671.    * Paul Dale for his work to add uClinux platform support to the m68k

  47672.      backend.

  47673. 

  47674.    * Palmer Dabbelt for his work maintaining the RISC-V port.

  47675. 

  47676.    * Dario Dariol contributed the four varieties of sample programs that

  47677.      print a copy of their source.

  47678. 

  47679.    * Russell Davidson for fstream and stringstream fixes in libstdc++.

  47680. 

  47681.    * Bud Davis for work on the G77 and GNU Fortran compilers.

  47682. 

  47683.    * Mo DeJong for GCJ and libgcj bug fixes.

  47684. 

  47685.    * Jerry DeLisle for contributions to GNU Fortran.

  47686. 

  47687.    * DJ Delorie for the DJGPP port, build and libiberty maintenance,

  47688.      various bug fixes, and the M32C, MeP, MSP430, and RL78 ports.

  47689. 

  47690.    * Arnaud Desitter for helping to debug GNU Fortran.

  47691. 

  47692.    * Gabriel Dos Reis for contributions to G++, contributions and

  47693.      maintenance of GCC diagnostics infrastructure, libstdc++-v3,

  47694.      including 'valarray<>', 'complex<>', maintaining the numerics

  47695.      library (including that pesky '<limits>' :-) and keeping up-to-date

  47696.      anything to do with numbers.

  47697. 

  47698.    * Ulrich Drepper for his work on glibc, testing of GCC using glibc,

  47699.      ISO C99 support, CFG dumping support, etc., plus support of the C++

  47700.      runtime libraries including for all kinds of C interface issues,

  47701.      contributing and maintaining 'complex<>', sanity checking and

  47702.      disbursement, configuration architecture, libio maintenance, and

  47703.      early math work.

  47704. 

  47705.    * Franc,ois Dumont for his work on libstdc++-v3, especially

  47706.      maintaining and improving 'debug-mode' and associative and

  47707.      unordered containers.

  47708. 

  47709.    * Zdenek Dvorak for a new loop unroller and various fixes.

  47710. 

  47711.    * Michael Eager for his work on the Xilinx MicroBlaze port.

  47712. 

  47713.    * Richard Earnshaw for his ongoing work with the ARM.

  47714. 

  47715.    * David Edelsohn for his direction via the steering committee,

  47716.      ongoing work with the RS6000/PowerPC port, help cleaning up Haifa

  47717.      loop changes, doing the entire AIX port of libstdc++ with his bare

  47718.      hands, and for ensuring GCC properly keeps working on AIX.

  47719. 

  47720.    * Kevin Ediger for the floating point formatting of num_put::do_put

  47721.      in libstdc++.

  47722. 

  47723.    * Phil Edwards for libstdc++ work including configuration hackery,

  47724.      documentation maintainer, chief breaker of the web pages, the

  47725.      occasional iostream bug fix, and work on shared library symbol

  47726.      versioning.

  47727. 

  47728.    * Paul Eggert for random hacking all over GCC.

  47729. 

  47730.    * Mark Elbrecht for various DJGPP improvements, and for libstdc++

  47731.      configuration support for locales and fstream-related fixes.

  47732. 

  47733.    * Vadim Egorov for libstdc++ fixes in strings, streambufs, and

  47734.      iostreams.

  47735. 

  47736.    * Christian Ehrhardt for dealing with bug reports.

  47737. 

  47738.    * Ben Elliston for his work to move the Objective-C runtime into its

  47739.      own subdirectory and for his work on autoconf.

  47740. 

  47741.    * Revital Eres for work on the PowerPC 750CL port.

  47742. 

  47743.    * Marc Espie for OpenBSD support.

  47744. 

  47745.    * Doug Evans for much of the global optimization framework, arc,

  47746.      m32r, and SPARC work.

  47747. 

  47748.    * Christopher Faylor for his work on the Cygwin port and for caring

  47749.      and feeding the gcc.gnu.org box and saving its users tons of spam.

  47750. 

  47751.    * Fred Fish for BeOS support and Ada fixes.

  47752. 

  47753.    * Ivan Fontes Garcia for the Portuguese translation of the GCJ FAQ.

  47754. 

  47755.    * Peter Gerwinski for various bug fixes and the Pascal front end.

  47756. 

  47757.    * Kaveh R. Ghazi for his direction via the steering committee,

  47758.      amazing work to make '-W -Wall -W* -Werror' useful, and testing GCC

  47759.      on a plethora of platforms.  Kaveh extends his gratitude to the

  47760.      CAIP Center at Rutgers University for providing him with computing

  47761.      resources to work on Free Software from the late 1980s to 2010.

  47762. 

  47763.    * John Gilmore for a donation to the FSF earmarked improving GNU

  47764.      Java.

  47765. 

  47766.    * Judy Goldberg for c++ contributions.

  47767. 

  47768.    * Torbjorn Granlund for various fixes and the c-torture testsuite,

  47769.      multiply- and divide-by-constant optimization, improved long long

  47770.      support, improved leaf function register allocation, and his

  47771.      direction via the steering committee.

  47772. 

  47773.    * Jonny Grant for improvements to 'collect2's' '--help'

  47774.      documentation.

  47775. 

  47776.    * Anthony Green for his '-Os' contributions, the moxie port, and Java

  47777.      front end work.

  47778. 

  47779.    * Stu Grossman for gdb hacking, allowing GCJ developers to debug Java

  47780.      code.

  47781. 

  47782.    * Michael K. Gschwind contributed the port to the PDP-11.

  47783. 

  47784.    * Richard Biener for his ongoing middle-end contributions and bug

  47785.      fixes and for release management.

  47786. 

  47787.    * Ron Guilmette implemented the 'protoize' and 'unprotoize' tools,

  47788.      the support for DWARF 1 symbolic debugging information, and much of

  47789.      the support for System V Release 4.  He has also worked heavily on

  47790.      the Intel 386 and 860 support.

  47791. 

  47792.    * Sumanth Gundapaneni for contributing the CR16 port.

  47793. 

  47794.    * Mostafa Hagog for Swing Modulo Scheduling (SMS) and post reload

  47795.      GCSE.

  47796. 

  47797.    * Bruno Haible for improvements in the runtime overhead for EH, new

  47798.      warnings and assorted bug fixes.

  47799. 

  47800.    * Andrew Haley for his amazing Java compiler and library efforts.

  47801. 

  47802.    * Chris Hanson assisted in making GCC work on HP-UX for the 9000

  47803.      series 300.

  47804. 

  47805.    * Michael Hayes for various thankless work he's done trying to get

  47806.      the c30/c40 ports functional.  Lots of loop and unroll improvements

  47807.      and fixes.

  47808. 

  47809.    * Dara Hazeghi for wading through myriads of target-specific bug

  47810.      reports.

  47811. 

  47812.    * Kate Hedstrom for staking the G77 folks with an initial testsuite.

  47813. 

  47814.    * Richard Henderson for his ongoing SPARC, alpha, ia32, and ia64

  47815.      work, loop opts, and generally fixing lots of old problems we've

  47816.      ignored for years, flow rewrite and lots of further stuff,

  47817.      including reviewing tons of patches.

  47818. 

  47819.    * Aldy Hernandez for working on the PowerPC port, SIMD support, and

  47820.      various fixes.

  47821. 

  47822.    * Nobuyuki Hikichi of Software Research Associates, Tokyo,

  47823.      contributed the support for the Sony NEWS machine.

  47824. 

  47825.    * Kazu Hirata for caring and feeding the Renesas H8/300 port and

  47826.      various fixes.

  47827. 

  47828.    * Katherine Holcomb for work on GNU Fortran.

  47829. 

  47830.    * Manfred Hollstein for his ongoing work to keep the m88k alive, lots

  47831.      of testing and bug fixing, particularly of GCC configury code.

  47832. 

  47833.    * Steve Holmgren for MachTen patches.

  47834. 

  47835.    * Mat Hostetter for work on the TILE-Gx and TILEPro ports.

  47836. 

  47837.    * Jan Hubicka for his x86 port improvements.

  47838. 

  47839.    * Falk Hueffner for working on C and optimization bug reports.

  47840. 

  47841.    * Bernardo Innocenti for his m68k work, including merging of ColdFire

  47842.      improvements and uClinux support.

  47843. 

  47844.    * Christian Iseli for various bug fixes.

  47845. 

  47846.    * Kamil Iskra for general m68k hacking.

  47847. 

  47848.    * Lee Iverson for random fixes and MIPS testing.

  47849. 

  47850.    * Balaji V. Iyer for Cilk+ development and merging.

  47851. 

  47852.    * Andreas Jaeger for testing and benchmarking of GCC and various bug

  47853.      fixes.

  47854. 

  47855.    * Martin Jambor for his work on inter-procedural optimizations, the

  47856.      switch conversion pass, and scalar replacement of aggregates.

  47857. 

  47858.    * Jakub Jelinek for his SPARC work and sibling call optimizations as

  47859.      well as lots of bug fixes and test cases, and for improving the

  47860.      Java build system.

  47861. 

  47862.    * Janis Johnson for ia64 testing and fixes, her quality improvement

  47863.      sidetracks, and web page maintenance.

  47864. 

  47865.    * Kean Johnston for SCO OpenServer support and various fixes.

  47866. 

  47867.    * Tim Josling for the sample language treelang based originally on

  47868.      Richard Kenner's "toy" language.

  47869. 

  47870.    * Nicolai Josuttis for additional libstdc++ documentation.

  47871. 

  47872.    * Klaus Kaempf for his ongoing work to make alpha-vms a viable

  47873.      target.

  47874. 

  47875.    * Steven G. Kargl for work on GNU Fortran.

  47876. 

  47877.    * David Kashtan of SRI adapted GCC to VMS.

  47878. 

  47879.    * Ryszard Kabatek for many, many libstdc++ bug fixes and

  47880.      optimizations of strings, especially member functions, and for

  47881.      auto_ptr fixes.

  47882. 

  47883.    * Geoffrey Keating for his ongoing work to make the PPC work for

  47884.      GNU/Linux and his automatic regression tester.

  47885. 

  47886.    * Brendan Kehoe for his ongoing work with G++ and for a lot of early

  47887.      work in just about every part of libstdc++.

  47888. 

  47889.    * Oliver M. Kellogg of Deutsche Aerospace contributed the port to the

  47890.      MIL-STD-1750A.

  47891. 

  47892.    * Richard Kenner of the New York University Ultracomputer Research

  47893.      Laboratory wrote the machine descriptions for the AMD 29000, the

  47894.      DEC Alpha, the IBM RT PC, and the IBM RS/6000 as well as the

  47895.      support for instruction attributes.  He also made changes to better

  47896.      support RISC processors including changes to common subexpression

  47897.      elimination, strength reduction, function calling sequence

  47898.      handling, and condition code support, in addition to generalizing

  47899.      the code for frame pointer elimination and delay slot scheduling.

  47900.      Richard Kenner was also the head maintainer of GCC for several

  47901.      years.

  47902. 

  47903.    * Mumit Khan for various contributions to the Cygwin and Mingw32

  47904.      ports and maintaining binary releases for Microsoft Windows hosts,

  47905.      and for massive libstdc++ porting work to Cygwin/Mingw32.

  47906. 

  47907.    * Robin Kirkham for cpu32 support.

  47908. 

  47909.    * Mark Klein for PA improvements.

  47910. 

  47911.    * Thomas Koenig for various bug fixes.

  47912. 

  47913.    * Bruce Korb for the new and improved fixincludes code.

  47914. 

  47915.    * Benjamin Kosnik for his G++ work and for leading the libstdc++-v3

  47916.      effort.

  47917. 

  47918.    * Maxim Kuvyrkov for contributions to the instruction scheduler, the

  47919.      Android and m68k/Coldfire ports, and optimizations.

  47920. 

  47921.    * Charles LaBrec contributed the support for the Integrated Solutions

  47922.      68020 system.

  47923. 

  47924.    * Asher Langton and Mike Kumbera for contributing Cray pointer

  47925.      support to GNU Fortran, and for other GNU Fortran improvements.

  47926. 

  47927.    * Jeff Law for his direction via the steering committee, coordinating

  47928.      the entire egcs project and GCC 2.95, rolling out snapshots and

  47929.      releases, handling merges from GCC2, reviewing tons of patches that

  47930.      might have fallen through the cracks else, and random but extensive

  47931.      hacking.

  47932. 

  47933.    * Walter Lee for work on the TILE-Gx and TILEPro ports.

  47934. 

  47935.    * Marc Lehmann for his direction via the steering committee and

  47936.      helping with analysis and improvements of x86 performance.

  47937. 

  47938.    * Victor Leikehman for work on GNU Fortran.

  47939. 

  47940.    * Ted Lemon wrote parts of the RTL reader and printer.

  47941. 

  47942.    * Kriang Lerdsuwanakij for C++ improvements including template as

  47943.      template parameter support, and many C++ fixes.

  47944. 

  47945.    * Warren Levy for tremendous work on libgcj (Java Runtime Library)

  47946.      and random work on the Java front end.

  47947. 

  47948.    * Alain Lichnewsky ported GCC to the MIPS CPU.

  47949. 

  47950.    * Oskar Liljeblad for hacking on AWT and his many Java bug reports

  47951.      and patches.

  47952. 

  47953.    * Robert Lipe for OpenServer support, new testsuites, testing, etc.

  47954. 

  47955.    * Chen Liqin for various S+core related fixes/improvement, and for

  47956.      maintaining the S+core port.

  47957. 

  47958.    * Martin Liska for his work on identical code folding, the

  47959.      sanitizers, HSA, general bug fixing and for running automated

  47960.      regression testing of GCC and reporting numerous bugs.

  47961. 

  47962.    * Weiwen Liu for testing and various bug fixes.

  47963. 

  47964.    * Manuel Lo'pez-Iba'n~ez for improving '-Wconversion' and many other

  47965.      diagnostics fixes and improvements.

  47966. 

  47967.    * Dave Love for his ongoing work with the Fortran front end and

  47968.      runtime libraries.

  47969. 

  47970.    * Martin von Lo"wis for internal consistency checking infrastructure,

  47971.      various C++ improvements including namespace support, and tons of

  47972.      assistance with libstdc++/compiler merges.

  47973. 

  47974.    * H.J. Lu for his previous contributions to the steering committee,

  47975.      many x86 bug reports, prototype patches, and keeping the GNU/Linux

  47976.      ports working.

  47977. 

  47978.    * Greg McGary for random fixes and (someday) bounded pointers.

  47979. 

  47980.    * Andrew MacLeod for his ongoing work in building a real EH system,

  47981.      various code generation improvements, work on the global optimizer,

  47982.      etc.

  47983. 

  47984.    * Vladimir Makarov for hacking some ugly i960 problems, PowerPC

  47985.      hacking improvements to compile-time performance, overall knowledge

  47986.      and direction in the area of instruction scheduling, design and

  47987.      implementation of the automaton based instruction scheduler and

  47988.      design and implementation of the integrated and local register

  47989.      allocators.

  47990. 

  47991.    * David Malcolm for his work on improving GCC diagnostics, JIT,

  47992.      self-tests and unit testing.

  47993. 

  47994.    * Bob Manson for his behind the scenes work on dejagnu.

  47995. 

  47996.    * John Marino for contributing the DragonFly BSD port.

  47997. 

  47998.    * Philip Martin for lots of libstdc++ string and vector iterator

  47999.      fixes and improvements, and string clean up and testsuites.

  48000. 

  48001.    * Michael Matz for his work on dominance tree discovery, the x86-64

  48002.      port, link-time optimization framework and general optimization

  48003.      improvements.

  48004. 

  48005.    * All of the Mauve project contributors for Java test code.

  48006. 

  48007.    * Bryce McKinlay for numerous GCJ and libgcj fixes and improvements.

  48008. 

  48009.    * Adam Megacz for his work on the Microsoft Windows port of GCJ.

  48010. 

  48011.    * Michael Meissner for LRS framework, ia32, m32r, v850, m88k, MIPS,

  48012.      powerpc, haifa, ECOFF debug support, and other assorted hacking.

  48013. 

  48014.    * Jason Merrill for his direction via the steering committee and

  48015.      leading the G++ effort.

  48016. 

  48017.    * Martin Michlmayr for testing GCC on several architectures using the

  48018.      entire Debian archive.

  48019. 

  48020.    * David Miller for his direction via the steering committee, lots of

  48021.      SPARC work, improvements in jump.c and interfacing with the Linux

  48022.      kernel developers.

  48023. 

  48024.    * Gary Miller ported GCC to Charles River Data Systems machines.

  48025. 

  48026.    * Alfred Minarik for libstdc++ string and ios bug fixes, and turning

  48027.      the entire libstdc++ testsuite namespace-compatible.

  48028. 

  48029.    * Mark Mitchell for his direction via the steering committee,

  48030.      mountains of C++ work, load/store hoisting out of loops, alias

  48031.      analysis improvements, ISO C 'restrict' support, and serving as

  48032.      release manager from 2000 to 2011.

  48033. 

  48034.    * Alan Modra for various GNU/Linux bits and testing.

  48035. 

  48036.    * Toon Moene for his direction via the steering committee, Fortran

  48037.      maintenance, and his ongoing work to make us make Fortran run fast.

  48038. 

  48039.    * Jason Molenda for major help in the care and feeding of all the

  48040.      services on the gcc.gnu.org (formerly egcs.cygnus.com)

  48041.      machine--mail, web services, ftp services, etc etc.  Doing all this

  48042.      work on scrap paper and the backs of envelopes would have been...

  48043.      difficult.

  48044. 

  48045.    * Catherine Moore for fixing various ugly problems we have sent her

  48046.      way, including the haifa bug which was killing the Alpha & PowerPC

  48047.      Linux kernels.

  48048. 

  48049.    * Mike Moreton for his various Java patches.

  48050. 

  48051.    * David Mosberger-Tang for various Alpha improvements, and for the

  48052.      initial IA-64 port.

  48053. 

  48054.    * Stephen Moshier contributed the floating point emulator that

  48055.      assists in cross-compilation and permits support for floating point

  48056.      numbers wider than 64 bits and for ISO C99 support.

  48057. 

  48058.    * Bill Moyer for his behind the scenes work on various issues.

  48059. 

  48060.    * Philippe De Muyter for his work on the m68k port.

  48061. 

  48062.    * Joseph S. Myers for his work on the PDP-11 port, format checking

  48063.      and ISO C99 support, and continuous emphasis on (and contributions

  48064.      to) documentation.

  48065. 

  48066.    * Nathan Myers for his work on libstdc++-v3: architecture and

  48067.      authorship through the first three snapshots, including

  48068.      implementation of locale infrastructure, string, shadow C headers,

  48069.      and the initial project documentation (DESIGN, CHECKLIST, and so

  48070.      forth).  Later, more work on MT-safe string and shadow headers.

  48071. 

  48072.    * Felix Natter for documentation on porting libstdc++.

  48073. 

  48074.    * Nathanael Nerode for cleaning up the configuration/build process.

  48075. 

  48076.    * NeXT, Inc. donated the front end that supports the Objective-C

  48077.      language.

  48078. 

  48079.    * Hans-Peter Nilsson for the CRIS and MMIX ports, improvements to the

  48080.      search engine setup, various documentation fixes and other small

  48081.      fixes.

  48082. 

  48083.    * Geoff Noer for his work on getting cygwin native builds working.

  48084. 

  48085.    * Vegard Nossum for running automated regression testing of GCC and

  48086.      reporting numerous bugs.

  48087. 

  48088.    * Diego Novillo for his work on Tree SSA, OpenMP, SPEC performance

  48089.      tracking web pages, GIMPLE tuples, and assorted fixes.

  48090. 

  48091.    * David O'Brien for the FreeBSD/alpha, FreeBSD/AMD x86-64,

  48092.      FreeBSD/ARM, FreeBSD/PowerPC, and FreeBSD/SPARC64 ports and related

  48093.      infrastructure improvements.

  48094. 

  48095.    * Alexandre Oliva for various build infrastructure improvements,

  48096.      scripts and amazing testing work, including keeping libtool issues

  48097.      sane and happy.

  48098. 

  48099.    * Stefan Olsson for work on mt_alloc.

  48100. 

  48101.    * Melissa O'Neill for various NeXT fixes.

  48102. 

  48103.    * Rainer Orth for random MIPS work, including improvements to GCC's

  48104.      o32 ABI support, improvements to dejagnu's MIPS support, Java

  48105.      configuration clean-ups and porting work, and maintaining the IRIX,

  48106.      Solaris 2, and Tru64 UNIX ports.

  48107. 

  48108.    * Steven Pemberton for his contribution of 'enquire' which allowed

  48109.      GCC to determine various properties of the floating point unit and

  48110.      generate 'float.h' in older versions of GCC.

  48111. 

  48112.    * Hartmut Penner for work on the s390 port.

  48113. 

  48114.    * Paul Petersen wrote the machine description for the Alliant FX/8.

  48115. 

  48116.    * Alexandre Petit-Bianco for implementing much of the Java compiler

  48117.      and continued Java maintainership.

  48118. 

  48119.    * Matthias Pfaller for major improvements to the NS32k port.

  48120. 

  48121.    * Gerald Pfeifer for his direction via the steering committee,

  48122.      pointing out lots of problems we need to solve, maintenance of the

  48123.      web pages, and taking care of documentation maintenance in general.

  48124. 

  48125.    * Marek Polacek for his work on the C front end, the sanitizers and

  48126.      general bug fixing.

  48127. 

  48128.    * Andrew Pinski for processing bug reports by the dozen.

  48129. 

  48130.    * Ovidiu Predescu for his work on the Objective-C front end and

  48131.      runtime libraries.

  48132. 

  48133.    * Jerry Quinn for major performance improvements in C++ formatted

  48134.      I/O.

  48135. 

  48136.    * Ken Raeburn for various improvements to checker, MIPS ports and

  48137.      various cleanups in the compiler.

  48138. 

  48139.    * Rolf W. Rasmussen for hacking on AWT.

  48140. 

  48141.    * David Reese of Sun Microsystems contributed to the Solaris on

  48142.      PowerPC port.

  48143. 

  48144.    * John Regehr for running automated regression testing of GCC and

  48145.      reporting numerous bugs.

  48146. 

  48147.    * Volker Reichelt for running automated regression testing of GCC and

  48148.      reporting numerous bugs and for keeping up with the problem

  48149.      reports.

  48150. 

  48151.    * Joern Rennecke for maintaining the sh port, loop, regmove & reload

  48152.      hacking and developing and maintaining the Epiphany port.

  48153. 

  48154.    * Loren J. Rittle for improvements to libstdc++-v3 including the

  48155.      FreeBSD port, threading fixes, thread-related configury changes,

  48156.      critical threading documentation, and solutions to really tricky

  48157.      I/O problems, as well as keeping GCC properly working on FreeBSD

  48158.      and continuous testing.

  48159. 

  48160.    * Craig Rodrigues for processing tons of bug reports.

  48161. 

  48162.    * Ola Ro"nnerup for work on mt_alloc.

  48163. 

  48164.    * Gavin Romig-Koch for lots of behind the scenes MIPS work.

  48165. 

  48166.    * David Ronis inspired and encouraged Craig to rewrite the G77

  48167.      documentation in texinfo format by contributing a first pass at a

  48168.      translation of the old 'g77-0.5.16/f/DOC' file.

  48169. 

  48170.    * Ken Rose for fixes to GCC's delay slot filling code.

  48171. 

  48172.    * Ira Rosen for her contributions to the auto-vectorizer.

  48173. 

  48174.    * Paul Rubin wrote most of the preprocessor.

  48175. 

  48176.    * Pe'tur Runo'lfsson for major performance improvements in C++

  48177.      formatted I/O and large file support in C++ filebuf.

  48178. 

  48179.    * Chip Salzenberg for libstdc++ patches and improvements to locales,

  48180.      traits, Makefiles, libio, libtool hackery, and "long long" support.

  48181. 

  48182.    * Juha Sarlin for improvements to the H8 code generator.

  48183. 

  48184.    * Greg Satz assisted in making GCC work on HP-UX for the 9000 series

  48185.      300.

  48186. 

  48187.    * Roger Sayle for improvements to constant folding and GCC's RTL

  48188.      optimizers as well as for fixing numerous bugs.

  48189. 

  48190.    * Bradley Schatz for his work on the GCJ FAQ.

  48191. 

  48192.    * Peter Schauer wrote the code to allow debugging to work on the

  48193.      Alpha.

  48194. 

  48195.    * William Schelter did most of the work on the Intel 80386 support.

  48196. 

  48197.    * Tobias Schlu"ter for work on GNU Fortran.

  48198. 

  48199.    * Bernd Schmidt for various code generation improvements and major

  48200.      work in the reload pass, serving as release manager for GCC 2.95.3,

  48201.      and work on the Blackfin and C6X ports.

  48202. 

  48203.    * Peter Schmid for constant testing of libstdc++--especially

  48204.      application testing, going above and beyond what was requested for

  48205.      the release criteria--and libstdc++ header file tweaks.

  48206. 

  48207.    * Jason Schroeder for jcf-dump patches.

  48208. 

  48209.    * Andreas Schwab for his work on the m68k port.

  48210. 

  48211.    * Lars Segerlund for work on GNU Fortran.

  48212. 

  48213.    * Dodji Seketeli for numerous C++ bug fixes and debug info

  48214.      improvements.

  48215. 

  48216.    * Tim Shen for major work on '<regex>'.

  48217. 

  48218.    * Joel Sherrill for his direction via the steering committee, RTEMS

  48219.      contributions and RTEMS testing.

  48220. 

  48221.    * Nathan Sidwell for many C++ fixes/improvements.

  48222. 

  48223.    * Jeffrey Siegal for helping RMS with the original design of GCC,

  48224.      some code which handles the parse tree and RTL data structures,

  48225.      constant folding and help with the original VAX & m68k ports.

  48226. 

  48227.    * Kenny Simpson for prompting libstdc++ fixes due to defect reports

  48228.      from the LWG (thereby keeping GCC in line with updates from the

  48229.      ISO).

  48230. 

  48231.    * Franz Sirl for his ongoing work with making the PPC port stable for

  48232.      GNU/Linux.

  48233. 

  48234.    * Andrey Slepuhin for assorted AIX hacking.

  48235. 

  48236.    * Trevor Smigiel for contributing the SPU port.

  48237. 

  48238.    * Christopher Smith did the port for Convex machines.

  48239. 

  48240.    * Danny Smith for his major efforts on the Mingw (and Cygwin) ports.

  48241.      Retired from GCC maintainership August 2010, having mentored two

  48242.      new maintainers into the role.

  48243. 

  48244.    * Randy Smith finished the Sun FPA support.

  48245. 

  48246.    * Ed Smith-Rowland for his continuous work on libstdc++-v3, special

  48247.      functions, '<random>', and various improvements to C++11 features.

  48248. 

  48249.    * Scott Snyder for queue, iterator, istream, and string fixes and

  48250.      libstdc++ testsuite entries.  Also for providing the patch to G77

  48251.      to add rudimentary support for 'INTEGER*1', 'INTEGER*2', and

  48252.      'LOGICAL*1'.

  48253. 

  48254.    * Zdenek Sojka for running automated regression testing of GCC and

  48255.      reporting numerous bugs.

  48256. 

  48257.    * Arseny Solokha for running automated regression testing of GCC and

  48258.      reporting numerous bugs.

  48259. 

  48260.    * Jayant Sonar for contributing the CR16 port.

  48261. 

  48262.    * Brad Spencer for contributions to the GLIBCPP_FORCE_NEW technique.

  48263. 

  48264.    * Richard Stallman, for writing the original GCC and launching the

  48265.      GNU project.

  48266. 

  48267.    * Jan Stein of the Chalmers Computer Society provided support for

  48268.      Genix, as well as part of the 32000 machine description.

  48269. 

  48270.    * Gerhard Steinmetz for running automated regression testing of GCC

  48271.      and reporting numerous bugs.

  48272. 

  48273.    * Nigel Stephens for various mips16 related fixes/improvements.

  48274. 

  48275.    * Jonathan Stone wrote the machine description for the Pyramid

  48276.      computer.

  48277. 

  48278.    * Graham Stott for various infrastructure improvements.

  48279. 

  48280.    * John Stracke for his Java HTTP protocol fixes.

  48281. 

  48282.    * Mike Stump for his Elxsi port, G++ contributions over the years and

  48283.      more recently his vxworks contributions

  48284. 

  48285.    * Jeff Sturm for Java porting help, bug fixes, and encouragement.

  48286. 

  48287.    * Zhendong Su for running automated regression testing of GCC and

  48288.      reporting numerous bugs.

  48289. 

  48290.    * Chengnian Sun for running automated regression testing of GCC and

  48291.      reporting numerous bugs.

  48292. 

  48293.    * Shigeya Suzuki for this fixes for the bsdi platforms.

  48294. 

  48295.    * Ian Lance Taylor for the Go frontend, the initial mips16 and mips64

  48296.      support, general configury hacking, fixincludes, etc.

  48297. 

  48298.    * Holger Teutsch provided the support for the Clipper CPU.

  48299. 

  48300.    * Gary Thomas for his ongoing work to make the PPC work for

  48301.      GNU/Linux.

  48302. 

  48303.    * Paul Thomas for contributions to GNU Fortran.

  48304. 

  48305.    * Philipp Thomas for random bug fixes throughout the compiler

  48306. 

  48307.    * Jason Thorpe for thread support in libstdc++ on NetBSD.

  48308. 

  48309.    * Kresten Krab Thorup wrote the run time support for the Objective-C

  48310.      language and the fantastic Java bytecode interpreter.

  48311. 

  48312.    * Michael Tiemann for random bug fixes, the first instruction

  48313.      scheduler, initial C++ support, function integration, NS32k, SPARC

  48314.      and M88k machine description work, delay slot scheduling.

  48315. 

  48316.    * Andreas Tobler for his work porting libgcj to Darwin.

  48317. 

  48318.    * Teemu Torma for thread safe exception handling support.

  48319. 

  48320.    * Leonard Tower wrote parts of the parser, RTL generator, and RTL

  48321.      definitions, and of the VAX machine description.

  48322. 

  48323.    * Daniel Towner and Hariharan Sandanagobalane contributed and

  48324.      maintain the picoChip port.

  48325. 

  48326.    * Tom Tromey for internationalization support and for his many Java

  48327.      contributions and libgcj maintainership.

  48328. 

  48329.    * Lassi Tuura for improvements to config.guess to determine HP

  48330.      processor types.

  48331. 

  48332.    * Petter Urkedal for libstdc++ CXXFLAGS, math, and algorithms fixes.

  48333. 

  48334.    * Andy Vaught for the design and initial implementation of the GNU

  48335.      Fortran front end.

  48336. 

  48337.    * Brent Verner for work with the libstdc++ cshadow files and their

  48338.      associated configure steps.

  48339. 

  48340.    * Todd Vierling for contributions for NetBSD ports.

  48341. 

  48342.    * Andrew Waterman for contributing the RISC-V port, as well as

  48343.      maintaining it.

  48344. 

  48345.    * Jonathan Wakely for contributing libstdc++ Doxygen notes and XHTML

  48346.      guidance and maintaining libstdc++.

  48347. 

  48348.    * Dean Wakerley for converting the install documentation from HTML to

  48349.      texinfo in time for GCC 3.0.

  48350. 

  48351.    * Krister Walfridsson for random bug fixes.

  48352. 

  48353.    * Feng Wang for contributions to GNU Fortran.

  48354. 

  48355.    * Stephen M. Webb for time and effort on making libstdc++ shadow

  48356.      files work with the tricky Solaris 8+ headers, and for pushing the

  48357.      build-time header tree.  Also, for starting and driving the

  48358.      '<regex>' effort.

  48359. 

  48360.    * John Wehle for various improvements for the x86 code generator,

  48361.      related infrastructure improvements to help x86 code generation,

  48362.      value range propagation and other work, WE32k port.

  48363. 

  48364.    * Ulrich Weigand for work on the s390 port.

  48365. 

  48366.    * Janus Weil for contributions to GNU Fortran.

  48367. 

  48368.    * Zack Weinberg for major work on cpplib and various other bug fixes.

  48369. 

  48370.    * Matt Welsh for help with Linux Threads support in GCJ.

  48371. 

  48372.    * Urban Widmark for help fixing java.io.

  48373. 

  48374.    * Mark Wielaard for new Java library code and his work integrating

  48375.      with Classpath.

  48376. 

  48377.    * Dale Wiles helped port GCC to the Tahoe.

  48378. 

  48379.    * Bob Wilson from Tensilica, Inc. for the Xtensa port.

  48380. 

  48381.    * Jim Wilson for his direction via the steering committee, tackling

  48382.      hard problems in various places that nobody else wanted to work on,

  48383.      strength reduction and other loop optimizations.

  48384. 

  48385.    * Paul Woegerer and Tal Agmon for the CRX port.

  48386. 

  48387.    * Carlo Wood for various fixes.

  48388. 

  48389.    * Tom Wood for work on the m88k port.

  48390. 

  48391.    * Chung-Ju Wu for his work on the Andes NDS32 port.

  48392. 

  48393.    * Canqun Yang for work on GNU Fortran.

  48394. 

  48395.    * Masanobu Yuhara of Fujitsu Laboratories implemented the machine

  48396.      description for the Tron architecture (specifically, the Gmicro).

  48397. 

  48398.    * Kevin Zachmann helped port GCC to the Tahoe.

  48399. 

  48400.    * Ayal Zaks for Swing Modulo Scheduling (SMS).

  48401. 

  48402.    * Qirun Zhang for running automated regression testing of GCC and

  48403.      reporting numerous bugs.

  48404. 

  48405.    * Xiaoqiang Zhang for work on GNU Fortran.

  48406. 

  48407.    * Gilles Zunino for help porting Java to Irix.

  48408. 

  48409.  The following people are recognized for their contributions to GNAT,

  48410. the Ada front end of GCC:

  48411.    * Bernard Banner

  48412. 

  48413.    * Romain Berrendonner

  48414. 

  48415.    * Geert Bosch

  48416. 

  48417.    * Emmanuel Briot

  48418. 

  48419.    * Joel Brobecker

  48420. 

  48421.    * Ben Brosgol

  48422. 

  48423.    * Vincent Celier

  48424. 

  48425.    * Arnaud Charlet

  48426. 

  48427.    * Chien Chieng

  48428. 

  48429.    * Cyrille Comar

  48430. 

  48431.    * Cyrille Crozes

  48432. 

  48433.    * Robert Dewar

  48434. 

  48435.    * Gary Dismukes

  48436. 

  48437.    * Robert Duff

  48438. 

  48439.    * Ed Falis

  48440. 

  48441.    * Ramon Fernandez

  48442. 

  48443.    * Sam Figueroa

  48444. 

  48445.    * Vasiliy Fofanov

  48446. 

  48447.    * Michael Friess

  48448. 

  48449.    * Franco Gasperoni

  48450. 

  48451.    * Ted Giering

  48452. 

  48453.    * Matthew Gingell

  48454. 

  48455.    * Laurent Guerby

  48456. 

  48457.    * Jerome Guitton

  48458. 

  48459.    * Olivier Hainque

  48460. 

  48461.    * Jerome Hugues

  48462. 

  48463.    * Hristian Kirtchev

  48464. 

  48465.    * Jerome Lambourg

  48466. 

  48467.    * Bruno Leclerc

  48468. 

  48469.    * Albert Lee

  48470. 

  48471.    * Sean McNeil

  48472. 

  48473.    * Javier Miranda

  48474. 

  48475.    * Laurent Nana

  48476. 

  48477.    * Pascal Obry

  48478. 

  48479.    * Dong-Ik Oh

  48480. 

  48481.    * Laurent Pautet

  48482. 

  48483.    * Brett Porter

  48484. 

  48485.    * Thomas Quinot

  48486. 

  48487.    * Nicolas Roche

  48488. 

  48489.    * Pat Rogers

  48490. 

  48491.    * Jose Ruiz

  48492. 

  48493.    * Douglas Rupp

  48494. 

  48495.    * Sergey Rybin

  48496. 

  48497.    * Gail Schenker

  48498. 

  48499.    * Ed Schonberg

  48500. 

  48501.    * Nicolas Setton

  48502. 

  48503.    * Samuel Tardieu

  48504. 

  48505.  The following people are recognized for their contributions of new

  48506. features, bug reports, testing and integration of classpath/libgcj for

  48507. GCC version 4.1:

  48508.    * Lillian Angel for 'JTree' implementation and lots Free Swing

  48509.      additions and bug fixes.

  48510. 

  48511.    * Wolfgang Baer for 'GapContent' bug fixes.

  48512. 

  48513.    * Anthony Balkissoon for 'JList', Free Swing 1.5 updates and mouse

  48514.      event fixes, lots of Free Swing work including 'JTable' editing.

  48515. 

  48516.    * Stuart Ballard for RMI constant fixes.

  48517. 

  48518.    * Goffredo Baroncelli for 'HTTPURLConnection' fixes.

  48519. 

  48520.    * Gary Benson for 'MessageFormat' fixes.

  48521. 

  48522.    * Daniel Bonniot for 'Serialization' fixes.

  48523. 

  48524.    * Chris Burdess for lots of gnu.xml and http protocol fixes, 'StAX'

  48525.      and 'DOM xml:id' support.

  48526. 

  48527.    * Ka-Hing Cheung for 'TreePath' and 'TreeSelection' fixes.

  48528. 

  48529.    * Archie Cobbs for build fixes, VM interface updates,

  48530.      'URLClassLoader' updates.

  48531. 

  48532.    * Kelley Cook for build fixes.

  48533. 

  48534.    * Martin Cordova for Suggestions for better 'SocketTimeoutException'.

  48535. 

  48536.    * David Daney for 'BitSet' bug fixes, 'HttpURLConnection' rewrite and

  48537.      improvements.

  48538. 

  48539.    * Thomas Fitzsimmons for lots of upgrades to the gtk+ AWT and Cairo

  48540.      2D support.  Lots of imageio framework additions, lots of AWT and

  48541.      Free Swing bug fixes.

  48542. 

  48543.    * Jeroen Frijters for 'ClassLoader' and nio cleanups, serialization

  48544.      fixes, better 'Proxy' support, bug fixes and IKVM integration.

  48545. 

  48546.    * Santiago Gala for 'AccessControlContext' fixes.

  48547. 

  48548.    * Nicolas Geoffray for 'VMClassLoader' and 'AccessController'

  48549.      improvements.

  48550. 

  48551.    * David Gilbert for 'basic' and 'metal' icon and plaf support and

  48552.      lots of documenting, Lots of Free Swing and metal theme additions.

  48553.      'MetalIconFactory' implementation.

  48554. 

  48555.    * Anthony Green for 'MIDI' framework, 'ALSA' and 'DSSI' providers.

  48556. 

  48557.    * Andrew Haley for 'Serialization' and 'URLClassLoader' fixes, gcj

  48558.      build speedups.

  48559. 

  48560.    * Kim Ho for 'JFileChooser' implementation.

  48561. 

  48562.    * Andrew John Hughes for 'Locale' and net fixes, URI RFC2986 updates,

  48563.      'Serialization' fixes, 'Properties' XML support and generic branch

  48564.      work, VMIntegration guide update.

  48565. 

  48566.    * Bastiaan Huisman for 'TimeZone' bug fixing.

  48567. 

  48568.    * Andreas Jaeger for mprec updates.

  48569. 

  48570.    * Paul Jenner for better '-Werror' support.

  48571. 

  48572.    * Ito Kazumitsu for 'NetworkInterface' implementation and updates.

  48573. 

  48574.    * Roman Kennke for 'BoxLayout', 'GrayFilter' and 'SplitPane', plus

  48575.      bug fixes all over.  Lots of Free Swing work including styled text.

  48576. 

  48577.    * Simon Kitching for 'String' cleanups and optimization suggestions.

  48578. 

  48579.    * Michael Koch for configuration fixes, 'Locale' updates, bug and

  48580.      build fixes.

  48581. 

  48582.    * Guilhem Lavaux for configuration, thread and channel fixes and

  48583.      Kaffe integration.  JCL native 'Pointer' updates.  Logger bug

  48584.      fixes.

  48585. 

  48586.    * David Lichteblau for JCL support library global/local reference

  48587.      cleanups.

  48588. 

  48589.    * Aaron Luchko for JDWP updates and documentation fixes.

  48590. 

  48591.    * Ziga Mahkovec for 'Graphics2D' upgraded to Cairo 0.5 and new regex

  48592.      features.

  48593. 

  48594.    * Sven de Marothy for BMP imageio support, CSS and 'TextLayout'

  48595.      fixes.  'GtkImage' rewrite, 2D, awt, free swing and date/time fixes

  48596.      and implementing the Qt4 peers.

  48597. 

  48598.    * Casey Marshall for crypto algorithm fixes, 'FileChannel' lock,

  48599.      'SystemLogger' and 'FileHandler' rotate implementations, NIO

  48600.      'FileChannel.map' support, security and policy updates.

  48601. 

  48602.    * Bryce McKinlay for RMI work.

  48603. 

  48604.    * Audrius Meskauskas for lots of Free Corba, RMI and HTML work plus

  48605.      testing and documenting.

  48606. 

  48607.    * Kalle Olavi Niemitalo for build fixes.

  48608. 

  48609.    * Rainer Orth for build fixes.

  48610. 

  48611.    * Andrew Overholt for 'File' locking fixes.

  48612. 

  48613.    * Ingo Proetel for 'Image', 'Logger' and 'URLClassLoader' updates.

  48614. 

  48615.    * Olga Rodimina for 'MenuSelectionManager' implementation.

  48616. 

  48617.    * Jan Roehrich for 'BasicTreeUI' and 'JTree' fixes.

  48618. 

  48619.    * Julian Scheid for documentation updates and gjdoc support.

  48620. 

  48621.    * Christian Schlichtherle for zip fixes and cleanups.

  48622. 

  48623.    * Robert Schuster for documentation updates and beans fixes,

  48624.      'TreeNode' enumerations and 'ActionCommand' and various fixes, XML

  48625.      and URL, AWT and Free Swing bug fixes.

  48626. 

  48627.    * Keith Seitz for lots of JDWP work.

  48628. 

  48629.    * Christian Thalinger for 64-bit cleanups, Configuration and VM

  48630.      interface fixes and 'CACAO' integration, 'fdlibm' updates.

  48631. 

  48632.    * Gael Thomas for 'VMClassLoader' boot packages support suggestions.

  48633. 

  48634.    * Andreas Tobler for Darwin and Solaris testing and fixing, 'Qt4'

  48635.      support for Darwin/OS X, 'Graphics2D' support, 'gtk+' updates.

  48636. 

  48637.    * Dalibor Topic for better 'DEBUG' support, build cleanups and Kaffe

  48638.      integration.  'Qt4' build infrastructure, 'SHA1PRNG' and

  48639.      'GdkPixbugDecoder' updates.

  48640. 

  48641.    * Tom Tromey for Eclipse integration, generics work, lots of bug

  48642.      fixes and gcj integration including coordinating The Big Merge.

  48643. 

  48644.    * Mark Wielaard for bug fixes, packaging and release management,

  48645.      'Clipboard' implementation, system call interrupts and network

  48646.      timeouts and 'GdkPixpufDecoder' fixes.

  48647. 

  48648.  In addition to the above, all of which also contributed time and energy

  48649. in testing GCC, we would like to thank the following for their

  48650. contributions to testing:

  48651. 

  48652.    * Michael Abd-El-Malek

  48653. 

  48654.    * Thomas Arend

  48655. 

  48656.    * Bonzo Armstrong

  48657. 

  48658.    * Steven Ashe

  48659. 

  48660.    * Chris Baldwin

  48661. 

  48662.    * David Billinghurst

  48663. 

  48664.    * Jim Blandy

  48665. 

  48666.    * Stephane Bortzmeyer

  48667. 

  48668.    * Horst von Brand

  48669. 

  48670.    * Frank Braun

  48671. 

  48672.    * Rodney Brown

  48673. 

  48674.    * Sidney Cadot

  48675. 

  48676.    * Bradford Castalia

  48677. 

  48678.    * Robert Clark

  48679. 

  48680.    * Jonathan Corbet

  48681. 

  48682.    * Ralph Doncaster

  48683. 

  48684.    * Richard Emberson

  48685. 

  48686.    * Levente Farkas

  48687. 

  48688.    * Graham Fawcett

  48689. 

  48690.    * Mark Fernyhough

  48691. 

  48692.    * Robert A. French

  48693. 

  48694.    * Jo"rgen Freyh

  48695. 

  48696.    * Mark K. Gardner

  48697. 

  48698.    * Charles-Antoine Gauthier

  48699. 

  48700.    * Yung Shing Gene

  48701. 

  48702.    * David Gilbert

  48703. 

  48704.    * Simon Gornall

  48705. 

  48706.    * Fred Gray

  48707. 

  48708.    * John Griffin

  48709. 

  48710.    * Patrik Hagglund

  48711. 

  48712.    * Phil Hargett

  48713. 

  48714.    * Amancio Hasty

  48715. 

  48716.    * Takafumi Hayashi

  48717. 

  48718.    * Bryan W. Headley

  48719. 

  48720.    * Kevin B. Hendricks

  48721. 

  48722.    * Joep Jansen

  48723. 

  48724.    * Christian Joensson

  48725. 

  48726.    * Michel Kern

  48727. 

  48728.    * David Kidd

  48729. 

  48730.    * Tobias Kuipers

  48731. 

  48732.    * Anand Krishnaswamy

  48733. 

  48734.    * A. O. V. Le Blanc

  48735. 

  48736.    * llewelly

  48737. 

  48738.    * Damon Love

  48739. 

  48740.    * Brad Lucier

  48741. 

  48742.    * Matthias Klose

  48743. 

  48744.    * Martin Knoblauch

  48745. 

  48746.    * Rick Lutowski

  48747. 

  48748.    * Jesse Macnish

  48749. 

  48750.    * Stefan Morrell

  48751. 

  48752.    * Anon A. Mous

  48753. 

  48754.    * Matthias Mueller

  48755. 

  48756.    * Pekka Nikander

  48757. 

  48758.    * Rick Niles

  48759. 

  48760.    * Jon Olson

  48761. 

  48762.    * Magnus Persson

  48763. 

  48764.    * Chris Pollard

  48765. 

  48766.    * Richard Polton

  48767. 

  48768.    * Derk Reefman

  48769. 

  48770.    * David Rees

  48771. 

  48772.    * Paul Reilly

  48773. 

  48774.    * Tom Reilly

  48775. 

  48776.    * Torsten Rueger

  48777. 

  48778.    * Danny Sadinoff

  48779. 

  48780.    * Marc Schifer

  48781. 

  48782.    * Erik Schnetter

  48783. 

  48784.    * Wayne K. Schroll

  48785. 

  48786.    * David Schuler

  48787. 

  48788.    * Vin Shelton

  48789. 

  48790.    * Tim Souder

  48791. 

  48792.    * Adam Sulmicki

  48793. 

  48794.    * Bill Thorson

  48795. 

  48796.    * George Talbot

  48797. 

  48798.    * Pedro A. M. Vazquez

  48799. 

  48800.    * Gregory Warnes

  48801. 

  48802.    * Ian Watson

  48803. 

  48804.    * David E. Young

  48805. 

  48806.    * And many others

  48807. 

  48808.  And finally we'd like to thank everyone who uses the compiler, provides

  48809. feedback and generally reminds us why we're doing this work in the first

  48810. place.

  48811. 

  48812. 

  48813. File: gccint.info,  Node: Option Index,  Next: Concept Index,  Prev: Contributors,  Up: Top

  48814. 

  48815. Option Index

  48816. ************

  48817. 

  48818. GCC's command line options are indexed here without any initial '-' or

  48819. '--'.  Where an option has both positive and negative forms (such as

  48820. '-fOPTION' and '-fno-OPTION'), relevant entries in the manual are

  48821. indexed under the most appropriate form; it may sometimes be useful to

  48822. look up both forms.

  48823. 

  48824. �[index�]

  48825. * Menu:

  48826. 

  48827. * fltrans:                               Internal flags.       (line 18)

  48828. * fltrans-output-list:                   Internal flags.       (line 23)

  48829. * fresolution:                           Internal flags.       (line 27)

  48830. * fwpa:                                  Internal flags.       (line  9)

  48831. * msoft-float:                           Soft float library routines.

  48832.                                                                (line  6)

  48833. 

  48834. 

  48835. File: gccint.info,  Node: Concept Index,  Prev: Option Index,  Up: Top

  48836. 

  48837. Concept Index

  48838. *************

  48839. 

  48840. �[index�]

  48841. * Menu:

  48842. 

  48843. * ! in constraint:                       Multi-Alternative.  (line   48)

  48844. * # in constraint:                       Modifiers.          (line   78)

  48845. * # in template:                         Output Template.    (line   66)

  48846. * #pragma:                               Misc.               (line  420)

  48847. * $ in constraint:                       Multi-Alternative.  (line   57)

  48848. * % in constraint:                       Modifiers.          (line   52)

  48849. * % in GTY option:                       GTY Options.        (line   18)

  48850. * % in template:                         Output Template.    (line    6)

  48851. * & in constraint:                       Modifiers.          (line   25)

  48852. * (gimple:                               Logical Operators.  (line  169)

  48853. * (gimple <1>:                           Logical Operators.  (line  173)

  48854. * (gimple <2>:                           Logical Operators.  (line  177)

  48855. * (gimple_stmt_iterator:                 GIMPLE API.         (line   30)

  48856. * (nil):                                 RTL Objects.        (line   73)

  48857. * * in constraint:                       Modifiers.          (line   83)

  48858. * * in template:                         Output Statement.   (line   29)

  48859. * *gimple_build_asm_vec:                 GIMPLE_ASM.         (line    6)

  48860. * *gimple_build_assign:                  GIMPLE_ASSIGN.      (line    6)

  48861. * *gimple_build_assign <1>:              GIMPLE_ASSIGN.      (line   18)

  48862. * *gimple_build_assign <2>:              GIMPLE_ASSIGN.      (line   29)

  48863. * *gimple_build_assign <3>:              GIMPLE_ASSIGN.      (line   35)

  48864. * *gimple_build_bind:                    GIMPLE_BIND.        (line    6)

  48865. * *gimple_build_call:                    GIMPLE_CALL.        (line    6)

  48866. * *gimple_build_call_from_tree:          GIMPLE_CALL.        (line   15)

  48867. * *gimple_build_call_vec:                GIMPLE_CALL.        (line   25)

  48868. * *gimple_build_catch:                   GIMPLE_CATCH.       (line    6)

  48869. * *gimple_build_cond:                    GIMPLE_COND.        (line    6)

  48870. * *gimple_build_cond_from_tree:          GIMPLE_COND.        (line   14)

  48871. * *gimple_build_debug_bind:              GIMPLE_DEBUG.       (line    6)

  48872. * *gimple_build_eh_filter:               GIMPLE_EH_FILTER.   (line    6)

  48873. * *gimple_build_goto:                    GIMPLE_GOTO.        (line    6)

  48874. * *gimple_build_label:                   GIMPLE_LABEL.       (line    6)

  48875. * *gimple_build_omp_atomic_load:         GIMPLE_OMP_ATOMIC_LOAD.

  48876.                                                              (line    6)

  48877. * *gimple_build_omp_atomic_store:        GIMPLE_OMP_ATOMIC_STORE.

  48878.                                                              (line    6)

  48879. * *gimple_build_omp_continue:            GIMPLE_OMP_CONTINUE.

  48880.                                                              (line    6)

  48881. * *gimple_build_omp_critical:            GIMPLE_OMP_CRITICAL.

  48882.                                                              (line    6)

  48883. * *gimple_build_omp_for:                 GIMPLE_OMP_FOR.     (line    6)

  48884. * *gimple_build_omp_parallel:            GIMPLE_OMP_PARALLEL.

  48885.                                                              (line    6)

  48886. * *gimple_build_omp_sections:            GIMPLE_OMP_SECTIONS.

  48887.                                                              (line    6)

  48888. * *gimple_build_omp_single:              GIMPLE_OMP_SINGLE.  (line    6)

  48889. * *gimple_build_resx:                    GIMPLE_RESX.        (line    6)

  48890. * *gimple_build_return:                  GIMPLE_RETURN.      (line    6)

  48891. * *gimple_build_switch:                  GIMPLE_SWITCH.      (line    6)

  48892. * *gimple_build_try:                     GIMPLE_TRY.         (line    6)

  48893. * + in constraint:                       Modifiers.          (line   12)

  48894. * -fsection-anchors:                     Special Accessors.  (line  117)

  48895. * -fsection-anchors <1>:                 Anchored Addresses. (line    6)

  48896. * /c in RTL dump:                        Flags.              (line  230)

  48897. * /f in RTL dump:                        Flags.              (line  238)

  48898. * /i in RTL dump:                        Flags.              (line  283)

  48899. * /j in RTL dump:                        Flags.              (line  295)

  48900. * /s in RTL dump:                        Flags.              (line  254)

  48901. * /u in RTL dump:                        Flags.              (line  307)

  48902. * /v in RTL dump:                        Flags.              (line  339)

  48903. * 0 in constraint:                       Simple Constraints. (line  128)

  48904. * < in constraint:                       Simple Constraints. (line   47)

  48905. * = in constraint:                       Modifiers.          (line    8)

  48906. * > in constraint:                       Simple Constraints. (line   59)

  48907. * ? in constraint:                       Multi-Alternative.  (line   42)

  48908. * @ in instruction pattern names:        Parameterized Names.

  48909.                                                              (line    6)

  48910. * \:                                     Output Template.    (line   46)

  48911. * ^ in constraint:                       Multi-Alternative.  (line   53)

  48912. * __absvdi2:                             Integer library routines.

  48913.                                                              (line  106)

  48914. * __absvsi2:                             Integer library routines.

  48915.                                                              (line  105)

  48916. * __addda3:                              Fixed-point fractional library routines.

  48917.                                                              (line   52)

  48918. * __adddf3:                              Soft float library routines.

  48919.                                                              (line   22)

  48920. * __adddq3:                              Fixed-point fractional library routines.

  48921.                                                              (line   39)

  48922. * __addha3:                              Fixed-point fractional library routines.

  48923.                                                              (line   49)

  48924. * __addhq3:                              Fixed-point fractional library routines.

  48925.                                                              (line   37)

  48926. * __addqq3:                              Fixed-point fractional library routines.

  48927.                                                              (line   35)

  48928. * __addsa3:                              Fixed-point fractional library routines.

  48929.                                                              (line   51)

  48930. * __addsf3:                              Soft float library routines.

  48931.                                                              (line   21)

  48932. * __addsq3:                              Fixed-point fractional library routines.

  48933.                                                              (line   38)

  48934. * __addta3:                              Fixed-point fractional library routines.

  48935.                                                              (line   53)

  48936. * __addtf3:                              Soft float library routines.

  48937.                                                              (line   23)

  48938. * __adduda3:                             Fixed-point fractional library routines.

  48939.                                                              (line   59)

  48940. * __addudq3:                             Fixed-point fractional library routines.

  48941.                                                              (line   47)

  48942. * __adduha3:                             Fixed-point fractional library routines.

  48943.                                                              (line   55)

  48944. * __adduhq3:                             Fixed-point fractional library routines.

  48945.                                                              (line   43)

  48946. * __adduqq3:                             Fixed-point fractional library routines.

  48947.                                                              (line   41)

  48948. * __addusa3:                             Fixed-point fractional library routines.

  48949.                                                              (line   57)

  48950. * __addusq3:                             Fixed-point fractional library routines.

  48951.                                                              (line   45)

  48952. * __adduta3:                             Fixed-point fractional library routines.

  48953.                                                              (line   61)

  48954. * __addvdi3:                             Integer library routines.

  48955.                                                              (line  110)

  48956. * __addvsi3:                             Integer library routines.

  48957.                                                              (line  109)

  48958. * __addxf3:                              Soft float library routines.

  48959.                                                              (line   25)

  48960. * __ashlda3:                             Fixed-point fractional library routines.

  48961.                                                              (line  358)

  48962. * __ashldi3:                             Integer library routines.

  48963.                                                              (line   13)

  48964. * __ashldq3:                             Fixed-point fractional library routines.

  48965.                                                              (line  346)

  48966. * __ashlha3:                             Fixed-point fractional library routines.

  48967.                                                              (line  356)

  48968. * __ashlhq3:                             Fixed-point fractional library routines.

  48969.                                                              (line  344)

  48970. * __ashlqq3:                             Fixed-point fractional library routines.

  48971.                                                              (line  343)

  48972. * __ashlsa3:                             Fixed-point fractional library routines.

  48973.                                                              (line  357)

  48974. * __ashlsi3:                             Integer library routines.

  48975.                                                              (line   12)

  48976. * __ashlsq3:                             Fixed-point fractional library routines.

  48977.                                                              (line  345)

  48978. * __ashlta3:                             Fixed-point fractional library routines.

  48979.                                                              (line  359)

  48980. * __ashlti3:                             Integer library routines.

  48981.                                                              (line   14)

  48982. * __ashluda3:                            Fixed-point fractional library routines.

  48983.                                                              (line  365)

  48984. * __ashludq3:                            Fixed-point fractional library routines.

  48985.                                                              (line  354)

  48986. * __ashluha3:                            Fixed-point fractional library routines.

  48987.                                                              (line  361)

  48988. * __ashluhq3:                            Fixed-point fractional library routines.

  48989.                                                              (line  350)

  48990. * __ashluqq3:                            Fixed-point fractional library routines.

  48991.                                                              (line  348)

  48992. * __ashlusa3:                            Fixed-point fractional library routines.

  48993.                                                              (line  363)

  48994. * __ashlusq3:                            Fixed-point fractional library routines.

  48995.                                                              (line  352)

  48996. * __ashluta3:                            Fixed-point fractional library routines.

  48997.                                                              (line  367)

  48998. * __ashrda3:                             Fixed-point fractional library routines.

  48999.                                                              (line  378)

  49000. * __ashrdi3:                             Integer library routines.

  49001.                                                              (line   18)

  49002. * __ashrdq3:                             Fixed-point fractional library routines.

  49003.                                                              (line  374)

  49004. * __ashrha3:                             Fixed-point fractional library routines.

  49005.                                                              (line  376)

  49006. * __ashrhq3:                             Fixed-point fractional library routines.

  49007.                                                              (line  372)

  49008. * __ashrqq3:                             Fixed-point fractional library routines.

  49009.                                                              (line  371)

  49010. * __ashrsa3:                             Fixed-point fractional library routines.

  49011.                                                              (line  377)

  49012. * __ashrsi3:                             Integer library routines.

  49013.                                                              (line   17)

  49014. * __ashrsq3:                             Fixed-point fractional library routines.

  49015.                                                              (line  373)

  49016. * __ashrta3:                             Fixed-point fractional library routines.

  49017.                                                              (line  379)

  49018. * __ashrti3:                             Integer library routines.

  49019.                                                              (line   19)

  49020. * __bid_adddd3:                          Decimal float library routines.

  49021.                                                              (line   23)

  49022. * __bid_addsd3:                          Decimal float library routines.

  49023.                                                              (line   19)

  49024. * __bid_addtd3:                          Decimal float library routines.

  49025.                                                              (line   27)

  49026. * __bid_divdd3:                          Decimal float library routines.

  49027.                                                              (line   66)

  49028. * __bid_divsd3:                          Decimal float library routines.

  49029.                                                              (line   62)

  49030. * __bid_divtd3:                          Decimal float library routines.

  49031.                                                              (line   70)

  49032. * __bid_eqdd2:                           Decimal float library routines.

  49033.                                                              (line  258)

  49034. * __bid_eqsd2:                           Decimal float library routines.

  49035.                                                              (line  256)

  49036. * __bid_eqtd2:                           Decimal float library routines.

  49037.                                                              (line  260)

  49038. * __bid_extendddtd2:                     Decimal float library routines.

  49039.                                                              (line   91)

  49040. * __bid_extendddtf:                      Decimal float library routines.

  49041.                                                              (line  139)

  49042. * __bid_extendddxf:                      Decimal float library routines.

  49043.                                                              (line  133)

  49044. * __bid_extenddfdd:                      Decimal float library routines.

  49045.                                                              (line  146)

  49046. * __bid_extenddftd:                      Decimal float library routines.

  49047.                                                              (line  106)

  49048. * __bid_extendsddd2:                     Decimal float library routines.

  49049.                                                              (line   87)

  49050. * __bid_extendsddf:                      Decimal float library routines.

  49051.                                                              (line  127)

  49052. * __bid_extendsdtd2:                     Decimal float library routines.

  49053.                                                              (line   89)

  49054. * __bid_extendsdtf:                      Decimal float library routines.

  49055.                                                              (line  137)

  49056. * __bid_extendsdxf:                      Decimal float library routines.

  49057.                                                              (line  131)

  49058. * __bid_extendsfdd:                      Decimal float library routines.

  49059.                                                              (line  102)

  49060. * __bid_extendsfsd:                      Decimal float library routines.

  49061.                                                              (line  144)

  49062. * __bid_extendsftd:                      Decimal float library routines.

  49063.                                                              (line  104)

  49064. * __bid_extendtftd:                      Decimal float library routines.

  49065.                                                              (line  148)

  49066. * __bid_extendxftd:                      Decimal float library routines.

  49067.                                                              (line  108)

  49068. * __bid_fixdddi:                         Decimal float library routines.

  49069.                                                              (line  169)

  49070. * __bid_fixddsi:                         Decimal float library routines.

  49071.                                                              (line  161)

  49072. * __bid_fixsddi:                         Decimal float library routines.

  49073.                                                              (line  167)

  49074. * __bid_fixsdsi:                         Decimal float library routines.

  49075.                                                              (line  159)

  49076. * __bid_fixtddi:                         Decimal float library routines.

  49077.                                                              (line  171)

  49078. * __bid_fixtdsi:                         Decimal float library routines.

  49079.                                                              (line  163)

  49080. * __bid_fixunsdddi:                      Decimal float library routines.

  49081.                                                              (line  186)

  49082. * __bid_fixunsddsi:                      Decimal float library routines.

  49083.                                                              (line  177)

  49084. * __bid_fixunssddi:                      Decimal float library routines.

  49085.                                                              (line  184)

  49086. * __bid_fixunssdsi:                      Decimal float library routines.

  49087.                                                              (line  175)

  49088. * __bid_fixunstddi:                      Decimal float library routines.

  49089.                                                              (line  188)

  49090. * __bid_fixunstdsi:                      Decimal float library routines.

  49091.                                                              (line  179)

  49092. * __bid_floatdidd:                       Decimal float library routines.

  49093.                                                              (line  204)

  49094. * __bid_floatdisd:                       Decimal float library routines.

  49095.                                                              (line  202)

  49096. * __bid_floatditd:                       Decimal float library routines.

  49097.                                                              (line  206)

  49098. * __bid_floatsidd:                       Decimal float library routines.

  49099.                                                              (line  195)

  49100. * __bid_floatsisd:                       Decimal float library routines.

  49101.                                                              (line  193)

  49102. * __bid_floatsitd:                       Decimal float library routines.

  49103.                                                              (line  197)

  49104. * __bid_floatunsdidd:                    Decimal float library routines.

  49105.                                                              (line  222)

  49106. * __bid_floatunsdisd:                    Decimal float library routines.

  49107.                                                              (line  220)

  49108. * __bid_floatunsditd:                    Decimal float library routines.

  49109.                                                              (line  224)

  49110. * __bid_floatunssidd:                    Decimal float library routines.

  49111.                                                              (line  213)

  49112. * __bid_floatunssisd:                    Decimal float library routines.

  49113.                                                              (line  211)

  49114. * __bid_floatunssitd:                    Decimal float library routines.

  49115.                                                              (line  215)

  49116. * __bid_gedd2:                           Decimal float library routines.

  49117.                                                              (line  276)

  49118. * __bid_gesd2:                           Decimal float library routines.

  49119.                                                              (line  274)

  49120. * __bid_getd2:                           Decimal float library routines.

  49121.                                                              (line  278)

  49122. * __bid_gtdd2:                           Decimal float library routines.

  49123.                                                              (line  303)

  49124. * __bid_gtsd2:                           Decimal float library routines.

  49125.                                                              (line  301)

  49126. * __bid_gttd2:                           Decimal float library routines.

  49127.                                                              (line  305)

  49128. * __bid_ledd2:                           Decimal float library routines.

  49129.                                                              (line  294)

  49130. * __bid_lesd2:                           Decimal float library routines.

  49131.                                                              (line  292)

  49132. * __bid_letd2:                           Decimal float library routines.

  49133.                                                              (line  296)

  49134. * __bid_ltdd2:                           Decimal float library routines.

  49135.                                                              (line  285)

  49136. * __bid_ltsd2:                           Decimal float library routines.

  49137.                                                              (line  283)

  49138. * __bid_lttd2:                           Decimal float library routines.

  49139.                                                              (line  287)

  49140. * __bid_muldd3:                          Decimal float library routines.

  49141.                                                              (line   52)

  49142. * __bid_mulsd3:                          Decimal float library routines.

  49143.                                                              (line   48)

  49144. * __bid_multd3:                          Decimal float library routines.

  49145.                                                              (line   56)

  49146. * __bid_nedd2:                           Decimal float library routines.

  49147.                                                              (line  267)

  49148. * __bid_negdd2:                          Decimal float library routines.

  49149.                                                              (line   77)

  49150. * __bid_negsd2:                          Decimal float library routines.

  49151.                                                              (line   75)

  49152. * __bid_negtd2:                          Decimal float library routines.

  49153.                                                              (line   79)

  49154. * __bid_nesd2:                           Decimal float library routines.

  49155.                                                              (line  265)

  49156. * __bid_netd2:                           Decimal float library routines.

  49157.                                                              (line  269)

  49158. * __bid_subdd3:                          Decimal float library routines.

  49159.                                                              (line   37)

  49160. * __bid_subsd3:                          Decimal float library routines.

  49161.                                                              (line   33)

  49162. * __bid_subtd3:                          Decimal float library routines.

  49163.                                                              (line   41)

  49164. * __bid_truncdddf:                       Decimal float library routines.

  49165.                                                              (line  152)

  49166. * __bid_truncddsd2:                      Decimal float library routines.

  49167.                                                              (line   93)

  49168. * __bid_truncddsf:                       Decimal float library routines.

  49169.                                                              (line  123)

  49170. * __bid_truncdfsd:                       Decimal float library routines.

  49171.                                                              (line  110)

  49172. * __bid_truncsdsf:                       Decimal float library routines.

  49173.                                                              (line  150)

  49174. * __bid_trunctddd2:                      Decimal float library routines.

  49175.                                                              (line   97)

  49176. * __bid_trunctddf:                       Decimal float library routines.

  49177.                                                              (line  129)

  49178. * __bid_trunctdsd2:                      Decimal float library routines.

  49179.                                                              (line   95)

  49180. * __bid_trunctdsf:                       Decimal float library routines.

  49181.                                                              (line  125)

  49182. * __bid_trunctdtf:                       Decimal float library routines.

  49183.                                                              (line  154)

  49184. * __bid_trunctdxf:                       Decimal float library routines.

  49185.                                                              (line  135)

  49186. * __bid_trunctfdd:                       Decimal float library routines.

  49187.                                                              (line  118)

  49188. * __bid_trunctfsd:                       Decimal float library routines.

  49189.                                                              (line  114)

  49190. * __bid_truncxfdd:                       Decimal float library routines.

  49191.                                                              (line  116)

  49192. * __bid_truncxfsd:                       Decimal float library routines.

  49193.                                                              (line  112)

  49194. * __bid_unorddd2:                        Decimal float library routines.

  49195.                                                              (line  234)

  49196. * __bid_unordsd2:                        Decimal float library routines.

  49197.                                                              (line  232)

  49198. * __bid_unordtd2:                        Decimal float library routines.

  49199.                                                              (line  236)

  49200. * __bswapdi2:                            Integer library routines.

  49201.                                                              (line  161)

  49202. * __bswapsi2:                            Integer library routines.

  49203.                                                              (line  160)

  49204. * __builtin_classify_type:               Varargs.            (line   48)

  49205. * __builtin_next_arg:                    Varargs.            (line   39)

  49206. * __builtin_saveregs:                    Varargs.            (line   22)

  49207. * __clear_cache:                         Miscellaneous routines.

  49208.                                                              (line    9)

  49209. * __clzdi2:                              Integer library routines.

  49210.                                                              (line  130)

  49211. * __clzsi2:                              Integer library routines.

  49212.                                                              (line  129)

  49213. * __clzti2:                              Integer library routines.

  49214.                                                              (line  131)

  49215. * __cmpda2:                              Fixed-point fractional library routines.

  49216.                                                              (line  458)

  49217. * __cmpdf2:                              Soft float library routines.

  49218.                                                              (line  163)

  49219. * __cmpdi2:                              Integer library routines.

  49220.                                                              (line   86)

  49221. * __cmpdq2:                              Fixed-point fractional library routines.

  49222.                                                              (line  447)

  49223. * __cmpha2:                              Fixed-point fractional library routines.

  49224.                                                              (line  456)

  49225. * __cmphq2:                              Fixed-point fractional library routines.

  49226.                                                              (line  445)

  49227. * __cmpqq2:                              Fixed-point fractional library routines.

  49228.                                                              (line  444)

  49229. * __cmpsa2:                              Fixed-point fractional library routines.

  49230.                                                              (line  457)

  49231. * __cmpsf2:                              Soft float library routines.

  49232.                                                              (line  162)

  49233. * __cmpsq2:                              Fixed-point fractional library routines.

  49234.                                                              (line  446)

  49235. * __cmpta2:                              Fixed-point fractional library routines.

  49236.                                                              (line  459)

  49237. * __cmptf2:                              Soft float library routines.

  49238.                                                              (line  164)

  49239. * __cmpti2:                              Integer library routines.

  49240.                                                              (line   87)

  49241. * __cmpuda2:                             Fixed-point fractional library routines.

  49242.                                                              (line  464)

  49243. * __cmpudq2:                             Fixed-point fractional library routines.

  49244.                                                              (line  454)

  49245. * __cmpuha2:                             Fixed-point fractional library routines.

  49246.                                                              (line  461)

  49247. * __cmpuhq2:                             Fixed-point fractional library routines.

  49248.                                                              (line  451)

  49249. * __cmpuqq2:                             Fixed-point fractional library routines.

  49250.                                                              (line  449)

  49251. * __cmpusa2:                             Fixed-point fractional library routines.

  49252.                                                              (line  463)

  49253. * __cmpusq2:                             Fixed-point fractional library routines.

  49254.                                                              (line  452)

  49255. * __cmputa2:                             Fixed-point fractional library routines.

  49256.                                                              (line  466)

  49257. * __CTOR_LIST__:                         Initialization.     (line   25)

  49258. * __ctzdi2:                              Integer library routines.

  49259.                                                              (line  137)

  49260. * __ctzsi2:                              Integer library routines.

  49261.                                                              (line  136)

  49262. * __ctzti2:                              Integer library routines.

  49263.                                                              (line  138)

  49264. * __divda3:                              Fixed-point fractional library routines.

  49265.                                                              (line  234)

  49266. * __divdc3:                              Soft float library routines.

  49267.                                                              (line  250)

  49268. * __divdf3:                              Soft float library routines.

  49269.                                                              (line   47)

  49270. * __divdi3:                              Integer library routines.

  49271.                                                              (line   24)

  49272. * __divdq3:                              Fixed-point fractional library routines.

  49273.                                                              (line  229)

  49274. * __divha3:                              Fixed-point fractional library routines.

  49275.                                                              (line  231)

  49276. * __divhq3:                              Fixed-point fractional library routines.

  49277.                                                              (line  227)

  49278. * __divqq3:                              Fixed-point fractional library routines.

  49279.                                                              (line  225)

  49280. * __divsa3:                              Fixed-point fractional library routines.

  49281.                                                              (line  233)

  49282. * __divsc3:                              Soft float library routines.

  49283.                                                              (line  248)

  49284. * __divsf3:                              Soft float library routines.

  49285.                                                              (line   46)

  49286. * __divsi3:                              Integer library routines.

  49287.                                                              (line   23)

  49288. * __divsq3:                              Fixed-point fractional library routines.

  49289.                                                              (line  228)

  49290. * __divta3:                              Fixed-point fractional library routines.

  49291.                                                              (line  235)

  49292. * __divtc3:                              Soft float library routines.

  49293.                                                              (line  252)

  49294. * __divtf3:                              Soft float library routines.

  49295.                                                              (line   48)

  49296. * __divti3:                              Integer library routines.

  49297.                                                              (line   25)

  49298. * __divxc3:                              Soft float library routines.

  49299.                                                              (line  254)

  49300. * __divxf3:                              Soft float library routines.

  49301.                                                              (line   50)

  49302. * __dpd_adddd3:                          Decimal float library routines.

  49303.                                                              (line   21)

  49304. * __dpd_addsd3:                          Decimal float library routines.

  49305.                                                              (line   17)

  49306. * __dpd_addtd3:                          Decimal float library routines.

  49307.                                                              (line   25)

  49308. * __dpd_divdd3:                          Decimal float library routines.

  49309.                                                              (line   64)

  49310. * __dpd_divsd3:                          Decimal float library routines.

  49311.                                                              (line   60)

  49312. * __dpd_divtd3:                          Decimal float library routines.

  49313.                                                              (line   68)

  49314. * __dpd_eqdd2:                           Decimal float library routines.

  49315.                                                              (line  257)

  49316. * __dpd_eqsd2:                           Decimal float library routines.

  49317.                                                              (line  255)

  49318. * __dpd_eqtd2:                           Decimal float library routines.

  49319.                                                              (line  259)

  49320. * __dpd_extendddtd2:                     Decimal float library routines.

  49321.                                                              (line   90)

  49322. * __dpd_extendddtf:                      Decimal float library routines.

  49323.                                                              (line  138)

  49324. * __dpd_extendddxf:                      Decimal float library routines.

  49325.                                                              (line  132)

  49326. * __dpd_extenddfdd:                      Decimal float library routines.

  49327.                                                              (line  145)

  49328. * __dpd_extenddftd:                      Decimal float library routines.

  49329.                                                              (line  105)

  49330. * __dpd_extendsddd2:                     Decimal float library routines.

  49331.                                                              (line   86)

  49332. * __dpd_extendsddf:                      Decimal float library routines.

  49333.                                                              (line  126)

  49334. * __dpd_extendsdtd2:                     Decimal float library routines.

  49335.                                                              (line   88)

  49336. * __dpd_extendsdtf:                      Decimal float library routines.

  49337.                                                              (line  136)

  49338. * __dpd_extendsdxf:                      Decimal float library routines.

  49339.                                                              (line  130)

  49340. * __dpd_extendsfdd:                      Decimal float library routines.

  49341.                                                              (line  101)

  49342. * __dpd_extendsfsd:                      Decimal float library routines.

  49343.                                                              (line  143)

  49344. * __dpd_extendsftd:                      Decimal float library routines.

  49345.                                                              (line  103)

  49346. * __dpd_extendtftd:                      Decimal float library routines.

  49347.                                                              (line  147)

  49348. * __dpd_extendxftd:                      Decimal float library routines.

  49349.                                                              (line  107)

  49350. * __dpd_fixdddi:                         Decimal float library routines.

  49351.                                                              (line  168)

  49352. * __dpd_fixddsi:                         Decimal float library routines.

  49353.                                                              (line  160)

  49354. * __dpd_fixsddi:                         Decimal float library routines.

  49355.                                                              (line  166)

  49356. * __dpd_fixsdsi:                         Decimal float library routines.

  49357.                                                              (line  158)

  49358. * __dpd_fixtddi:                         Decimal float library routines.

  49359.                                                              (line  170)

  49360. * __dpd_fixtdsi:                         Decimal float library routines.

  49361.                                                              (line  162)

  49362. * __dpd_fixunsdddi:                      Decimal float library routines.

  49363.                                                              (line  185)

  49364. * __dpd_fixunsddsi:                      Decimal float library routines.

  49365.                                                              (line  176)

  49366. * __dpd_fixunssddi:                      Decimal float library routines.

  49367.                                                              (line  183)

  49368. * __dpd_fixunssdsi:                      Decimal float library routines.

  49369.                                                              (line  174)

  49370. * __dpd_fixunstddi:                      Decimal float library routines.

  49371.                                                              (line  187)

  49372. * __dpd_fixunstdsi:                      Decimal float library routines.

  49373.                                                              (line  178)

  49374. * __dpd_floatdidd:                       Decimal float library routines.

  49375.                                                              (line  203)

  49376. * __dpd_floatdisd:                       Decimal float library routines.

  49377.                                                              (line  201)

  49378. * __dpd_floatditd:                       Decimal float library routines.

  49379.                                                              (line  205)

  49380. * __dpd_floatsidd:                       Decimal float library routines.

  49381.                                                              (line  194)

  49382. * __dpd_floatsisd:                       Decimal float library routines.

  49383.                                                              (line  192)

  49384. * __dpd_floatsitd:                       Decimal float library routines.

  49385.                                                              (line  196)

  49386. * __dpd_floatunsdidd:                    Decimal float library routines.

  49387.                                                              (line  221)

  49388. * __dpd_floatunsdisd:                    Decimal float library routines.

  49389.                                                              (line  219)

  49390. * __dpd_floatunsditd:                    Decimal float library routines.

  49391.                                                              (line  223)

  49392. * __dpd_floatunssidd:                    Decimal float library routines.

  49393.                                                              (line  212)

  49394. * __dpd_floatunssisd:                    Decimal float library routines.

  49395.                                                              (line  210)

  49396. * __dpd_floatunssitd:                    Decimal float library routines.

  49397.                                                              (line  214)

  49398. * __dpd_gedd2:                           Decimal float library routines.

  49399.                                                              (line  275)

  49400. * __dpd_gesd2:                           Decimal float library routines.

  49401.                                                              (line  273)

  49402. * __dpd_getd2:                           Decimal float library routines.

  49403.                                                              (line  277)

  49404. * __dpd_gtdd2:                           Decimal float library routines.

  49405.                                                              (line  302)

  49406. * __dpd_gtsd2:                           Decimal float library routines.

  49407.                                                              (line  300)

  49408. * __dpd_gttd2:                           Decimal float library routines.

  49409.                                                              (line  304)

  49410. * __dpd_ledd2:                           Decimal float library routines.

  49411.                                                              (line  293)

  49412. * __dpd_lesd2:                           Decimal float library routines.

  49413.                                                              (line  291)

  49414. * __dpd_letd2:                           Decimal float library routines.

  49415.                                                              (line  295)

  49416. * __dpd_ltdd2:                           Decimal float library routines.

  49417.                                                              (line  284)

  49418. * __dpd_ltsd2:                           Decimal float library routines.

  49419.                                                              (line  282)

  49420. * __dpd_lttd2:                           Decimal float library routines.

  49421.                                                              (line  286)

  49422. * __dpd_muldd3:                          Decimal float library routines.

  49423.                                                              (line   50)

  49424. * __dpd_mulsd3:                          Decimal float library routines.

  49425.                                                              (line   46)

  49426. * __dpd_multd3:                          Decimal float library routines.

  49427.                                                              (line   54)

  49428. * __dpd_nedd2:                           Decimal float library routines.

  49429.                                                              (line  266)

  49430. * __dpd_negdd2:                          Decimal float library routines.

  49431.                                                              (line   76)

  49432. * __dpd_negsd2:                          Decimal float library routines.

  49433.                                                              (line   74)

  49434. * __dpd_negtd2:                          Decimal float library routines.

  49435.                                                              (line   78)

  49436. * __dpd_nesd2:                           Decimal float library routines.

  49437.                                                              (line  264)

  49438. * __dpd_netd2:                           Decimal float library routines.

  49439.                                                              (line  268)

  49440. * __dpd_subdd3:                          Decimal float library routines.

  49441.                                                              (line   35)

  49442. * __dpd_subsd3:                          Decimal float library routines.

  49443.                                                              (line   31)

  49444. * __dpd_subtd3:                          Decimal float library routines.

  49445.                                                              (line   39)

  49446. * __dpd_truncdddf:                       Decimal float library routines.

  49447.                                                              (line  151)

  49448. * __dpd_truncddsd2:                      Decimal float library routines.

  49449.                                                              (line   92)

  49450. * __dpd_truncddsf:                       Decimal float library routines.

  49451.                                                              (line  122)

  49452. * __dpd_truncdfsd:                       Decimal float library routines.

  49453.                                                              (line  109)

  49454. * __dpd_truncsdsf:                       Decimal float library routines.

  49455.                                                              (line  149)

  49456. * __dpd_trunctddd2:                      Decimal float library routines.

  49457.                                                              (line   96)

  49458. * __dpd_trunctddf:                       Decimal float library routines.

  49459.                                                              (line  128)

  49460. * __dpd_trunctdsd2:                      Decimal float library routines.

  49461.                                                              (line   94)

  49462. * __dpd_trunctdsf:                       Decimal float library routines.

  49463.                                                              (line  124)

  49464. * __dpd_trunctdtf:                       Decimal float library routines.

  49465.                                                              (line  153)

  49466. * __dpd_trunctdxf:                       Decimal float library routines.

  49467.                                                              (line  134)

  49468. * __dpd_trunctfdd:                       Decimal float library routines.

  49469.                                                              (line  117)

  49470. * __dpd_trunctfsd:                       Decimal float library routines.

  49471.                                                              (line  113)

  49472. * __dpd_truncxfdd:                       Decimal float library routines.

  49473.                                                              (line  115)

  49474. * __dpd_truncxfsd:                       Decimal float library routines.

  49475.                                                              (line  111)

  49476. * __dpd_unorddd2:                        Decimal float library routines.

  49477.                                                              (line  233)

  49478. * __dpd_unordsd2:                        Decimal float library routines.

  49479.                                                              (line  231)

  49480. * __dpd_unordtd2:                        Decimal float library routines.

  49481.                                                              (line  235)

  49482. * __DTOR_LIST__:                         Initialization.     (line   25)

  49483. * __eqdf2:                               Soft float library routines.

  49484.                                                              (line  193)

  49485. * __eqsf2:                               Soft float library routines.

  49486.                                                              (line  192)

  49487. * __eqtf2:                               Soft float library routines.

  49488.                                                              (line  194)

  49489. * __extenddftf2:                         Soft float library routines.

  49490.                                                              (line   67)

  49491. * __extenddfxf2:                         Soft float library routines.

  49492.                                                              (line   68)

  49493. * __extendsfdf2:                         Soft float library routines.

  49494.                                                              (line   64)

  49495. * __extendsftf2:                         Soft float library routines.

  49496.                                                              (line   65)

  49497. * __extendsfxf2:                         Soft float library routines.

  49498.                                                              (line   66)

  49499. * __ffsdi2:                              Integer library routines.

  49500.                                                              (line  143)

  49501. * __ffsti2:                              Integer library routines.

  49502.                                                              (line  144)

  49503. * __fixdfdi:                             Soft float library routines.

  49504.                                                              (line   87)

  49505. * __fixdfsi:                             Soft float library routines.

  49506.                                                              (line   80)

  49507. * __fixdfti:                             Soft float library routines.

  49508.                                                              (line   93)

  49509. * __fixsfdi:                             Soft float library routines.

  49510.                                                              (line   86)

  49511. * __fixsfsi:                             Soft float library routines.

  49512.                                                              (line   79)

  49513. * __fixsfti:                             Soft float library routines.

  49514.                                                              (line   92)

  49515. * __fixtfdi:                             Soft float library routines.

  49516.                                                              (line   88)

  49517. * __fixtfsi:                             Soft float library routines.

  49518.                                                              (line   81)

  49519. * __fixtfti:                             Soft float library routines.

  49520.                                                              (line   94)

  49521. * __fixunsdfdi:                          Soft float library routines.

  49522.                                                              (line  107)

  49523. * __fixunsdfsi:                          Soft float library routines.

  49524.                                                              (line  100)

  49525. * __fixunsdfti:                          Soft float library routines.

  49526.                                                              (line  114)

  49527. * __fixunssfdi:                          Soft float library routines.

  49528.                                                              (line  106)

  49529. * __fixunssfsi:                          Soft float library routines.

  49530.                                                              (line   99)

  49531. * __fixunssfti:                          Soft float library routines.

  49532.                                                              (line  113)

  49533. * __fixunstfdi:                          Soft float library routines.

  49534.                                                              (line  108)

  49535. * __fixunstfsi:                          Soft float library routines.

  49536.                                                              (line  101)

  49537. * __fixunstfti:                          Soft float library routines.

  49538.                                                              (line  115)

  49539. * __fixunsxfdi:                          Soft float library routines.

  49540.                                                              (line  109)

  49541. * __fixunsxfsi:                          Soft float library routines.

  49542.                                                              (line  102)

  49543. * __fixunsxfti:                          Soft float library routines.

  49544.                                                              (line  116)

  49545. * __fixxfdi:                             Soft float library routines.

  49546.                                                              (line   89)

  49547. * __fixxfsi:                             Soft float library routines.

  49548.                                                              (line   82)

  49549. * __fixxfti:                             Soft float library routines.

  49550.                                                              (line   95)

  49551. * __floatdidf:                           Soft float library routines.

  49552.                                                              (line  127)

  49553. * __floatdisf:                           Soft float library routines.

  49554.                                                              (line  126)

  49555. * __floatditf:                           Soft float library routines.

  49556.                                                              (line  128)

  49557. * __floatdixf:                           Soft float library routines.

  49558.                                                              (line  129)

  49559. * __floatsidf:                           Soft float library routines.

  49560.                                                              (line  121)

  49561. * __floatsisf:                           Soft float library routines.

  49562.                                                              (line  120)

  49563. * __floatsitf:                           Soft float library routines.

  49564.                                                              (line  122)

  49565. * __floatsixf:                           Soft float library routines.

  49566.                                                              (line  123)

  49567. * __floattidf:                           Soft float library routines.

  49568.                                                              (line  133)

  49569. * __floattisf:                           Soft float library routines.

  49570.                                                              (line  132)

  49571. * __floattitf:                           Soft float library routines.

  49572.                                                              (line  134)

  49573. * __floattixf:                           Soft float library routines.

  49574.                                                              (line  135)

  49575. * __floatundidf:                         Soft float library routines.

  49576.                                                              (line  145)

  49577. * __floatundisf:                         Soft float library routines.

  49578.                                                              (line  144)

  49579. * __floatunditf:                         Soft float library routines.

  49580.                                                              (line  146)

  49581. * __floatundixf:                         Soft float library routines.

  49582.                                                              (line  147)

  49583. * __floatunsidf:                         Soft float library routines.

  49584.                                                              (line  139)

  49585. * __floatunsisf:                         Soft float library routines.

  49586.                                                              (line  138)

  49587. * __floatunsitf:                         Soft float library routines.

  49588.                                                              (line  140)

  49589. * __floatunsixf:                         Soft float library routines.

  49590.                                                              (line  141)

  49591. * __floatuntidf:                         Soft float library routines.

  49592.                                                              (line  151)

  49593. * __floatuntisf:                         Soft float library routines.

  49594.                                                              (line  150)

  49595. * __floatuntitf:                         Soft float library routines.

  49596.                                                              (line  152)

  49597. * __floatuntixf:                         Soft float library routines.

  49598.                                                              (line  153)

  49599. * __fractdadf:                           Fixed-point fractional library routines.

  49600.                                                              (line  643)

  49601. * __fractdadi:                           Fixed-point fractional library routines.

  49602.                                                              (line  640)

  49603. * __fractdadq:                           Fixed-point fractional library routines.

  49604.                                                              (line  623)

  49605. * __fractdaha2:                          Fixed-point fractional library routines.

  49606.                                                              (line  624)

  49607. * __fractdahi:                           Fixed-point fractional library routines.

  49608.                                                              (line  638)

  49609. * __fractdahq:                           Fixed-point fractional library routines.

  49610.                                                              (line  621)

  49611. * __fractdaqi:                           Fixed-point fractional library routines.

  49612.                                                              (line  637)

  49613. * __fractdaqq:                           Fixed-point fractional library routines.

  49614.                                                              (line  620)

  49615. * __fractdasa2:                          Fixed-point fractional library routines.

  49616.                                                              (line  625)

  49617. * __fractdasf:                           Fixed-point fractional library routines.

  49618.                                                              (line  642)

  49619. * __fractdasi:                           Fixed-point fractional library routines.

  49620.                                                              (line  639)

  49621. * __fractdasq:                           Fixed-point fractional library routines.

  49622.                                                              (line  622)

  49623. * __fractdata2:                          Fixed-point fractional library routines.

  49624.                                                              (line  626)

  49625. * __fractdati:                           Fixed-point fractional library routines.

  49626.                                                              (line  641)

  49627. * __fractdauda:                          Fixed-point fractional library routines.

  49628.                                                              (line  634)

  49629. * __fractdaudq:                          Fixed-point fractional library routines.

  49630.                                                              (line  630)

  49631. * __fractdauha:                          Fixed-point fractional library routines.

  49632.                                                              (line  632)

  49633. * __fractdauhq:                          Fixed-point fractional library routines.

  49634.                                                              (line  628)

  49635. * __fractdauqq:                          Fixed-point fractional library routines.

  49636.                                                              (line  627)

  49637. * __fractdausa:                          Fixed-point fractional library routines.

  49638.                                                              (line  633)

  49639. * __fractdausq:                          Fixed-point fractional library routines.

  49640.                                                              (line  629)

  49641. * __fractdauta:                          Fixed-point fractional library routines.

  49642.                                                              (line  635)

  49643. * __fractdfda:                           Fixed-point fractional library routines.

  49644.                                                              (line 1032)

  49645. * __fractdfdq:                           Fixed-point fractional library routines.

  49646.                                                              (line 1029)

  49647. * __fractdfha:                           Fixed-point fractional library routines.

  49648.                                                              (line 1030)

  49649. * __fractdfhq:                           Fixed-point fractional library routines.

  49650.                                                              (line 1027)

  49651. * __fractdfqq:                           Fixed-point fractional library routines.

  49652.                                                              (line 1026)

  49653. * __fractdfsa:                           Fixed-point fractional library routines.

  49654.                                                              (line 1031)

  49655. * __fractdfsq:                           Fixed-point fractional library routines.

  49656.                                                              (line 1028)

  49657. * __fractdfta:                           Fixed-point fractional library routines.

  49658.                                                              (line 1033)

  49659. * __fractdfuda:                          Fixed-point fractional library routines.

  49660.                                                              (line 1040)

  49661. * __fractdfudq:                          Fixed-point fractional library routines.

  49662.                                                              (line 1037)

  49663. * __fractdfuha:                          Fixed-point fractional library routines.

  49664.                                                              (line 1038)

  49665. * __fractdfuhq:                          Fixed-point fractional library routines.

  49666.                                                              (line 1035)

  49667. * __fractdfuqq:                          Fixed-point fractional library routines.

  49668.                                                              (line 1034)

  49669. * __fractdfusa:                          Fixed-point fractional library routines.

  49670.                                                              (line 1039)

  49671. * __fractdfusq:                          Fixed-point fractional library routines.

  49672.                                                              (line 1036)

  49673. * __fractdfuta:                          Fixed-point fractional library routines.

  49674.                                                              (line 1041)

  49675. * __fractdida:                           Fixed-point fractional library routines.

  49676.                                                              (line  982)

  49677. * __fractdidq:                           Fixed-point fractional library routines.

  49678.                                                              (line  979)

  49679. * __fractdiha:                           Fixed-point fractional library routines.

  49680.                                                              (line  980)

  49681. * __fractdihq:                           Fixed-point fractional library routines.

  49682.                                                              (line  977)

  49683. * __fractdiqq:                           Fixed-point fractional library routines.

  49684.                                                              (line  976)

  49685. * __fractdisa:                           Fixed-point fractional library routines.

  49686.                                                              (line  981)

  49687. * __fractdisq:                           Fixed-point fractional library routines.

  49688.                                                              (line  978)

  49689. * __fractdita:                           Fixed-point fractional library routines.

  49690.                                                              (line  983)

  49691. * __fractdiuda:                          Fixed-point fractional library routines.

  49692.                                                              (line  990)

  49693. * __fractdiudq:                          Fixed-point fractional library routines.

  49694.                                                              (line  987)

  49695. * __fractdiuha:                          Fixed-point fractional library routines.

  49696.                                                              (line  988)

  49697. * __fractdiuhq:                          Fixed-point fractional library routines.

  49698.                                                              (line  985)

  49699. * __fractdiuqq:                          Fixed-point fractional library routines.

  49700.                                                              (line  984)

  49701. * __fractdiusa:                          Fixed-point fractional library routines.

  49702.                                                              (line  989)

  49703. * __fractdiusq:                          Fixed-point fractional library routines.

  49704.                                                              (line  986)

  49705. * __fractdiuta:                          Fixed-point fractional library routines.

  49706.                                                              (line  991)

  49707. * __fractdqda:                           Fixed-point fractional library routines.

  49708.                                                              (line  551)

  49709. * __fractdqdf:                           Fixed-point fractional library routines.

  49710.                                                              (line  573)

  49711. * __fractdqdi:                           Fixed-point fractional library routines.

  49712.                                                              (line  570)

  49713. * __fractdqha:                           Fixed-point fractional library routines.

  49714.                                                              (line  549)

  49715. * __fractdqhi:                           Fixed-point fractional library routines.

  49716.                                                              (line  568)

  49717. * __fractdqhq2:                          Fixed-point fractional library routines.

  49718.                                                              (line  547)

  49719. * __fractdqqi:                           Fixed-point fractional library routines.

  49720.                                                              (line  567)

  49721. * __fractdqqq2:                          Fixed-point fractional library routines.

  49722.                                                              (line  546)

  49723. * __fractdqsa:                           Fixed-point fractional library routines.

  49724.                                                              (line  550)

  49725. * __fractdqsf:                           Fixed-point fractional library routines.

  49726.                                                              (line  572)

  49727. * __fractdqsi:                           Fixed-point fractional library routines.

  49728.                                                              (line  569)

  49729. * __fractdqsq2:                          Fixed-point fractional library routines.

  49730.                                                              (line  548)

  49731. * __fractdqta:                           Fixed-point fractional library routines.

  49732.                                                              (line  552)

  49733. * __fractdqti:                           Fixed-point fractional library routines.

  49734.                                                              (line  571)

  49735. * __fractdquda:                          Fixed-point fractional library routines.

  49736.                                                              (line  563)

  49737. * __fractdqudq:                          Fixed-point fractional library routines.

  49738.                                                              (line  558)

  49739. * __fractdquha:                          Fixed-point fractional library routines.

  49740.                                                              (line  560)

  49741. * __fractdquhq:                          Fixed-point fractional library routines.

  49742.                                                              (line  555)

  49743. * __fractdquqq:                          Fixed-point fractional library routines.

  49744.                                                              (line  553)

  49745. * __fractdqusa:                          Fixed-point fractional library routines.

  49746.                                                              (line  562)

  49747. * __fractdqusq:                          Fixed-point fractional library routines.

  49748.                                                              (line  556)

  49749. * __fractdquta:                          Fixed-point fractional library routines.

  49750.                                                              (line  565)

  49751. * __fracthada2:                          Fixed-point fractional library routines.

  49752.                                                              (line  579)

  49753. * __fracthadf:                           Fixed-point fractional library routines.

  49754.                                                              (line  597)

  49755. * __fracthadi:                           Fixed-point fractional library routines.

  49756.                                                              (line  594)

  49757. * __fracthadq:                           Fixed-point fractional library routines.

  49758.                                                              (line  577)

  49759. * __fracthahi:                           Fixed-point fractional library routines.

  49760.                                                              (line  592)

  49761. * __fracthahq:                           Fixed-point fractional library routines.

  49762.                                                              (line  575)

  49763. * __fracthaqi:                           Fixed-point fractional library routines.

  49764.                                                              (line  591)

  49765. * __fracthaqq:                           Fixed-point fractional library routines.

  49766.                                                              (line  574)

  49767. * __fracthasa2:                          Fixed-point fractional library routines.

  49768.                                                              (line  578)

  49769. * __fracthasf:                           Fixed-point fractional library routines.

  49770.                                                              (line  596)

  49771. * __fracthasi:                           Fixed-point fractional library routines.

  49772.                                                              (line  593)

  49773. * __fracthasq:                           Fixed-point fractional library routines.

  49774.                                                              (line  576)

  49775. * __fracthata2:                          Fixed-point fractional library routines.

  49776.                                                              (line  580)

  49777. * __fracthati:                           Fixed-point fractional library routines.

  49778.                                                              (line  595)

  49779. * __fracthauda:                          Fixed-point fractional library routines.

  49780.                                                              (line  588)

  49781. * __fracthaudq:                          Fixed-point fractional library routines.

  49782.                                                              (line  584)

  49783. * __fracthauha:                          Fixed-point fractional library routines.

  49784.                                                              (line  586)

  49785. * __fracthauhq:                          Fixed-point fractional library routines.

  49786.                                                              (line  582)

  49787. * __fracthauqq:                          Fixed-point fractional library routines.

  49788.                                                              (line  581)

  49789. * __fracthausa:                          Fixed-point fractional library routines.

  49790.                                                              (line  587)

  49791. * __fracthausq:                          Fixed-point fractional library routines.

  49792.                                                              (line  583)

  49793. * __fracthauta:                          Fixed-point fractional library routines.

  49794.                                                              (line  589)

  49795. * __fracthida:                           Fixed-point fractional library routines.

  49796.                                                              (line  950)

  49797. * __fracthidq:                           Fixed-point fractional library routines.

  49798.                                                              (line  947)

  49799. * __fracthiha:                           Fixed-point fractional library routines.

  49800.                                                              (line  948)

  49801. * __fracthihq:                           Fixed-point fractional library routines.

  49802.                                                              (line  945)

  49803. * __fracthiqq:                           Fixed-point fractional library routines.

  49804.                                                              (line  944)

  49805. * __fracthisa:                           Fixed-point fractional library routines.

  49806.                                                              (line  949)

  49807. * __fracthisq:                           Fixed-point fractional library routines.

  49808.                                                              (line  946)

  49809. * __fracthita:                           Fixed-point fractional library routines.

  49810.                                                              (line  951)

  49811. * __fracthiuda:                          Fixed-point fractional library routines.

  49812.                                                              (line  958)

  49813. * __fracthiudq:                          Fixed-point fractional library routines.

  49814.                                                              (line  955)

  49815. * __fracthiuha:                          Fixed-point fractional library routines.

  49816.                                                              (line  956)

  49817. * __fracthiuhq:                          Fixed-point fractional library routines.

  49818.                                                              (line  953)

  49819. * __fracthiuqq:                          Fixed-point fractional library routines.

  49820.                                                              (line  952)

  49821. * __fracthiusa:                          Fixed-point fractional library routines.

  49822.                                                              (line  957)

  49823. * __fracthiusq:                          Fixed-point fractional library routines.

  49824.                                                              (line  954)

  49825. * __fracthiuta:                          Fixed-point fractional library routines.

  49826.                                                              (line  959)

  49827. * __fracthqda:                           Fixed-point fractional library routines.

  49828.                                                              (line  505)

  49829. * __fracthqdf:                           Fixed-point fractional library routines.

  49830.                                                              (line  521)

  49831. * __fracthqdi:                           Fixed-point fractional library routines.

  49832.                                                              (line  518)

  49833. * __fracthqdq2:                          Fixed-point fractional library routines.

  49834.                                                              (line  502)

  49835. * __fracthqha:                           Fixed-point fractional library routines.

  49836.                                                              (line  503)

  49837. * __fracthqhi:                           Fixed-point fractional library routines.

  49838.                                                              (line  516)

  49839. * __fracthqqi:                           Fixed-point fractional library routines.

  49840.                                                              (line  515)

  49841. * __fracthqqq2:                          Fixed-point fractional library routines.

  49842.                                                              (line  500)

  49843. * __fracthqsa:                           Fixed-point fractional library routines.

  49844.                                                              (line  504)

  49845. * __fracthqsf:                           Fixed-point fractional library routines.

  49846.                                                              (line  520)

  49847. * __fracthqsi:                           Fixed-point fractional library routines.

  49848.                                                              (line  517)

  49849. * __fracthqsq2:                          Fixed-point fractional library routines.

  49850.                                                              (line  501)

  49851. * __fracthqta:                           Fixed-point fractional library routines.

  49852.                                                              (line  506)

  49853. * __fracthqti:                           Fixed-point fractional library routines.

  49854.                                                              (line  519)

  49855. * __fracthquda:                          Fixed-point fractional library routines.

  49856.                                                              (line  513)

  49857. * __fracthqudq:                          Fixed-point fractional library routines.

  49858.                                                              (line  510)

  49859. * __fracthquha:                          Fixed-point fractional library routines.

  49860.                                                              (line  511)

  49861. * __fracthquhq:                          Fixed-point fractional library routines.

  49862.                                                              (line  508)

  49863. * __fracthquqq:                          Fixed-point fractional library routines.

  49864.                                                              (line  507)

  49865. * __fracthqusa:                          Fixed-point fractional library routines.

  49866.                                                              (line  512)

  49867. * __fracthqusq:                          Fixed-point fractional library routines.

  49868.                                                              (line  509)

  49869. * __fracthquta:                          Fixed-point fractional library routines.

  49870.                                                              (line  514)

  49871. * __fractqida:                           Fixed-point fractional library routines.

  49872.                                                              (line  932)

  49873. * __fractqidq:                           Fixed-point fractional library routines.

  49874.                                                              (line  929)

  49875. * __fractqiha:                           Fixed-point fractional library routines.

  49876.                                                              (line  930)

  49877. * __fractqihq:                           Fixed-point fractional library routines.

  49878.                                                              (line  927)

  49879. * __fractqiqq:                           Fixed-point fractional library routines.

  49880.                                                              (line  926)

  49881. * __fractqisa:                           Fixed-point fractional library routines.

  49882.                                                              (line  931)

  49883. * __fractqisq:                           Fixed-point fractional library routines.

  49884.                                                              (line  928)

  49885. * __fractqita:                           Fixed-point fractional library routines.

  49886.                                                              (line  933)

  49887. * __fractqiuda:                          Fixed-point fractional library routines.

  49888.                                                              (line  941)

  49889. * __fractqiudq:                          Fixed-point fractional library routines.

  49890.                                                              (line  937)

  49891. * __fractqiuha:                          Fixed-point fractional library routines.

  49892.                                                              (line  939)

  49893. * __fractqiuhq:                          Fixed-point fractional library routines.

  49894.                                                              (line  935)

  49895. * __fractqiuqq:                          Fixed-point fractional library routines.

  49896.                                                              (line  934)

  49897. * __fractqiusa:                          Fixed-point fractional library routines.

  49898.                                                              (line  940)

  49899. * __fractqiusq:                          Fixed-point fractional library routines.

  49900.                                                              (line  936)

  49901. * __fractqiuta:                          Fixed-point fractional library routines.

  49902.                                                              (line  942)

  49903. * __fractqqda:                           Fixed-point fractional library routines.

  49904.                                                              (line  481)

  49905. * __fractqqdf:                           Fixed-point fractional library routines.

  49906.                                                              (line  499)

  49907. * __fractqqdi:                           Fixed-point fractional library routines.

  49908.                                                              (line  496)

  49909. * __fractqqdq2:                          Fixed-point fractional library routines.

  49910.                                                              (line  478)

  49911. * __fractqqha:                           Fixed-point fractional library routines.

  49912.                                                              (line  479)

  49913. * __fractqqhi:                           Fixed-point fractional library routines.

  49914.                                                              (line  494)

  49915. * __fractqqhq2:                          Fixed-point fractional library routines.

  49916.                                                              (line  476)

  49917. * __fractqqqi:                           Fixed-point fractional library routines.

  49918.                                                              (line  493)

  49919. * __fractqqsa:                           Fixed-point fractional library routines.

  49920.                                                              (line  480)

  49921. * __fractqqsf:                           Fixed-point fractional library routines.

  49922.                                                              (line  498)

  49923. * __fractqqsi:                           Fixed-point fractional library routines.

  49924.                                                              (line  495)

  49925. * __fractqqsq2:                          Fixed-point fractional library routines.

  49926.                                                              (line  477)

  49927. * __fractqqta:                           Fixed-point fractional library routines.

  49928.                                                              (line  482)

  49929. * __fractqqti:                           Fixed-point fractional library routines.

  49930.                                                              (line  497)

  49931. * __fractqquda:                          Fixed-point fractional library routines.

  49932.                                                              (line  490)

  49933. * __fractqqudq:                          Fixed-point fractional library routines.

  49934.                                                              (line  486)

  49935. * __fractqquha:                          Fixed-point fractional library routines.

  49936.                                                              (line  488)

  49937. * __fractqquhq:                          Fixed-point fractional library routines.

  49938.                                                              (line  484)

  49939. * __fractqquqq:                          Fixed-point fractional library routines.

  49940.                                                              (line  483)

  49941. * __fractqqusa:                          Fixed-point fractional library routines.

  49942.                                                              (line  489)

  49943. * __fractqqusq:                          Fixed-point fractional library routines.

  49944.                                                              (line  485)

  49945. * __fractqquta:                          Fixed-point fractional library routines.

  49946.                                                              (line  491)

  49947. * __fractsada2:                          Fixed-point fractional library routines.

  49948.                                                              (line  603)

  49949. * __fractsadf:                           Fixed-point fractional library routines.

  49950.                                                              (line  619)

  49951. * __fractsadi:                           Fixed-point fractional library routines.

  49952.                                                              (line  616)

  49953. * __fractsadq:                           Fixed-point fractional library routines.

  49954.                                                              (line  601)

  49955. * __fractsaha2:                          Fixed-point fractional library routines.

  49956.                                                              (line  602)

  49957. * __fractsahi:                           Fixed-point fractional library routines.

  49958.                                                              (line  614)

  49959. * __fractsahq:                           Fixed-point fractional library routines.

  49960.                                                              (line  599)

  49961. * __fractsaqi:                           Fixed-point fractional library routines.

  49962.                                                              (line  613)

  49963. * __fractsaqq:                           Fixed-point fractional library routines.

  49964.                                                              (line  598)

  49965. * __fractsasf:                           Fixed-point fractional library routines.

  49966.                                                              (line  618)

  49967. * __fractsasi:                           Fixed-point fractional library routines.

  49968.                                                              (line  615)

  49969. * __fractsasq:                           Fixed-point fractional library routines.

  49970.                                                              (line  600)

  49971. * __fractsata2:                          Fixed-point fractional library routines.

  49972.                                                              (line  604)

  49973. * __fractsati:                           Fixed-point fractional library routines.

  49974.                                                              (line  617)

  49975. * __fractsauda:                          Fixed-point fractional library routines.

  49976.                                                              (line  611)

  49977. * __fractsaudq:                          Fixed-point fractional library routines.

  49978.                                                              (line  608)

  49979. * __fractsauha:                          Fixed-point fractional library routines.

  49980.                                                              (line  609)

  49981. * __fractsauhq:                          Fixed-point fractional library routines.

  49982.                                                              (line  606)

  49983. * __fractsauqq:                          Fixed-point fractional library routines.

  49984.                                                              (line  605)

  49985. * __fractsausa:                          Fixed-point fractional library routines.

  49986.                                                              (line  610)

  49987. * __fractsausq:                          Fixed-point fractional library routines.

  49988.                                                              (line  607)

  49989. * __fractsauta:                          Fixed-point fractional library routines.

  49990.                                                              (line  612)

  49991. * __fractsfda:                           Fixed-point fractional library routines.

  49992.                                                              (line 1016)

  49993. * __fractsfdq:                           Fixed-point fractional library routines.

  49994.                                                              (line 1013)

  49995. * __fractsfha:                           Fixed-point fractional library routines.

  49996.                                                              (line 1014)

  49997. * __fractsfhq:                           Fixed-point fractional library routines.

  49998.                                                              (line 1011)

  49999. * __fractsfqq:                           Fixed-point fractional library routines.

  50000.                                                              (line 1010)

  50001. * __fractsfsa:                           Fixed-point fractional library routines.

  50002.                                                              (line 1015)

  50003. * __fractsfsq:                           Fixed-point fractional library routines.

  50004.                                                              (line 1012)

  50005. * __fractsfta:                           Fixed-point fractional library routines.

  50006.                                                              (line 1017)

  50007. * __fractsfuda:                          Fixed-point fractional library routines.

  50008.                                                              (line 1024)

  50009. * __fractsfudq:                          Fixed-point fractional library routines.

  50010.                                                              (line 1021)

  50011. * __fractsfuha:                          Fixed-point fractional library routines.

  50012.                                                              (line 1022)

  50013. * __fractsfuhq:                          Fixed-point fractional library routines.

  50014.                                                              (line 1019)

  50015. * __fractsfuqq:                          Fixed-point fractional library routines.

  50016.                                                              (line 1018)

  50017. * __fractsfusa:                          Fixed-point fractional library routines.

  50018.                                                              (line 1023)

  50019. * __fractsfusq:                          Fixed-point fractional library routines.

  50020.                                                              (line 1020)

  50021. * __fractsfuta:                          Fixed-point fractional library routines.

  50022.                                                              (line 1025)

  50023. * __fractsida:                           Fixed-point fractional library routines.

  50024.                                                              (line  966)

  50025. * __fractsidq:                           Fixed-point fractional library routines.

  50026.                                                              (line  963)

  50027. * __fractsiha:                           Fixed-point fractional library routines.

  50028.                                                              (line  964)

  50029. * __fractsihq:                           Fixed-point fractional library routines.

  50030.                                                              (line  961)

  50031. * __fractsiqq:                           Fixed-point fractional library routines.

  50032.                                                              (line  960)

  50033. * __fractsisa:                           Fixed-point fractional library routines.

  50034.                                                              (line  965)

  50035. * __fractsisq:                           Fixed-point fractional library routines.

  50036.                                                              (line  962)

  50037. * __fractsita:                           Fixed-point fractional library routines.

  50038.                                                              (line  967)

  50039. * __fractsiuda:                          Fixed-point fractional library routines.

  50040.                                                              (line  974)

  50041. * __fractsiudq:                          Fixed-point fractional library routines.

  50042.                                                              (line  971)

  50043. * __fractsiuha:                          Fixed-point fractional library routines.

  50044.                                                              (line  972)

  50045. * __fractsiuhq:                          Fixed-point fractional library routines.

  50046.                                                              (line  969)

  50047. * __fractsiuqq:                          Fixed-point fractional library routines.

  50048.                                                              (line  968)

  50049. * __fractsiusa:                          Fixed-point fractional library routines.

  50050.                                                              (line  973)

  50051. * __fractsiusq:                          Fixed-point fractional library routines.

  50052.                                                              (line  970)

  50053. * __fractsiuta:                          Fixed-point fractional library routines.

  50054.                                                              (line  975)

  50055. * __fractsqda:                           Fixed-point fractional library routines.

  50056.                                                              (line  527)

  50057. * __fractsqdf:                           Fixed-point fractional library routines.

  50058.                                                              (line  545)

  50059. * __fractsqdi:                           Fixed-point fractional library routines.

  50060.                                                              (line  542)

  50061. * __fractsqdq2:                          Fixed-point fractional library routines.

  50062.                                                              (line  524)

  50063. * __fractsqha:                           Fixed-point fractional library routines.

  50064.                                                              (line  525)

  50065. * __fractsqhi:                           Fixed-point fractional library routines.

  50066.                                                              (line  540)

  50067. * __fractsqhq2:                          Fixed-point fractional library routines.

  50068.                                                              (line  523)

  50069. * __fractsqqi:                           Fixed-point fractional library routines.

  50070.                                                              (line  539)

  50071. * __fractsqqq2:                          Fixed-point fractional library routines.

  50072.                                                              (line  522)

  50073. * __fractsqsa:                           Fixed-point fractional library routines.

  50074.                                                              (line  526)

  50075. * __fractsqsf:                           Fixed-point fractional library routines.

  50076.                                                              (line  544)

  50077. * __fractsqsi:                           Fixed-point fractional library routines.

  50078.                                                              (line  541)

  50079. * __fractsqta:                           Fixed-point fractional library routines.

  50080.                                                              (line  528)

  50081. * __fractsqti:                           Fixed-point fractional library routines.

  50082.                                                              (line  543)

  50083. * __fractsquda:                          Fixed-point fractional library routines.

  50084.                                                              (line  536)

  50085. * __fractsqudq:                          Fixed-point fractional library routines.

  50086.                                                              (line  532)

  50087. * __fractsquha:                          Fixed-point fractional library routines.

  50088.                                                              (line  534)

  50089. * __fractsquhq:                          Fixed-point fractional library routines.

  50090.                                                              (line  530)

  50091. * __fractsquqq:                          Fixed-point fractional library routines.

  50092.                                                              (line  529)

  50093. * __fractsqusa:                          Fixed-point fractional library routines.

  50094.                                                              (line  535)

  50095. * __fractsqusq:                          Fixed-point fractional library routines.

  50096.                                                              (line  531)

  50097. * __fractsquta:                          Fixed-point fractional library routines.

  50098.                                                              (line  537)

  50099. * __fracttada2:                          Fixed-point fractional library routines.

  50100.                                                              (line  650)

  50101. * __fracttadf:                           Fixed-point fractional library routines.

  50102.                                                              (line  671)

  50103. * __fracttadi:                           Fixed-point fractional library routines.

  50104.                                                              (line  668)

  50105. * __fracttadq:                           Fixed-point fractional library routines.

  50106.                                                              (line  647)

  50107. * __fracttaha2:                          Fixed-point fractional library routines.

  50108.                                                              (line  648)

  50109. * __fracttahi:                           Fixed-point fractional library routines.

  50110.                                                              (line  666)

  50111. * __fracttahq:                           Fixed-point fractional library routines.

  50112.                                                              (line  645)

  50113. * __fracttaqi:                           Fixed-point fractional library routines.

  50114.                                                              (line  665)

  50115. * __fracttaqq:                           Fixed-point fractional library routines.

  50116.                                                              (line  644)

  50117. * __fracttasa2:                          Fixed-point fractional library routines.

  50118.                                                              (line  649)

  50119. * __fracttasf:                           Fixed-point fractional library routines.

  50120.                                                              (line  670)

  50121. * __fracttasi:                           Fixed-point fractional library routines.

  50122.                                                              (line  667)

  50123. * __fracttasq:                           Fixed-point fractional library routines.

  50124.                                                              (line  646)

  50125. * __fracttati:                           Fixed-point fractional library routines.

  50126.                                                              (line  669)

  50127. * __fracttauda:                          Fixed-point fractional library routines.

  50128.                                                              (line  661)

  50129. * __fracttaudq:                          Fixed-point fractional library routines.

  50130.                                                              (line  656)

  50131. * __fracttauha:                          Fixed-point fractional library routines.

  50132.                                                              (line  658)

  50133. * __fracttauhq:                          Fixed-point fractional library routines.

  50134.                                                              (line  653)

  50135. * __fracttauqq:                          Fixed-point fractional library routines.

  50136.                                                              (line  651)

  50137. * __fracttausa:                          Fixed-point fractional library routines.

  50138.                                                              (line  660)

  50139. * __fracttausq:                          Fixed-point fractional library routines.

  50140.                                                              (line  654)

  50141. * __fracttauta:                          Fixed-point fractional library routines.

  50142.                                                              (line  663)

  50143. * __fracttida:                           Fixed-point fractional library routines.

  50144.                                                              (line  998)

  50145. * __fracttidq:                           Fixed-point fractional library routines.

  50146.                                                              (line  995)

  50147. * __fracttiha:                           Fixed-point fractional library routines.

  50148.                                                              (line  996)

  50149. * __fracttihq:                           Fixed-point fractional library routines.

  50150.                                                              (line  993)

  50151. * __fracttiqq:                           Fixed-point fractional library routines.

  50152.                                                              (line  992)

  50153. * __fracttisa:                           Fixed-point fractional library routines.

  50154.                                                              (line  997)

  50155. * __fracttisq:                           Fixed-point fractional library routines.

  50156.                                                              (line  994)

  50157. * __fracttita:                           Fixed-point fractional library routines.

  50158.                                                              (line  999)

  50159. * __fracttiuda:                          Fixed-point fractional library routines.

  50160.                                                              (line 1007)

  50161. * __fracttiudq:                          Fixed-point fractional library routines.

  50162.                                                              (line 1003)

  50163. * __fracttiuha:                          Fixed-point fractional library routines.

  50164.                                                              (line 1005)

  50165. * __fracttiuhq:                          Fixed-point fractional library routines.

  50166.                                                              (line 1001)

  50167. * __fracttiuqq:                          Fixed-point fractional library routines.

  50168.                                                              (line 1000)

  50169. * __fracttiusa:                          Fixed-point fractional library routines.

  50170.                                                              (line 1006)

  50171. * __fracttiusq:                          Fixed-point fractional library routines.

  50172.                                                              (line 1002)

  50173. * __fracttiuta:                          Fixed-point fractional library routines.

  50174.                                                              (line 1008)

  50175. * __fractudada:                          Fixed-point fractional library routines.

  50176.                                                              (line  865)

  50177. * __fractudadf:                          Fixed-point fractional library routines.

  50178.                                                              (line  888)

  50179. * __fractudadi:                          Fixed-point fractional library routines.

  50180.                                                              (line  885)

  50181. * __fractudadq:                          Fixed-point fractional library routines.

  50182.                                                              (line  861)

  50183. * __fractudaha:                          Fixed-point fractional library routines.

  50184.                                                              (line  863)

  50185. * __fractudahi:                          Fixed-point fractional library routines.

  50186.                                                              (line  883)

  50187. * __fractudahq:                          Fixed-point fractional library routines.

  50188.                                                              (line  859)

  50189. * __fractudaqi:                          Fixed-point fractional library routines.

  50190.                                                              (line  882)

  50191. * __fractudaqq:                          Fixed-point fractional library routines.

  50192.                                                              (line  858)

  50193. * __fractudasa:                          Fixed-point fractional library routines.

  50194.                                                              (line  864)

  50195. * __fractudasf:                          Fixed-point fractional library routines.

  50196.                                                              (line  887)

  50197. * __fractudasi:                          Fixed-point fractional library routines.

  50198.                                                              (line  884)

  50199. * __fractudasq:                          Fixed-point fractional library routines.

  50200.                                                              (line  860)

  50201. * __fractudata:                          Fixed-point fractional library routines.

  50202.                                                              (line  866)

  50203. * __fractudati:                          Fixed-point fractional library routines.

  50204.                                                              (line  886)

  50205. * __fractudaudq:                         Fixed-point fractional library routines.

  50206.                                                              (line  874)

  50207. * __fractudauha2:                        Fixed-point fractional library routines.

  50208.                                                              (line  876)

  50209. * __fractudauhq:                         Fixed-point fractional library routines.

  50210.                                                              (line  870)

  50211. * __fractudauqq:                         Fixed-point fractional library routines.

  50212.                                                              (line  868)

  50213. * __fractudausa2:                        Fixed-point fractional library routines.

  50214.                                                              (line  878)

  50215. * __fractudausq:                         Fixed-point fractional library routines.

  50216.                                                              (line  872)

  50217. * __fractudauta2:                        Fixed-point fractional library routines.

  50218.                                                              (line  880)

  50219. * __fractudqda:                          Fixed-point fractional library routines.

  50220.                                                              (line  772)

  50221. * __fractudqdf:                          Fixed-point fractional library routines.

  50222.                                                              (line  798)

  50223. * __fractudqdi:                          Fixed-point fractional library routines.

  50224.                                                              (line  794)

  50225. * __fractudqdq:                          Fixed-point fractional library routines.

  50226.                                                              (line  767)

  50227. * __fractudqha:                          Fixed-point fractional library routines.

  50228.                                                              (line  769)

  50229. * __fractudqhi:                          Fixed-point fractional library routines.

  50230.                                                              (line  792)

  50231. * __fractudqhq:                          Fixed-point fractional library routines.

  50232.                                                              (line  764)

  50233. * __fractudqqi:                          Fixed-point fractional library routines.

  50234.                                                              (line  790)

  50235. * __fractudqqq:                          Fixed-point fractional library routines.

  50236.                                                              (line  762)

  50237. * __fractudqsa:                          Fixed-point fractional library routines.

  50238.                                                              (line  771)

  50239. * __fractudqsf:                          Fixed-point fractional library routines.

  50240.                                                              (line  797)

  50241. * __fractudqsi:                          Fixed-point fractional library routines.

  50242.                                                              (line  793)

  50243. * __fractudqsq:                          Fixed-point fractional library routines.

  50244.                                                              (line  765)

  50245. * __fractudqta:                          Fixed-point fractional library routines.

  50246.                                                              (line  774)

  50247. * __fractudqti:                          Fixed-point fractional library routines.

  50248.                                                              (line  795)

  50249. * __fractudquda:                         Fixed-point fractional library routines.

  50250.                                                              (line  786)

  50251. * __fractudquha:                         Fixed-point fractional library routines.

  50252.                                                              (line  782)

  50253. * __fractudquhq2:                        Fixed-point fractional library routines.

  50254.                                                              (line  778)

  50255. * __fractudquqq2:                        Fixed-point fractional library routines.

  50256.                                                              (line  776)

  50257. * __fractudqusa:                         Fixed-point fractional library routines.

  50258.                                                              (line  784)

  50259. * __fractudqusq2:                        Fixed-point fractional library routines.

  50260.                                                              (line  780)

  50261. * __fractudquta:                         Fixed-point fractional library routines.

  50262.                                                              (line  788)

  50263. * __fractuhada:                          Fixed-point fractional library routines.

  50264.                                                              (line  806)

  50265. * __fractuhadf:                          Fixed-point fractional library routines.

  50266.                                                              (line  829)

  50267. * __fractuhadi:                          Fixed-point fractional library routines.

  50268.                                                              (line  826)

  50269. * __fractuhadq:                          Fixed-point fractional library routines.

  50270.                                                              (line  802)

  50271. * __fractuhaha:                          Fixed-point fractional library routines.

  50272.                                                              (line  804)

  50273. * __fractuhahi:                          Fixed-point fractional library routines.

  50274.                                                              (line  824)

  50275. * __fractuhahq:                          Fixed-point fractional library routines.

  50276.                                                              (line  800)

  50277. * __fractuhaqi:                          Fixed-point fractional library routines.

  50278.                                                              (line  823)

  50279. * __fractuhaqq:                          Fixed-point fractional library routines.

  50280.                                                              (line  799)

  50281. * __fractuhasa:                          Fixed-point fractional library routines.

  50282.                                                              (line  805)

  50283. * __fractuhasf:                          Fixed-point fractional library routines.

  50284.                                                              (line  828)

  50285. * __fractuhasi:                          Fixed-point fractional library routines.

  50286.                                                              (line  825)

  50287. * __fractuhasq:                          Fixed-point fractional library routines.

  50288.                                                              (line  801)

  50289. * __fractuhata:                          Fixed-point fractional library routines.

  50290.                                                              (line  807)

  50291. * __fractuhati:                          Fixed-point fractional library routines.

  50292.                                                              (line  827)

  50293. * __fractuhauda2:                        Fixed-point fractional library routines.

  50294.                                                              (line  819)

  50295. * __fractuhaudq:                         Fixed-point fractional library routines.

  50296.                                                              (line  815)

  50297. * __fractuhauhq:                         Fixed-point fractional library routines.

  50298.                                                              (line  811)

  50299. * __fractuhauqq:                         Fixed-point fractional library routines.

  50300.                                                              (line  809)

  50301. * __fractuhausa2:                        Fixed-point fractional library routines.

  50302.                                                              (line  817)

  50303. * __fractuhausq:                         Fixed-point fractional library routines.

  50304.                                                              (line  813)

  50305. * __fractuhauta2:                        Fixed-point fractional library routines.

  50306.                                                              (line  821)

  50307. * __fractuhqda:                          Fixed-point fractional library routines.

  50308.                                                              (line  709)

  50309. * __fractuhqdf:                          Fixed-point fractional library routines.

  50310.                                                              (line  730)

  50311. * __fractuhqdi:                          Fixed-point fractional library routines.

  50312.                                                              (line  727)

  50313. * __fractuhqdq:                          Fixed-point fractional library routines.

  50314.                                                              (line  706)

  50315. * __fractuhqha:                          Fixed-point fractional library routines.

  50316.                                                              (line  707)

  50317. * __fractuhqhi:                          Fixed-point fractional library routines.

  50318.                                                              (line  725)

  50319. * __fractuhqhq:                          Fixed-point fractional library routines.

  50320.                                                              (line  704)

  50321. * __fractuhqqi:                          Fixed-point fractional library routines.

  50322.                                                              (line  724)

  50323. * __fractuhqqq:                          Fixed-point fractional library routines.

  50324.                                                              (line  703)

  50325. * __fractuhqsa:                          Fixed-point fractional library routines.

  50326.                                                              (line  708)

  50327. * __fractuhqsf:                          Fixed-point fractional library routines.

  50328.                                                              (line  729)

  50329. * __fractuhqsi:                          Fixed-point fractional library routines.

  50330.                                                              (line  726)

  50331. * __fractuhqsq:                          Fixed-point fractional library routines.

  50332.                                                              (line  705)

  50333. * __fractuhqta:                          Fixed-point fractional library routines.

  50334.                                                              (line  710)

  50335. * __fractuhqti:                          Fixed-point fractional library routines.

  50336.                                                              (line  728)

  50337. * __fractuhquda:                         Fixed-point fractional library routines.

  50338.                                                              (line  720)

  50339. * __fractuhqudq2:                        Fixed-point fractional library routines.

  50340.                                                              (line  715)

  50341. * __fractuhquha:                         Fixed-point fractional library routines.

  50342.                                                              (line  717)

  50343. * __fractuhquqq2:                        Fixed-point fractional library routines.

  50344.                                                              (line  711)

  50345. * __fractuhqusa:                         Fixed-point fractional library routines.

  50346.                                                              (line  719)

  50347. * __fractuhqusq2:                        Fixed-point fractional library routines.

  50348.                                                              (line  713)

  50349. * __fractuhquta:                         Fixed-point fractional library routines.

  50350.                                                              (line  722)

  50351. * __fractunsdadi:                        Fixed-point fractional library routines.

  50352.                                                              (line 1562)

  50353. * __fractunsdahi:                        Fixed-point fractional library routines.

  50354.                                                              (line 1560)

  50355. * __fractunsdaqi:                        Fixed-point fractional library routines.

  50356.                                                              (line 1559)

  50357. * __fractunsdasi:                        Fixed-point fractional library routines.

  50358.                                                              (line 1561)

  50359. * __fractunsdati:                        Fixed-point fractional library routines.

  50360.                                                              (line 1563)

  50361. * __fractunsdida:                        Fixed-point fractional library routines.

  50362.                                                              (line 1714)

  50363. * __fractunsdidq:                        Fixed-point fractional library routines.

  50364.                                                              (line 1711)

  50365. * __fractunsdiha:                        Fixed-point fractional library routines.

  50366.                                                              (line 1712)

  50367. * __fractunsdihq:                        Fixed-point fractional library routines.

  50368.                                                              (line 1709)

  50369. * __fractunsdiqq:                        Fixed-point fractional library routines.

  50370.                                                              (line 1708)

  50371. * __fractunsdisa:                        Fixed-point fractional library routines.

  50372.                                                              (line 1713)

  50373. * __fractunsdisq:                        Fixed-point fractional library routines.

  50374.                                                              (line 1710)

  50375. * __fractunsdita:                        Fixed-point fractional library routines.

  50376.                                                              (line 1715)

  50377. * __fractunsdiuda:                       Fixed-point fractional library routines.

  50378.                                                              (line 1726)

  50379. * __fractunsdiudq:                       Fixed-point fractional library routines.

  50380.                                                              (line 1721)

  50381. * __fractunsdiuha:                       Fixed-point fractional library routines.

  50382.                                                              (line 1723)

  50383. * __fractunsdiuhq:                       Fixed-point fractional library routines.

  50384.                                                              (line 1718)

  50385. * __fractunsdiuqq:                       Fixed-point fractional library routines.

  50386.                                                              (line 1716)

  50387. * __fractunsdiusa:                       Fixed-point fractional library routines.

  50388.                                                              (line 1725)

  50389. * __fractunsdiusq:                       Fixed-point fractional library routines.

  50390.                                                              (line 1719)

  50391. * __fractunsdiuta:                       Fixed-point fractional library routines.

  50392.                                                              (line 1728)

  50393. * __fractunsdqdi:                        Fixed-point fractional library routines.

  50394.                                                              (line 1546)

  50395. * __fractunsdqhi:                        Fixed-point fractional library routines.

  50396.                                                              (line 1544)

  50397. * __fractunsdqqi:                        Fixed-point fractional library routines.

  50398.                                                              (line 1543)

  50399. * __fractunsdqsi:                        Fixed-point fractional library routines.

  50400.                                                              (line 1545)

  50401. * __fractunsdqti:                        Fixed-point fractional library routines.

  50402.                                                              (line 1547)

  50403. * __fractunshadi:                        Fixed-point fractional library routines.

  50404.                                                              (line 1552)

  50405. * __fractunshahi:                        Fixed-point fractional library routines.

  50406.                                                              (line 1550)

  50407. * __fractunshaqi:                        Fixed-point fractional library routines.

  50408.                                                              (line 1549)

  50409. * __fractunshasi:                        Fixed-point fractional library routines.

  50410.                                                              (line 1551)

  50411. * __fractunshati:                        Fixed-point fractional library routines.

  50412.                                                              (line 1553)

  50413. * __fractunshida:                        Fixed-point fractional library routines.

  50414.                                                              (line 1670)

  50415. * __fractunshidq:                        Fixed-point fractional library routines.

  50416.                                                              (line 1667)

  50417. * __fractunshiha:                        Fixed-point fractional library routines.

  50418.                                                              (line 1668)

  50419. * __fractunshihq:                        Fixed-point fractional library routines.

  50420.                                                              (line 1665)

  50421. * __fractunshiqq:                        Fixed-point fractional library routines.

  50422.                                                              (line 1664)

  50423. * __fractunshisa:                        Fixed-point fractional library routines.

  50424.                                                              (line 1669)

  50425. * __fractunshisq:                        Fixed-point fractional library routines.

  50426.                                                              (line 1666)

  50427. * __fractunshita:                        Fixed-point fractional library routines.

  50428.                                                              (line 1671)

  50429. * __fractunshiuda:                       Fixed-point fractional library routines.

  50430.                                                              (line 1682)

  50431. * __fractunshiudq:                       Fixed-point fractional library routines.

  50432.                                                              (line 1677)

  50433. * __fractunshiuha:                       Fixed-point fractional library routines.

  50434.                                                              (line 1679)

  50435. * __fractunshiuhq:                       Fixed-point fractional library routines.

  50436.                                                              (line 1674)

  50437. * __fractunshiuqq:                       Fixed-point fractional library routines.

  50438.                                                              (line 1672)

  50439. * __fractunshiusa:                       Fixed-point fractional library routines.

  50440.                                                              (line 1681)

  50441. * __fractunshiusq:                       Fixed-point fractional library routines.

  50442.                                                              (line 1675)

  50443. * __fractunshiuta:                       Fixed-point fractional library routines.

  50444.                                                              (line 1684)

  50445. * __fractunshqdi:                        Fixed-point fractional library routines.

  50446.                                                              (line 1536)

  50447. * __fractunshqhi:                        Fixed-point fractional library routines.

  50448.                                                              (line 1534)

  50449. * __fractunshqqi:                        Fixed-point fractional library routines.

  50450.                                                              (line 1533)

  50451. * __fractunshqsi:                        Fixed-point fractional library routines.

  50452.                                                              (line 1535)

  50453. * __fractunshqti:                        Fixed-point fractional library routines.

  50454.                                                              (line 1537)

  50455. * __fractunsqida:                        Fixed-point fractional library routines.

  50456.                                                              (line 1648)

  50457. * __fractunsqidq:                        Fixed-point fractional library routines.

  50458.                                                              (line 1645)

  50459. * __fractunsqiha:                        Fixed-point fractional library routines.

  50460.                                                              (line 1646)

  50461. * __fractunsqihq:                        Fixed-point fractional library routines.

  50462.                                                              (line 1643)

  50463. * __fractunsqiqq:                        Fixed-point fractional library routines.

  50464.                                                              (line 1642)

  50465. * __fractunsqisa:                        Fixed-point fractional library routines.

  50466.                                                              (line 1647)

  50467. * __fractunsqisq:                        Fixed-point fractional library routines.

  50468.                                                              (line 1644)

  50469. * __fractunsqita:                        Fixed-point fractional library routines.

  50470.                                                              (line 1649)

  50471. * __fractunsqiuda:                       Fixed-point fractional library routines.

  50472.                                                              (line 1660)

  50473. * __fractunsqiudq:                       Fixed-point fractional library routines.

  50474.                                                              (line 1655)

  50475. * __fractunsqiuha:                       Fixed-point fractional library routines.

  50476.                                                              (line 1657)

  50477. * __fractunsqiuhq:                       Fixed-point fractional library routines.

  50478.                                                              (line 1652)

  50479. * __fractunsqiuqq:                       Fixed-point fractional library routines.

  50480.                                                              (line 1650)

  50481. * __fractunsqiusa:                       Fixed-point fractional library routines.

  50482.                                                              (line 1659)

  50483. * __fractunsqiusq:                       Fixed-point fractional library routines.

  50484.                                                              (line 1653)

  50485. * __fractunsqiuta:                       Fixed-point fractional library routines.

  50486.                                                              (line 1662)

  50487. * __fractunsqqdi:                        Fixed-point fractional library routines.

  50488.                                                              (line 1531)

  50489. * __fractunsqqhi:                        Fixed-point fractional library routines.

  50490.                                                              (line 1529)

  50491. * __fractunsqqqi:                        Fixed-point fractional library routines.

  50492.                                                              (line 1528)

  50493. * __fractunsqqsi:                        Fixed-point fractional library routines.

  50494.                                                              (line 1530)

  50495. * __fractunsqqti:                        Fixed-point fractional library routines.

  50496.                                                              (line 1532)

  50497. * __fractunssadi:                        Fixed-point fractional library routines.

  50498.                                                              (line 1557)

  50499. * __fractunssahi:                        Fixed-point fractional library routines.

  50500.                                                              (line 1555)

  50501. * __fractunssaqi:                        Fixed-point fractional library routines.

  50502.                                                              (line 1554)

  50503. * __fractunssasi:                        Fixed-point fractional library routines.

  50504.                                                              (line 1556)

  50505. * __fractunssati:                        Fixed-point fractional library routines.

  50506.                                                              (line 1558)

  50507. * __fractunssida:                        Fixed-point fractional library routines.

  50508.                                                              (line 1692)

  50509. * __fractunssidq:                        Fixed-point fractional library routines.

  50510.                                                              (line 1689)

  50511. * __fractunssiha:                        Fixed-point fractional library routines.

  50512.                                                              (line 1690)

  50513. * __fractunssihq:                        Fixed-point fractional library routines.

  50514.                                                              (line 1687)

  50515. * __fractunssiqq:                        Fixed-point fractional library routines.

  50516.                                                              (line 1686)

  50517. * __fractunssisa:                        Fixed-point fractional library routines.

  50518.                                                              (line 1691)

  50519. * __fractunssisq:                        Fixed-point fractional library routines.

  50520.                                                              (line 1688)

  50521. * __fractunssita:                        Fixed-point fractional library routines.

  50522.                                                              (line 1693)

  50523. * __fractunssiuda:                       Fixed-point fractional library routines.

  50524.                                                              (line 1704)

  50525. * __fractunssiudq:                       Fixed-point fractional library routines.

  50526.                                                              (line 1699)

  50527. * __fractunssiuha:                       Fixed-point fractional library routines.

  50528.                                                              (line 1701)

  50529. * __fractunssiuhq:                       Fixed-point fractional library routines.

  50530.                                                              (line 1696)

  50531. * __fractunssiuqq:                       Fixed-point fractional library routines.

  50532.                                                              (line 1694)

  50533. * __fractunssiusa:                       Fixed-point fractional library routines.

  50534.                                                              (line 1703)

  50535. * __fractunssiusq:                       Fixed-point fractional library routines.

  50536.                                                              (line 1697)

  50537. * __fractunssiuta:                       Fixed-point fractional library routines.

  50538.                                                              (line 1706)

  50539. * __fractunssqdi:                        Fixed-point fractional library routines.

  50540.                                                              (line 1541)

  50541. * __fractunssqhi:                        Fixed-point fractional library routines.

  50542.                                                              (line 1539)

  50543. * __fractunssqqi:                        Fixed-point fractional library routines.

  50544.                                                              (line 1538)

  50545. * __fractunssqsi:                        Fixed-point fractional library routines.

  50546.                                                              (line 1540)

  50547. * __fractunssqti:                        Fixed-point fractional library routines.

  50548.                                                              (line 1542)

  50549. * __fractunstadi:                        Fixed-point fractional library routines.

  50550.                                                              (line 1567)

  50551. * __fractunstahi:                        Fixed-point fractional library routines.

  50552.                                                              (line 1565)

  50553. * __fractunstaqi:                        Fixed-point fractional library routines.

  50554.                                                              (line 1564)

  50555. * __fractunstasi:                        Fixed-point fractional library routines.

  50556.                                                              (line 1566)

  50557. * __fractunstati:                        Fixed-point fractional library routines.

  50558.                                                              (line 1568)

  50559. * __fractunstida:                        Fixed-point fractional library routines.

  50560.                                                              (line 1737)

  50561. * __fractunstidq:                        Fixed-point fractional library routines.

  50562.                                                              (line 1733)

  50563. * __fractunstiha:                        Fixed-point fractional library routines.

  50564.                                                              (line 1735)

  50565. * __fractunstihq:                        Fixed-point fractional library routines.

  50566.                                                              (line 1731)

  50567. * __fractunstiqq:                        Fixed-point fractional library routines.

  50568.                                                              (line 1730)

  50569. * __fractunstisa:                        Fixed-point fractional library routines.

  50570.                                                              (line 1736)

  50571. * __fractunstisq:                        Fixed-point fractional library routines.

  50572.                                                              (line 1732)

  50573. * __fractunstita:                        Fixed-point fractional library routines.

  50574.                                                              (line 1738)

  50575. * __fractunstiuda:                       Fixed-point fractional library routines.

  50576.                                                              (line 1752)

  50577. * __fractunstiudq:                       Fixed-point fractional library routines.

  50578.                                                              (line 1746)

  50579. * __fractunstiuha:                       Fixed-point fractional library routines.

  50580.                                                              (line 1748)

  50581. * __fractunstiuhq:                       Fixed-point fractional library routines.

  50582.                                                              (line 1742)

  50583. * __fractunstiuqq:                       Fixed-point fractional library routines.

  50584.                                                              (line 1740)

  50585. * __fractunstiusa:                       Fixed-point fractional library routines.

  50586.                                                              (line 1750)

  50587. * __fractunstiusq:                       Fixed-point fractional library routines.

  50588.                                                              (line 1744)

  50589. * __fractunstiuta:                       Fixed-point fractional library routines.

  50590.                                                              (line 1754)

  50591. * __fractunsudadi:                       Fixed-point fractional library routines.

  50592.                                                              (line 1628)

  50593. * __fractunsudahi:                       Fixed-point fractional library routines.

  50594.                                                              (line 1624)

  50595. * __fractunsudaqi:                       Fixed-point fractional library routines.

  50596.                                                              (line 1622)

  50597. * __fractunsudasi:                       Fixed-point fractional library routines.

  50598.                                                              (line 1626)

  50599. * __fractunsudati:                       Fixed-point fractional library routines.

  50600.                                                              (line 1630)

  50601. * __fractunsudqdi:                       Fixed-point fractional library routines.

  50602.                                                              (line 1602)

  50603. * __fractunsudqhi:                       Fixed-point fractional library routines.

  50604.                                                              (line 1598)

  50605. * __fractunsudqqi:                       Fixed-point fractional library routines.

  50606.                                                              (line 1596)

  50607. * __fractunsudqsi:                       Fixed-point fractional library routines.

  50608.                                                              (line 1600)

  50609. * __fractunsudqti:                       Fixed-point fractional library routines.

  50610.                                                              (line 1604)

  50611. * __fractunsuhadi:                       Fixed-point fractional library routines.

  50612.                                                              (line 1612)

  50613. * __fractunsuhahi:                       Fixed-point fractional library routines.

  50614.                                                              (line 1608)

  50615. * __fractunsuhaqi:                       Fixed-point fractional library routines.

  50616.                                                              (line 1606)

  50617. * __fractunsuhasi:                       Fixed-point fractional library routines.

  50618.                                                              (line 1610)

  50619. * __fractunsuhati:                       Fixed-point fractional library routines.

  50620.                                                              (line 1614)

  50621. * __fractunsuhqdi:                       Fixed-point fractional library routines.

  50622.                                                              (line 1583)

  50623. * __fractunsuhqhi:                       Fixed-point fractional library routines.

  50624.                                                              (line 1581)

  50625. * __fractunsuhqqi:                       Fixed-point fractional library routines.

  50626.                                                              (line 1580)

  50627. * __fractunsuhqsi:                       Fixed-point fractional library routines.

  50628.                                                              (line 1582)

  50629. * __fractunsuhqti:                       Fixed-point fractional library routines.

  50630.                                                              (line 1584)

  50631. * __fractunsuqqdi:                       Fixed-point fractional library routines.

  50632.                                                              (line 1576)

  50633. * __fractunsuqqhi:                       Fixed-point fractional library routines.

  50634.                                                              (line 1572)

  50635. * __fractunsuqqqi:                       Fixed-point fractional library routines.

  50636.                                                              (line 1570)

  50637. * __fractunsuqqsi:                       Fixed-point fractional library routines.

  50638.                                                              (line 1574)

  50639. * __fractunsuqqti:                       Fixed-point fractional library routines.

  50640.                                                              (line 1578)

  50641. * __fractunsusadi:                       Fixed-point fractional library routines.

  50642.                                                              (line 1619)

  50643. * __fractunsusahi:                       Fixed-point fractional library routines.

  50644.                                                              (line 1617)

  50645. * __fractunsusaqi:                       Fixed-point fractional library routines.

  50646.                                                              (line 1616)

  50647. * __fractunsusasi:                       Fixed-point fractional library routines.

  50648.                                                              (line 1618)

  50649. * __fractunsusati:                       Fixed-point fractional library routines.

  50650.                                                              (line 1620)

  50651. * __fractunsusqdi:                       Fixed-point fractional library routines.

  50652.                                                              (line 1592)

  50653. * __fractunsusqhi:                       Fixed-point fractional library routines.

  50654.                                                              (line 1588)

  50655. * __fractunsusqqi:                       Fixed-point fractional library routines.

  50656.                                                              (line 1586)

  50657. * __fractunsusqsi:                       Fixed-point fractional library routines.

  50658.                                                              (line 1590)

  50659. * __fractunsusqti:                       Fixed-point fractional library routines.

  50660.                                                              (line 1594)

  50661. * __fractunsutadi:                       Fixed-point fractional library routines.

  50662.                                                              (line 1638)

  50663. * __fractunsutahi:                       Fixed-point fractional library routines.

  50664.                                                              (line 1634)

  50665. * __fractunsutaqi:                       Fixed-point fractional library routines.

  50666.                                                              (line 1632)

  50667. * __fractunsutasi:                       Fixed-point fractional library routines.

  50668.                                                              (line 1636)

  50669. * __fractunsutati:                       Fixed-point fractional library routines.

  50670.                                                              (line 1640)

  50671. * __fractuqqda:                          Fixed-point fractional library routines.

  50672.                                                              (line  679)

  50673. * __fractuqqdf:                          Fixed-point fractional library routines.

  50674.                                                              (line  702)

  50675. * __fractuqqdi:                          Fixed-point fractional library routines.

  50676.                                                              (line  699)

  50677. * __fractuqqdq:                          Fixed-point fractional library routines.

  50678.                                                              (line  675)

  50679. * __fractuqqha:                          Fixed-point fractional library routines.

  50680.                                                              (line  677)

  50681. * __fractuqqhi:                          Fixed-point fractional library routines.

  50682.                                                              (line  697)

  50683. * __fractuqqhq:                          Fixed-point fractional library routines.

  50684.                                                              (line  673)

  50685. * __fractuqqqi:                          Fixed-point fractional library routines.

  50686.                                                              (line  696)

  50687. * __fractuqqqq:                          Fixed-point fractional library routines.

  50688.                                                              (line  672)

  50689. * __fractuqqsa:                          Fixed-point fractional library routines.

  50690.                                                              (line  678)

  50691. * __fractuqqsf:                          Fixed-point fractional library routines.

  50692.                                                              (line  701)

  50693. * __fractuqqsi:                          Fixed-point fractional library routines.

  50694.                                                              (line  698)

  50695. * __fractuqqsq:                          Fixed-point fractional library routines.

  50696.                                                              (line  674)

  50697. * __fractuqqta:                          Fixed-point fractional library routines.

  50698.                                                              (line  680)

  50699. * __fractuqqti:                          Fixed-point fractional library routines.

  50700.                                                              (line  700)

  50701. * __fractuqquda:                         Fixed-point fractional library routines.

  50702.                                                              (line  692)

  50703. * __fractuqqudq2:                        Fixed-point fractional library routines.

  50704.                                                              (line  686)

  50705. * __fractuqquha:                         Fixed-point fractional library routines.

  50706.                                                              (line  688)

  50707. * __fractuqquhq2:                        Fixed-point fractional library routines.

  50708.                                                              (line  682)

  50709. * __fractuqqusa:                         Fixed-point fractional library routines.

  50710.                                                              (line  690)

  50711. * __fractuqqusq2:                        Fixed-point fractional library routines.

  50712.                                                              (line  684)

  50713. * __fractuqquta:                         Fixed-point fractional library routines.

  50714.                                                              (line  694)

  50715. * __fractusada:                          Fixed-point fractional library routines.

  50716.                                                              (line  836)

  50717. * __fractusadf:                          Fixed-point fractional library routines.

  50718.                                                              (line  857)

  50719. * __fractusadi:                          Fixed-point fractional library routines.

  50720.                                                              (line  854)

  50721. * __fractusadq:                          Fixed-point fractional library routines.

  50722.                                                              (line  833)

  50723. * __fractusaha:                          Fixed-point fractional library routines.

  50724.                                                              (line  834)

  50725. * __fractusahi:                          Fixed-point fractional library routines.

  50726.                                                              (line  852)

  50727. * __fractusahq:                          Fixed-point fractional library routines.

  50728.                                                              (line  831)

  50729. * __fractusaqi:                          Fixed-point fractional library routines.

  50730.                                                              (line  851)

  50731. * __fractusaqq:                          Fixed-point fractional library routines.

  50732.                                                              (line  830)

  50733. * __fractusasa:                          Fixed-point fractional library routines.

  50734.                                                              (line  835)

  50735. * __fractusasf:                          Fixed-point fractional library routines.

  50736.                                                              (line  856)

  50737. * __fractusasi:                          Fixed-point fractional library routines.

  50738.                                                              (line  853)

  50739. * __fractusasq:                          Fixed-point fractional library routines.

  50740.                                                              (line  832)

  50741. * __fractusata:                          Fixed-point fractional library routines.

  50742.                                                              (line  837)

  50743. * __fractusati:                          Fixed-point fractional library routines.

  50744.                                                              (line  855)

  50745. * __fractusauda2:                        Fixed-point fractional library routines.

  50746.                                                              (line  847)

  50747. * __fractusaudq:                         Fixed-point fractional library routines.

  50748.                                                              (line  843)

  50749. * __fractusauha2:                        Fixed-point fractional library routines.

  50750.                                                              (line  845)

  50751. * __fractusauhq:                         Fixed-point fractional library routines.

  50752.                                                              (line  840)

  50753. * __fractusauqq:                         Fixed-point fractional library routines.

  50754.                                                              (line  838)

  50755. * __fractusausq:                         Fixed-point fractional library routines.

  50756.                                                              (line  841)

  50757. * __fractusauta2:                        Fixed-point fractional library routines.

  50758.                                                              (line  849)

  50759. * __fractusqda:                          Fixed-point fractional library routines.

  50760.                                                              (line  738)

  50761. * __fractusqdf:                          Fixed-point fractional library routines.

  50762.                                                              (line  761)

  50763. * __fractusqdi:                          Fixed-point fractional library routines.

  50764.                                                              (line  758)

  50765. * __fractusqdq:                          Fixed-point fractional library routines.

  50766.                                                              (line  734)

  50767. * __fractusqha:                          Fixed-point fractional library routines.

  50768.                                                              (line  736)

  50769. * __fractusqhi:                          Fixed-point fractional library routines.

  50770.                                                              (line  756)

  50771. * __fractusqhq:                          Fixed-point fractional library routines.

  50772.                                                              (line  732)

  50773. * __fractusqqi:                          Fixed-point fractional library routines.

  50774.                                                              (line  755)

  50775. * __fractusqqq:                          Fixed-point fractional library routines.

  50776.                                                              (line  731)

  50777. * __fractusqsa:                          Fixed-point fractional library routines.

  50778.                                                              (line  737)

  50779. * __fractusqsf:                          Fixed-point fractional library routines.

  50780.                                                              (line  760)

  50781. * __fractusqsi:                          Fixed-point fractional library routines.

  50782.                                                              (line  757)

  50783. * __fractusqsq:                          Fixed-point fractional library routines.

  50784.                                                              (line  733)

  50785. * __fractusqta:                          Fixed-point fractional library routines.

  50786.                                                              (line  739)

  50787. * __fractusqti:                          Fixed-point fractional library routines.

  50788.                                                              (line  759)

  50789. * __fractusquda:                         Fixed-point fractional library routines.

  50790.                                                              (line  751)

  50791. * __fractusqudq2:                        Fixed-point fractional library routines.

  50792.                                                              (line  745)

  50793. * __fractusquha:                         Fixed-point fractional library routines.

  50794.                                                              (line  747)

  50795. * __fractusquhq2:                        Fixed-point fractional library routines.

  50796.                                                              (line  743)

  50797. * __fractusquqq2:                        Fixed-point fractional library routines.

  50798.                                                              (line  741)

  50799. * __fractusqusa:                         Fixed-point fractional library routines.

  50800.                                                              (line  749)

  50801. * __fractusquta:                         Fixed-point fractional library routines.

  50802.                                                              (line  753)

  50803. * __fractutada:                          Fixed-point fractional library routines.

  50804.                                                              (line  899)

  50805. * __fractutadf:                          Fixed-point fractional library routines.

  50806.                                                              (line  925)

  50807. * __fractutadi:                          Fixed-point fractional library routines.

  50808.                                                              (line  921)

  50809. * __fractutadq:                          Fixed-point fractional library routines.

  50810.                                                              (line  894)

  50811. * __fractutaha:                          Fixed-point fractional library routines.

  50812.                                                              (line  896)

  50813. * __fractutahi:                          Fixed-point fractional library routines.

  50814.                                                              (line  919)

  50815. * __fractutahq:                          Fixed-point fractional library routines.

  50816.                                                              (line  891)

  50817. * __fractutaqi:                          Fixed-point fractional library routines.

  50818.                                                              (line  917)

  50819. * __fractutaqq:                          Fixed-point fractional library routines.

  50820.                                                              (line  889)

  50821. * __fractutasa:                          Fixed-point fractional library routines.

  50822.                                                              (line  898)

  50823. * __fractutasf:                          Fixed-point fractional library routines.

  50824.                                                              (line  924)

  50825. * __fractutasi:                          Fixed-point fractional library routines.

  50826.                                                              (line  920)

  50827. * __fractutasq:                          Fixed-point fractional library routines.

  50828.                                                              (line  892)

  50829. * __fractutata:                          Fixed-point fractional library routines.

  50830.                                                              (line  901)

  50831. * __fractutati:                          Fixed-point fractional library routines.

  50832.                                                              (line  922)

  50833. * __fractutauda2:                        Fixed-point fractional library routines.

  50834.                                                              (line  915)

  50835. * __fractutaudq:                         Fixed-point fractional library routines.

  50836.                                                              (line  909)

  50837. * __fractutauha2:                        Fixed-point fractional library routines.

  50838.                                                              (line  911)

  50839. * __fractutauhq:                         Fixed-point fractional library routines.

  50840.                                                              (line  905)

  50841. * __fractutauqq:                         Fixed-point fractional library routines.

  50842.                                                              (line  903)

  50843. * __fractutausa2:                        Fixed-point fractional library routines.

  50844.                                                              (line  913)

  50845. * __fractutausq:                         Fixed-point fractional library routines.

  50846.                                                              (line  907)

  50847. * __gedf2:                               Soft float library routines.

  50848.                                                              (line  205)

  50849. * __gesf2:                               Soft float library routines.

  50850.                                                              (line  204)

  50851. * __getf2:                               Soft float library routines.

  50852.                                                              (line  206)

  50853. * __gtdf2:                               Soft float library routines.

  50854.                                                              (line  223)

  50855. * __gtsf2:                               Soft float library routines.

  50856.                                                              (line  222)

  50857. * __gttf2:                               Soft float library routines.

  50858.                                                              (line  224)

  50859. * __ledf2:                               Soft float library routines.

  50860.                                                              (line  217)

  50861. * __lesf2:                               Soft float library routines.

  50862.                                                              (line  216)

  50863. * __letf2:                               Soft float library routines.

  50864.                                                              (line  218)

  50865. * __lshrdi3:                             Integer library routines.

  50866.                                                              (line   30)

  50867. * __lshrsi3:                             Integer library routines.

  50868.                                                              (line   29)

  50869. * __lshrti3:                             Integer library routines.

  50870.                                                              (line   31)

  50871. * __lshruda3:                            Fixed-point fractional library routines.

  50872.                                                              (line  396)

  50873. * __lshrudq3:                            Fixed-point fractional library routines.

  50874.                                                              (line  390)

  50875. * __lshruha3:                            Fixed-point fractional library routines.

  50876.                                                              (line  392)

  50877. * __lshruhq3:                            Fixed-point fractional library routines.

  50878.                                                              (line  386)

  50879. * __lshruqq3:                            Fixed-point fractional library routines.

  50880.                                                              (line  384)

  50881. * __lshrusa3:                            Fixed-point fractional library routines.

  50882.                                                              (line  394)

  50883. * __lshrusq3:                            Fixed-point fractional library routines.

  50884.                                                              (line  388)

  50885. * __lshruta3:                            Fixed-point fractional library routines.

  50886.                                                              (line  398)

  50887. * __ltdf2:                               Soft float library routines.

  50888.                                                              (line  211)

  50889. * __ltsf2:                               Soft float library routines.

  50890.                                                              (line  210)

  50891. * __lttf2:                               Soft float library routines.

  50892.                                                              (line  212)

  50893. * __main:                                Collect2.           (line   15)

  50894. * __moddi3:                              Integer library routines.

  50895.                                                              (line   36)

  50896. * __modsi3:                              Integer library routines.

  50897.                                                              (line   35)

  50898. * __modti3:                              Integer library routines.

  50899.                                                              (line   37)

  50900. * __morestack_current_segment:           Miscellaneous routines.

  50901.                                                              (line   45)

  50902. * __morestack_initial_sp:                Miscellaneous routines.

  50903.                                                              (line   46)

  50904. * __morestack_segments:                  Miscellaneous routines.

  50905.                                                              (line   44)

  50906. * __mulda3:                              Fixed-point fractional library routines.

  50907.                                                              (line  178)

  50908. * __muldc3:                              Soft float library routines.

  50909.                                                              (line  239)

  50910. * __muldf3:                              Soft float library routines.

  50911.                                                              (line   39)

  50912. * __muldi3:                              Integer library routines.

  50913.                                                              (line   42)

  50914. * __muldq3:                              Fixed-point fractional library routines.

  50915.                                                              (line  165)

  50916. * __mulha3:                              Fixed-point fractional library routines.

  50917.                                                              (line  175)

  50918. * __mulhq3:                              Fixed-point fractional library routines.

  50919.                                                              (line  163)

  50920. * __mulqq3:                              Fixed-point fractional library routines.

  50921.                                                              (line  161)

  50922. * __mulsa3:                              Fixed-point fractional library routines.

  50923.                                                              (line  177)

  50924. * __mulsc3:                              Soft float library routines.

  50925.                                                              (line  237)

  50926. * __mulsf3:                              Soft float library routines.

  50927.                                                              (line   38)

  50928. * __mulsi3:                              Integer library routines.

  50929.                                                              (line   41)

  50930. * __mulsq3:                              Fixed-point fractional library routines.

  50931.                                                              (line  164)

  50932. * __multa3:                              Fixed-point fractional library routines.

  50933.                                                              (line  179)

  50934. * __multc3:                              Soft float library routines.

  50935.                                                              (line  241)

  50936. * __multf3:                              Soft float library routines.

  50937.                                                              (line   40)

  50938. * __multi3:                              Integer library routines.

  50939.                                                              (line   43)

  50940. * __muluda3:                             Fixed-point fractional library routines.

  50941.                                                              (line  185)

  50942. * __muludq3:                             Fixed-point fractional library routines.

  50943.                                                              (line  173)

  50944. * __muluha3:                             Fixed-point fractional library routines.

  50945.                                                              (line  181)

  50946. * __muluhq3:                             Fixed-point fractional library routines.

  50947.                                                              (line  169)

  50948. * __muluqq3:                             Fixed-point fractional library routines.

  50949.                                                              (line  167)

  50950. * __mulusa3:                             Fixed-point fractional library routines.

  50951.                                                              (line  183)

  50952. * __mulusq3:                             Fixed-point fractional library routines.

  50953.                                                              (line  171)

  50954. * __muluta3:                             Fixed-point fractional library routines.

  50955.                                                              (line  187)

  50956. * __mulvdi3:                             Integer library routines.

  50957.                                                              (line  114)

  50958. * __mulvsi3:                             Integer library routines.

  50959.                                                              (line  113)

  50960. * __mulxc3:                              Soft float library routines.

  50961.                                                              (line  243)

  50962. * __mulxf3:                              Soft float library routines.

  50963.                                                              (line   42)

  50964. * __nedf2:                               Soft float library routines.

  50965.                                                              (line  199)

  50966. * __negda2:                              Fixed-point fractional library routines.

  50967.                                                              (line  306)

  50968. * __negdf2:                              Soft float library routines.

  50969.                                                              (line   55)

  50970. * __negdi2:                              Integer library routines.

  50971.                                                              (line   46)

  50972. * __negdq2:                              Fixed-point fractional library routines.

  50973.                                                              (line  296)

  50974. * __negha2:                              Fixed-point fractional library routines.

  50975.                                                              (line  304)

  50976. * __neghq2:                              Fixed-point fractional library routines.

  50977.                                                              (line  294)

  50978. * __negqq2:                              Fixed-point fractional library routines.

  50979.                                                              (line  293)

  50980. * __negsa2:                              Fixed-point fractional library routines.

  50981.                                                              (line  305)

  50982. * __negsf2:                              Soft float library routines.

  50983.                                                              (line   54)

  50984. * __negsq2:                              Fixed-point fractional library routines.

  50985.                                                              (line  295)

  50986. * __negta2:                              Fixed-point fractional library routines.

  50987.                                                              (line  307)

  50988. * __negtf2:                              Soft float library routines.

  50989.                                                              (line   56)

  50990. * __negti2:                              Integer library routines.

  50991.                                                              (line   47)

  50992. * __neguda2:                             Fixed-point fractional library routines.

  50993.                                                              (line  311)

  50994. * __negudq2:                             Fixed-point fractional library routines.

  50995.                                                              (line  302)

  50996. * __neguha2:                             Fixed-point fractional library routines.

  50997.                                                              (line  308)

  50998. * __neguhq2:                             Fixed-point fractional library routines.

  50999.                                                              (line  299)

  51000. * __neguqq2:                             Fixed-point fractional library routines.

  51001.                                                              (line  297)

  51002. * __negusa2:                             Fixed-point fractional library routines.

  51003.                                                              (line  310)

  51004. * __negusq2:                             Fixed-point fractional library routines.

  51005.                                                              (line  300)

  51006. * __neguta2:                             Fixed-point fractional library routines.

  51007.                                                              (line  313)

  51008. * __negvdi2:                             Integer library routines.

  51009.                                                              (line  118)

  51010. * __negvsi2:                             Integer library routines.

  51011.                                                              (line  117)

  51012. * __negxf2:                              Soft float library routines.

  51013.                                                              (line   57)

  51014. * __nesf2:                               Soft float library routines.

  51015.                                                              (line  198)

  51016. * __netf2:                               Soft float library routines.

  51017.                                                              (line  200)

  51018. * __paritydi2:                           Integer library routines.

  51019.                                                              (line  150)

  51020. * __paritysi2:                           Integer library routines.

  51021.                                                              (line  149)

  51022. * __parityti2:                           Integer library routines.

  51023.                                                              (line  151)

  51024. * __popcountdi2:                         Integer library routines.

  51025.                                                              (line  156)

  51026. * __popcountsi2:                         Integer library routines.

  51027.                                                              (line  155)

  51028. * __popcountti2:                         Integer library routines.

  51029.                                                              (line  157)

  51030. * __powidf2:                             Soft float library routines.

  51031.                                                              (line  232)

  51032. * __powisf2:                             Soft float library routines.

  51033.                                                              (line  231)

  51034. * __powitf2:                             Soft float library routines.

  51035.                                                              (line  233)

  51036. * __powixf2:                             Soft float library routines.

  51037.                                                              (line  234)

  51038. * __satfractdadq:                        Fixed-point fractional library routines.

  51039.                                                              (line 1160)

  51040. * __satfractdaha2:                       Fixed-point fractional library routines.

  51041.                                                              (line 1161)

  51042. * __satfractdahq:                        Fixed-point fractional library routines.

  51043.                                                              (line 1158)

  51044. * __satfractdaqq:                        Fixed-point fractional library routines.

  51045.                                                              (line 1157)

  51046. * __satfractdasa2:                       Fixed-point fractional library routines.

  51047.                                                              (line 1162)

  51048. * __satfractdasq:                        Fixed-point fractional library routines.

  51049.                                                              (line 1159)

  51050. * __satfractdata2:                       Fixed-point fractional library routines.

  51051.                                                              (line 1163)

  51052. * __satfractdauda:                       Fixed-point fractional library routines.

  51053.                                                              (line 1173)

  51054. * __satfractdaudq:                       Fixed-point fractional library routines.

  51055.                                                              (line 1168)

  51056. * __satfractdauha:                       Fixed-point fractional library routines.

  51057.                                                              (line 1170)

  51058. * __satfractdauhq:                       Fixed-point fractional library routines.

  51059.                                                              (line 1166)

  51060. * __satfractdauqq:                       Fixed-point fractional library routines.

  51061.                                                              (line 1164)

  51062. * __satfractdausa:                       Fixed-point fractional library routines.

  51063.                                                              (line 1172)

  51064. * __satfractdausq:                       Fixed-point fractional library routines.

  51065.                                                              (line 1167)

  51066. * __satfractdauta:                       Fixed-point fractional library routines.

  51067.                                                              (line 1174)

  51068. * __satfractdfda:                        Fixed-point fractional library routines.

  51069.                                                              (line 1513)

  51070. * __satfractdfdq:                        Fixed-point fractional library routines.

  51071.                                                              (line 1510)

  51072. * __satfractdfha:                        Fixed-point fractional library routines.

  51073.                                                              (line 1511)

  51074. * __satfractdfhq:                        Fixed-point fractional library routines.

  51075.                                                              (line 1508)

  51076. * __satfractdfqq:                        Fixed-point fractional library routines.

  51077.                                                              (line 1507)

  51078. * __satfractdfsa:                        Fixed-point fractional library routines.

  51079.                                                              (line 1512)

  51080. * __satfractdfsq:                        Fixed-point fractional library routines.

  51081.                                                              (line 1509)

  51082. * __satfractdfta:                        Fixed-point fractional library routines.

  51083.                                                              (line 1514)

  51084. * __satfractdfuda:                       Fixed-point fractional library routines.

  51085.                                                              (line 1522)

  51086. * __satfractdfudq:                       Fixed-point fractional library routines.

  51087.                                                              (line 1518)

  51088. * __satfractdfuha:                       Fixed-point fractional library routines.

  51089.                                                              (line 1520)

  51090. * __satfractdfuhq:                       Fixed-point fractional library routines.

  51091.                                                              (line 1516)

  51092. * __satfractdfuqq:                       Fixed-point fractional library routines.

  51093.                                                              (line 1515)

  51094. * __satfractdfusa:                       Fixed-point fractional library routines.

  51095.                                                              (line 1521)

  51096. * __satfractdfusq:                       Fixed-point fractional library routines.

  51097.                                                              (line 1517)

  51098. * __satfractdfuta:                       Fixed-point fractional library routines.

  51099.                                                              (line 1523)

  51100. * __satfractdida:                        Fixed-point fractional library routines.

  51101.                                                              (line 1463)

  51102. * __satfractdidq:                        Fixed-point fractional library routines.

  51103.                                                              (line 1460)

  51104. * __satfractdiha:                        Fixed-point fractional library routines.

  51105.                                                              (line 1461)

  51106. * __satfractdihq:                        Fixed-point fractional library routines.

  51107.                                                              (line 1458)

  51108. * __satfractdiqq:                        Fixed-point fractional library routines.

  51109.                                                              (line 1457)

  51110. * __satfractdisa:                        Fixed-point fractional library routines.

  51111.                                                              (line 1462)

  51112. * __satfractdisq:                        Fixed-point fractional library routines.

  51113.                                                              (line 1459)

  51114. * __satfractdita:                        Fixed-point fractional library routines.

  51115.                                                              (line 1464)

  51116. * __satfractdiuda:                       Fixed-point fractional library routines.

  51117.                                                              (line 1471)

  51118. * __satfractdiudq:                       Fixed-point fractional library routines.

  51119.                                                              (line 1468)

  51120. * __satfractdiuha:                       Fixed-point fractional library routines.

  51121.                                                              (line 1469)

  51122. * __satfractdiuhq:                       Fixed-point fractional library routines.

  51123.                                                              (line 1466)

  51124. * __satfractdiuqq:                       Fixed-point fractional library routines.

  51125.                                                              (line 1465)

  51126. * __satfractdiusa:                       Fixed-point fractional library routines.

  51127.                                                              (line 1470)

  51128. * __satfractdiusq:                       Fixed-point fractional library routines.

  51129.                                                              (line 1467)

  51130. * __satfractdiuta:                       Fixed-point fractional library routines.

  51131.                                                              (line 1472)

  51132. * __satfractdqda:                        Fixed-point fractional library routines.

  51133.                                                              (line 1105)

  51134. * __satfractdqha:                        Fixed-point fractional library routines.

  51135.                                                              (line 1103)

  51136. * __satfractdqhq2:                       Fixed-point fractional library routines.

  51137.                                                              (line 1101)

  51138. * __satfractdqqq2:                       Fixed-point fractional library routines.

  51139.                                                              (line 1100)

  51140. * __satfractdqsa:                        Fixed-point fractional library routines.

  51141.                                                              (line 1104)

  51142. * __satfractdqsq2:                       Fixed-point fractional library routines.

  51143.                                                              (line 1102)

  51144. * __satfractdqta:                        Fixed-point fractional library routines.

  51145.                                                              (line 1106)

  51146. * __satfractdquda:                       Fixed-point fractional library routines.

  51147.                                                              (line 1117)

  51148. * __satfractdqudq:                       Fixed-point fractional library routines.

  51149.                                                              (line 1112)

  51150. * __satfractdquha:                       Fixed-point fractional library routines.

  51151.                                                              (line 1114)

  51152. * __satfractdquhq:                       Fixed-point fractional library routines.

  51153.                                                              (line 1109)

  51154. * __satfractdquqq:                       Fixed-point fractional library routines.

  51155.                                                              (line 1107)

  51156. * __satfractdqusa:                       Fixed-point fractional library routines.

  51157.                                                              (line 1116)

  51158. * __satfractdqusq:                       Fixed-point fractional library routines.

  51159.                                                              (line 1110)

  51160. * __satfractdquta:                       Fixed-point fractional library routines.

  51161.                                                              (line 1119)

  51162. * __satfracthada2:                       Fixed-point fractional library routines.

  51163.                                                              (line 1126)

  51164. * __satfracthadq:                        Fixed-point fractional library routines.

  51165.                                                              (line 1124)

  51166. * __satfracthahq:                        Fixed-point fractional library routines.

  51167.                                                              (line 1122)

  51168. * __satfracthaqq:                        Fixed-point fractional library routines.

  51169.                                                              (line 1121)

  51170. * __satfracthasa2:                       Fixed-point fractional library routines.

  51171.                                                              (line 1125)

  51172. * __satfracthasq:                        Fixed-point fractional library routines.

  51173.                                                              (line 1123)

  51174. * __satfracthata2:                       Fixed-point fractional library routines.

  51175.                                                              (line 1127)

  51176. * __satfracthauda:                       Fixed-point fractional library routines.

  51177.                                                              (line 1138)

  51178. * __satfracthaudq:                       Fixed-point fractional library routines.

  51179.                                                              (line 1133)

  51180. * __satfracthauha:                       Fixed-point fractional library routines.

  51181.                                                              (line 1135)

  51182. * __satfracthauhq:                       Fixed-point fractional library routines.

  51183.                                                              (line 1130)

  51184. * __satfracthauqq:                       Fixed-point fractional library routines.

  51185.                                                              (line 1128)

  51186. * __satfracthausa:                       Fixed-point fractional library routines.

  51187.                                                              (line 1137)

  51188. * __satfracthausq:                       Fixed-point fractional library routines.

  51189.                                                              (line 1131)

  51190. * __satfracthauta:                       Fixed-point fractional library routines.

  51191.                                                              (line 1140)

  51192. * __satfracthida:                        Fixed-point fractional library routines.

  51193.                                                              (line 1431)

  51194. * __satfracthidq:                        Fixed-point fractional library routines.

  51195.                                                              (line 1428)

  51196. * __satfracthiha:                        Fixed-point fractional library routines.

  51197.                                                              (line 1429)

  51198. * __satfracthihq:                        Fixed-point fractional library routines.

  51199.                                                              (line 1426)

  51200. * __satfracthiqq:                        Fixed-point fractional library routines.

  51201.                                                              (line 1425)

  51202. * __satfracthisa:                        Fixed-point fractional library routines.

  51203.                                                              (line 1430)

  51204. * __satfracthisq:                        Fixed-point fractional library routines.

  51205.                                                              (line 1427)

  51206. * __satfracthita:                        Fixed-point fractional library routines.

  51207.                                                              (line 1432)

  51208. * __satfracthiuda:                       Fixed-point fractional library routines.

  51209.                                                              (line 1439)

  51210. * __satfracthiudq:                       Fixed-point fractional library routines.

  51211.                                                              (line 1436)

  51212. * __satfracthiuha:                       Fixed-point fractional library routines.

  51213.                                                              (line 1437)

  51214. * __satfracthiuhq:                       Fixed-point fractional library routines.

  51215.                                                              (line 1434)

  51216. * __satfracthiuqq:                       Fixed-point fractional library routines.

  51217.                                                              (line 1433)

  51218. * __satfracthiusa:                       Fixed-point fractional library routines.

  51219.                                                              (line 1438)

  51220. * __satfracthiusq:                       Fixed-point fractional library routines.

  51221.                                                              (line 1435)

  51222. * __satfracthiuta:                       Fixed-point fractional library routines.

  51223.                                                              (line 1440)

  51224. * __satfracthqda:                        Fixed-point fractional library routines.

  51225.                                                              (line 1071)

  51226. * __satfracthqdq2:                       Fixed-point fractional library routines.

  51227.                                                              (line 1068)

  51228. * __satfracthqha:                        Fixed-point fractional library routines.

  51229.                                                              (line 1069)

  51230. * __satfracthqqq2:                       Fixed-point fractional library routines.

  51231.                                                              (line 1066)

  51232. * __satfracthqsa:                        Fixed-point fractional library routines.

  51233.                                                              (line 1070)

  51234. * __satfracthqsq2:                       Fixed-point fractional library routines.

  51235.                                                              (line 1067)

  51236. * __satfracthqta:                        Fixed-point fractional library routines.

  51237.                                                              (line 1072)

  51238. * __satfracthquda:                       Fixed-point fractional library routines.

  51239.                                                              (line 1079)

  51240. * __satfracthqudq:                       Fixed-point fractional library routines.

  51241.                                                              (line 1076)

  51242. * __satfracthquha:                       Fixed-point fractional library routines.

  51243.                                                              (line 1077)

  51244. * __satfracthquhq:                       Fixed-point fractional library routines.

  51245.                                                              (line 1074)

  51246. * __satfracthquqq:                       Fixed-point fractional library routines.

  51247.                                                              (line 1073)

  51248. * __satfracthqusa:                       Fixed-point fractional library routines.

  51249.                                                              (line 1078)

  51250. * __satfracthqusq:                       Fixed-point fractional library routines.

  51251.                                                              (line 1075)

  51252. * __satfracthquta:                       Fixed-point fractional library routines.

  51253.                                                              (line 1080)

  51254. * __satfractqida:                        Fixed-point fractional library routines.

  51255.                                                              (line 1409)

  51256. * __satfractqidq:                        Fixed-point fractional library routines.

  51257.                                                              (line 1406)

  51258. * __satfractqiha:                        Fixed-point fractional library routines.

  51259.                                                              (line 1407)

  51260. * __satfractqihq:                        Fixed-point fractional library routines.

  51261.                                                              (line 1404)

  51262. * __satfractqiqq:                        Fixed-point fractional library routines.

  51263.                                                              (line 1403)

  51264. * __satfractqisa:                        Fixed-point fractional library routines.

  51265.                                                              (line 1408)

  51266. * __satfractqisq:                        Fixed-point fractional library routines.

  51267.                                                              (line 1405)

  51268. * __satfractqita:                        Fixed-point fractional library routines.

  51269.                                                              (line 1410)

  51270. * __satfractqiuda:                       Fixed-point fractional library routines.

  51271.                                                              (line 1421)

  51272. * __satfractqiudq:                       Fixed-point fractional library routines.

  51273.                                                              (line 1416)

  51274. * __satfractqiuha:                       Fixed-point fractional library routines.

  51275.                                                              (line 1418)

  51276. * __satfractqiuhq:                       Fixed-point fractional library routines.

  51277.                                                              (line 1413)

  51278. * __satfractqiuqq:                       Fixed-point fractional library routines.

  51279.                                                              (line 1411)

  51280. * __satfractqiusa:                       Fixed-point fractional library routines.

  51281.                                                              (line 1420)

  51282. * __satfractqiusq:                       Fixed-point fractional library routines.

  51283.                                                              (line 1414)

  51284. * __satfractqiuta:                       Fixed-point fractional library routines.

  51285.                                                              (line 1423)

  51286. * __satfractqqda:                        Fixed-point fractional library routines.

  51287.                                                              (line 1050)

  51288. * __satfractqqdq2:                       Fixed-point fractional library routines.

  51289.                                                              (line 1047)

  51290. * __satfractqqha:                        Fixed-point fractional library routines.

  51291.                                                              (line 1048)

  51292. * __satfractqqhq2:                       Fixed-point fractional library routines.

  51293.                                                              (line 1045)

  51294. * __satfractqqsa:                        Fixed-point fractional library routines.

  51295.                                                              (line 1049)

  51296. * __satfractqqsq2:                       Fixed-point fractional library routines.

  51297.                                                              (line 1046)

  51298. * __satfractqqta:                        Fixed-point fractional library routines.

  51299.                                                              (line 1051)

  51300. * __satfractqquda:                       Fixed-point fractional library routines.

  51301.                                                              (line 1062)

  51302. * __satfractqqudq:                       Fixed-point fractional library routines.

  51303.                                                              (line 1057)

  51304. * __satfractqquha:                       Fixed-point fractional library routines.

  51305.                                                              (line 1059)

  51306. * __satfractqquhq:                       Fixed-point fractional library routines.

  51307.                                                              (line 1054)

  51308. * __satfractqquqq:                       Fixed-point fractional library routines.

  51309.                                                              (line 1052)

  51310. * __satfractqqusa:                       Fixed-point fractional library routines.

  51311.                                                              (line 1061)

  51312. * __satfractqqusq:                       Fixed-point fractional library routines.

  51313.                                                              (line 1055)

  51314. * __satfractqquta:                       Fixed-point fractional library routines.

  51315.                                                              (line 1064)

  51316. * __satfractsada2:                       Fixed-point fractional library routines.

  51317.                                                              (line 1147)

  51318. * __satfractsadq:                        Fixed-point fractional library routines.

  51319.                                                              (line 1145)

  51320. * __satfractsaha2:                       Fixed-point fractional library routines.

  51321.                                                              (line 1146)

  51322. * __satfractsahq:                        Fixed-point fractional library routines.

  51323.                                                              (line 1143)

  51324. * __satfractsaqq:                        Fixed-point fractional library routines.

  51325.                                                              (line 1142)

  51326. * __satfractsasq:                        Fixed-point fractional library routines.

  51327.                                                              (line 1144)

  51328. * __satfractsata2:                       Fixed-point fractional library routines.

  51329.                                                              (line 1148)

  51330. * __satfractsauda:                       Fixed-point fractional library routines.

  51331.                                                              (line 1155)

  51332. * __satfractsaudq:                       Fixed-point fractional library routines.

  51333.                                                              (line 1152)

  51334. * __satfractsauha:                       Fixed-point fractional library routines.

  51335.                                                              (line 1153)

  51336. * __satfractsauhq:                       Fixed-point fractional library routines.

  51337.                                                              (line 1150)

  51338. * __satfractsauqq:                       Fixed-point fractional library routines.

  51339.                                                              (line 1149)

  51340. * __satfractsausa:                       Fixed-point fractional library routines.

  51341.                                                              (line 1154)

  51342. * __satfractsausq:                       Fixed-point fractional library routines.

  51343.                                                              (line 1151)

  51344. * __satfractsauta:                       Fixed-point fractional library routines.

  51345.                                                              (line 1156)

  51346. * __satfractsfda:                        Fixed-point fractional library routines.

  51347.                                                              (line 1497)

  51348. * __satfractsfdq:                        Fixed-point fractional library routines.

  51349.                                                              (line 1494)

  51350. * __satfractsfha:                        Fixed-point fractional library routines.

  51351.                                                              (line 1495)

  51352. * __satfractsfhq:                        Fixed-point fractional library routines.

  51353.                                                              (line 1492)

  51354. * __satfractsfqq:                        Fixed-point fractional library routines.

  51355.                                                              (line 1491)

  51356. * __satfractsfsa:                        Fixed-point fractional library routines.

  51357.                                                              (line 1496)

  51358. * __satfractsfsq:                        Fixed-point fractional library routines.

  51359.                                                              (line 1493)

  51360. * __satfractsfta:                        Fixed-point fractional library routines.

  51361.                                                              (line 1498)

  51362. * __satfractsfuda:                       Fixed-point fractional library routines.

  51363.                                                              (line 1505)

  51364. * __satfractsfudq:                       Fixed-point fractional library routines.

  51365.                                                              (line 1502)

  51366. * __satfractsfuha:                       Fixed-point fractional library routines.

  51367.                                                              (line 1503)

  51368. * __satfractsfuhq:                       Fixed-point fractional library routines.

  51369.                                                              (line 1500)

  51370. * __satfractsfuqq:                       Fixed-point fractional library routines.

  51371.                                                              (line 1499)

  51372. * __satfractsfusa:                       Fixed-point fractional library routines.

  51373.                                                              (line 1504)

  51374. * __satfractsfusq:                       Fixed-point fractional library routines.

  51375.                                                              (line 1501)

  51376. * __satfractsfuta:                       Fixed-point fractional library routines.

  51377.                                                              (line 1506)

  51378. * __satfractsida:                        Fixed-point fractional library routines.

  51379.                                                              (line 1447)

  51380. * __satfractsidq:                        Fixed-point fractional library routines.

  51381.                                                              (line 1444)

  51382. * __satfractsiha:                        Fixed-point fractional library routines.

  51383.                                                              (line 1445)

  51384. * __satfractsihq:                        Fixed-point fractional library routines.

  51385.                                                              (line 1442)

  51386. * __satfractsiqq:                        Fixed-point fractional library routines.

  51387.                                                              (line 1441)

  51388. * __satfractsisa:                        Fixed-point fractional library routines.

  51389.                                                              (line 1446)

  51390. * __satfractsisq:                        Fixed-point fractional library routines.

  51391.                                                              (line 1443)

  51392. * __satfractsita:                        Fixed-point fractional library routines.

  51393.                                                              (line 1448)

  51394. * __satfractsiuda:                       Fixed-point fractional library routines.

  51395.                                                              (line 1455)

  51396. * __satfractsiudq:                       Fixed-point fractional library routines.

  51397.                                                              (line 1452)

  51398. * __satfractsiuha:                       Fixed-point fractional library routines.

  51399.                                                              (line 1453)

  51400. * __satfractsiuhq:                       Fixed-point fractional library routines.

  51401.                                                              (line 1450)

  51402. * __satfractsiuqq:                       Fixed-point fractional library routines.

  51403.                                                              (line 1449)

  51404. * __satfractsiusa:                       Fixed-point fractional library routines.

  51405.                                                              (line 1454)

  51406. * __satfractsiusq:                       Fixed-point fractional library routines.

  51407.                                                              (line 1451)

  51408. * __satfractsiuta:                       Fixed-point fractional library routines.

  51409.                                                              (line 1456)

  51410. * __satfractsqda:                        Fixed-point fractional library routines.

  51411.                                                              (line 1086)

  51412. * __satfractsqdq2:                       Fixed-point fractional library routines.

  51413.                                                              (line 1083)

  51414. * __satfractsqha:                        Fixed-point fractional library routines.

  51415.                                                              (line 1084)

  51416. * __satfractsqhq2:                       Fixed-point fractional library routines.

  51417.                                                              (line 1082)

  51418. * __satfractsqqq2:                       Fixed-point fractional library routines.

  51419.                                                              (line 1081)

  51420. * __satfractsqsa:                        Fixed-point fractional library routines.

  51421.                                                              (line 1085)

  51422. * __satfractsqta:                        Fixed-point fractional library routines.

  51423.                                                              (line 1087)

  51424. * __satfractsquda:                       Fixed-point fractional library routines.

  51425.                                                              (line 1097)

  51426. * __satfractsqudq:                       Fixed-point fractional library routines.

  51427.                                                              (line 1092)

  51428. * __satfractsquha:                       Fixed-point fractional library routines.

  51429.                                                              (line 1094)

  51430. * __satfractsquhq:                       Fixed-point fractional library routines.

  51431.                                                              (line 1090)

  51432. * __satfractsquqq:                       Fixed-point fractional library routines.

  51433.                                                              (line 1088)

  51434. * __satfractsqusa:                       Fixed-point fractional library routines.

  51435.                                                              (line 1096)

  51436. * __satfractsqusq:                       Fixed-point fractional library routines.

  51437.                                                              (line 1091)

  51438. * __satfractsquta:                       Fixed-point fractional library routines.

  51439.                                                              (line 1098)

  51440. * __satfracttada2:                       Fixed-point fractional library routines.

  51441.                                                              (line 1182)

  51442. * __satfracttadq:                        Fixed-point fractional library routines.

  51443.                                                              (line 1179)

  51444. * __satfracttaha2:                       Fixed-point fractional library routines.

  51445.                                                              (line 1180)

  51446. * __satfracttahq:                        Fixed-point fractional library routines.

  51447.                                                              (line 1177)

  51448. * __satfracttaqq:                        Fixed-point fractional library routines.

  51449.                                                              (line 1176)

  51450. * __satfracttasa2:                       Fixed-point fractional library routines.

  51451.                                                              (line 1181)

  51452. * __satfracttasq:                        Fixed-point fractional library routines.

  51453.                                                              (line 1178)

  51454. * __satfracttauda:                       Fixed-point fractional library routines.

  51455.                                                              (line 1193)

  51456. * __satfracttaudq:                       Fixed-point fractional library routines.

  51457.                                                              (line 1188)

  51458. * __satfracttauha:                       Fixed-point fractional library routines.

  51459.                                                              (line 1190)

  51460. * __satfracttauhq:                       Fixed-point fractional library routines.

  51461.                                                              (line 1185)

  51462. * __satfracttauqq:                       Fixed-point fractional library routines.

  51463.                                                              (line 1183)

  51464. * __satfracttausa:                       Fixed-point fractional library routines.

  51465.                                                              (line 1192)

  51466. * __satfracttausq:                       Fixed-point fractional library routines.

  51467.                                                              (line 1186)

  51468. * __satfracttauta:                       Fixed-point fractional library routines.

  51469.                                                              (line 1195)

  51470. * __satfracttida:                        Fixed-point fractional library routines.

  51471.                                                              (line 1479)

  51472. * __satfracttidq:                        Fixed-point fractional library routines.

  51473.                                                              (line 1476)

  51474. * __satfracttiha:                        Fixed-point fractional library routines.

  51475.                                                              (line 1477)

  51476. * __satfracttihq:                        Fixed-point fractional library routines.

  51477.                                                              (line 1474)

  51478. * __satfracttiqq:                        Fixed-point fractional library routines.

  51479.                                                              (line 1473)

  51480. * __satfracttisa:                        Fixed-point fractional library routines.

  51481.                                                              (line 1478)

  51482. * __satfracttisq:                        Fixed-point fractional library routines.

  51483.                                                              (line 1475)

  51484. * __satfracttita:                        Fixed-point fractional library routines.

  51485.                                                              (line 1480)

  51486. * __satfracttiuda:                       Fixed-point fractional library routines.

  51487.                                                              (line 1488)

  51488. * __satfracttiudq:                       Fixed-point fractional library routines.

  51489.                                                              (line 1484)

  51490. * __satfracttiuha:                       Fixed-point fractional library routines.

  51491.                                                              (line 1486)

  51492. * __satfracttiuhq:                       Fixed-point fractional library routines.

  51493.                                                              (line 1482)

  51494. * __satfracttiuqq:                       Fixed-point fractional library routines.

  51495.                                                              (line 1481)

  51496. * __satfracttiusa:                       Fixed-point fractional library routines.

  51497.                                                              (line 1487)

  51498. * __satfracttiusq:                       Fixed-point fractional library routines.

  51499.                                                              (line 1483)

  51500. * __satfracttiuta:                       Fixed-point fractional library routines.

  51501.                                                              (line 1489)

  51502. * __satfractudada:                       Fixed-point fractional library routines.

  51503.                                                              (line 1358)

  51504. * __satfractudadq:                       Fixed-point fractional library routines.

  51505.                                                              (line 1353)

  51506. * __satfractudaha:                       Fixed-point fractional library routines.

  51507.                                                              (line 1355)

  51508. * __satfractudahq:                       Fixed-point fractional library routines.

  51509.                                                              (line 1351)

  51510. * __satfractudaqq:                       Fixed-point fractional library routines.

  51511.                                                              (line 1349)

  51512. * __satfractudasa:                       Fixed-point fractional library routines.

  51513.                                                              (line 1357)

  51514. * __satfractudasq:                       Fixed-point fractional library routines.

  51515.                                                              (line 1352)

  51516. * __satfractudata:                       Fixed-point fractional library routines.

  51517.                                                              (line 1359)

  51518. * __satfractudaudq:                      Fixed-point fractional library routines.

  51519.                                                              (line 1367)

  51520. * __satfractudauha2:                     Fixed-point fractional library routines.

  51521.                                                              (line 1369)

  51522. * __satfractudauhq:                      Fixed-point fractional library routines.

  51523.                                                              (line 1363)

  51524. * __satfractudauqq:                      Fixed-point fractional library routines.

  51525.                                                              (line 1361)

  51526. * __satfractudausa2:                     Fixed-point fractional library routines.

  51527.                                                              (line 1371)

  51528. * __satfractudausq:                      Fixed-point fractional library routines.

  51529.                                                              (line 1365)

  51530. * __satfractudauta2:                     Fixed-point fractional library routines.

  51531.                                                              (line 1373)

  51532. * __satfractudqda:                       Fixed-point fractional library routines.

  51533.                                                              (line 1282)

  51534. * __satfractudqdq:                       Fixed-point fractional library routines.

  51535.                                                              (line 1277)

  51536. * __satfractudqha:                       Fixed-point fractional library routines.

  51537.                                                              (line 1279)

  51538. * __satfractudqhq:                       Fixed-point fractional library routines.

  51539.                                                              (line 1274)

  51540. * __satfractudqqq:                       Fixed-point fractional library routines.

  51541.                                                              (line 1272)

  51542. * __satfractudqsa:                       Fixed-point fractional library routines.

  51543.                                                              (line 1281)

  51544. * __satfractudqsq:                       Fixed-point fractional library routines.

  51545.                                                              (line 1275)

  51546. * __satfractudqta:                       Fixed-point fractional library routines.

  51547.                                                              (line 1284)

  51548. * __satfractudquda:                      Fixed-point fractional library routines.

  51549.                                                              (line 1296)

  51550. * __satfractudquha:                      Fixed-point fractional library routines.

  51551.                                                              (line 1292)

  51552. * __satfractudquhq2:                     Fixed-point fractional library routines.

  51553.                                                              (line 1288)

  51554. * __satfractudquqq2:                     Fixed-point fractional library routines.

  51555.                                                              (line 1286)

  51556. * __satfractudqusa:                      Fixed-point fractional library routines.

  51557.                                                              (line 1294)

  51558. * __satfractudqusq2:                     Fixed-point fractional library routines.

  51559.                                                              (line 1290)

  51560. * __satfractudquta:                      Fixed-point fractional library routines.

  51561.                                                              (line 1298)

  51562. * __satfractuhada:                       Fixed-point fractional library routines.

  51563.                                                              (line 1310)

  51564. * __satfractuhadq:                       Fixed-point fractional library routines.

  51565.                                                              (line 1305)

  51566. * __satfractuhaha:                       Fixed-point fractional library routines.

  51567.                                                              (line 1307)

  51568. * __satfractuhahq:                       Fixed-point fractional library routines.

  51569.                                                              (line 1302)

  51570. * __satfractuhaqq:                       Fixed-point fractional library routines.

  51571.                                                              (line 1300)

  51572. * __satfractuhasa:                       Fixed-point fractional library routines.

  51573.                                                              (line 1309)

  51574. * __satfractuhasq:                       Fixed-point fractional library routines.

  51575.                                                              (line 1303)

  51576. * __satfractuhata:                       Fixed-point fractional library routines.

  51577.                                                              (line 1312)

  51578. * __satfractuhauda2:                     Fixed-point fractional library routines.

  51579.                                                              (line 1324)

  51580. * __satfractuhaudq:                      Fixed-point fractional library routines.

  51581.                                                              (line 1320)

  51582. * __satfractuhauhq:                      Fixed-point fractional library routines.

  51583.                                                              (line 1316)

  51584. * __satfractuhauqq:                      Fixed-point fractional library routines.

  51585.                                                              (line 1314)

  51586. * __satfractuhausa2:                     Fixed-point fractional library routines.

  51587.                                                              (line 1322)

  51588. * __satfractuhausq:                      Fixed-point fractional library routines.

  51589.                                                              (line 1318)

  51590. * __satfractuhauta2:                     Fixed-point fractional library routines.

  51591.                                                              (line 1326)

  51592. * __satfractuhqda:                       Fixed-point fractional library routines.

  51593.                                                              (line 1231)

  51594. * __satfractuhqdq:                       Fixed-point fractional library routines.

  51595.                                                              (line 1228)

  51596. * __satfractuhqha:                       Fixed-point fractional library routines.

  51597.                                                              (line 1229)

  51598. * __satfractuhqhq:                       Fixed-point fractional library routines.

  51599.                                                              (line 1226)

  51600. * __satfractuhqqq:                       Fixed-point fractional library routines.

  51601.                                                              (line 1225)

  51602. * __satfractuhqsa:                       Fixed-point fractional library routines.

  51603.                                                              (line 1230)

  51604. * __satfractuhqsq:                       Fixed-point fractional library routines.

  51605.                                                              (line 1227)

  51606. * __satfractuhqta:                       Fixed-point fractional library routines.

  51607.                                                              (line 1232)

  51608. * __satfractuhquda:                      Fixed-point fractional library routines.

  51609.                                                              (line 1242)

  51610. * __satfractuhqudq2:                     Fixed-point fractional library routines.

  51611.                                                              (line 1237)

  51612. * __satfractuhquha:                      Fixed-point fractional library routines.

  51613.                                                              (line 1239)

  51614. * __satfractuhquqq2:                     Fixed-point fractional library routines.

  51615.                                                              (line 1233)

  51616. * __satfractuhqusa:                      Fixed-point fractional library routines.

  51617.                                                              (line 1241)

  51618. * __satfractuhqusq2:                     Fixed-point fractional library routines.

  51619.                                                              (line 1235)

  51620. * __satfractuhquta:                      Fixed-point fractional library routines.

  51621.                                                              (line 1244)

  51622. * __satfractunsdida:                     Fixed-point fractional library routines.

  51623.                                                              (line 1841)

  51624. * __satfractunsdidq:                     Fixed-point fractional library routines.

  51625.                                                              (line 1837)

  51626. * __satfractunsdiha:                     Fixed-point fractional library routines.

  51627.                                                              (line 1839)

  51628. * __satfractunsdihq:                     Fixed-point fractional library routines.

  51629.                                                              (line 1835)

  51630. * __satfractunsdiqq:                     Fixed-point fractional library routines.

  51631.                                                              (line 1834)

  51632. * __satfractunsdisa:                     Fixed-point fractional library routines.

  51633.                                                              (line 1840)

  51634. * __satfractunsdisq:                     Fixed-point fractional library routines.

  51635.                                                              (line 1836)

  51636. * __satfractunsdita:                     Fixed-point fractional library routines.

  51637.                                                              (line 1842)

  51638. * __satfractunsdiuda:                    Fixed-point fractional library routines.

  51639.                                                              (line 1856)

  51640. * __satfractunsdiudq:                    Fixed-point fractional library routines.

  51641.                                                              (line 1850)

  51642. * __satfractunsdiuha:                    Fixed-point fractional library routines.

  51643.                                                              (line 1852)

  51644. * __satfractunsdiuhq:                    Fixed-point fractional library routines.

  51645.                                                              (line 1846)

  51646. * __satfractunsdiuqq:                    Fixed-point fractional library routines.

  51647.                                                              (line 1844)

  51648. * __satfractunsdiusa:                    Fixed-point fractional library routines.

  51649.                                                              (line 1854)

  51650. * __satfractunsdiusq:                    Fixed-point fractional library routines.

  51651.                                                              (line 1848)

  51652. * __satfractunsdiuta:                    Fixed-point fractional library routines.

  51653.                                                              (line 1858)

  51654. * __satfractunshida:                     Fixed-point fractional library routines.

  51655.                                                              (line 1793)

  51656. * __satfractunshidq:                     Fixed-point fractional library routines.

  51657.                                                              (line 1789)

  51658. * __satfractunshiha:                     Fixed-point fractional library routines.

  51659.                                                              (line 1791)

  51660. * __satfractunshihq:                     Fixed-point fractional library routines.

  51661.                                                              (line 1787)

  51662. * __satfractunshiqq:                     Fixed-point fractional library routines.

  51663.                                                              (line 1786)

  51664. * __satfractunshisa:                     Fixed-point fractional library routines.

  51665.                                                              (line 1792)

  51666. * __satfractunshisq:                     Fixed-point fractional library routines.

  51667.                                                              (line 1788)

  51668. * __satfractunshita:                     Fixed-point fractional library routines.

  51669.                                                              (line 1794)

  51670. * __satfractunshiuda:                    Fixed-point fractional library routines.

  51671.                                                              (line 1808)

  51672. * __satfractunshiudq:                    Fixed-point fractional library routines.

  51673.                                                              (line 1802)

  51674. * __satfractunshiuha:                    Fixed-point fractional library routines.

  51675.                                                              (line 1804)

  51676. * __satfractunshiuhq:                    Fixed-point fractional library routines.

  51677.                                                              (line 1798)

  51678. * __satfractunshiuqq:                    Fixed-point fractional library routines.

  51679.                                                              (line 1796)

  51680. * __satfractunshiusa:                    Fixed-point fractional library routines.

  51681.                                                              (line 1806)

  51682. * __satfractunshiusq:                    Fixed-point fractional library routines.

  51683.                                                              (line 1800)

  51684. * __satfractunshiuta:                    Fixed-point fractional library routines.

  51685.                                                              (line 1810)

  51686. * __satfractunsqida:                     Fixed-point fractional library routines.

  51687.                                                              (line 1767)

  51688. * __satfractunsqidq:                     Fixed-point fractional library routines.

  51689.                                                              (line 1763)

  51690. * __satfractunsqiha:                     Fixed-point fractional library routines.

  51691.                                                              (line 1765)

  51692. * __satfractunsqihq:                     Fixed-point fractional library routines.

  51693.                                                              (line 1761)

  51694. * __satfractunsqiqq:                     Fixed-point fractional library routines.

  51695.                                                              (line 1760)

  51696. * __satfractunsqisa:                     Fixed-point fractional library routines.

  51697.                                                              (line 1766)

  51698. * __satfractunsqisq:                     Fixed-point fractional library routines.

  51699.                                                              (line 1762)

  51700. * __satfractunsqita:                     Fixed-point fractional library routines.

  51701.                                                              (line 1768)

  51702. * __satfractunsqiuda:                    Fixed-point fractional library routines.

  51703.                                                              (line 1782)

  51704. * __satfractunsqiudq:                    Fixed-point fractional library routines.

  51705.                                                              (line 1776)

  51706. * __satfractunsqiuha:                    Fixed-point fractional library routines.

  51707.                                                              (line 1778)

  51708. * __satfractunsqiuhq:                    Fixed-point fractional library routines.

  51709.                                                              (line 1772)

  51710. * __satfractunsqiuqq:                    Fixed-point fractional library routines.

  51711.                                                              (line 1770)

  51712. * __satfractunsqiusa:                    Fixed-point fractional library routines.

  51713.                                                              (line 1780)

  51714. * __satfractunsqiusq:                    Fixed-point fractional library routines.

  51715.                                                              (line 1774)

  51716. * __satfractunsqiuta:                    Fixed-point fractional library routines.

  51717.                                                              (line 1784)

  51718. * __satfractunssida:                     Fixed-point fractional library routines.

  51719.                                                              (line 1818)

  51720. * __satfractunssidq:                     Fixed-point fractional library routines.

  51721.                                                              (line 1815)

  51722. * __satfractunssiha:                     Fixed-point fractional library routines.

  51723.                                                              (line 1816)

  51724. * __satfractunssihq:                     Fixed-point fractional library routines.

  51725.                                                              (line 1813)

  51726. * __satfractunssiqq:                     Fixed-point fractional library routines.

  51727.                                                              (line 1812)

  51728. * __satfractunssisa:                     Fixed-point fractional library routines.

  51729.                                                              (line 1817)

  51730. * __satfractunssisq:                     Fixed-point fractional library routines.

  51731.                                                              (line 1814)

  51732. * __satfractunssita:                     Fixed-point fractional library routines.

  51733.                                                              (line 1819)

  51734. * __satfractunssiuda:                    Fixed-point fractional library routines.

  51735.                                                              (line 1830)

  51736. * __satfractunssiudq:                    Fixed-point fractional library routines.

  51737.                                                              (line 1825)

  51738. * __satfractunssiuha:                    Fixed-point fractional library routines.

  51739.                                                              (line 1827)

  51740. * __satfractunssiuhq:                    Fixed-point fractional library routines.

  51741.                                                              (line 1822)

  51742. * __satfractunssiuqq:                    Fixed-point fractional library routines.

  51743.                                                              (line 1820)

  51744. * __satfractunssiusa:                    Fixed-point fractional library routines.

  51745.                                                              (line 1829)

  51746. * __satfractunssiusq:                    Fixed-point fractional library routines.

  51747.                                                              (line 1823)

  51748. * __satfractunssiuta:                    Fixed-point fractional library routines.

  51749.                                                              (line 1832)

  51750. * __satfractunstida:                     Fixed-point fractional library routines.

  51751.                                                              (line 1870)

  51752. * __satfractunstidq:                     Fixed-point fractional library routines.

  51753.                                                              (line 1865)

  51754. * __satfractunstiha:                     Fixed-point fractional library routines.

  51755.                                                              (line 1867)

  51756. * __satfractunstihq:                     Fixed-point fractional library routines.

  51757.                                                              (line 1862)

  51758. * __satfractunstiqq:                     Fixed-point fractional library routines.

  51759.                                                              (line 1860)

  51760. * __satfractunstisa:                     Fixed-point fractional library routines.

  51761.                                                              (line 1869)

  51762. * __satfractunstisq:                     Fixed-point fractional library routines.

  51763.                                                              (line 1863)

  51764. * __satfractunstita:                     Fixed-point fractional library routines.

  51765.                                                              (line 1872)

  51766. * __satfractunstiuda:                    Fixed-point fractional library routines.

  51767.                                                              (line 1886)

  51768. * __satfractunstiudq:                    Fixed-point fractional library routines.

  51769.                                                              (line 1880)

  51770. * __satfractunstiuha:                    Fixed-point fractional library routines.

  51771.                                                              (line 1882)

  51772. * __satfractunstiuhq:                    Fixed-point fractional library routines.

  51773.                                                              (line 1876)

  51774. * __satfractunstiuqq:                    Fixed-point fractional library routines.

  51775.                                                              (line 1874)

  51776. * __satfractunstiusa:                    Fixed-point fractional library routines.

  51777.                                                              (line 1884)

  51778. * __satfractunstiusq:                    Fixed-point fractional library routines.

  51779.                                                              (line 1878)

  51780. * __satfractunstiuta:                    Fixed-point fractional library routines.

  51781.                                                              (line 1888)

  51782. * __satfractuqqda:                       Fixed-point fractional library routines.

  51783.                                                              (line 1207)

  51784. * __satfractuqqdq:                       Fixed-point fractional library routines.

  51785.                                                              (line 1202)

  51786. * __satfractuqqha:                       Fixed-point fractional library routines.

  51787.                                                              (line 1204)

  51788. * __satfractuqqhq:                       Fixed-point fractional library routines.

  51789.                                                              (line 1199)

  51790. * __satfractuqqqq:                       Fixed-point fractional library routines.

  51791.                                                              (line 1197)

  51792. * __satfractuqqsa:                       Fixed-point fractional library routines.

  51793.                                                              (line 1206)

  51794. * __satfractuqqsq:                       Fixed-point fractional library routines.

  51795.                                                              (line 1200)

  51796. * __satfractuqqta:                       Fixed-point fractional library routines.

  51797.                                                              (line 1209)

  51798. * __satfractuqquda:                      Fixed-point fractional library routines.

  51799.                                                              (line 1221)

  51800. * __satfractuqqudq2:                     Fixed-point fractional library routines.

  51801.                                                              (line 1215)

  51802. * __satfractuqquha:                      Fixed-point fractional library routines.

  51803.                                                              (line 1217)

  51804. * __satfractuqquhq2:                     Fixed-point fractional library routines.

  51805.                                                              (line 1211)

  51806. * __satfractuqqusa:                      Fixed-point fractional library routines.

  51807.                                                              (line 1219)

  51808. * __satfractuqqusq2:                     Fixed-point fractional library routines.

  51809.                                                              (line 1213)

  51810. * __satfractuqquta:                      Fixed-point fractional library routines.

  51811.                                                              (line 1223)

  51812. * __satfractusada:                       Fixed-point fractional library routines.

  51813.                                                              (line 1334)

  51814. * __satfractusadq:                       Fixed-point fractional library routines.

  51815.                                                              (line 1331)

  51816. * __satfractusaha:                       Fixed-point fractional library routines.

  51817.                                                              (line 1332)

  51818. * __satfractusahq:                       Fixed-point fractional library routines.

  51819.                                                              (line 1329)

  51820. * __satfractusaqq:                       Fixed-point fractional library routines.

  51821.                                                              (line 1328)

  51822. * __satfractusasa:                       Fixed-point fractional library routines.

  51823.                                                              (line 1333)

  51824. * __satfractusasq:                       Fixed-point fractional library routines.

  51825.                                                              (line 1330)

  51826. * __satfractusata:                       Fixed-point fractional library routines.

  51827.                                                              (line 1335)

  51828. * __satfractusauda2:                     Fixed-point fractional library routines.

  51829.                                                              (line 1345)

  51830. * __satfractusaudq:                      Fixed-point fractional library routines.

  51831.                                                              (line 1341)

  51832. * __satfractusauha2:                     Fixed-point fractional library routines.

  51833.                                                              (line 1343)

  51834. * __satfractusauhq:                      Fixed-point fractional library routines.

  51835.                                                              (line 1338)

  51836. * __satfractusauqq:                      Fixed-point fractional library routines.

  51837.                                                              (line 1336)

  51838. * __satfractusausq:                      Fixed-point fractional library routines.

  51839.                                                              (line 1339)

  51840. * __satfractusauta2:                     Fixed-point fractional library routines.

  51841.                                                              (line 1347)

  51842. * __satfractusqda:                       Fixed-point fractional library routines.

  51843.                                                              (line 1255)

  51844. * __satfractusqdq:                       Fixed-point fractional library routines.

  51845.                                                              (line 1250)

  51846. * __satfractusqha:                       Fixed-point fractional library routines.

  51847.                                                              (line 1252)

  51848. * __satfractusqhq:                       Fixed-point fractional library routines.

  51849.                                                              (line 1248)

  51850. * __satfractusqqq:                       Fixed-point fractional library routines.

  51851.                                                              (line 1246)

  51852. * __satfractusqsa:                       Fixed-point fractional library routines.

  51853.                                                              (line 1254)

  51854. * __satfractusqsq:                       Fixed-point fractional library routines.

  51855.                                                              (line 1249)

  51856. * __satfractusqta:                       Fixed-point fractional library routines.

  51857.                                                              (line 1256)

  51858. * __satfractusquda:                      Fixed-point fractional library routines.

  51859.                                                              (line 1268)

  51860. * __satfractusqudq2:                     Fixed-point fractional library routines.

  51861.                                                              (line 1262)

  51862. * __satfractusquha:                      Fixed-point fractional library routines.

  51863.                                                              (line 1264)

  51864. * __satfractusquhq2:                     Fixed-point fractional library routines.

  51865.                                                              (line 1260)

  51866. * __satfractusquqq2:                     Fixed-point fractional library routines.

  51867.                                                              (line 1258)

  51868. * __satfractusqusa:                      Fixed-point fractional library routines.

  51869.                                                              (line 1266)

  51870. * __satfractusquta:                      Fixed-point fractional library routines.

  51871.                                                              (line 1270)

  51872. * __satfractutada:                       Fixed-point fractional library routines.

  51873.                                                              (line 1385)

  51874. * __satfractutadq:                       Fixed-point fractional library routines.

  51875.                                                              (line 1380)

  51876. * __satfractutaha:                       Fixed-point fractional library routines.

  51877.                                                              (line 1382)

  51878. * __satfractutahq:                       Fixed-point fractional library routines.

  51879.                                                              (line 1377)

  51880. * __satfractutaqq:                       Fixed-point fractional library routines.

  51881.                                                              (line 1375)

  51882. * __satfractutasa:                       Fixed-point fractional library routines.

  51883.                                                              (line 1384)

  51884. * __satfractutasq:                       Fixed-point fractional library routines.

  51885.                                                              (line 1378)

  51886. * __satfractutata:                       Fixed-point fractional library routines.

  51887.                                                              (line 1387)

  51888. * __satfractutauda2:                     Fixed-point fractional library routines.

  51889.                                                              (line 1401)

  51890. * __satfractutaudq:                      Fixed-point fractional library routines.

  51891.                                                              (line 1395)

  51892. * __satfractutauha2:                     Fixed-point fractional library routines.

  51893.                                                              (line 1397)

  51894. * __satfractutauhq:                      Fixed-point fractional library routines.

  51895.                                                              (line 1391)

  51896. * __satfractutauqq:                      Fixed-point fractional library routines.

  51897.                                                              (line 1389)

  51898. * __satfractutausa2:                     Fixed-point fractional library routines.

  51899.                                                              (line 1399)

  51900. * __satfractutausq:                      Fixed-point fractional library routines.

  51901.                                                              (line 1393)

  51902. * __splitstack_find:                     Miscellaneous routines.

  51903.                                                              (line   15)

  51904. * __ssaddda3:                            Fixed-point fractional library routines.

  51905.                                                              (line   74)

  51906. * __ssadddq3:                            Fixed-point fractional library routines.

  51907.                                                              (line   69)

  51908. * __ssaddha3:                            Fixed-point fractional library routines.

  51909.                                                              (line   71)

  51910. * __ssaddhq3:                            Fixed-point fractional library routines.

  51911.                                                              (line   67)

  51912. * __ssaddqq3:                            Fixed-point fractional library routines.

  51913.                                                              (line   65)

  51914. * __ssaddsa3:                            Fixed-point fractional library routines.

  51915.                                                              (line   73)

  51916. * __ssaddsq3:                            Fixed-point fractional library routines.

  51917.                                                              (line   68)

  51918. * __ssaddta3:                            Fixed-point fractional library routines.

  51919.                                                              (line   75)

  51920. * __ssashlda3:                           Fixed-point fractional library routines.

  51921.                                                              (line  409)

  51922. * __ssashldq3:                           Fixed-point fractional library routines.

  51923.                                                              (line  405)

  51924. * __ssashlha3:                           Fixed-point fractional library routines.

  51925.                                                              (line  407)

  51926. * __ssashlhq3:                           Fixed-point fractional library routines.

  51927.                                                              (line  403)

  51928. * __ssashlsa3:                           Fixed-point fractional library routines.

  51929.                                                              (line  408)

  51930. * __ssashlsq3:                           Fixed-point fractional library routines.

  51931.                                                              (line  404)

  51932. * __ssashlta3:                           Fixed-point fractional library routines.

  51933.                                                              (line  410)

  51934. * __ssdivda3:                            Fixed-point fractional library routines.

  51935.                                                              (line  268)

  51936. * __ssdivdq3:                            Fixed-point fractional library routines.

  51937.                                                              (line  263)

  51938. * __ssdivha3:                            Fixed-point fractional library routines.

  51939.                                                              (line  265)

  51940. * __ssdivhq3:                            Fixed-point fractional library routines.

  51941.                                                              (line  261)

  51942. * __ssdivqq3:                            Fixed-point fractional library routines.

  51943.                                                              (line  259)

  51944. * __ssdivsa3:                            Fixed-point fractional library routines.

  51945.                                                              (line  267)

  51946. * __ssdivsq3:                            Fixed-point fractional library routines.

  51947.                                                              (line  262)

  51948. * __ssdivta3:                            Fixed-point fractional library routines.

  51949.                                                              (line  269)

  51950. * __ssmulda3:                            Fixed-point fractional library routines.

  51951.                                                              (line  200)

  51952. * __ssmuldq3:                            Fixed-point fractional library routines.

  51953.                                                              (line  195)

  51954. * __ssmulha3:                            Fixed-point fractional library routines.

  51955.                                                              (line  197)

  51956. * __ssmulhq3:                            Fixed-point fractional library routines.

  51957.                                                              (line  193)

  51958. * __ssmulqq3:                            Fixed-point fractional library routines.

  51959.                                                              (line  191)

  51960. * __ssmulsa3:                            Fixed-point fractional library routines.

  51961.                                                              (line  199)

  51962. * __ssmulsq3:                            Fixed-point fractional library routines.

  51963.                                                              (line  194)

  51964. * __ssmulta3:                            Fixed-point fractional library routines.

  51965.                                                              (line  201)

  51966. * __ssnegda2:                            Fixed-point fractional library routines.

  51967.                                                              (line  323)

  51968. * __ssnegdq2:                            Fixed-point fractional library routines.

  51969.                                                              (line  320)

  51970. * __ssnegha2:                            Fixed-point fractional library routines.

  51971.                                                              (line  321)

  51972. * __ssneghq2:                            Fixed-point fractional library routines.

  51973.                                                              (line  318)

  51974. * __ssnegqq2:                            Fixed-point fractional library routines.

  51975.                                                              (line  317)

  51976. * __ssnegsa2:                            Fixed-point fractional library routines.

  51977.                                                              (line  322)

  51978. * __ssnegsq2:                            Fixed-point fractional library routines.

  51979.                                                              (line  319)

  51980. * __ssnegta2:                            Fixed-point fractional library routines.

  51981.                                                              (line  324)

  51982. * __sssubda3:                            Fixed-point fractional library routines.

  51983.                                                              (line  136)

  51984. * __sssubdq3:                            Fixed-point fractional library routines.

  51985.                                                              (line  131)

  51986. * __sssubha3:                            Fixed-point fractional library routines.

  51987.                                                              (line  133)

  51988. * __sssubhq3:                            Fixed-point fractional library routines.

  51989.                                                              (line  129)

  51990. * __sssubqq3:                            Fixed-point fractional library routines.

  51991.                                                              (line  127)

  51992. * __sssubsa3:                            Fixed-point fractional library routines.

  51993.                                                              (line  135)

  51994. * __sssubsq3:                            Fixed-point fractional library routines.

  51995.                                                              (line  130)

  51996. * __sssubta3:                            Fixed-point fractional library routines.

  51997.                                                              (line  137)

  51998. * __subda3:                              Fixed-point fractional library routines.

  51999.                                                              (line  114)

  52000. * __subdf3:                              Soft float library routines.

  52001.                                                              (line   30)

  52002. * __subdq3:                              Fixed-point fractional library routines.

  52003.                                                              (line  101)

  52004. * __subha3:                              Fixed-point fractional library routines.

  52005.                                                              (line  111)

  52006. * __subhq3:                              Fixed-point fractional library routines.

  52007.                                                              (line   99)

  52008. * __subqq3:                              Fixed-point fractional library routines.

  52009.                                                              (line   97)

  52010. * __subsa3:                              Fixed-point fractional library routines.

  52011.                                                              (line  113)

  52012. * __subsf3:                              Soft float library routines.

  52013.                                                              (line   29)

  52014. * __subsq3:                              Fixed-point fractional library routines.

  52015.                                                              (line  100)

  52016. * __subta3:                              Fixed-point fractional library routines.

  52017.                                                              (line  115)

  52018. * __subtf3:                              Soft float library routines.

  52019.                                                              (line   31)

  52020. * __subuda3:                             Fixed-point fractional library routines.

  52021.                                                              (line  121)

  52022. * __subudq3:                             Fixed-point fractional library routines.

  52023.                                                              (line  109)

  52024. * __subuha3:                             Fixed-point fractional library routines.

  52025.                                                              (line  117)

  52026. * __subuhq3:                             Fixed-point fractional library routines.

  52027.                                                              (line  105)

  52028. * __subuqq3:                             Fixed-point fractional library routines.

  52029.                                                              (line  103)

  52030. * __subusa3:                             Fixed-point fractional library routines.

  52031.                                                              (line  119)

  52032. * __subusq3:                             Fixed-point fractional library routines.

  52033.                                                              (line  107)

  52034. * __subuta3:                             Fixed-point fractional library routines.

  52035.                                                              (line  123)

  52036. * __subvdi3:                             Integer library routines.

  52037.                                                              (line  122)

  52038. * __subvsi3:                             Integer library routines.

  52039.                                                              (line  121)

  52040. * __subxf3:                              Soft float library routines.

  52041.                                                              (line   33)

  52042. * __truncdfsf2:                          Soft float library routines.

  52043.                                                              (line   75)

  52044. * __trunctfdf2:                          Soft float library routines.

  52045.                                                              (line   72)

  52046. * __trunctfsf2:                          Soft float library routines.

  52047.                                                              (line   74)

  52048. * __truncxfdf2:                          Soft float library routines.

  52049.                                                              (line   71)

  52050. * __truncxfsf2:                          Soft float library routines.

  52051.                                                              (line   73)

  52052. * __ucmpdi2:                             Integer library routines.

  52053.                                                              (line   92)

  52054. * __ucmpti2:                             Integer library routines.

  52055.                                                              (line   93)

  52056. * __udivdi3:                             Integer library routines.

  52057.                                                              (line   52)

  52058. * __udivmoddi4:                          Integer library routines.

  52059.                                                              (line   59)

  52060. * __udivmodti4:                          Integer library routines.

  52061.                                                              (line   61)

  52062. * __udivsi3:                             Integer library routines.

  52063.                                                              (line   50)

  52064. * __udivti3:                             Integer library routines.

  52065.                                                              (line   54)

  52066. * __udivuda3:                            Fixed-point fractional library routines.

  52067.                                                              (line  252)

  52068. * __udivudq3:                            Fixed-point fractional library routines.

  52069.                                                              (line  246)

  52070. * __udivuha3:                            Fixed-point fractional library routines.

  52071.                                                              (line  248)

  52072. * __udivuhq3:                            Fixed-point fractional library routines.

  52073.                                                              (line  242)

  52074. * __udivuqq3:                            Fixed-point fractional library routines.

  52075.                                                              (line  240)

  52076. * __udivusa3:                            Fixed-point fractional library routines.

  52077.                                                              (line  250)

  52078. * __udivusq3:                            Fixed-point fractional library routines.

  52079.                                                              (line  244)

  52080. * __udivuta3:                            Fixed-point fractional library routines.

  52081.                                                              (line  254)

  52082. * __umoddi3:                             Integer library routines.

  52083.                                                              (line   69)

  52084. * __umodsi3:                             Integer library routines.

  52085.                                                              (line   67)

  52086. * __umodti3:                             Integer library routines.

  52087.                                                              (line   71)

  52088. * __unorddf2:                            Soft float library routines.

  52089.                                                              (line  172)

  52090. * __unordsf2:                            Soft float library routines.

  52091.                                                              (line  171)

  52092. * __unordtf2:                            Soft float library routines.

  52093.                                                              (line  173)

  52094. * __usadduda3:                           Fixed-point fractional library routines.

  52095.                                                              (line   91)

  52096. * __usaddudq3:                           Fixed-point fractional library routines.

  52097.                                                              (line   85)

  52098. * __usadduha3:                           Fixed-point fractional library routines.

  52099.                                                              (line   87)

  52100. * __usadduhq3:                           Fixed-point fractional library routines.

  52101.                                                              (line   81)

  52102. * __usadduqq3:                           Fixed-point fractional library routines.

  52103.                                                              (line   79)

  52104. * __usaddusa3:                           Fixed-point fractional library routines.

  52105.                                                              (line   89)

  52106. * __usaddusq3:                           Fixed-point fractional library routines.

  52107.                                                              (line   83)

  52108. * __usadduta3:                           Fixed-point fractional library routines.

  52109.                                                              (line   93)

  52110. * __usashluda3:                          Fixed-point fractional library routines.

  52111.                                                              (line  427)

  52112. * __usashludq3:                          Fixed-point fractional library routines.

  52113.                                                              (line  421)

  52114. * __usashluha3:                          Fixed-point fractional library routines.

  52115.                                                              (line  423)

  52116. * __usashluhq3:                          Fixed-point fractional library routines.

  52117.                                                              (line  417)

  52118. * __usashluqq3:                          Fixed-point fractional library routines.

  52119.                                                              (line  415)

  52120. * __usashlusa3:                          Fixed-point fractional library routines.

  52121.                                                              (line  425)

  52122. * __usashlusq3:                          Fixed-point fractional library routines.

  52123.                                                              (line  419)

  52124. * __usashluta3:                          Fixed-point fractional library routines.

  52125.                                                              (line  429)

  52126. * __usdivuda3:                           Fixed-point fractional library routines.

  52127.                                                              (line  286)

  52128. * __usdivudq3:                           Fixed-point fractional library routines.

  52129.                                                              (line  280)

  52130. * __usdivuha3:                           Fixed-point fractional library routines.

  52131.                                                              (line  282)

  52132. * __usdivuhq3:                           Fixed-point fractional library routines.

  52133.                                                              (line  276)

  52134. * __usdivuqq3:                           Fixed-point fractional library routines.

  52135.                                                              (line  274)

  52136. * __usdivusa3:                           Fixed-point fractional library routines.

  52137.                                                              (line  284)

  52138. * __usdivusq3:                           Fixed-point fractional library routines.

  52139.                                                              (line  278)

  52140. * __usdivuta3:                           Fixed-point fractional library routines.

  52141.                                                              (line  288)

  52142. * __usmuluda3:                           Fixed-point fractional library routines.

  52143.                                                              (line  218)

  52144. * __usmuludq3:                           Fixed-point fractional library routines.

  52145.                                                              (line  212)

  52146. * __usmuluha3:                           Fixed-point fractional library routines.

  52147.                                                              (line  214)

  52148. * __usmuluhq3:                           Fixed-point fractional library routines.

  52149.                                                              (line  208)

  52150. * __usmuluqq3:                           Fixed-point fractional library routines.

  52151.                                                              (line  206)

  52152. * __usmulusa3:                           Fixed-point fractional library routines.

  52153.                                                              (line  216)

  52154. * __usmulusq3:                           Fixed-point fractional library routines.

  52155.                                                              (line  210)

  52156. * __usmuluta3:                           Fixed-point fractional library routines.

  52157.                                                              (line  220)

  52158. * __usneguda2:                           Fixed-point fractional library routines.

  52159.                                                              (line  337)

  52160. * __usnegudq2:                           Fixed-point fractional library routines.

  52161.                                                              (line  332)

  52162. * __usneguha2:                           Fixed-point fractional library routines.

  52163.                                                              (line  334)

  52164. * __usneguhq2:                           Fixed-point fractional library routines.

  52165.                                                              (line  329)

  52166. * __usneguqq2:                           Fixed-point fractional library routines.

  52167.                                                              (line  327)

  52168. * __usnegusa2:                           Fixed-point fractional library routines.

  52169.                                                              (line  336)

  52170. * __usnegusq2:                           Fixed-point fractional library routines.

  52171.                                                              (line  330)

  52172. * __usneguta2:                           Fixed-point fractional library routines.

  52173.                                                              (line  339)

  52174. * __ussubuda3:                           Fixed-point fractional library routines.

  52175.                                                              (line  154)

  52176. * __ussubudq3:                           Fixed-point fractional library routines.

  52177.                                                              (line  148)

  52178. * __ussubuha3:                           Fixed-point fractional library routines.

  52179.                                                              (line  150)

  52180. * __ussubuhq3:                           Fixed-point fractional library routines.

  52181.                                                              (line  144)

  52182. * __ussubuqq3:                           Fixed-point fractional library routines.

  52183.                                                              (line  142)

  52184. * __ussubusa3:                           Fixed-point fractional library routines.

  52185.                                                              (line  152)

  52186. * __ussubusq3:                           Fixed-point fractional library routines.

  52187.                                                              (line  146)

  52188. * __ussubuta3:                           Fixed-point fractional library routines.

  52189.                                                              (line  156)

  52190. * abort:                                 Portability.        (line   20)

  52191. * abs:                                   Arithmetic.         (line  200)

  52192. * abs and attributes:                    Expressions.        (line   83)

  52193. * absence_set:                           Processor pipeline description.

  52194.                                                              (line  223)

  52195. * absM2 instruction pattern:             Standard Names.     (line  879)

  52196. * absolute value:                        Arithmetic.         (line  200)

  52197. * ABSU_EXPR:                             Unary and Binary Expressions.

  52198.                                                              (line    6)

  52199. * ABS_EXPR:                              Unary and Binary Expressions.

  52200.                                                              (line    6)

  52201. * access to operands:                    Accessors.          (line    6)

  52202. * access to special operands:            Special Accessors.  (line    6)

  52203. * accessors:                             Accessors.          (line    6)

  52204. * ACCUMULATE_OUTGOING_ARGS:              Stack Arguments.    (line   48)

  52205. * ACCUMULATE_OUTGOING_ARGS and stack frames: Function Entry. (line  140)

  52206. * ACCUM_TYPE_SIZE:                       Type Layout.        (line   87)

  52207. * acosM2 instruction pattern:            Standard Names.     (line  966)

  52208. * ADA_LONG_TYPE_SIZE:                    Type Layout.        (line   25)

  52209. * Adding a new GIMPLE statement code:    Adding a new GIMPLE statement code.

  52210.                                                              (line    6)

  52211. * ADDITIONAL_REGISTER_NAMES:             Instruction Output. (line   14)

  52212. * addM3 instruction pattern:             Standard Names.     (line  436)

  52213. * addMODEcc instruction pattern:         Standard Names.     (line 1589)

  52214. * addptrM3 instruction pattern:          Standard Names.     (line  469)

  52215. * address constraints:                   Simple Constraints. (line  162)

  52216. * addressing modes:                      Addressing Modes.   (line    6)

  52217. * address_operand:                       Machine-Independent Predicates.

  52218.                                                              (line   62)

  52219. * address_operand <1>:                   Simple Constraints. (line  166)

  52220. * addr_diff_vec:                         Side Effects.       (line  314)

  52221. * addr_diff_vec, length of:              Insn Lengths.       (line   26)

  52222. * ADDR_EXPR:                             Storage References. (line    6)

  52223. * addr_vec:                              Side Effects.       (line  309)

  52224. * addr_vec, length of:                   Insn Lengths.       (line   26)

  52225. * addvM4 instruction pattern:            Standard Names.     (line  452)

  52226. * ADJUST_FIELD_ALIGN:                    Storage Layout.     (line  212)

  52227. * ADJUST_INSN_LENGTH:                    Insn Lengths.       (line   41)

  52228. * ADJUST_REG_ALLOC_ORDER:                Allocation Order.   (line   22)

  52229. * aggregates as return values:           Aggregate Return.   (line    6)

  52230. * alias:                                 Alias analysis.     (line    6)

  52231. * allocate_stack instruction pattern:    Standard Names.     (line 1956)

  52232. * ALL_REGS:                              Register Classes.   (line   17)

  52233. * alternate entry points:                Insns.              (line  146)

  52234. * analyzer:                              Static Analyzer.    (line    6)

  52235. * analyzer, debugging:                   Debugging the Analyzer.

  52236.                                                              (line    6)

  52237. * analyzer, internals:                   Analyzer Internals. (line    6)

  52238. * anchored addresses:                    Anchored Addresses. (line    6)

  52239. * and:                                   Arithmetic.         (line  158)

  52240. * and and attributes:                    Expressions.        (line   50)

  52241. * and, canonicalization of:              Insn Canonicalizations.

  52242.                                                              (line   67)

  52243. * andM3 instruction pattern:             Standard Names.     (line  442)

  52244. * ANNOTATE_EXPR:                         Unary and Binary Expressions.

  52245.                                                              (line    6)

  52246. * annotations:                           Annotations.        (line    6)

  52247. * APPLY_RESULT_SIZE:                     Scalar Return.      (line  112)

  52248. * ARGS_GROW_DOWNWARD:                    Frame Layout.       (line   30)

  52249. * argument passing:                      Interface.          (line   36)

  52250. * arguments in registers:                Register Arguments. (line    6)

  52251. * arguments on stack:                    Stack Arguments.    (line    6)

  52252. * ARG_POINTER_CFA_OFFSET:                Frame Layout.       (line  207)

  52253. * ARG_POINTER_REGNUM:                    Frame Registers.    (line   40)

  52254. * ARG_POINTER_REGNUM and virtual registers: Regs and Memory. (line   65)

  52255. * arg_pointer_rtx:                       Frame Registers.    (line  104)

  52256. * arithmetic library:                    Soft float library routines.

  52257.                                                              (line    6)

  52258. * arithmetic shift:                      Arithmetic.         (line  173)

  52259. * arithmetic shift with signed saturation: Arithmetic.       (line  173)

  52260. * arithmetic shift with unsigned saturation: Arithmetic.     (line  173)

  52261. * arithmetic, in RTL:                    Arithmetic.         (line    6)

  52262. * ARITHMETIC_TYPE_P:                     Types for C++.      (line   59)

  52263. * array:                                 Types.              (line    6)

  52264. * ARRAY_RANGE_REF:                       Storage References. (line    6)

  52265. * ARRAY_REF:                             Storage References. (line    6)

  52266. * ARRAY_TYPE:                            Types.              (line    6)

  52267. * ashift:                                Arithmetic.         (line  173)

  52268. * ashift and attributes:                 Expressions.        (line   83)

  52269. * ashiftrt:                              Arithmetic.         (line  190)

  52270. * ashiftrt and attributes:               Expressions.        (line   83)

  52271. * ashlM3 instruction pattern:            Standard Names.     (line  828)

  52272. * ashrM3 instruction pattern:            Standard Names.     (line  840)

  52273. * asinM2 instruction pattern:            Standard Names.     (line  960)

  52274. * ASM_APP_OFF:                           File Framework.     (line   76)

  52275. * ASM_APP_ON:                            File Framework.     (line   69)

  52276. * ASM_COMMENT_START:                     File Framework.     (line   64)

  52277. * ASM_DECLARE_COLD_FUNCTION_NAME:        Label Output.       (line  136)

  52278. * ASM_DECLARE_COLD_FUNCTION_SIZE:        Label Output.       (line  151)

  52279. * ASM_DECLARE_FUNCTION_NAME:             Label Output.       (line  108)

  52280. * ASM_DECLARE_FUNCTION_SIZE:             Label Output.       (line  123)

  52281. * ASM_DECLARE_OBJECT_NAME:               Label Output.       (line  164)

  52282. * ASM_DECLARE_REGISTER_GLOBAL:           Label Output.       (line  192)

  52283. * ASM_FINAL_SPEC:                        Driver.             (line   81)

  52284. * ASM_FINISH_DECLARE_OBJECT:             Label Output.       (line  200)

  52285. * ASM_FORMAT_PRIVATE_NAME:               Label Output.       (line  426)

  52286. * asm_fprintf:                           Instruction Output. (line  150)

  52287. * ASM_FPRINTF_EXTENSIONS:                Instruction Output. (line  160)

  52288. * ASM_GENERATE_INTERNAL_LABEL:           Label Output.       (line  410)

  52289. * asm_input:                             Side Effects.       (line  296)

  52290. * asm_input and /v:                      Flags.              (line   65)

  52291. * ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX:     Exception Handling. (line   80)

  52292. * asm_noperands:                         Insns.              (line  327)

  52293. * ASM_NO_SKIP_IN_TEXT:                   Alignment Output.   (line   59)

  52294. * asm_operands and /v:                   Flags.              (line   65)

  52295. * asm_operands, RTL sharing:             Sharing.            (line   48)

  52296. * asm_operands, usage:                   Assembler.          (line    6)

  52297. * ASM_OUTPUT_ADDR_DIFF_ELT:              Dispatch Tables.    (line    8)

  52298. * ASM_OUTPUT_ADDR_VEC_ELT:               Dispatch Tables.    (line   25)

  52299. * ASM_OUTPUT_ALIGN:                      Alignment Output.   (line   66)

  52300. * ASM_OUTPUT_ALIGNED_BSS:                Uninitialized Data. (line   45)

  52301. * ASM_OUTPUT_ALIGNED_COMMON:             Uninitialized Data. (line   29)

  52302. * ASM_OUTPUT_ALIGNED_DECL_COMMON:        Uninitialized Data. (line   36)

  52303. * ASM_OUTPUT_ALIGNED_DECL_LOCAL:         Uninitialized Data. (line   89)

  52304. * ASM_OUTPUT_ALIGNED_LOCAL:              Uninitialized Data. (line   82)

  52305. * ASM_OUTPUT_ALIGN_WITH_NOP:             Alignment Output.   (line   71)

  52306. * ASM_OUTPUT_ASCII:                      Data Output.        (line   60)

  52307. * ASM_OUTPUT_CASE_END:                   Dispatch Tables.    (line   50)

  52308. * ASM_OUTPUT_CASE_LABEL:                 Dispatch Tables.    (line   37)

  52309. * ASM_OUTPUT_COMMON:                     Uninitialized Data. (line    9)

  52310. * ASM_OUTPUT_DEBUG_LABEL:                Label Output.       (line  398)

  52311. * ASM_OUTPUT_DEF:                        Label Output.       (line  447)

  52312. * ASM_OUTPUT_DEF_FROM_DECLS:             Label Output.       (line  454)

  52313. * ASM_OUTPUT_DWARF_DATAREL:              DWARF.              (line  110)

  52314. * ASM_OUTPUT_DWARF_DELTA:                DWARF.              (line   89)

  52315. * ASM_OUTPUT_DWARF_OFFSET:               DWARF.              (line   98)

  52316. * ASM_OUTPUT_DWARF_PCREL:                DWARF.              (line  105)

  52317. * ASM_OUTPUT_DWARF_TABLE_REF:            DWARF.              (line  115)

  52318. * ASM_OUTPUT_DWARF_VMS_DELTA:            DWARF.              (line   93)

  52319. * ASM_OUTPUT_EXTERNAL:                   Label Output.       (line  327)

  52320. * ASM_OUTPUT_FDESC:                      Data Output.        (line   69)

  52321. * ASM_OUTPUT_FUNCTION_LABEL:             Label Output.       (line   16)

  52322. * ASM_OUTPUT_INTERNAL_LABEL:             Label Output.       (line   27)

  52323. * ASM_OUTPUT_LABEL:                      Label Output.       (line    8)

  52324. * ASM_OUTPUT_LABELREF:                   Label Output.       (line  349)

  52325. * ASM_OUTPUT_LABEL_REF:                  Label Output.       (line  371)

  52326. * ASM_OUTPUT_LOCAL:                      Uninitialized Data. (line   69)

  52327. * ASM_OUTPUT_MAX_SKIP_ALIGN:             Alignment Output.   (line   75)

  52328. * ASM_OUTPUT_MEASURED_SIZE:              Label Output.       (line   51)

  52329. * ASM_OUTPUT_OPCODE:                     Instruction Output. (line   35)

  52330. * ASM_OUTPUT_POOL_EPILOGUE:              Data Output.        (line  118)

  52331. * ASM_OUTPUT_POOL_PROLOGUE:              Data Output.        (line   82)

  52332. * ASM_OUTPUT_REG_POP:                    Instruction Output. (line  206)

  52333. * ASM_OUTPUT_REG_PUSH:                   Instruction Output. (line  201)

  52334. * ASM_OUTPUT_SIZE_DIRECTIVE:             Label Output.       (line   45)

  52335. * ASM_OUTPUT_SKIP:                       Alignment Output.   (line   53)

  52336. * ASM_OUTPUT_SOURCE_FILENAME:            File Framework.     (line   83)

  52337. * ASM_OUTPUT_SPECIAL_POOL_ENTRY:         Data Output.        (line   93)

  52338. * ASM_OUTPUT_SYMBOL_REF:                 Label Output.       (line  364)

  52339. * ASM_OUTPUT_TYPE_DIRECTIVE:             Label Output.       (line   98)

  52340. * ASM_OUTPUT_WEAKREF:                    Label Output.       (line  259)

  52341. * ASM_OUTPUT_WEAK_ALIAS:                 Label Output.       (line  473)

  52342. * ASM_PREFERRED_EH_DATA_FORMAT:          Exception Handling. (line   66)

  52343. * ASM_SPEC:                              Driver.             (line   73)

  52344. * ASM_STABD_OP:                          DBX Options.        (line   34)

  52345. * ASM_STABN_OP:                          DBX Options.        (line   41)

  52346. * ASM_STABS_OP:                          DBX Options.        (line   28)

  52347. * ASM_WEAKEN_DECL:                       Label Output.       (line  251)

  52348. * ASM_WEAKEN_LABEL:                      Label Output.       (line  238)

  52349. * assembler format:                      File Framework.     (line    6)

  52350. * assembler instructions in RTL:         Assembler.          (line    6)

  52351. * ASSEMBLER_DIALECT:                     Instruction Output. (line  172)

  52352. * assemble_name:                         Label Output.       (line    8)

  52353. * assemble_name_raw:                     Label Output.       (line   27)

  52354. * assigning attribute values to insns:   Tagging Insns.      (line    6)

  52355. * ASSUME_EXTENDED_UNWIND_CONTEXT:        Frame Registers.    (line  163)

  52356. * asterisk in template:                  Output Statement.   (line   29)

  52357. * AS_NEEDS_DASH_FOR_PIPED_INPUT:         Driver.             (line   88)

  52358. * atan2M3 instruction pattern:           Standard Names.     (line 1061)

  52359. * atanM2 instruction pattern:            Standard Names.     (line  972)

  52360. * atomic:                                GTY Options.        (line  197)

  52361. * atomic_addMODE instruction pattern:    Standard Names.     (line 2380)

  52362. * atomic_add_fetchMODE instruction pattern: Standard Names.  (line 2409)

  52363. * atomic_andMODE instruction pattern:    Standard Names.     (line 2380)

  52364. * atomic_and_fetchMODE instruction pattern: Standard Names.  (line 2409)

  52365. * atomic_bit_test_and_complementMODE instruction pattern: Standard Names.

  52366.                                                              (line 2437)

  52367. * atomic_bit_test_and_resetMODE instruction pattern: Standard Names.

  52368.                                                              (line 2437)

  52369. * atomic_bit_test_and_setMODE instruction pattern: Standard Names.

  52370.                                                              (line 2437)

  52371. * atomic_compare_and_swapMODE instruction pattern: Standard Names.

  52372.                                                              (line 2316)

  52373. * atomic_exchangeMODE instruction pattern: Standard Names.   (line 2368)

  52374. * atomic_fetch_addMODE instruction pattern: Standard Names.  (line 2394)

  52375. * atomic_fetch_andMODE instruction pattern: Standard Names.  (line 2394)

  52376. * atomic_fetch_nandMODE instruction pattern: Standard Names. (line 2394)

  52377. * atomic_fetch_orMODE instruction pattern: Standard Names.   (line 2394)

  52378. * atomic_fetch_subMODE instruction pattern: Standard Names.  (line 2394)

  52379. * atomic_fetch_xorMODE instruction pattern: Standard Names.  (line 2394)

  52380. * atomic_loadMODE instruction pattern:   Standard Names.     (line 2347)

  52381. * atomic_nandMODE instruction pattern:   Standard Names.     (line 2380)

  52382. * atomic_nand_fetchMODE instruction pattern: Standard Names. (line 2409)

  52383. * atomic_orMODE instruction pattern:     Standard Names.     (line 2380)

  52384. * atomic_or_fetchMODE instruction pattern: Standard Names.   (line 2409)

  52385. * atomic_storeMODE instruction pattern:  Standard Names.     (line 2357)

  52386. * atomic_subMODE instruction pattern:    Standard Names.     (line 2380)

  52387. * atomic_sub_fetchMODE instruction pattern: Standard Names.  (line 2409)

  52388. * atomic_test_and_set instruction pattern: Standard Names.   (line 2426)

  52389. * atomic_xorMODE instruction pattern:    Standard Names.     (line 2380)

  52390. * atomic_xor_fetchMODE instruction pattern: Standard Names.  (line 2409)

  52391. * attr:                                  Expressions.        (line  163)

  52392. * attr <1>:                              Tagging Insns.      (line   54)

  52393. * attribute expressions:                 Expressions.        (line    6)

  52394. * attribute specifications:              Attr Example.       (line    6)

  52395. * attribute specifications example:      Attr Example.       (line    6)

  52396. * attributes:                            Attributes.         (line    6)

  52397. * attributes, defining:                  Defining Attributes.

  52398.                                                              (line    6)

  52399. * attributes, target-specific:           Target Attributes.  (line    6)

  52400. * ATTRIBUTE_ALIGNED_VALUE:               Storage Layout.     (line  194)

  52401. * attr_flag:                             Expressions.        (line  138)

  52402. * autoincrement addressing, availability: Portability.       (line   20)

  52403. * autoincrement/decrement addressing:    Simple Constraints. (line   30)

  52404. * automata_option:                       Processor pipeline description.

  52405.                                                              (line  304)

  52406. * automaton based pipeline description:  Processor pipeline description.

  52407.                                                              (line    6)

  52408. * automaton based pipeline description <1>: Processor pipeline description.

  52409.                                                              (line   49)

  52410. * automaton based scheduler:             Processor pipeline description.

  52411.                                                              (line    6)

  52412. * avgM3_ceil instruction pattern:        Standard Names.     (line  860)

  52413. * avgM3_floor instruction pattern:       Standard Names.     (line  848)

  52414. * AVOID_CCMODE_COPIES:                   Values in Registers.

  52415.                                                              (line  148)

  52416. * backslash:                             Output Template.    (line   46)

  52417. * barrier:                               Insns.              (line  176)

  52418. * barrier and /f:                        Flags.              (line  135)

  52419. * barrier and /v:                        Flags.              (line   33)

  52420. * BASE_REG_CLASS:                        Register Classes.   (line  111)

  52421. * basic block:                           Basic Blocks.       (line    6)

  52422. * Basic Statements:                      Basic Statements.   (line    6)

  52423. * basic-block.h:                         Control Flow.       (line    6)

  52424. * basic_block:                           Basic Blocks.       (line    6)

  52425. * BASIC_BLOCK:                           Basic Blocks.       (line   14)

  52426. * BB_HEAD, BB_END:                       Maintaining the CFG.

  52427.                                                              (line   76)

  52428. * bb_seq:                                GIMPLE sequences.   (line   72)

  52429. * BIGGEST_ALIGNMENT:                     Storage Layout.     (line  179)

  52430. * BIGGEST_FIELD_ALIGNMENT:               Storage Layout.     (line  205)

  52431. * BImode:                                Machine Modes.      (line   22)

  52432. * BIND_EXPR:                             Unary and Binary Expressions.

  52433.                                                              (line    6)

  52434. * BINFO_TYPE:                            Classes.            (line    6)

  52435. * bit-fields:                            Bit-Fields.         (line    6)

  52436. * BITFIELD_NBYTES_LIMITED:               Storage Layout.     (line  428)

  52437. * BITS_BIG_ENDIAN:                       Storage Layout.     (line   11)

  52438. * BITS_BIG_ENDIAN, effect on sign_extract: Bit-Fields.       (line    8)

  52439. * BITS_PER_UNIT:                         Machine Modes.      (line  440)

  52440. * BITS_PER_WORD:                         Storage Layout.     (line   50)

  52441. * bitwise complement:                    Arithmetic.         (line  154)

  52442. * bitwise exclusive-or:                  Arithmetic.         (line  168)

  52443. * bitwise inclusive-or:                  Arithmetic.         (line  163)

  52444. * bitwise logical-and:                   Arithmetic.         (line  158)

  52445. * BIT_AND_EXPR:                          Unary and Binary Expressions.

  52446.                                                              (line    6)

  52447. * BIT_IOR_EXPR:                          Unary and Binary Expressions.

  52448.                                                              (line    6)

  52449. * BIT_NOT_EXPR:                          Unary and Binary Expressions.

  52450.                                                              (line    6)

  52451. * BIT_XOR_EXPR:                          Unary and Binary Expressions.

  52452.                                                              (line    6)

  52453. * BLKmode:                               Machine Modes.      (line  185)

  52454. * BLKmode, and function return values:   Calls.              (line   23)

  52455. * blockage instruction pattern:          Standard Names.     (line 2156)

  52456. * Blocks:                                Blocks.             (line    6)

  52457. * BLOCK_FOR_INSN, gimple_bb:             Maintaining the CFG.

  52458.                                                              (line   28)

  52459. * BLOCK_REG_PADDING:                     Register Arguments. (line  238)

  52460. * BND32mode:                             Machine Modes.      (line  210)

  52461. * BND64mode:                             Machine Modes.      (line  210)

  52462. * bool:                                  Misc.               (line  958)

  52463. * BOOLEAN_TYPE:                          Types.              (line    6)

  52464. * BOOL_TYPE_SIZE:                        Type Layout.        (line   43)

  52465. * branch prediction:                     Profile information.

  52466.                                                              (line   24)

  52467. * BRANCH_COST:                           Costs.              (line  104)

  52468. * break_out_memory_refs:                 Addressing Modes.   (line  134)

  52469. * BREAK_STMT:                            Statements for C++. (line    6)

  52470. * BSS_SECTION_ASM_OP:                    Sections.           (line   67)

  52471. * bswap:                                 Arithmetic.         (line  246)

  52472. * bswapM2 instruction pattern:           Standard Names.     (line  868)

  52473. * btruncM2 instruction pattern:          Standard Names.     (line 1078)

  52474. * build0:                                Macros and Functions.

  52475.                                                              (line   16)

  52476. * build1:                                Macros and Functions.

  52477.                                                              (line   17)

  52478. * build2:                                Macros and Functions.

  52479.                                                              (line   18)

  52480. * build3:                                Macros and Functions.

  52481.                                                              (line   19)

  52482. * build4:                                Macros and Functions.

  52483.                                                              (line   20)

  52484. * build5:                                Macros and Functions.

  52485.                                                              (line   21)

  52486. * build6:                                Macros and Functions.

  52487.                                                              (line   22)

  52488. * builtin_longjmp instruction pattern:   Standard Names.     (line 2054)

  52489. * builtin_setjmp_receiver instruction pattern: Standard Names.

  52490.                                                              (line 2044)

  52491. * builtin_setjmp_setup instruction pattern: Standard Names.  (line 2033)

  52492. * BYTES_BIG_ENDIAN:                      Storage Layout.     (line   23)

  52493. * BYTES_BIG_ENDIAN, effect on subreg:    Regs and Memory.    (line  229)

  52494. * byte_mode:                             Machine Modes.      (line  458)

  52495. * C statements for assembler output:     Output Statement.   (line    6)

  52496. * cache:                                 GTY Options.        (line  127)

  52497. * call:                                  Flags.              (line  230)

  52498. * call <1>:                              Side Effects.       (line   92)

  52499. * call instruction pattern:              Standard Names.     (line 1709)

  52500. * call usage:                            Calls.              (line   10)

  52501. * call, in call_insn:                    Flags.              (line  129)

  52502. * call, in mem:                          Flags.              (line   70)

  52503. * call-clobbered register:               Register Basics.    (line   35)

  52504. * call-clobbered register <1>:           Register Basics.    (line   50)

  52505. * call-clobbered register <2>:           Register Basics.    (line   58)

  52506. * call-clobbered register <3>:           Register Basics.    (line   76)

  52507. * call-saved register:                   Register Basics.    (line   35)

  52508. * call-saved register <1>:               Register Basics.    (line   50)

  52509. * call-saved register <2>:               Register Basics.    (line   58)

  52510. * call-saved register <3>:               Register Basics.    (line   76)

  52511. * call-used register:                    Register Basics.    (line   35)

  52512. * call-used register <1>:                Register Basics.    (line   50)

  52513. * call-used register <2>:                Register Basics.    (line   58)

  52514. * call-used register <3>:                Register Basics.    (line   76)

  52515. * calling conventions:                   Stack and Calling.  (line    6)

  52516. * calling functions in RTL:              Calls.              (line    6)

  52517. * CALL_EXPR:                             Unary and Binary Expressions.

  52518.                                                              (line    6)

  52519. * call_insn:                             Insns.              (line   95)

  52520. * call_insn and /c:                      Flags.              (line  129)

  52521. * call_insn and /f:                      Flags.              (line  135)

  52522. * call_insn and /i:                      Flags.              (line  120)

  52523. * call_insn and /j:                      Flags.              (line  175)

  52524. * call_insn and /s:                      Flags.              (line   38)

  52525. * call_insn and /s <1>:                  Flags.              (line  162)

  52526. * call_insn and /u:                      Flags.              (line   28)

  52527. * call_insn and /u <1>:                  Flags.              (line  115)

  52528. * call_insn and /u or /i:                Flags.              (line  125)

  52529. * call_insn and /v:                      Flags.              (line   33)

  52530. * CALL_INSN_FUNCTION_USAGE:              Insns.              (line  101)

  52531. * call_pop instruction pattern:          Standard Names.     (line 1737)

  52532. * CALL_POPS_ARGS:                        Stack Arguments.    (line  138)

  52533. * CALL_REALLY_USED_REGISTERS:            Register Basics.    (line   49)

  52534. * CALL_USED_REGISTERS:                   Register Basics.    (line   34)

  52535. * call_used_regs:                        Register Basics.    (line  102)

  52536. * call_value instruction pattern:        Standard Names.     (line 1729)

  52537. * call_value_pop instruction pattern:    Standard Names.     (line 1737)

  52538. * canadian:                              Configure Terms.    (line    6)

  52539. * canonicalization of instructions:      Insn Canonicalizations.

  52540.                                                              (line    6)

  52541. * canonicalize_funcptr_for_compare instruction pattern: Standard Names.

  52542.                                                              (line 1888)

  52543. * can_create_pseudo_p:                   Standard Names.     (line   75)

  52544. * can_fallthru:                          Basic Blocks.       (line   67)

  52545. * caret:                                 Multi-Alternative.  (line   53)

  52546. * caret <1>:                             Guidelines for Diagnostics.

  52547.                                                              (line  159)

  52548. * casesi instruction pattern:            Standard Names.     (line 1830)

  52549. * CASE_VECTOR_MODE:                      Misc.               (line   26)

  52550. * CASE_VECTOR_PC_RELATIVE:               Misc.               (line   39)

  52551. * CASE_VECTOR_SHORTEN_MODE:              Misc.               (line   30)

  52552. * cbranchMODE4 instruction pattern:      Standard Names.     (line 1698)

  52553. * cc0:                                   Regs and Memory.    (line  329)

  52554. * cc0 <1>:                               CC0 Condition Codes.

  52555.                                                              (line    6)

  52556. * cc0, RTL sharing:                      Sharing.            (line   30)

  52557. * cc0_rtx:                               Regs and Memory.    (line  355)

  52558. * CC1PLUS_SPEC:                          Driver.             (line   63)

  52559. * CC1_SPEC:                              Driver.             (line   55)

  52560. * CCmode:                                Machine Modes.      (line  178)

  52561. * CCmode <1>:                            MODE_CC Condition Codes.

  52562.                                                              (line    6)

  52563. * cc_status:                             CC0 Condition Codes.

  52564.                                                              (line    6)

  52565. * CC_STATUS_MDEP:                        CC0 Condition Codes.

  52566.                                                              (line   16)

  52567. * CC_STATUS_MDEP_INIT:                   CC0 Condition Codes.

  52568.                                                              (line   22)

  52569. * CDImode:                               Machine Modes.      (line  204)

  52570. * ceilM2 instruction pattern:            Standard Names.     (line 1097)

  52571. * CEIL_DIV_EXPR:                         Unary and Binary Expressions.

  52572.                                                              (line    6)

  52573. * CEIL_MOD_EXPR:                         Unary and Binary Expressions.

  52574.                                                              (line    6)

  52575. * CFA_FRAME_BASE_OFFSET:                 Frame Layout.       (line  239)

  52576. * CFG verification:                      Maintaining the CFG.

  52577.                                                              (line  116)

  52578. * CFG, Control Flow Graph:               Control Flow.       (line    6)

  52579. * cfghooks.h:                            Maintaining the CFG.

  52580.                                                              (line    6)

  52581. * cgraph_finalize_function:              Parsing pass.       (line   51)

  52582. * chain_circular:                        GTY Options.        (line  160)

  52583. * chain_next:                            GTY Options.        (line  160)

  52584. * chain_prev:                            GTY Options.        (line  160)

  52585. * change_address:                        Standard Names.     (line   47)

  52586. * CHAR_TYPE_SIZE:                        Type Layout.        (line   38)

  52587. * check_raw_ptrsM instruction pattern:   Standard Names.     (line  330)

  52588. * check_stack instruction pattern:       Standard Names.     (line 1974)

  52589. * check_war_ptrsM instruction pattern:   Standard Names.     (line  349)

  52590. * CHImode:                               Machine Modes.      (line  204)

  52591. * class definitions, register:           Register Classes.   (line    6)

  52592. * class preference constraints:          Class Preferences.  (line    6)

  52593. * class, scope:                          Classes.            (line    6)

  52594. * classes of RTX codes:                  RTL Classes.        (line    6)

  52595. * CLASSTYPE_DECLARED_CLASS:              Classes.            (line    6)

  52596. * CLASSTYPE_HAS_MUTABLE:                 Classes.            (line   82)

  52597. * CLASSTYPE_NON_POD_P:                   Classes.            (line   87)

  52598. * CLASS_MAX_NREGS:                       Register Classes.   (line  531)

  52599. * CLASS_TYPE_P:                          Types for C++.      (line   63)

  52600. * Cleanups:                              Cleanups.           (line    6)

  52601. * CLEANUP_DECL:                          Statements for C++. (line    6)

  52602. * CLEANUP_EXPR:                          Statements for C++. (line    6)

  52603. * CLEANUP_POINT_EXPR:                    Unary and Binary Expressions.

  52604.                                                              (line    6)

  52605. * CLEANUP_STMT:                          Statements for C++. (line    6)

  52606. * clear_cache instruction pattern:       Standard Names.     (line 2543)

  52607. * CLEAR_INSN_CACHE:                      Trampolines.        (line  151)

  52608. * CLEAR_RATIO:                           Costs.              (line  225)

  52609. * clobber:                               Side Effects.       (line  106)

  52610. * clrsb:                                 Arithmetic.         (line  215)

  52611. * clrsbM2 instruction pattern:           Standard Names.     (line 1170)

  52612. * clz:                                   Arithmetic.         (line  222)

  52613. * clzM2 instruction pattern:             Standard Names.     (line 1186)

  52614. * CLZ_DEFINED_VALUE_AT_ZERO:             Misc.               (line  338)

  52615. * cmpmemM instruction pattern:           Standard Names.     (line 1389)

  52616. * cmpstrM instruction pattern:           Standard Names.     (line 1368)

  52617. * cmpstrnM instruction pattern:          Standard Names.     (line 1355)

  52618. * code generation RTL sequences:         Expander Definitions.

  52619.                                                              (line    6)

  52620. * code iterators in .md files:           Code Iterators.     (line    6)

  52621. * codes, RTL expression:                 RTL Objects.        (line   47)

  52622. * code_label:                            Insns.              (line  125)

  52623. * CODE_LABEL:                            Basic Blocks.       (line   50)

  52624. * code_label and /i:                     Flags.              (line   48)

  52625. * code_label and /v:                     Flags.              (line   33)

  52626. * CODE_LABEL_NUMBER:                     Insns.              (line  125)

  52627. * COImode:                               Machine Modes.      (line  204)

  52628. * COLLECT2_HOST_INITIALIZATION:          Host Misc.          (line   32)

  52629. * COLLECT_EXPORT_LIST:                   Misc.               (line  864)

  52630. * COLLECT_SHARED_FINI_FUNC:              Macros for Initialization.

  52631.                                                              (line   43)

  52632. * COLLECT_SHARED_INIT_FUNC:              Macros for Initialization.

  52633.                                                              (line   32)

  52634. * command-line options, guidelines for:  Guidelines for Options.

  52635.                                                              (line    6)

  52636. * commit_edge_insertions:                Maintaining the CFG.

  52637.                                                              (line  104)

  52638. * compare:                               Arithmetic.         (line   46)

  52639. * compare, canonicalization of:          Insn Canonicalizations.

  52640.                                                              (line   36)

  52641. * COMPARE_MAX_PIECES:                    Costs.              (line  220)

  52642. * comparison_operator:                   Machine-Independent Predicates.

  52643.                                                              (line  110)

  52644. * compiler passes and files:             Passes.             (line    6)

  52645. * complement, bitwise:                   Arithmetic.         (line  154)

  52646. * COMPLEX_CST:                           Constant expressions.

  52647.                                                              (line    6)

  52648. * COMPLEX_EXPR:                          Unary and Binary Expressions.

  52649.                                                              (line    6)

  52650. * complex_mode:                          Machine Modes.      (line  302)

  52651. * COMPLEX_TYPE:                          Types.              (line    6)

  52652. * COMPONENT_REF:                         Storage References. (line    6)

  52653. * Compound Expressions:                  Compound Expressions.

  52654.                                                              (line    6)

  52655. * Compound Lvalues:                      Compound Lvalues.   (line    6)

  52656. * COMPOUND_EXPR:                         Unary and Binary Expressions.

  52657.                                                              (line    6)

  52658. * COMPOUND_LITERAL_EXPR:                 Unary and Binary Expressions.

  52659.                                                              (line    6)

  52660. * COMPOUND_LITERAL_EXPR_DECL:            Unary and Binary Expressions.

  52661.                                                              (line  392)

  52662. * COMPOUND_LITERAL_EXPR_DECL_EXPR:       Unary and Binary Expressions.

  52663.                                                              (line  392)

  52664. * computed jump:                         Edges.              (line  127)

  52665. * computing the length of an insn:       Insn Lengths.       (line    6)

  52666. * concat:                                Regs and Memory.    (line  407)

  52667. * concatn:                               Regs and Memory.    (line  413)

  52668. * cond:                                  Comparisons.        (line   90)

  52669. * cond and attributes:                   Expressions.        (line   37)

  52670. * condition code register:               Regs and Memory.    (line  329)

  52671. * condition code status:                 Condition Code.     (line    6)

  52672. * condition codes:                       Comparisons.        (line   20)

  52673. * conditional execution:                 Conditional Execution.

  52674.                                                              (line    6)

  52675. * Conditional Expressions:               Conditional Expressions.

  52676.                                                              (line    6)

  52677. * conditions, in patterns:               Patterns.           (line   55)

  52678. * cond_addMODE instruction pattern:      Standard Names.     (line 1596)

  52679. * cond_andMODE instruction pattern:      Standard Names.     (line 1596)

  52680. * cond_divMODE instruction pattern:      Standard Names.     (line 1596)

  52681. * cond_exec:                             Side Effects.       (line  254)

  52682. * COND_EXPR:                             Unary and Binary Expressions.

  52683.                                                              (line    6)

  52684. * cond_fmaMODE instruction pattern:      Standard Names.     (line 1634)

  52685. * cond_fmsMODE instruction pattern:      Standard Names.     (line 1634)

  52686. * cond_fnmaMODE instruction pattern:     Standard Names.     (line 1634)

  52687. * cond_fnmsMODE instruction pattern:     Standard Names.     (line 1634)

  52688. * cond_iorMODE instruction pattern:      Standard Names.     (line 1596)

  52689. * cond_modMODE instruction pattern:      Standard Names.     (line 1596)

  52690. * cond_mulMODE instruction pattern:      Standard Names.     (line 1596)

  52691. * cond_smaxMODE instruction pattern:     Standard Names.     (line 1596)

  52692. * cond_sminMODE instruction pattern:     Standard Names.     (line 1596)

  52693. * cond_subMODE instruction pattern:      Standard Names.     (line 1596)

  52694. * cond_udivMODE instruction pattern:     Standard Names.     (line 1596)

  52695. * cond_umaxMODE instruction pattern:     Standard Names.     (line 1596)

  52696. * cond_uminMODE instruction pattern:     Standard Names.     (line 1596)

  52697. * cond_umodMODE instruction pattern:     Standard Names.     (line 1596)

  52698. * cond_xorMODE instruction pattern:      Standard Names.     (line 1596)

  52699. * configuration file:                    Filesystem.         (line    6)

  52700. * configuration file <1>:                Host Misc.          (line    6)

  52701. * configure terms:                       Configure Terms.    (line    6)

  52702. * CONJ_EXPR:                             Unary and Binary Expressions.

  52703.                                                              (line    6)

  52704. * const:                                 Constants.          (line  212)

  52705. * const0_rtx:                            Constants.          (line   21)

  52706. * CONST0_RTX:                            Constants.          (line  230)

  52707. * const1_rtx:                            Constants.          (line   21)

  52708. * CONST1_RTX:                            Constants.          (line  230)

  52709. * const2_rtx:                            Constants.          (line   21)

  52710. * CONST2_RTX:                            Constants.          (line  230)

  52711. * constant attributes:                   Constant Attributes.

  52712.                                                              (line    6)

  52713. * constant definitions:                  Constant Definitions.

  52714.                                                              (line    6)

  52715. * constants in constraints:              Simple Constraints. (line   68)

  52716. * CONSTANT_ADDRESS_P:                    Addressing Modes.   (line   28)

  52717. * CONSTANT_P:                            Addressing Modes.   (line   35)

  52718. * CONSTANT_POOL_ADDRESS_P:               Flags.              (line   19)

  52719. * CONSTANT_POOL_BEFORE_FUNCTION:         Data Output.        (line   74)

  52720. * constm1_rtx:                           Constants.          (line   21)

  52721. * constraint modifier characters:        Modifiers.          (line    6)

  52722. * constraint, matching:                  Simple Constraints. (line  140)

  52723. * constraints:                           Constraints.        (line    6)

  52724. * constraints, defining:                 Define Constraints. (line    6)

  52725. * constraints, machine specific:         Machine Constraints.

  52726.                                                              (line    6)

  52727. * constraints, testing:                  C Constraint Interface.

  52728.                                                              (line    6)

  52729. * constraint_num:                        C Constraint Interface.

  52730.                                                              (line   30)

  52731. * constraint_satisfied_p:                C Constraint Interface.

  52732.                                                              (line   42)

  52733. * CONSTRUCTOR:                           Unary and Binary Expressions.

  52734.                                                              (line    6)

  52735. * constructors, automatic calls:         Collect2.           (line   15)

  52736. * constructors, output of:               Initialization.     (line    6)

  52737. * CONST_DECL:                            Declarations.       (line    6)

  52738. * const_double:                          Constants.          (line   37)

  52739. * const_double, RTL sharing:             Sharing.            (line   32)

  52740. * CONST_DOUBLE_LOW:                      Constants.          (line   54)

  52741. * const_double_operand:                  Machine-Independent Predicates.

  52742.                                                              (line   20)

  52743. * const_fixed:                           Constants.          (line   93)

  52744. * const_int:                             Constants.          (line    8)

  52745. * const_int and attribute tests:         Expressions.        (line   47)

  52746. * const_int and attributes:              Expressions.        (line   10)

  52747. * const_int, RTL sharing:                Sharing.            (line   23)

  52748. * const_int_operand:                     Machine-Independent Predicates.

  52749.                                                              (line   15)

  52750. * const_poly_int:                        Constants.          (line  100)

  52751. * const_poly_int, RTL sharing:           Sharing.            (line   25)

  52752. * const_string:                          Constants.          (line  184)

  52753. * const_string and attributes:           Expressions.        (line   20)

  52754. * const_true_rtx:                        Constants.          (line   31)

  52755. * const_vector:                          Constants.          (line  107)

  52756. * const_vector, RTL sharing:             Sharing.            (line   35)

  52757. * CONST_WIDE_INT:                        Constants.          (line   67)

  52758. * CONST_WIDE_INT_ELT:                    Constants.          (line   89)

  52759. * CONST_WIDE_INT_NUNITS:                 Constants.          (line   84)

  52760. * CONST_WIDE_INT_VEC:                    Constants.          (line   80)

  52761. * container:                             Containers.         (line    6)

  52762. * CONTINUE_STMT:                         Statements for C++. (line    6)

  52763. * contributors:                          Contributors.       (line    6)

  52764. * controlling register usage:            Register Basics.    (line  116)

  52765. * controlling the compilation driver:    Driver.             (line    6)

  52766. * conventions, run-time:                 Interface.          (line    6)

  52767. * conversions:                           Conversions.        (line    6)

  52768. * CONVERT_EXPR:                          Unary and Binary Expressions.

  52769.                                                              (line    6)

  52770. * copysignM3 instruction pattern:        Standard Names.     (line 1142)

  52771. * copy_rtx:                              Addressing Modes.   (line  189)

  52772. * copy_rtx_if_shared:                    Sharing.            (line   67)

  52773. * cosM2 instruction pattern:             Standard Names.     (line  931)

  52774. * costs of instructions:                 Costs.              (line    6)

  52775. * CPLUSPLUS_CPP_SPEC:                    Driver.             (line   50)

  52776. * CPP_SPEC:                              Driver.             (line   43)

  52777. * CPSImode:                              Machine Modes.      (line  204)

  52778. * cpymemM instruction pattern:           Standard Names.     (line 1244)

  52779. * CP_INTEGRAL_TYPE:                      Types for C++.      (line   55)

  52780. * cp_namespace_decls:                    Namespaces.         (line   49)

  52781. * CP_TYPE_CONST_NON_VOLATILE_P:          Types for C++.      (line   33)

  52782. * CP_TYPE_CONST_P:                       Types for C++.      (line   24)

  52783. * cp_type_quals:                         Types for C++.      (line    6)

  52784. * cp_type_quals <1>:                     Types for C++.      (line   16)

  52785. * CP_TYPE_RESTRICT_P:                    Types for C++.      (line   30)

  52786. * CP_TYPE_VOLATILE_P:                    Types for C++.      (line   27)

  52787. * CQImode:                               Machine Modes.      (line  204)

  52788. * cross compilation and floating point:  Floating Point.     (line    6)

  52789. * CROSSING_JUMP_P:                       Flags.              (line   10)

  52790. * crtl->args.pops_args:                  Function Entry.     (line  111)

  52791. * crtl->args.pretend_args_size:          Function Entry.     (line  117)

  52792. * crtl->outgoing_args_size:              Stack Arguments.    (line   48)

  52793. * CRTSTUFF_T_CFLAGS:                     Target Fragment.    (line   15)

  52794. * CRTSTUFF_T_CFLAGS_S:                   Target Fragment.    (line   19)

  52795. * CRT_CALL_STATIC_FUNCTION:              Sections.           (line  125)

  52796. * CSImode:                               Machine Modes.      (line  204)

  52797. * cstoreMODE4 instruction pattern:       Standard Names.     (line 1659)

  52798. * CTImode:                               Machine Modes.      (line  204)

  52799. * ctrapMM4 instruction pattern:          Standard Names.     (line 2125)

  52800. * ctz:                                   Arithmetic.         (line  230)

  52801. * ctzM2 instruction pattern:             Standard Names.     (line 1201)

  52802. * CTZ_DEFINED_VALUE_AT_ZERO:             Misc.               (line  339)

  52803. * CUMULATIVE_ARGS:                       Register Arguments. (line  137)

  52804. * current_function_is_leaf:              Leaf Functions.     (line   50)

  52805. * current_function_uses_only_leaf_regs:  Leaf Functions.     (line   50)

  52806. * current_insn_predicate:                Conditional Execution.

  52807.                                                              (line   27)

  52808. * C_COMMON_OVERRIDE_OPTIONS:             Run-time Target.    (line  136)

  52809. * c_register_pragma:                     Misc.               (line  440)

  52810. * c_register_pragma_with_expansion:      Misc.               (line  442)

  52811. * DAmode:                                Machine Modes.      (line  154)

  52812. * data bypass:                           Processor pipeline description.

  52813.                                                              (line  105)

  52814. * data bypass <1>:                       Processor pipeline description.

  52815.                                                              (line  196)

  52816. * data dependence delays:                Processor pipeline description.

  52817.                                                              (line    6)

  52818. * Data Dependency Analysis:              Dependency analysis.

  52819.                                                              (line    6)

  52820. * data structures:                       Per-Function Data.  (line    6)

  52821. * DATA_ABI_ALIGNMENT:                    Storage Layout.     (line  263)

  52822. * DATA_ALIGNMENT:                        Storage Layout.     (line  250)

  52823. * DATA_SECTION_ASM_OP:                   Sections.           (line   52)

  52824. * DBR_OUTPUT_SEQEND:                     Instruction Output. (line  133)

  52825. * dbr_sequence_length:                   Instruction Output. (line  133)

  52826. * DBX_BLOCKS_FUNCTION_RELATIVE:          DBX Options.        (line  100)

  52827. * DBX_CONTIN_CHAR:                       DBX Options.        (line   63)

  52828. * DBX_CONTIN_LENGTH:                     DBX Options.        (line   53)

  52829. * DBX_DEBUGGING_INFO:                    DBX Options.        (line    8)

  52830. * DBX_FUNCTION_FIRST:                    DBX Options.        (line   94)

  52831. * DBX_LINES_FUNCTION_RELATIVE:           DBX Options.        (line  106)

  52832. * DBX_NO_XREFS:                          DBX Options.        (line   47)

  52833. * DBX_OUTPUT_MAIN_SOURCE_FILENAME:       File Names and DBX. (line    8)

  52834. * DBX_OUTPUT_MAIN_SOURCE_FILE_END:       File Names and DBX. (line   33)

  52835. * DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END: File Names and DBX.

  52836.                                                              (line   41)

  52837. * DBX_OUTPUT_SOURCE_LINE:                DBX Hooks.          (line    8)

  52838. * DBX_REGISTER_NUMBER:                   All Debuggers.      (line    8)

  52839. * DBX_REGPARM_STABS_CODE:                DBX Options.        (line   84)

  52840. * DBX_REGPARM_STABS_LETTER:              DBX Options.        (line   89)

  52841. * DBX_STATIC_CONST_VAR_CODE:             DBX Options.        (line   79)

  52842. * DBX_STATIC_STAB_DATA_SECTION:          DBX Options.        (line   70)

  52843. * DBX_TYPE_DECL_STABS_CODE:              DBX Options.        (line   75)

  52844. * DBX_USE_BINCL:                         DBX Options.        (line  112)

  52845. * DCmode:                                Machine Modes.      (line  199)

  52846. * DDmode:                                Machine Modes.      (line   93)

  52847. * De Morgan's law:                       Insn Canonicalizations.

  52848.                                                              (line   67)

  52849. * dead_or_set_p:                         define_peephole.    (line   65)

  52850. * DEBUGGER_ARG_OFFSET:                   All Debuggers.      (line   35)

  52851. * DEBUGGER_AUTO_OFFSET:                  All Debuggers.      (line   27)

  52852. * debug_expr:                            Debug Information.  (line   22)

  52853. * DEBUG_EXPR_DECL:                       Declarations.       (line    6)

  52854. * debug_implicit_ptr:                    Debug Information.  (line   27)

  52855. * debug_insn:                            Insns.              (line  247)

  52856. * debug_marker:                          Debug Information.  (line   37)

  52857. * debug_parameter_ref:                   Debug Information.  (line   34)

  52858. * DEBUG_SYMS_TEXT:                       DBX Options.        (line   24)

  52859. * decimal float library:                 Decimal float library routines.

  52860.                                                              (line    6)

  52861. * declaration:                           Declarations.       (line    6)

  52862. * declarations, RTL:                     RTL Declarations.   (line    6)

  52863. * DECLARE_LIBRARY_RENAMES:               Library Calls.      (line    8)

  52864. * DECL_ALIGN:                            Declarations.       (line    6)

  52865. * DECL_ANTICIPATED:                      Functions for C++.  (line   42)

  52866. * DECL_ARGUMENTS:                        Function Basics.    (line   36)

  52867. * DECL_ARRAY_DELETE_OPERATOR_P:          Functions for C++.  (line  158)

  52868. * DECL_ARTIFICIAL:                       Working with declarations.

  52869.                                                              (line   24)

  52870. * DECL_ARTIFICIAL <1>:                   Function Basics.    (line    6)

  52871. * DECL_ARTIFICIAL <2>:                   Function Properties.

  52872.                                                              (line   47)

  52873. * DECL_ASSEMBLER_NAME:                   Function Basics.    (line    6)

  52874. * DECL_ASSEMBLER_NAME <1>:               Function Basics.    (line   19)

  52875. * DECL_ATTRIBUTES:                       Attributes.         (line   21)

  52876. * DECL_BASE_CONSTRUCTOR_P:               Functions for C++.  (line   88)

  52877. * DECL_COMPLETE_CONSTRUCTOR_P:           Functions for C++.  (line   84)

  52878. * DECL_COMPLETE_DESTRUCTOR_P:            Functions for C++.  (line   98)

  52879. * DECL_CONSTRUCTOR_P:                    Functions for C++.  (line   77)

  52880. * DECL_CONST_MEMFUNC_P:                  Functions for C++.  (line   71)

  52881. * DECL_CONTEXT:                          Namespaces.         (line   31)

  52882. * DECL_CONV_FN_P:                        Functions for C++.  (line  105)

  52883. * DECL_COPY_CONSTRUCTOR_P:               Functions for C++.  (line   92)

  52884. * DECL_DESTRUCTOR_P:                     Functions for C++.  (line   95)

  52885. * DECL_EXTERNAL:                         Declarations.       (line    6)

  52886. * DECL_EXTERNAL <1>:                     Function Properties.

  52887.                                                              (line   25)

  52888. * DECL_EXTERN_C_FUNCTION_P:              Functions for C++.  (line   46)

  52889. * DECL_FUNCTION_MEMBER_P:                Functions for C++.  (line   61)

  52890. * DECL_FUNCTION_SPECIFIC_OPTIMIZATION:   Function Basics.    (line    6)

  52891. * DECL_FUNCTION_SPECIFIC_OPTIMIZATION <1>: Function Properties.

  52892.                                                              (line   61)

  52893. * DECL_FUNCTION_SPECIFIC_TARGET:         Function Basics.    (line    6)

  52894. * DECL_FUNCTION_SPECIFIC_TARGET <1>:     Function Properties.

  52895.                                                              (line   55)

  52896. * DECL_GLOBAL_CTOR_P:                    Functions for C++.  (line  108)

  52897. * DECL_GLOBAL_DTOR_P:                    Functions for C++.  (line  112)

  52898. * DECL_INITIAL:                          Declarations.       (line    6)

  52899. * DECL_INITIAL <1>:                      Function Basics.    (line   51)

  52900. * DECL_LINKONCE_P:                       Functions for C++.  (line   50)

  52901. * DECL_LOCAL_FUNCTION_P:                 Functions for C++.  (line   38)

  52902. * DECL_MAIN_P:                           Functions for C++.  (line   34)

  52903. * DECL_NAME:                             Working with declarations.

  52904.                                                              (line    7)

  52905. * DECL_NAME <1>:                         Function Basics.    (line    6)

  52906. * DECL_NAME <2>:                         Function Basics.    (line    9)

  52907. * DECL_NAME <3>:                         Namespaces.         (line   20)

  52908. * DECL_NAMESPACE_ALIAS:                  Namespaces.         (line   35)

  52909. * DECL_NAMESPACE_STD_P:                  Namespaces.         (line   45)

  52910. * DECL_NONCONVERTING_P:                  Functions for C++.  (line   80)

  52911. * DECL_NONSTATIC_MEMBER_FUNCTION_P:      Functions for C++.  (line   68)

  52912. * DECL_NON_THUNK_FUNCTION_P:             Functions for C++.  (line  138)

  52913. * DECL_OVERLOADED_OPERATOR_P:            Functions for C++.  (line  102)

  52914. * DECL_PURE_P:                           Function Properties.

  52915.                                                              (line   40)

  52916. * DECL_RESULT:                           Function Basics.    (line   41)

  52917. * DECL_SAVED_TREE:                       Function Basics.    (line   44)

  52918. * DECL_SIZE:                             Declarations.       (line    6)

  52919. * DECL_STATIC_FUNCTION_P:                Functions for C++.  (line   65)

  52920. * DECL_STMT:                             Statements for C++. (line    6)

  52921. * DECL_STMT_DECL:                        Statements for C++. (line    6)

  52922. * DECL_THUNK_P:                          Functions for C++.  (line  116)

  52923. * DECL_VIRTUAL_P:                        Function Properties.

  52924.                                                              (line   44)

  52925. * DECL_VOLATILE_MEMFUNC_P:               Functions for C++.  (line   74)

  52926. * default:                               GTY Options.        (line   90)

  52927. * default_file_start:                    File Framework.     (line    8)

  52928. * DEFAULT_GDB_EXTENSIONS:                DBX Options.        (line   17)

  52929. * DEFAULT_INCOMING_FRAME_SP_OFFSET:      Frame Layout.       (line  199)

  52930. * DEFAULT_PCC_STRUCT_RETURN:             Aggregate Return.   (line   34)

  52931. * DEFAULT_SIGNED_CHAR:                   Type Layout.        (line  117)

  52932. * define_address_constraint:             Define Constraints. (line  113)

  52933. * define_asm_attributes:                 Tagging Insns.      (line   73)

  52934. * define_attr:                           Defining Attributes.

  52935.                                                              (line    6)

  52936. * define_automaton:                      Processor pipeline description.

  52937.                                                              (line   53)

  52938. * define_bypass:                         Processor pipeline description.

  52939.                                                              (line  196)

  52940. * define_code_attr:                      Code Iterators.     (line    6)

  52941. * define_code_iterator:                  Code Iterators.     (line    6)

  52942. * define_cond_exec:                      Conditional Execution.

  52943.                                                              (line   13)

  52944. * define_constants:                      Constant Definitions.

  52945.                                                              (line    6)

  52946. * define_constraint:                     Define Constraints. (line   45)

  52947. * define_cpu_unit:                       Processor pipeline description.

  52948.                                                              (line   68)

  52949. * define_c_enum:                         Constant Definitions.

  52950.                                                              (line   49)

  52951. * define_delay:                          Delay Slots.        (line   25)

  52952. * define_enum:                           Constant Definitions.

  52953.                                                              (line  118)

  52954. * define_enum_attr:                      Defining Attributes.

  52955.                                                              (line   83)

  52956. * define_enum_attr <1>:                  Constant Definitions.

  52957.                                                              (line  136)

  52958. * define_expand:                         Expander Definitions.

  52959.                                                              (line   11)

  52960. * define_insn:                           Patterns.           (line    6)

  52961. * define_insn example:                   Example.            (line    6)

  52962. * define_insn_and_rewrite:               Insn Splitting.     (line  236)

  52963. * define_insn_and_split:                 Insn Splitting.     (line  190)

  52964. * define_insn_reservation:               Processor pipeline description.

  52965.                                                              (line  105)

  52966. * define_int_attr:                       Int Iterators.      (line    6)

  52967. * define_int_iterator:                   Int Iterators.      (line    6)

  52968. * define_memory_constraint:              Define Constraints. (line   80)

  52969. * define_mode_attr:                      Substitutions.      (line    6)

  52970. * define_mode_iterator:                  Defining Mode Iterators.

  52971.                                                              (line    6)

  52972. * define_peephole:                       define_peephole.    (line    6)

  52973. * define_peephole2:                      define_peephole2.   (line    6)

  52974. * define_predicate:                      Defining Predicates.

  52975.                                                              (line    6)

  52976. * define_query_cpu_unit:                 Processor pipeline description.

  52977.                                                              (line   90)

  52978. * define_register_constraint:            Define Constraints. (line   26)

  52979. * define_reservation:                    Processor pipeline description.

  52980.                                                              (line  185)

  52981. * define_special_memory_constraint:      Define Constraints. (line   99)

  52982. * define_special_predicate:              Defining Predicates.

  52983.                                                              (line    6)

  52984. * define_split:                          Insn Splitting.     (line   32)

  52985. * define_subst:                          Define Subst.       (line    6)

  52986. * define_subst <1>:                      Define Subst Example.

  52987.                                                              (line    6)

  52988. * define_subst <2>:                      Define Subst Pattern Matching.

  52989.                                                              (line    6)

  52990. * define_subst <3>:                      Define Subst Output Template.

  52991.                                                              (line    6)

  52992. * define_subst <4>:                      Define Subst.       (line   14)

  52993. * define_subst <5>:                      Subst Iterators.    (line    6)

  52994. * define_subst_attr:                     Subst Iterators.    (line    6)

  52995. * define_subst_attr <1>:                 Subst Iterators.    (line   26)

  52996. * defining attributes and their values:  Defining Attributes.

  52997.                                                              (line    6)

  52998. * defining constraints:                  Define Constraints. (line    6)

  52999. * defining jump instruction patterns:    Jump Patterns.      (line    6)

  53000. * defining looping instruction patterns: Looping Patterns.   (line    6)

  53001. * defining peephole optimizers:          Peephole Definitions.

  53002.                                                              (line    6)

  53003. * defining predicates:                   Defining Predicates.

  53004.                                                              (line    6)

  53005. * defining RTL sequences for code generation: Expander Definitions.

  53006.                                                              (line    6)

  53007. * delay slots, defining:                 Delay Slots.        (line    6)

  53008. * deletable:                             GTY Options.        (line  134)

  53009. * DELETE_IF_ORDINARY:                    Filesystem.         (line   79)

  53010. * Dependent Patterns:                    Dependent Patterns. (line    6)

  53011. * desc:                                  GTY Options.        (line   90)

  53012. * descriptors for nested functions:      Trampolines.        (line    6)

  53013. * destructors, output of:                Initialization.     (line    6)

  53014. * deterministic finite state automaton:  Processor pipeline description.

  53015.                                                              (line    6)

  53016. * deterministic finite state automaton <1>: Processor pipeline description.

  53017.                                                              (line  304)

  53018. * DFmode:                                Machine Modes.      (line   76)

  53019. * diagnostics guidelines, fix-it hints:  Guidelines for Diagnostics.

  53020.                                                              (line  320)

  53021. * diagnostics, actionable:               Guidelines for Diagnostics.

  53022.                                                              (line   15)

  53023. * diagnostics, false positive:           Guidelines for Diagnostics.

  53024.                                                              (line   39)

  53025. * diagnostics, guidelines for:           Guidelines for Diagnostics.

  53026.                                                              (line    5)

  53027. * diagnostics, locations:                Guidelines for Diagnostics.

  53028.                                                              (line  159)

  53029. * diagnostics, true positive:            Guidelines for Diagnostics.

  53030.                                                              (line   39)

  53031. * digits in constraint:                  Simple Constraints. (line  128)

  53032. * DImode:                                Machine Modes.      (line   45)

  53033. * directory options .md:                 Including Patterns. (line   47)

  53034. * DIR_SEPARATOR:                         Filesystem.         (line   18)

  53035. * DIR_SEPARATOR_2:                       Filesystem.         (line   19)

  53036. * disabling certain registers:           Register Basics.    (line  116)

  53037. * dispatch table:                        Dispatch Tables.    (line    8)

  53038. * div:                                   Arithmetic.         (line  116)

  53039. * div and attributes:                    Expressions.        (line   83)

  53040. * division:                              Arithmetic.         (line  116)

  53041. * division <1>:                          Arithmetic.         (line  130)

  53042. * division <2>:                          Arithmetic.         (line  136)

  53043. * divM3 instruction pattern:             Standard Names.     (line  442)

  53044. * divmodM4 instruction pattern:          Standard Names.     (line  808)

  53045. * dollar sign:                           Multi-Alternative.  (line   57)

  53046. * DOLLARS_IN_IDENTIFIERS:                Misc.               (line  485)

  53047. * doloop_begin instruction pattern:      Standard Names.     (line 1879)

  53048. * doloop_end instruction pattern:        Standard Names.     (line 1867)

  53049. * DONE:                                  Expander Definitions.

  53050.                                                              (line   77)

  53051. * DONE <1>:                              Insn Splitting.     (line   61)

  53052. * DONE <2>:                              define_peephole2.   (line   76)

  53053. * DONT_USE_BUILTIN_SETJMP:               Exception Region Output.

  53054.                                                              (line   78)

  53055. * DOUBLE_TYPE_SIZE:                      Type Layout.        (line   52)

  53056. * DO_BODY:                               Statements for C++. (line    6)

  53057. * DO_COND:                               Statements for C++. (line    6)

  53058. * DO_STMT:                               Statements for C++. (line    6)

  53059. * DQmode:                                Machine Modes.      (line  118)

  53060. * driver:                                Driver.             (line    6)

  53061. * DRIVER_SELF_SPECS:                     Driver.             (line    8)

  53062. * dump examples:                         Dump examples.      (line    6)

  53063. * dump setup:                            Dump setup.         (line    6)

  53064. * dump types:                            Dump types.         (line    6)

  53065. * dump verbosity:                        Dump output verbosity.

  53066.                                                              (line    6)

  53067. * DUMPFILE_FORMAT:                       Filesystem.         (line   67)

  53068. * dump_basic_block:                      Dump types.         (line   29)

  53069. * dump_generic_expr:                     Dump types.         (line   31)

  53070. * dump_gimple_stmt:                      Dump types.         (line   33)

  53071. * dump_printf:                           Dump types.         (line    6)

  53072. * DWARF2_ASM_LINE_DEBUG_INFO:            DWARF.              (line   45)

  53073. * DWARF2_ASM_VIEW_DEBUG_INFO:            DWARF.              (line   51)

  53074. * DWARF2_DEBUGGING_INFO:                 DWARF.              (line    8)

  53075. * DWARF2_FRAME_INFO:                     DWARF.              (line   25)

  53076. * DWARF2_FRAME_REG_OUT:                  Frame Registers.    (line  149)

  53077. * DWARF2_UNWIND_INFO:                    Exception Region Output.

  53078.                                                              (line   39)

  53079. * DWARF_ALT_FRAME_RETURN_COLUMN:         Frame Layout.       (line  146)

  53080. * DWARF_CIE_DATA_ALIGNMENT:              Exception Region Output.

  53081.                                                              (line   90)

  53082. * DWARF_FRAME_REGISTERS:                 Frame Registers.    (line  109)

  53083. * DWARF_FRAME_REGNUM:                    Frame Registers.    (line  141)

  53084. * DWARF_LAZY_REGISTER_VALUE:             Frame Registers.    (line  170)

  53085. * DWARF_REG_TO_UNWIND_COLUMN:            Frame Registers.    (line  134)

  53086. * DWARF_ZERO_REG:                        Frame Layout.       (line  157)

  53087. * DYNAMIC_CHAIN_ADDRESS:                 Frame Layout.       (line   84)

  53088. * E in constraint:                       Simple Constraints. (line   87)

  53089. * earlyclobber operand:                  Modifiers.          (line   25)

  53090. * edge:                                  Edges.              (line    6)

  53091. * edge in the flow graph:                Edges.              (line    6)

  53092. * edge iterators:                        Edges.              (line   15)

  53093. * edge splitting:                        Maintaining the CFG.

  53094.                                                              (line  104)

  53095. * EDGE_ABNORMAL:                         Edges.              (line  127)

  53096. * EDGE_ABNORMAL, EDGE_ABNORMAL_CALL:     Edges.              (line  171)

  53097. * EDGE_ABNORMAL, EDGE_EH:                Edges.              (line   95)

  53098. * EDGE_ABNORMAL, EDGE_SIBCALL:           Edges.              (line  121)

  53099. * EDGE_FALLTHRU, force_nonfallthru:      Edges.              (line   85)

  53100. * EDOM, implicit usage:                  Library Calls.      (line   59)

  53101. * EH_FRAME_SECTION_NAME:                 Exception Region Output.

  53102.                                                              (line    9)

  53103. * EH_FRAME_THROUGH_COLLECT2:             Exception Region Output.

  53104.                                                              (line   19)

  53105. * eh_return instruction pattern:         Standard Names.     (line 2060)

  53106. * EH_RETURN_DATA_REGNO:                  Exception Handling. (line    6)

  53107. * EH_RETURN_HANDLER_RTX:                 Exception Handling. (line   38)

  53108. * EH_RETURN_STACKADJ_RTX:                Exception Handling. (line   21)

  53109. * EH_TABLES_CAN_BE_READ_ONLY:            Exception Region Output.

  53110.                                                              (line   29)

  53111. * EH_USES:                               Function Entry.     (line  162)

  53112. * ei_edge:                               Edges.              (line   43)

  53113. * ei_end_p:                              Edges.              (line   27)

  53114. * ei_last:                               Edges.              (line   23)

  53115. * ei_next:                               Edges.              (line   35)

  53116. * ei_one_before_end_p:                   Edges.              (line   31)

  53117. * ei_prev:                               Edges.              (line   39)

  53118. * ei_safe_safe:                          Edges.              (line   47)

  53119. * ei_start:                              Edges.              (line   19)

  53120. * ELIMINABLE_REGS:                       Elimination.        (line   34)

  53121. * ELSE_CLAUSE:                           Statements for C++. (line    6)

  53122. * Embedded C:                            Fixed-point fractional library routines.

  53123.                                                              (line    6)

  53124. * Empty Statements:                      Empty Statements.   (line    6)

  53125. * EMPTY_CLASS_EXPR:                      Statements for C++. (line    6)

  53126. * EMPTY_FIELD_BOUNDARY:                  Storage Layout.     (line  341)

  53127. * Emulated TLS:                          Emulated TLS.       (line    6)

  53128. * enabled:                               Disable Insn Alternatives.

  53129.                                                              (line    6)

  53130. * ENDFILE_SPEC:                          Driver.             (line  155)

  53131. * endianness:                            Portability.        (line   20)

  53132. * ENTRY_BLOCK_PTR, EXIT_BLOCK_PTR:       Basic Blocks.       (line   10)

  53133. * entry_value:                           Debug Information.  (line   30)

  53134. * enum reg_class:                        Register Classes.   (line   70)

  53135. * ENUMERAL_TYPE:                         Types.              (line    6)

  53136. * enumerations:                          Constant Definitions.

  53137.                                                              (line   49)

  53138. * epilogue:                              Function Entry.     (line    6)

  53139. * epilogue instruction pattern:          Standard Names.     (line 2098)

  53140. * EPILOGUE_USES:                         Function Entry.     (line  156)

  53141. * eq:                                    Comparisons.        (line   52)

  53142. * eq and attributes:                     Expressions.        (line   83)

  53143. * equal:                                 Comparisons.        (line   52)

  53144. * eq_attr:                               Expressions.        (line  104)

  53145. * EQ_EXPR:                               Unary and Binary Expressions.

  53146.                                                              (line    6)

  53147. * errno, implicit usage:                 Library Calls.      (line   71)

  53148. * EXACT_DIV_EXPR:                        Unary and Binary Expressions.

  53149.                                                              (line    6)

  53150. * examining SSA_NAMEs:                   SSA.                (line  182)

  53151. * exception handling:                    Edges.              (line   95)

  53152. * exception handling <1>:                Exception Handling. (line    6)

  53153. * exception_receiver instruction pattern: Standard Names.    (line 2025)

  53154. * exclamation point:                     Multi-Alternative.  (line   48)

  53155. * exclusion_set:                         Processor pipeline description.

  53156.                                                              (line  223)

  53157. * exclusive-or, bitwise:                 Arithmetic.         (line  168)

  53158. * EXIT_EXPR:                             Unary and Binary Expressions.

  53159.                                                              (line    6)

  53160. * EXIT_IGNORE_STACK:                     Function Entry.     (line  144)

  53161. * exp10M2 instruction pattern:           Standard Names.     (line  995)

  53162. * exp2M2 instruction pattern:            Standard Names.     (line 1002)

  53163. * expander definitions:                  Expander Definitions.

  53164.                                                              (line    6)

  53165. * expm1M2 instruction pattern:           Standard Names.     (line  985)

  53166. * expM2 instruction pattern:             Standard Names.     (line  978)

  53167. * expression:                            Expression trees.   (line    6)

  53168. * expression codes:                      RTL Objects.        (line   47)

  53169. * EXPR_FILENAME:                         Working with declarations.

  53170.                                                              (line   14)

  53171. * EXPR_LINENO:                           Working with declarations.

  53172.                                                              (line   20)

  53173. * expr_list:                             Insns.              (line  568)

  53174. * EXPR_STMT:                             Statements for C++. (line    6)

  53175. * EXPR_STMT_EXPR:                        Statements for C++. (line    6)

  53176. * extendMN2 instruction pattern:         Standard Names.     (line 1447)

  53177. * extensible constraints:                Simple Constraints. (line  171)

  53178. * extract_last_M instruction pattern:    Standard Names.     (line  543)

  53179. * EXTRA_SPECS:                           Driver.             (line  182)

  53180. * extv instruction pattern:              Standard Names.     (line 1538)

  53181. * extvM instruction pattern:             Standard Names.     (line 1483)

  53182. * extvmisalignM instruction pattern:     Standard Names.     (line 1493)

  53183. * extzv instruction pattern:             Standard Names.     (line 1556)

  53184. * extzvM instruction pattern:            Standard Names.     (line 1507)

  53185. * extzvmisalignM instruction pattern:    Standard Names.     (line 1510)

  53186. * F in constraint:                       Simple Constraints. (line   92)

  53187. * FAIL:                                  Expander Definitions.

  53188.                                                              (line   83)

  53189. * FAIL <1>:                              Insn Splitting.     (line   68)

  53190. * FAIL <2>:                              define_peephole2.   (line   83)

  53191. * fall-thru:                             Edges.              (line   68)

  53192. * false positive:                        Guidelines for Diagnostics.

  53193.                                                              (line   39)

  53194. * FATAL_EXIT_CODE:                       Host Misc.          (line    6)

  53195. * FDL, GNU Free Documentation License:   GNU Free Documentation License.

  53196.                                                              (line    6)

  53197. * features, optional, in system conventions: Run-time Target.

  53198.                                                              (line   59)

  53199. * ffs:                                   Arithmetic.         (line  210)

  53200. * ffsM2 instruction pattern:             Standard Names.     (line 1157)

  53201. * FIELD_DECL:                            Declarations.       (line    6)

  53202. * files and passes of the compiler:      Passes.             (line    6)

  53203. * files, generated:                      Files.              (line    6)

  53204. * file_end_indicate_exec_stack:          File Framework.     (line   39)

  53205. * final_absence_set:                     Processor pipeline description.

  53206.                                                              (line  223)

  53207. * FINAL_PRESCAN_INSN:                    Instruction Output. (line   60)

  53208. * final_presence_set:                    Processor pipeline description.

  53209.                                                              (line  223)

  53210. * final_sequence:                        Instruction Output. (line  144)

  53211. * FIND_BASE_TERM:                        Addressing Modes.   (line  117)

  53212. * finite state automaton minimization:   Processor pipeline description.

  53213.                                                              (line  304)

  53214. * FINI_ARRAY_SECTION_ASM_OP:             Sections.           (line  113)

  53215. * FINI_SECTION_ASM_OP:                   Sections.           (line   98)

  53216. * FIRST_PARM_OFFSET:                     Frame Layout.       (line   59)

  53217. * FIRST_PARM_OFFSET and virtual registers: Regs and Memory.  (line   65)

  53218. * FIRST_PSEUDO_REGISTER:                 Register Basics.    (line    8)

  53219. * FIRST_STACK_REG:                       Stack Registers.    (line   26)

  53220. * FIRST_VIRTUAL_REGISTER:                Regs and Memory.    (line   51)

  53221. * fix:                                   Conversions.        (line   66)

  53222. * fix-it hints:                          Guidelines for Diagnostics.

  53223.                                                              (line  320)

  53224. * fixed register:                        Register Basics.    (line   15)

  53225. * fixed-point fractional library:        Fixed-point fractional library routines.

  53226.                                                              (line    6)

  53227. * FIXED_CONVERT_EXPR:                    Unary and Binary Expressions.

  53228.                                                              (line    6)

  53229. * FIXED_CST:                             Constant expressions.

  53230.                                                              (line    6)

  53231. * FIXED_POINT_TYPE:                      Types.              (line    6)

  53232. * FIXED_REGISTERS:                       Register Basics.    (line   14)

  53233. * fixed_regs:                            Register Basics.    (line  102)

  53234. * fixed_size_mode:                       Machine Modes.      (line  305)

  53235. * fixMN2 instruction pattern:            Standard Names.     (line 1414)

  53236. * fixunsMN2 instruction pattern:         Standard Names.     (line 1423)

  53237. * fixuns_truncMN2 instruction pattern:   Standard Names.     (line 1438)

  53238. * fix_truncMN2 instruction pattern:      Standard Names.     (line 1434)

  53239. * FIX_TRUNC_EXPR:                        Unary and Binary Expressions.

  53240.                                                              (line    6)

  53241. * flags in RTL expression:               Flags.              (line    6)

  53242. * float:                                 Conversions.        (line   58)

  53243. * floating point and cross compilation:  Floating Point.     (line    6)

  53244. * floatMN2 instruction pattern:          Standard Names.     (line 1406)

  53245. * floatunsMN2 instruction pattern:       Standard Names.     (line 1410)

  53246. * FLOAT_EXPR:                            Unary and Binary Expressions.

  53247.                                                              (line    6)

  53248. * float_extend:                          Conversions.        (line   33)

  53249. * FLOAT_LIB_COMPARE_RETURNS_BOOL:        Library Calls.      (line   32)

  53250. * FLOAT_STORE_FLAG_VALUE:                Misc.               (line  320)

  53251. * float_truncate:                        Conversions.        (line   53)

  53252. * FLOAT_TYPE_SIZE:                       Type Layout.        (line   48)

  53253. * FLOAT_WORDS_BIG_ENDIAN:                Storage Layout.     (line   41)

  53254. * FLOAT_WORDS_BIG_ENDIAN, (lack of) effect on subreg: Regs and Memory.

  53255.                                                              (line  234)

  53256. * floorM2 instruction pattern:           Standard Names.     (line 1069)

  53257. * FLOOR_DIV_EXPR:                        Unary and Binary Expressions.

  53258.                                                              (line    6)

  53259. * FLOOR_MOD_EXPR:                        Unary and Binary Expressions.

  53260.                                                              (line    6)

  53261. * flow-insensitive alias analysis:       Alias analysis.     (line    6)

  53262. * flow-sensitive alias analysis:         Alias analysis.     (line    6)

  53263. * fma:                                   Arithmetic.         (line  112)

  53264. * fmaM4 instruction pattern:             Standard Names.     (line  478)

  53265. * fmaxM3 instruction pattern:            Standard Names.     (line  509)

  53266. * fminM3 instruction pattern:            Standard Names.     (line  509)

  53267. * fmodM3 instruction pattern:            Standard Names.     (line  901)

  53268. * fmsM4 instruction pattern:             Standard Names.     (line  485)

  53269. * fnmaM4 instruction pattern:            Standard Names.     (line  491)

  53270. * fnmsM4 instruction pattern:            Standard Names.     (line  497)

  53271. * fold_extract_last_M instruction pattern: Standard Names.   (line  550)

  53272. * fold_left_plus_M instruction pattern:  Standard Names.     (line  558)

  53273. * FORCE_CODE_SECTION_ALIGN:              Sections.           (line  149)

  53274. * force_reg:                             Standard Names.     (line   36)

  53275. * FOR_BODY:                              Statements for C++. (line    6)

  53276. * FOR_COND:                              Statements for C++. (line    6)

  53277. * FOR_EXPR:                              Statements for C++. (line    6)

  53278. * FOR_INIT_STMT:                         Statements for C++. (line    6)

  53279. * FOR_STMT:                              Statements for C++. (line    6)

  53280. * for_user:                              GTY Options.        (line   82)

  53281. * fractional types:                      Fixed-point fractional library routines.

  53282.                                                              (line    6)

  53283. * fractMN2 instruction pattern:          Standard Names.     (line 1456)

  53284. * fractunsMN2 instruction pattern:       Standard Names.     (line 1471)

  53285. * fract_convert:                         Conversions.        (line   82)

  53286. * FRACT_TYPE_SIZE:                       Type Layout.        (line   67)

  53287. * frame layout:                          Frame Layout.       (line    6)

  53288. * FRAME_ADDR_RTX:                        Frame Layout.       (line  108)

  53289. * FRAME_GROWS_DOWNWARD:                  Frame Layout.       (line   26)

  53290. * FRAME_GROWS_DOWNWARD and virtual registers: Regs and Memory.

  53291.                                                              (line   69)

  53292. * FRAME_POINTER_CFA_OFFSET:              Frame Layout.       (line  225)

  53293. * frame_pointer_needed:                  Function Entry.     (line   42)

  53294. * FRAME_POINTER_REGNUM:                  Frame Registers.    (line   13)

  53295. * FRAME_POINTER_REGNUM and virtual registers: Regs and Memory.

  53296.                                                              (line   74)

  53297. * frame_pointer_rtx:                     Frame Registers.    (line  104)

  53298. * frame_related:                         Flags.              (line  238)

  53299. * frame_related, in insn, call_insn, jump_insn, barrier, and set: Flags.

  53300.                                                              (line  135)

  53301. * frame_related, in mem:                 Flags.              (line   74)

  53302. * frame_related, in reg:                 Flags.              (line  102)

  53303. * frame_related, in symbol_ref:          Flags.              (line  179)

  53304. * frequency, count, BB_FREQ_BASE:        Profile information.

  53305.                                                              (line   30)

  53306. * ftruncM2 instruction pattern:          Standard Names.     (line 1429)

  53307. * function:                              Functions.          (line    6)

  53308. * function <1>:                          Functions for C++.  (line    6)

  53309. * function call conventions:             Interface.          (line    6)

  53310. * function entry and exit:               Function Entry.     (line    6)

  53311. * function entry point, alternate function entry point: Edges.

  53312.                                                              (line  180)

  53313. * function properties:                   Function Properties.

  53314.                                                              (line    6)

  53315. * function-call insns:                   Calls.              (line    6)

  53316. * functions, leaf:                       Leaf Functions.     (line    6)

  53317. * FUNCTION_ARG_REGNO_P:                  Register Arguments. (line  261)

  53318. * FUNCTION_BOUNDARY:                     Storage Layout.     (line  176)

  53319. * FUNCTION_DECL:                         Functions.          (line    6)

  53320. * FUNCTION_DECL <1>:                     Functions for C++.  (line    6)

  53321. * FUNCTION_MODE:                         Misc.               (line  375)

  53322. * FUNCTION_PROFILER:                     Profiling.          (line    8)

  53323. * FUNCTION_TYPE:                         Types.              (line    6)

  53324. * FUNCTION_VALUE:                        Scalar Return.      (line   52)

  53325. * FUNCTION_VALUE_REGNO_P:                Scalar Return.      (line   78)

  53326. * fundamental type:                      Types.              (line    6)

  53327. * G in constraint:                       Simple Constraints. (line   96)

  53328. * g in constraint:                       Simple Constraints. (line  118)

  53329. * garbage collector, invocation:         Invoking the garbage collector.

  53330.                                                              (line    6)

  53331. * garbage collector, troubleshooting:    Troubleshooting.    (line    6)

  53332. * gather_loadMN instruction pattern:     Standard Names.     (line  232)

  53333. * GCC and portability:                   Portability.        (line    6)

  53334. * GCC_DRIVER_HOST_INITIALIZATION:        Host Misc.          (line   36)

  53335. * gcov_type:                             Profile information.

  53336.                                                              (line   41)

  53337. * ge:                                    Comparisons.        (line   72)

  53338. * ge and attributes:                     Expressions.        (line   83)

  53339. * gencodes:                              RTL passes.         (line   18)

  53340. * general_operand:                       Machine-Independent Predicates.

  53341.                                                              (line  104)

  53342. * GENERAL_REGS:                          Register Classes.   (line   22)

  53343. * generated files:                       Files.              (line    6)

  53344. * generating assembler output:           Output Statement.   (line    6)

  53345. * generating insns:                      RTL Template.       (line    6)

  53346. * GENERIC:                               Parsing pass.       (line    6)

  53347. * GENERIC <1>:                           GENERIC.            (line    6)

  53348. * generic predicates:                    Machine-Independent Predicates.

  53349.                                                              (line    6)

  53350. * genflags:                              RTL passes.         (line   18)

  53351. * GEN_ERRNO_RTX:                         Library Calls.      (line   71)

  53352. * get_attr:                              Expressions.        (line   99)

  53353. * get_attr_length:                       Insn Lengths.       (line   52)

  53354. * GET_CLASS_NARROWEST_MODE:              Machine Modes.      (line  430)

  53355. * GET_CODE:                              RTL Objects.        (line   47)

  53356. * get_insns:                             Insns.              (line   34)

  53357. * get_last_insn:                         Insns.              (line   34)

  53358. * GET_MODE:                              Machine Modes.      (line  377)

  53359. * GET_MODE_ALIGNMENT:                    Machine Modes.      (line  417)

  53360. * GET_MODE_BITSIZE:                      Machine Modes.      (line  401)

  53361. * GET_MODE_CLASS:                        Machine Modes.      (line  391)

  53362. * GET_MODE_FBIT:                         Machine Modes.      (line  408)

  53363. * GET_MODE_IBIT:                         Machine Modes.      (line  404)

  53364. * GET_MODE_MASK:                         Machine Modes.      (line  412)

  53365. * GET_MODE_NAME:                         Machine Modes.      (line  388)

  53366. * GET_MODE_NUNITS:                       Machine Modes.      (line  426)

  53367. * GET_MODE_SIZE:                         Machine Modes.      (line  398)

  53368. * GET_MODE_UNIT_SIZE:                    Machine Modes.      (line  420)

  53369. * GET_MODE_WIDER_MODE:                   Machine Modes.      (line  394)

  53370. * GET_RTX_CLASS:                         RTL Classes.        (line    6)

  53371. * GET_RTX_FORMAT:                        RTL Classes.        (line  136)

  53372. * GET_RTX_LENGTH:                        RTL Classes.        (line  133)

  53373. * get_thread_pointerMODE instruction pattern: Standard Names.

  53374.                                                              (line 2477)

  53375. * geu:                                   Comparisons.        (line   72)

  53376. * geu and attributes:                    Expressions.        (line   83)

  53377. * GE_EXPR:                               Unary and Binary Expressions.

  53378.                                                              (line    6)

  53379. * GGC:                                   Type Information.   (line    6)

  53380. * ggc_collect:                           Invoking the garbage collector.

  53381.                                                              (line    6)

  53382. * GIMPLE:                                Parsing pass.       (line   13)

  53383. * GIMPLE <1>:                            Gimplification pass.

  53384.                                                              (line    6)

  53385. * GIMPLE <2>:                            GIMPLE.             (line    6)

  53386. * gimple:                                Tuple representation.

  53387.                                                              (line   14)

  53388. * GIMPLE API:                            GIMPLE API.         (line    6)

  53389. * GIMPLE class hierarchy:                Class hierarchy of GIMPLE statements.

  53390.                                                              (line    6)

  53391. * GIMPLE Exception Handling:             GIMPLE Exception Handling.

  53392.                                                              (line    6)

  53393. * GIMPLE instruction set:                GIMPLE instruction set.

  53394.                                                              (line    6)

  53395. * GIMPLE sequences:                      GIMPLE sequences.   (line    6)

  53396. * GIMPLE statement iterators:            Basic Blocks.       (line   78)

  53397. * GIMPLE statement iterators <1>:        Maintaining the CFG.

  53398.                                                              (line   33)

  53399. * gimple_addresses_taken:                Manipulating GIMPLE statements.

  53400.                                                              (line   89)

  53401. * GIMPLE_ASM:                            GIMPLE_ASM.         (line    6)

  53402. * gimple_asm_clobber_op:                 GIMPLE_ASM.         (line   39)

  53403. * gimple_asm_input_op:                   GIMPLE_ASM.         (line   23)

  53404. * gimple_asm_nclobbers:                  GIMPLE_ASM.         (line   20)

  53405. * gimple_asm_ninputs:                    GIMPLE_ASM.         (line   14)

  53406. * gimple_asm_noutputs:                   GIMPLE_ASM.         (line   17)

  53407. * gimple_asm_output_op:                  GIMPLE_ASM.         (line   31)

  53408. * gimple_asm_set_clobber_op:             GIMPLE_ASM.         (line   43)

  53409. * gimple_asm_set_input_op:               GIMPLE_ASM.         (line   27)

  53410. * gimple_asm_set_output_op:              GIMPLE_ASM.         (line   35)

  53411. * gimple_asm_set_volatile:               GIMPLE_ASM.         (line   54)

  53412. * gimple_asm_string:                     GIMPLE_ASM.         (line   47)

  53413. * gimple_asm_volatile_p:                 GIMPLE_ASM.         (line   51)

  53414. * GIMPLE_ASSIGN:                         GIMPLE_ASSIGN.      (line    6)

  53415. * gimple_assign_cast_p:                  Logical Operators.  (line  158)

  53416. * gimple_assign_cast_p <1>:              GIMPLE_ASSIGN.      (line  104)

  53417. * gimple_assign_lhs:                     GIMPLE_ASSIGN.      (line   62)

  53418. * gimple_assign_lhs_ptr:                 GIMPLE_ASSIGN.      (line   65)

  53419. * gimple_assign_rhs1:                    GIMPLE_ASSIGN.      (line   68)

  53420. * gimple_assign_rhs1_ptr:                GIMPLE_ASSIGN.      (line   71)

  53421. * gimple_assign_rhs2:                    GIMPLE_ASSIGN.      (line   75)

  53422. * gimple_assign_rhs2_ptr:                GIMPLE_ASSIGN.      (line   78)

  53423. * gimple_assign_rhs3:                    GIMPLE_ASSIGN.      (line   82)

  53424. * gimple_assign_rhs3_ptr:                GIMPLE_ASSIGN.      (line   85)

  53425. * gimple_assign_rhs_class:               GIMPLE_ASSIGN.      (line   56)

  53426. * gimple_assign_rhs_code:                GIMPLE_ASSIGN.      (line   52)

  53427. * gimple_assign_set_lhs:                 GIMPLE_ASSIGN.      (line   89)

  53428. * gimple_assign_set_rhs1:                GIMPLE_ASSIGN.      (line   92)

  53429. * gimple_assign_set_rhs2:                GIMPLE_ASSIGN.      (line   96)

  53430. * gimple_assign_set_rhs3:                GIMPLE_ASSIGN.      (line  100)

  53431. * gimple_bb:                             Manipulating GIMPLE statements.

  53432.                                                              (line   17)

  53433. * GIMPLE_BIND:                           GIMPLE_BIND.        (line    6)

  53434. * gimple_bind_add_seq:                   GIMPLE_BIND.        (line   34)

  53435. * gimple_bind_add_stmt:                  GIMPLE_BIND.        (line   31)

  53436. * gimple_bind_append_vars:               GIMPLE_BIND.        (line   18)

  53437. * gimple_bind_block:                     GIMPLE_BIND.        (line   39)

  53438. * gimple_bind_body:                      GIMPLE_BIND.        (line   22)

  53439. * gimple_bind_set_block:                 GIMPLE_BIND.        (line   44)

  53440. * gimple_bind_set_body:                  GIMPLE_BIND.        (line   26)

  53441. * gimple_bind_set_vars:                  GIMPLE_BIND.        (line   14)

  53442. * gimple_bind_vars:                      GIMPLE_BIND.        (line   11)

  53443. * gimple_block:                          Manipulating GIMPLE statements.

  53444.                                                              (line   20)

  53445. * gimple_build:                          GIMPLE API.         (line   34)

  53446. * gimple_build <1>:                      GIMPLE API.         (line   36)

  53447. * gimple_build <2>:                      GIMPLE API.         (line   38)

  53448. * gimple_build <3>:                      GIMPLE API.         (line   41)

  53449. * gimple_build <4>:                      GIMPLE API.         (line   44)

  53450. * gimple_build <5>:                      GIMPLE API.         (line   47)

  53451. * gimple_build_debug_begin_stmt:         GIMPLE_DEBUG.       (line   72)

  53452. * gimple_build_debug_inline_entry:       GIMPLE_DEBUG.       (line   82)

  53453. * gimple_build_nop:                      GIMPLE_NOP.         (line    6)

  53454. * gimple_build_omp_master:               GIMPLE_OMP_MASTER.  (line    6)

  53455. * gimple_build_omp_ordered:              GIMPLE_OMP_ORDERED. (line    6)

  53456. * gimple_build_omp_return:               GIMPLE_OMP_RETURN.  (line    6)

  53457. * gimple_build_omp_section:              GIMPLE_OMP_SECTION. (line    6)

  53458. * gimple_build_omp_sections_switch:      GIMPLE_OMP_SECTIONS.

  53459.                                                              (line   13)

  53460. * gimple_build_wce:                      GIMPLE_WITH_CLEANUP_EXPR.

  53461.                                                              (line    6)

  53462. * GIMPLE_CALL:                           GIMPLE_CALL.        (line    6)

  53463. * gimple_call_arg:                       GIMPLE_CALL.        (line   67)

  53464. * gimple_call_arg_ptr:                   GIMPLE_CALL.        (line   71)

  53465. * gimple_call_chain:                     GIMPLE_CALL.        (line   58)

  53466. * gimple_call_copy_skip_args:            GIMPLE_CALL.        (line   92)

  53467. * gimple_call_fn:                        GIMPLE_CALL.        (line   39)

  53468. * gimple_call_fndecl:                    GIMPLE_CALL.        (line   47)

  53469. * gimple_call_lhs:                       GIMPLE_CALL.        (line   30)

  53470. * gimple_call_lhs_ptr:                   GIMPLE_CALL.        (line   33)

  53471. * gimple_call_noreturn_p:                GIMPLE_CALL.        (line   89)

  53472. * gimple_call_num_args:                  GIMPLE_CALL.        (line   64)

  53473. * gimple_call_return_type:               GIMPLE_CALL.        (line   55)

  53474. * gimple_call_set_arg:                   GIMPLE_CALL.        (line   76)

  53475. * gimple_call_set_chain:                 GIMPLE_CALL.        (line   61)

  53476. * gimple_call_set_fn:                    GIMPLE_CALL.        (line   43)

  53477. * gimple_call_set_fndecl:                GIMPLE_CALL.        (line   52)

  53478. * gimple_call_set_lhs:                   GIMPLE_CALL.        (line   36)

  53479. * gimple_call_set_tail:                  GIMPLE_CALL.        (line   81)

  53480. * gimple_call_tail_p:                    GIMPLE_CALL.        (line   86)

  53481. * GIMPLE_CATCH:                          GIMPLE_CATCH.       (line    6)

  53482. * gimple_catch_handler:                  GIMPLE_CATCH.       (line   19)

  53483. * gimple_catch_set_handler:              GIMPLE_CATCH.       (line   26)

  53484. * gimple_catch_set_types:                GIMPLE_CATCH.       (line   23)

  53485. * gimple_catch_types:                    GIMPLE_CATCH.       (line   12)

  53486. * gimple_catch_types_ptr:                GIMPLE_CATCH.       (line   15)

  53487. * gimple_code:                           Manipulating GIMPLE statements.

  53488.                                                              (line   14)

  53489. * GIMPLE_COND:                           GIMPLE_COND.        (line    6)

  53490. * gimple_cond_code:                      GIMPLE_COND.        (line   20)

  53491. * gimple_cond_false_label:               GIMPLE_COND.        (line   59)

  53492. * gimple_cond_lhs:                       GIMPLE_COND.        (line   29)

  53493. * gimple_cond_make_false:                GIMPLE_COND.        (line   63)

  53494. * gimple_cond_make_true:                 GIMPLE_COND.        (line   66)

  53495. * gimple_cond_rhs:                       GIMPLE_COND.        (line   37)

  53496. * gimple_cond_set_code:                  GIMPLE_COND.        (line   24)

  53497. * gimple_cond_set_false_label:           GIMPLE_COND.        (line   54)

  53498. * gimple_cond_set_lhs:                   GIMPLE_COND.        (line   33)

  53499. * gimple_cond_set_rhs:                   GIMPLE_COND.        (line   41)

  53500. * gimple_cond_set_true_label:            GIMPLE_COND.        (line   49)

  53501. * gimple_cond_true_label:                GIMPLE_COND.        (line   45)

  53502. * gimple_convert:                        GIMPLE API.         (line   50)

  53503. * gimple_copy:                           Manipulating GIMPLE statements.

  53504.                                                              (line  146)

  53505. * GIMPLE_DEBUG:                          GIMPLE_DEBUG.       (line    6)

  53506. * GIMPLE_DEBUG_BEGIN_STMT:               GIMPLE_DEBUG.       (line    6)

  53507. * GIMPLE_DEBUG_BIND:                     GIMPLE_DEBUG.       (line    6)

  53508. * gimple_debug_bind_get_value:           GIMPLE_DEBUG.       (line   46)

  53509. * gimple_debug_bind_get_value_ptr:       GIMPLE_DEBUG.       (line   50)

  53510. * gimple_debug_bind_get_var:             GIMPLE_DEBUG.       (line   43)

  53511. * gimple_debug_bind_has_value_p:         GIMPLE_DEBUG.       (line   68)

  53512. * gimple_debug_bind_p:                   Logical Operators.  (line  162)

  53513. * gimple_debug_bind_reset_value:         GIMPLE_DEBUG.       (line   64)

  53514. * gimple_debug_bind_set_value:           GIMPLE_DEBUG.       (line   59)

  53515. * gimple_debug_bind_set_var:             GIMPLE_DEBUG.       (line   55)

  53516. * GIMPLE_DEBUG_INLINE_ENTRY:             GIMPLE_DEBUG.       (line    6)

  53517. * gimple_def_ops:                        Manipulating GIMPLE statements.

  53518.                                                              (line   93)

  53519. * GIMPLE_EH_FILTER:                      GIMPLE_EH_FILTER.   (line    6)

  53520. * gimple_eh_filter_failure:              GIMPLE_EH_FILTER.   (line   18)

  53521. * gimple_eh_filter_set_failure:          GIMPLE_EH_FILTER.   (line   27)

  53522. * gimple_eh_filter_set_types:            GIMPLE_EH_FILTER.   (line   22)

  53523. * gimple_eh_filter_types:                GIMPLE_EH_FILTER.   (line   11)

  53524. * gimple_eh_filter_types_ptr:            GIMPLE_EH_FILTER.   (line   14)

  53525. * gimple_eh_must_not_throw_fndecl:       GIMPLE_EH_FILTER.   (line   32)

  53526. * gimple_eh_must_not_throw_set_fndecl:   GIMPLE_EH_FILTER.   (line   36)

  53527. * gimple_expr_code:                      Manipulating GIMPLE statements.

  53528.                                                              (line   30)

  53529. * gimple_expr_type:                      Manipulating GIMPLE statements.

  53530.                                                              (line   23)

  53531. * GIMPLE_GOTO:                           GIMPLE_GOTO.        (line    6)

  53532. * gimple_goto_dest:                      GIMPLE_GOTO.        (line    9)

  53533. * gimple_goto_set_dest:                  GIMPLE_GOTO.        (line   12)

  53534. * gimple_has_mem_ops:                    Manipulating GIMPLE statements.

  53535.                                                              (line   71)

  53536. * gimple_has_ops:                        Manipulating GIMPLE statements.

  53537.                                                              (line   68)

  53538. * gimple_has_volatile_ops:               Manipulating GIMPLE statements.

  53539.                                                              (line  133)

  53540. * GIMPLE_LABEL:                          GIMPLE_LABEL.       (line    6)

  53541. * gimple_label_label:                    GIMPLE_LABEL.       (line   10)

  53542. * gimple_label_set_label:                GIMPLE_LABEL.       (line   13)

  53543. * gimple_loaded_syms:                    Manipulating GIMPLE statements.

  53544.                                                              (line  121)

  53545. * gimple_locus:                          Manipulating GIMPLE statements.

  53546.                                                              (line   41)

  53547. * gimple_locus_empty_p:                  Manipulating GIMPLE statements.

  53548.                                                              (line   47)

  53549. * gimple_modified_p:                     Manipulating GIMPLE statements.

  53550.                                                              (line  129)

  53551. * GIMPLE_NOP:                            GIMPLE_NOP.         (line    6)

  53552. * gimple_nop_p:                          GIMPLE_NOP.         (line    9)

  53553. * gimple_no_warning_p:                   Manipulating GIMPLE statements.

  53554.                                                              (line   50)

  53555. * gimple_num_ops:                        Logical Operators.  (line   76)

  53556. * gimple_num_ops <1>:                    Manipulating GIMPLE statements.

  53557.                                                              (line   74)

  53558. * GIMPLE_OMP_ATOMIC_LOAD:                GIMPLE_OMP_ATOMIC_LOAD.

  53559.                                                              (line    6)

  53560. * gimple_omp_atomic_load_lhs:            GIMPLE_OMP_ATOMIC_LOAD.

  53561.                                                              (line   16)

  53562. * gimple_omp_atomic_load_rhs:            GIMPLE_OMP_ATOMIC_LOAD.

  53563.                                                              (line   24)

  53564. * gimple_omp_atomic_load_set_lhs:        GIMPLE_OMP_ATOMIC_LOAD.

  53565.                                                              (line   12)

  53566. * gimple_omp_atomic_load_set_rhs:        GIMPLE_OMP_ATOMIC_LOAD.

  53567.                                                              (line   20)

  53568. * GIMPLE_OMP_ATOMIC_STORE:               GIMPLE_OMP_ATOMIC_STORE.

  53569.                                                              (line    6)

  53570. * gimple_omp_atomic_store_set_val:       GIMPLE_OMP_ATOMIC_STORE.

  53571.                                                              (line   11)

  53572. * gimple_omp_atomic_store_val:           GIMPLE_OMP_ATOMIC_STORE.

  53573.                                                              (line   15)

  53574. * gimple_omp_body:                       GIMPLE_OMP_PARALLEL.

  53575.                                                              (line   23)

  53576. * GIMPLE_OMP_CONTINUE:                   GIMPLE_OMP_CONTINUE.

  53577.                                                              (line    6)

  53578. * gimple_omp_continue_control_def:       GIMPLE_OMP_CONTINUE.

  53579.                                                              (line   12)

  53580. * gimple_omp_continue_control_def_ptr:   GIMPLE_OMP_CONTINUE.

  53581.                                                              (line   17)

  53582. * gimple_omp_continue_control_use:       GIMPLE_OMP_CONTINUE.

  53583.                                                              (line   26)

  53584. * gimple_omp_continue_control_use_ptr:   GIMPLE_OMP_CONTINUE.

  53585.                                                              (line   31)

  53586. * gimple_omp_continue_set_control_def:   GIMPLE_OMP_CONTINUE.

  53587.                                                              (line   21)

  53588. * gimple_omp_continue_set_control_use:   GIMPLE_OMP_CONTINUE.

  53589.                                                              (line   35)

  53590. * GIMPLE_OMP_CRITICAL:                   GIMPLE_OMP_CRITICAL.

  53591.                                                              (line    6)

  53592. * gimple_omp_critical_name:              GIMPLE_OMP_CRITICAL.

  53593.                                                              (line   12)

  53594. * gimple_omp_critical_name_ptr:          GIMPLE_OMP_CRITICAL.

  53595.                                                              (line   16)

  53596. * gimple_omp_critical_set_name:          GIMPLE_OMP_CRITICAL.

  53597.                                                              (line   21)

  53598. * GIMPLE_OMP_FOR:                        GIMPLE_OMP_FOR.     (line    6)

  53599. * gimple_omp_for_clauses:                GIMPLE_OMP_FOR.     (line   17)

  53600. * gimple_omp_for_clauses_ptr:            GIMPLE_OMP_FOR.     (line   20)

  53601. * gimple_omp_for_cond:                   GIMPLE_OMP_FOR.     (line   80)

  53602. * gimple_omp_for_final:                  GIMPLE_OMP_FOR.     (line   48)

  53603. * gimple_omp_for_final_ptr:              GIMPLE_OMP_FOR.     (line   51)

  53604. * gimple_omp_for_incr:                   GIMPLE_OMP_FOR.     (line   58)

  53605. * gimple_omp_for_incr_ptr:               GIMPLE_OMP_FOR.     (line   61)

  53606. * gimple_omp_for_index:                  GIMPLE_OMP_FOR.     (line   28)

  53607. * gimple_omp_for_index_ptr:              GIMPLE_OMP_FOR.     (line   31)

  53608. * gimple_omp_for_initial:                GIMPLE_OMP_FOR.     (line   38)

  53609. * gimple_omp_for_initial_ptr:            GIMPLE_OMP_FOR.     (line   41)

  53610. * gimple_omp_for_pre_body:               GIMPLE_OMP_FOR.     (line   67)

  53611. * gimple_omp_for_set_clauses:            GIMPLE_OMP_FOR.     (line   23)

  53612. * gimple_omp_for_set_cond:               GIMPLE_OMP_FOR.     (line   76)

  53613. * gimple_omp_for_set_final:              GIMPLE_OMP_FOR.     (line   54)

  53614. * gimple_omp_for_set_incr:               GIMPLE_OMP_FOR.     (line   64)

  53615. * gimple_omp_for_set_index:              GIMPLE_OMP_FOR.     (line   34)

  53616. * gimple_omp_for_set_initial:            GIMPLE_OMP_FOR.     (line   44)

  53617. * gimple_omp_for_set_pre_body:           GIMPLE_OMP_FOR.     (line   71)

  53618. * GIMPLE_OMP_MASTER:                     GIMPLE_OMP_MASTER.  (line    6)

  53619. * GIMPLE_OMP_ORDERED:                    GIMPLE_OMP_ORDERED. (line    6)

  53620. * GIMPLE_OMP_PARALLEL:                   GIMPLE_OMP_PARALLEL.

  53621.                                                              (line    6)

  53622. * gimple_omp_parallel_child_fn:          GIMPLE_OMP_PARALLEL.

  53623.                                                              (line   42)

  53624. * gimple_omp_parallel_child_fn_ptr:      GIMPLE_OMP_PARALLEL.

  53625.                                                              (line   47)

  53626. * gimple_omp_parallel_clauses:           GIMPLE_OMP_PARALLEL.

  53627.                                                              (line   30)

  53628. * gimple_omp_parallel_clauses_ptr:       GIMPLE_OMP_PARALLEL.

  53629.                                                              (line   33)

  53630. * gimple_omp_parallel_combined_p:        GIMPLE_OMP_PARALLEL.

  53631.                                                              (line   15)

  53632. * gimple_omp_parallel_data_arg:          GIMPLE_OMP_PARALLEL.

  53633.                                                              (line   56)

  53634. * gimple_omp_parallel_data_arg_ptr:      GIMPLE_OMP_PARALLEL.

  53635.                                                              (line   61)

  53636. * gimple_omp_parallel_set_child_fn:      GIMPLE_OMP_PARALLEL.

  53637.                                                              (line   52)

  53638. * gimple_omp_parallel_set_clauses:       GIMPLE_OMP_PARALLEL.

  53639.                                                              (line   37)

  53640. * gimple_omp_parallel_set_combined_p:    GIMPLE_OMP_PARALLEL.

  53641.                                                              (line   19)

  53642. * gimple_omp_parallel_set_data_arg:      GIMPLE_OMP_PARALLEL.

  53643.                                                              (line   65)

  53644. * GIMPLE_OMP_RETURN:                     GIMPLE_OMP_RETURN.  (line    6)

  53645. * gimple_omp_return_nowait_p:            GIMPLE_OMP_RETURN.  (line   13)

  53646. * gimple_omp_return_set_nowait:          GIMPLE_OMP_RETURN.  (line   10)

  53647. * GIMPLE_OMP_SECTION:                    GIMPLE_OMP_SECTION. (line    6)

  53648. * GIMPLE_OMP_SECTIONS:                   GIMPLE_OMP_SECTIONS.

  53649.                                                              (line    6)

  53650. * gimple_omp_sections_clauses:           GIMPLE_OMP_SECTIONS.

  53651.                                                              (line   29)

  53652. * gimple_omp_sections_clauses_ptr:       GIMPLE_OMP_SECTIONS.

  53653.                                                              (line   32)

  53654. * gimple_omp_sections_control:           GIMPLE_OMP_SECTIONS.

  53655.                                                              (line   16)

  53656. * gimple_omp_sections_control_ptr:       GIMPLE_OMP_SECTIONS.

  53657.                                                              (line   20)

  53658. * gimple_omp_sections_set_clauses:       GIMPLE_OMP_SECTIONS.

  53659.                                                              (line   35)

  53660. * gimple_omp_sections_set_control:       GIMPLE_OMP_SECTIONS.

  53661.                                                              (line   24)

  53662. * gimple_omp_section_last_p:             GIMPLE_OMP_SECTION. (line   11)

  53663. * gimple_omp_section_set_last:           GIMPLE_OMP_SECTION. (line   15)

  53664. * gimple_omp_set_body:                   GIMPLE_OMP_PARALLEL.

  53665.                                                              (line   26)

  53666. * GIMPLE_OMP_SINGLE:                     GIMPLE_OMP_SINGLE.  (line    6)

  53667. * gimple_omp_single_clauses:             GIMPLE_OMP_SINGLE.  (line   13)

  53668. * gimple_omp_single_clauses_ptr:         GIMPLE_OMP_SINGLE.  (line   16)

  53669. * gimple_omp_single_set_clauses:         GIMPLE_OMP_SINGLE.  (line   19)

  53670. * gimple_op:                             Logical Operators.  (line   79)

  53671. * gimple_op <1>:                         Manipulating GIMPLE statements.

  53672.                                                              (line   80)

  53673. * gimple_ops:                            Logical Operators.  (line   82)

  53674. * gimple_ops <1>:                        Manipulating GIMPLE statements.

  53675.                                                              (line   77)

  53676. * gimple_op_ptr:                         Manipulating GIMPLE statements.

  53677.                                                              (line   83)

  53678. * GIMPLE_PHI:                            GIMPLE_PHI.         (line    6)

  53679. * gimple_phi_arg:                        GIMPLE_PHI.         (line   24)

  53680. * gimple_phi_arg <1>:                    SSA.                (line   62)

  53681. * gimple_phi_arg_def:                    SSA.                (line   68)

  53682. * gimple_phi_arg_edge:                   SSA.                (line   65)

  53683. * gimple_phi_capacity:                   GIMPLE_PHI.         (line    6)

  53684. * gimple_phi_num_args:                   GIMPLE_PHI.         (line   10)

  53685. * gimple_phi_num_args <1>:               SSA.                (line   58)

  53686. * gimple_phi_result:                     GIMPLE_PHI.         (line   15)

  53687. * gimple_phi_result <1>:                 SSA.                (line   55)

  53688. * gimple_phi_result_ptr:                 GIMPLE_PHI.         (line   18)

  53689. * gimple_phi_set_arg:                    GIMPLE_PHI.         (line   28)

  53690. * gimple_phi_set_result:                 GIMPLE_PHI.         (line   21)

  53691. * gimple_plf:                            Manipulating GIMPLE statements.

  53692.                                                              (line   64)

  53693. * GIMPLE_RESX:                           GIMPLE_RESX.        (line    6)

  53694. * gimple_resx_region:                    GIMPLE_RESX.        (line   12)

  53695. * gimple_resx_set_region:                GIMPLE_RESX.        (line   15)

  53696. * GIMPLE_RETURN:                         GIMPLE_RETURN.      (line    6)

  53697. * gimple_return_retval:                  GIMPLE_RETURN.      (line    9)

  53698. * gimple_return_set_retval:              GIMPLE_RETURN.      (line   12)

  53699. * gimple_seq_add_seq:                    GIMPLE sequences.   (line   30)

  53700. * gimple_seq_add_stmt:                   GIMPLE sequences.   (line   24)

  53701. * gimple_seq_alloc:                      GIMPLE sequences.   (line   61)

  53702. * gimple_seq_copy:                       GIMPLE sequences.   (line   65)

  53703. * gimple_seq_deep_copy:                  GIMPLE sequences.   (line   36)

  53704. * gimple_seq_empty_p:                    GIMPLE sequences.   (line   69)

  53705. * gimple_seq_first:                      GIMPLE sequences.   (line   43)

  53706. * gimple_seq_init:                       GIMPLE sequences.   (line   58)

  53707. * gimple_seq_last:                       GIMPLE sequences.   (line   46)

  53708. * gimple_seq_reverse:                    GIMPLE sequences.   (line   39)

  53709. * gimple_seq_set_first:                  GIMPLE sequences.   (line   53)

  53710. * gimple_seq_set_last:                   GIMPLE sequences.   (line   49)

  53711. * gimple_seq_singleton_p:                GIMPLE sequences.   (line   78)

  53712. * gimple_set_block:                      Manipulating GIMPLE statements.

  53713.                                                              (line   38)

  53714. * gimple_set_def_ops:                    Manipulating GIMPLE statements.

  53715.                                                              (line   96)

  53716. * gimple_set_has_volatile_ops:           Manipulating GIMPLE statements.

  53717.                                                              (line  136)

  53718. * gimple_set_locus:                      Manipulating GIMPLE statements.

  53719.                                                              (line   44)

  53720. * gimple_set_op:                         Manipulating GIMPLE statements.

  53721.                                                              (line   86)

  53722. * gimple_set_plf:                        Manipulating GIMPLE statements.

  53723.                                                              (line   60)

  53724. * gimple_set_use_ops:                    Manipulating GIMPLE statements.

  53725.                                                              (line  103)

  53726. * gimple_set_vdef_ops:                   Manipulating GIMPLE statements.

  53727.                                                              (line  117)

  53728. * gimple_set_visited:                    Manipulating GIMPLE statements.

  53729.                                                              (line   53)

  53730. * gimple_set_vuse_ops:                   Manipulating GIMPLE statements.

  53731.                                                              (line  110)

  53732. * gimple_simplify:                       GIMPLE API.         (line    6)

  53733. * gimple_simplify <1>:                   GIMPLE API.         (line    8)

  53734. * gimple_simplify <2>:                   GIMPLE API.         (line   10)

  53735. * gimple_simplify <3>:                   GIMPLE API.         (line   12)

  53736. * gimple_simplify <4>:                   GIMPLE API.         (line   14)

  53737. * gimple_simplify <5>:                   GIMPLE API.         (line   16)

  53738. * gimple_statement_with_ops:             Tuple representation.

  53739.                                                              (line   96)

  53740. * gimple_stored_syms:                    Manipulating GIMPLE statements.

  53741.                                                              (line  125)

  53742. * GIMPLE_SWITCH:                         GIMPLE_SWITCH.      (line    6)

  53743. * gimple_switch_default_label:           GIMPLE_SWITCH.      (line   41)

  53744. * gimple_switch_index:                   GIMPLE_SWITCH.      (line   24)

  53745. * gimple_switch_label:                   GIMPLE_SWITCH.      (line   31)

  53746. * gimple_switch_num_labels:              GIMPLE_SWITCH.      (line   14)

  53747. * gimple_switch_set_default_label:       GIMPLE_SWITCH.      (line   45)

  53748. * gimple_switch_set_index:               GIMPLE_SWITCH.      (line   27)

  53749. * gimple_switch_set_label:               GIMPLE_SWITCH.      (line   36)

  53750. * gimple_switch_set_num_labels:          GIMPLE_SWITCH.      (line   19)

  53751. * GIMPLE_TRY:                            GIMPLE_TRY.         (line    6)

  53752. * gimple_try_catch_is_cleanup:           GIMPLE_TRY.         (line   19)

  53753. * gimple_try_cleanup:                    GIMPLE_TRY.         (line   26)

  53754. * gimple_try_eval:                       GIMPLE_TRY.         (line   22)

  53755. * gimple_try_kind:                       GIMPLE_TRY.         (line   15)

  53756. * gimple_try_set_catch_is_cleanup:       GIMPLE_TRY.         (line   30)

  53757. * gimple_try_set_cleanup:                GIMPLE_TRY.         (line   38)

  53758. * gimple_try_set_eval:                   GIMPLE_TRY.         (line   34)

  53759. * gimple_use_ops:                        Manipulating GIMPLE statements.

  53760.                                                              (line  100)

  53761. * gimple_vdef_ops:                       Manipulating GIMPLE statements.

  53762.                                                              (line  114)

  53763. * gimple_visited_p:                      Manipulating GIMPLE statements.

  53764.                                                              (line   57)

  53765. * gimple_vuse_ops:                       Manipulating GIMPLE statements.

  53766.                                                              (line  107)

  53767. * gimple_wce_cleanup:                    GIMPLE_WITH_CLEANUP_EXPR.

  53768.                                                              (line   10)

  53769. * gimple_wce_cleanup_eh_only:            GIMPLE_WITH_CLEANUP_EXPR.

  53770.                                                              (line   17)

  53771. * gimple_wce_set_cleanup:                GIMPLE_WITH_CLEANUP_EXPR.

  53772.                                                              (line   13)

  53773. * gimple_wce_set_cleanup_eh_only:        GIMPLE_WITH_CLEANUP_EXPR.

  53774.                                                              (line   20)

  53775. * GIMPLE_WITH_CLEANUP_EXPR:              GIMPLE_WITH_CLEANUP_EXPR.

  53776.                                                              (line    6)

  53777. * gimplification:                        Parsing pass.       (line   13)

  53778. * gimplification <1>:                    Gimplification pass.

  53779.                                                              (line    6)

  53780. * gimplifier:                            Parsing pass.       (line   13)

  53781. * gimplify_assign:                       GIMPLE_ASSIGN.      (line   41)

  53782. * gimplify_expr:                         Gimplification pass.

  53783.                                                              (line   18)

  53784. * gimplify_function_tree:                Gimplification pass.

  53785.                                                              (line   18)

  53786. * GLOBAL_INIT_PRIORITY:                  Functions for C++.  (line  141)

  53787. * global_regs:                           Register Basics.    (line  102)

  53788. * GO_IF_LEGITIMATE_ADDRESS:              Addressing Modes.   (line   90)

  53789. * greater than:                          Comparisons.        (line   60)

  53790. * greater than <1>:                      Comparisons.        (line   64)

  53791. * greater than <2>:                      Comparisons.        (line   72)

  53792. * gsi_after_labels:                      Sequence iterators. (line   74)

  53793. * gsi_bb:                                Sequence iterators. (line   82)

  53794. * gsi_commit_edge_inserts:               Sequence iterators. (line  193)

  53795. * gsi_commit_edge_inserts <1>:           Maintaining the CFG.

  53796.                                                              (line  104)

  53797. * gsi_commit_one_edge_insert:            Sequence iterators. (line  188)

  53798. * gsi_end_p:                             Sequence iterators. (line   59)

  53799. * gsi_end_p <1>:                         Maintaining the CFG.

  53800.                                                              (line   48)

  53801. * gsi_for_stmt:                          Sequence iterators. (line  156)

  53802. * gsi_insert_after:                      Sequence iterators. (line  145)

  53803. * gsi_insert_after <1>:                  Maintaining the CFG.

  53804.                                                              (line   60)

  53805. * gsi_insert_before:                     Sequence iterators. (line  134)

  53806. * gsi_insert_before <1>:                 Maintaining the CFG.

  53807.                                                              (line   66)

  53808. * gsi_insert_on_edge:                    Sequence iterators. (line  173)

  53809. * gsi_insert_on_edge <1>:                Maintaining the CFG.

  53810.                                                              (line  104)

  53811. * gsi_insert_on_edge_immediate:          Sequence iterators. (line  183)

  53812. * gsi_insert_seq_after:                  Sequence iterators. (line  152)

  53813. * gsi_insert_seq_before:                 Sequence iterators. (line  141)

  53814. * gsi_insert_seq_on_edge:                Sequence iterators. (line  177)

  53815. * gsi_last:                              Sequence iterators. (line   49)

  53816. * gsi_last <1>:                          Maintaining the CFG.

  53817.                                                              (line   44)

  53818. * gsi_last_bb:                           Sequence iterators. (line   55)

  53819. * gsi_link_after:                        Sequence iterators. (line  113)

  53820. * gsi_link_before:                       Sequence iterators. (line  103)

  53821. * gsi_link_seq_after:                    Sequence iterators. (line  108)

  53822. * gsi_link_seq_before:                   Sequence iterators. (line   97)

  53823. * gsi_move_after:                        Sequence iterators. (line  159)

  53824. * gsi_move_before:                       Sequence iterators. (line  164)

  53825. * gsi_move_to_bb_end:                    Sequence iterators. (line  169)

  53826. * gsi_next:                              Sequence iterators. (line   65)

  53827. * gsi_next <1>:                          Maintaining the CFG.

  53828.                                                              (line   52)

  53829. * gsi_one_before_end_p:                  Sequence iterators. (line   62)

  53830. * gsi_prev:                              Sequence iterators. (line   68)

  53831. * gsi_prev <1>:                          Maintaining the CFG.

  53832.                                                              (line   56)

  53833. * gsi_remove:                            Sequence iterators. (line   88)

  53834. * gsi_remove <1>:                        Maintaining the CFG.

  53835.                                                              (line   72)

  53836. * gsi_replace:                           Sequence iterators. (line  128)

  53837. * gsi_seq:                               Sequence iterators. (line   85)

  53838. * gsi_split_seq_after:                   Sequence iterators. (line  118)

  53839. * gsi_split_seq_before:                  Sequence iterators. (line  123)

  53840. * gsi_start:                             Sequence iterators. (line   39)

  53841. * gsi_start <1>:                         Maintaining the CFG.

  53842.                                                              (line   40)

  53843. * gsi_start_bb:                          Sequence iterators. (line   45)

  53844. * gsi_stmt:                              Sequence iterators. (line   71)

  53845. * gsi_stmt_ptr:                          Sequence iterators. (line   79)

  53846. * gt:                                    Comparisons.        (line   60)

  53847. * gt and attributes:                     Expressions.        (line   83)

  53848. * gtu:                                   Comparisons.        (line   64)

  53849. * gtu and attributes:                    Expressions.        (line   83)

  53850. * GTY:                                   Type Information.   (line    6)

  53851. * GT_EXPR:                               Unary and Binary Expressions.

  53852.                                                              (line    6)

  53853. * guidelines for diagnostics:            Guidelines for Diagnostics.

  53854.                                                              (line    6)

  53855. * guidelines for options:                Guidelines for Options.

  53856.                                                              (line    6)

  53857. * guidelines, user experience:           User Experience Guidelines.

  53858.                                                              (line    6)

  53859. * H in constraint:                       Simple Constraints. (line   96)

  53860. * HAmode:                                Machine Modes.      (line  146)

  53861. * HANDLER:                               Statements for C++. (line    6)

  53862. * HANDLER_BODY:                          Statements for C++. (line    6)

  53863. * HANDLER_PARMS:                         Statements for C++. (line    6)

  53864. * HANDLE_PRAGMA_PACK_WITH_EXPANSION:     Misc.               (line  475)

  53865. * hard registers:                        Regs and Memory.    (line    9)

  53866. * HARD_FRAME_POINTER_IS_ARG_POINTER:     Frame Registers.    (line   57)

  53867. * HARD_FRAME_POINTER_IS_FRAME_POINTER:   Frame Registers.    (line   50)

  53868. * HARD_FRAME_POINTER_REGNUM:             Frame Registers.    (line   19)

  53869. * HARD_REGNO_CALLER_SAVE_MODE:           Caller Saves.       (line   10)

  53870. * HARD_REGNO_NREGS_HAS_PADDING:          Values in Registers.

  53871.                                                              (line   21)

  53872. * HARD_REGNO_NREGS_WITH_PADDING:         Values in Registers.

  53873.                                                              (line   39)

  53874. * HARD_REGNO_RENAME_OK:                  Values in Registers.

  53875.                                                              (line  113)

  53876. * HAS_INIT_SECTION:                      Macros for Initialization.

  53877.                                                              (line   18)

  53878. * HAS_LONG_COND_BRANCH:                  Misc.               (line    8)

  53879. * HAS_LONG_UNCOND_BRANCH:                Misc.               (line   17)

  53880. * HAVE_DOS_BASED_FILE_SYSTEM:            Filesystem.         (line   11)

  53881. * HAVE_POST_DECREMENT:                   Addressing Modes.   (line   11)

  53882. * HAVE_POST_INCREMENT:                   Addressing Modes.   (line   10)

  53883. * HAVE_POST_MODIFY_DISP:                 Addressing Modes.   (line   17)

  53884. * HAVE_POST_MODIFY_REG:                  Addressing Modes.   (line   23)

  53885. * HAVE_PRE_DECREMENT:                    Addressing Modes.   (line    9)

  53886. * HAVE_PRE_INCREMENT:                    Addressing Modes.   (line    8)

  53887. * HAVE_PRE_MODIFY_DISP:                  Addressing Modes.   (line   16)

  53888. * HAVE_PRE_MODIFY_REG:                   Addressing Modes.   (line   22)

  53889. * HCmode:                                Machine Modes.      (line  199)

  53890. * HFmode:                                Machine Modes.      (line   61)

  53891. * high:                                  Constants.          (line  220)

  53892. * HImode:                                Machine Modes.      (line   29)

  53893. * HImode, in insn:                       Insns.              (line  291)

  53894. * HONOR_REG_ALLOC_ORDER:                 Allocation Order.   (line   36)

  53895. * host configuration:                    Host Config.        (line    6)

  53896. * host functions:                        Host Common.        (line    6)

  53897. * host hooks:                            Host Common.        (line    6)

  53898. * host makefile fragment:                Host Fragment.      (line    6)

  53899. * HOST_BIT_BUCKET:                       Filesystem.         (line   51)

  53900. * HOST_EXECUTABLE_SUFFIX:                Filesystem.         (line   45)

  53901. * HOST_HOOKS_EXTRA_SIGNALS:              Host Common.        (line   11)

  53902. * HOST_HOOKS_GT_PCH_ALLOC_GRANULARITY:   Host Common.        (line   43)

  53903. * HOST_HOOKS_GT_PCH_GET_ADDRESS:         Host Common.        (line   15)

  53904. * HOST_HOOKS_GT_PCH_USE_ADDRESS:         Host Common.        (line   24)

  53905. * HOST_LACKS_INODE_NUMBERS:              Filesystem.         (line   89)

  53906. * HOST_LONG_FORMAT:                      Host Misc.          (line   45)

  53907. * HOST_LONG_LONG_FORMAT:                 Host Misc.          (line   41)

  53908. * HOST_OBJECT_SUFFIX:                    Filesystem.         (line   40)

  53909. * HOST_PTR_PRINTF:                       Host Misc.          (line   49)

  53910. * HOT_TEXT_SECTION_NAME:                 Sections.           (line   42)

  53911. * HQmode:                                Machine Modes.      (line  110)

  53912. * i in constraint:                       Simple Constraints. (line   68)

  53913. * I in constraint:                       Simple Constraints. (line   79)

  53914. * identifier:                            Identifiers.        (line    6)

  53915. * IDENTIFIER_LENGTH:                     Identifiers.        (line   22)

  53916. * IDENTIFIER_NODE:                       Identifiers.        (line    6)

  53917. * IDENTIFIER_OPNAME_P:                   Identifiers.        (line   27)

  53918. * IDENTIFIER_POINTER:                    Identifiers.        (line   17)

  53919. * IDENTIFIER_TYPENAME_P:                 Identifiers.        (line   33)

  53920. * IEEE 754-2008:                         Decimal float library routines.

  53921.                                                              (line    6)

  53922. * IFCVT_MACHDEP_INIT:                    Misc.               (line  601)

  53923. * IFCVT_MODIFY_CANCEL:                   Misc.               (line  595)

  53924. * IFCVT_MODIFY_FINAL:                    Misc.               (line  589)

  53925. * IFCVT_MODIFY_INSN:                     Misc.               (line  583)

  53926. * IFCVT_MODIFY_MULTIPLE_TESTS:           Misc.               (line  575)

  53927. * IFCVT_MODIFY_TESTS:                    Misc.               (line  565)

  53928. * IF_COND:                               Statements for C++. (line    6)

  53929. * IF_STMT:                               Statements for C++. (line    6)

  53930. * if_then_else:                          Comparisons.        (line   80)

  53931. * if_then_else and attributes:           Expressions.        (line   32)

  53932. * if_then_else usage:                    Side Effects.       (line   56)

  53933. * IMAGPART_EXPR:                         Unary and Binary Expressions.

  53934.                                                              (line    6)

  53935. * Immediate Uses:                        SSA Operands.       (line  258)

  53936. * immediate_operand:                     Machine-Independent Predicates.

  53937.                                                              (line   10)

  53938. * IMMEDIATE_PREFIX:                      Instruction Output. (line  153)

  53939. * include:                               Including Patterns. (line    6)

  53940. * INCLUDE_DEFAULTS:                      Driver.             (line  331)

  53941. * inclusive-or, bitwise:                 Arithmetic.         (line  163)

  53942. * INCOMING_FRAME_SP_OFFSET:              Frame Layout.       (line  188)

  53943. * INCOMING_REGNO:                        Register Basics.    (line  129)

  53944. * INCOMING_REG_PARM_STACK_SPACE:         Stack Arguments.    (line   73)

  53945. * INCOMING_RETURN_ADDR_RTX:              Frame Layout.       (line  133)

  53946. * INCOMING_STACK_BOUNDARY:               Storage Layout.     (line  171)

  53947. * INDEX_REG_CLASS:                       Register Classes.   (line  140)

  53948. * indirect_jump instruction pattern:     Standard Names.     (line 1826)

  53949. * indirect_operand:                      Machine-Independent Predicates.

  53950.                                                              (line   70)

  53951. * INDIRECT_REF:                          Storage References. (line    6)

  53952. * initialization routines:               Initialization.     (line    6)

  53953. * INITIAL_ELIMINATION_OFFSET:            Elimination.        (line   68)

  53954. * INITIAL_FRAME_ADDRESS_RTX:             Frame Layout.       (line   75)

  53955. * INIT_ARRAY_SECTION_ASM_OP:             Sections.           (line  106)

  53956. * INIT_CUMULATIVE_ARGS:                  Register Arguments. (line  158)

  53957. * INIT_CUMULATIVE_INCOMING_ARGS:         Register Arguments. (line  186)

  53958. * INIT_CUMULATIVE_LIBCALL_ARGS:          Register Arguments. (line  180)

  53959. * INIT_ENVIRONMENT:                      Driver.             (line  309)

  53960. * INIT_EXPANDERS:                        Per-Function Data.  (line   36)

  53961. * INIT_EXPR:                             Unary and Binary Expressions.

  53962.                                                              (line    6)

  53963. * init_machine_status:                   Per-Function Data.  (line   42)

  53964. * init_one_libfunc:                      Library Calls.      (line   15)

  53965. * INIT_SECTION_ASM_OP:                   Sections.           (line   90)

  53966. * INIT_SECTION_ASM_OP <1>:               Macros for Initialization.

  53967.                                                              (line    9)

  53968. * inlining:                              Target Attributes.  (line  103)

  53969. * insert_insn_on_edge:                   Maintaining the CFG.

  53970.                                                              (line  104)

  53971. * insn:                                  Insns.              (line   63)

  53972. * insn and /f:                           Flags.              (line  135)

  53973. * insn and /j:                           Flags.              (line  171)

  53974. * insn and /s:                           Flags.              (line   38)

  53975. * insn and /s <1>:                       Flags.              (line  162)

  53976. * insn and /u:                           Flags.              (line   28)

  53977. * insn and /v:                           Flags.              (line   33)

  53978. * insn attributes:                       Insn Attributes.    (line    6)

  53979. * insn canonicalization:                 Insn Canonicalizations.

  53980.                                                              (line    6)

  53981. * insn includes:                         Including Patterns. (line    6)

  53982. * insn lengths, computing:               Insn Lengths.       (line    6)

  53983. * insn notes, notes:                     Basic Blocks.       (line   52)

  53984. * insn splitting:                        Insn Splitting.     (line    6)

  53985. * insn-attr.h:                           Defining Attributes.

  53986.                                                              (line   34)

  53987. * insns:                                 Insns.              (line    6)

  53988. * insns, generating:                     RTL Template.       (line    6)

  53989. * insns, recognizing:                    RTL Template.       (line    6)

  53990. * INSN_ANNULLED_BRANCH_P:                Flags.              (line   28)

  53991. * INSN_CODE:                             Insns.              (line  318)

  53992. * INSN_DELETED_P:                        Flags.              (line   33)

  53993. * INSN_FROM_TARGET_P:                    Flags.              (line   38)

  53994. * insn_list:                             Insns.              (line  568)

  53995. * INSN_REFERENCES_ARE_DELAYED:           Misc.               (line  502)

  53996. * INSN_SETS_ARE_DELAYED:                 Misc.               (line  491)

  53997. * INSN_UID:                              Insns.              (line   23)

  53998. * INSN_VAR_LOCATION:                     Insns.              (line  247)

  53999. * instruction attributes:                Insn Attributes.    (line    6)

  54000. * instruction latency time:              Processor pipeline description.

  54001.                                                              (line    6)

  54002. * instruction latency time <1>:          Processor pipeline description.

  54003.                                                              (line  105)

  54004. * instruction latency time <2>:          Processor pipeline description.

  54005.                                                              (line  196)

  54006. * instruction patterns:                  Patterns.           (line    6)

  54007. * instruction splitting:                 Insn Splitting.     (line    6)

  54008. * insv instruction pattern:              Standard Names.     (line 1562)

  54009. * insvM instruction pattern:             Standard Names.     (line 1514)

  54010. * insvmisalignM instruction pattern:     Standard Names.     (line 1524)

  54011. * int iterators in .md files:            Int Iterators.      (line    6)

  54012. * INT16_TYPE:                            Type Layout.        (line  210)

  54013. * INT32_TYPE:                            Type Layout.        (line  211)

  54014. * INT64_TYPE:                            Type Layout.        (line  212)

  54015. * INT8_TYPE:                             Type Layout.        (line  209)

  54016. * INTEGER_CST:                           Constant expressions.

  54017.                                                              (line    6)

  54018. * INTEGER_TYPE:                          Types.              (line    6)

  54019. * inter-procedural optimization passes:  IPA passes.         (line    6)

  54020. * Interdependence of Patterns:           Dependent Patterns. (line    6)

  54021. * interfacing to GCC output:             Interface.          (line    6)

  54022. * interlock delays:                      Processor pipeline description.

  54023.                                                              (line    6)

  54024. * intermediate representation lowering:  Parsing pass.       (line   13)

  54025. * INTMAX_TYPE:                           Type Layout.        (line  186)

  54026. * INTPTR_TYPE:                           Type Layout.        (line  233)

  54027. * introduction:                          Top.                (line    6)

  54028. * INT_FAST16_TYPE:                       Type Layout.        (line  226)

  54029. * INT_FAST32_TYPE:                       Type Layout.        (line  227)

  54030. * INT_FAST64_TYPE:                       Type Layout.        (line  228)

  54031. * INT_FAST8_TYPE:                        Type Layout.        (line  225)

  54032. * INT_LEAST16_TYPE:                      Type Layout.        (line  218)

  54033. * INT_LEAST32_TYPE:                      Type Layout.        (line  219)

  54034. * INT_LEAST64_TYPE:                      Type Layout.        (line  220)

  54035. * INT_LEAST8_TYPE:                       Type Layout.        (line  217)

  54036. * INT_TYPE_SIZE:                         Type Layout.        (line   11)

  54037. * INVOKE__main:                          Macros for Initialization.

  54038.                                                              (line   50)

  54039. * in_struct:                             Flags.              (line  254)

  54040. * in_struct, in code_label and note:     Flags.              (line   48)

  54041. * in_struct, in insn and jump_insn and call_insn: Flags.     (line   38)

  54042. * in_struct, in insn, call_insn, jump_insn and jump_table_data: Flags.

  54043.                                                              (line  162)

  54044. * in_struct, in subreg:                  Flags.              (line  201)

  54045. * ior:                                   Arithmetic.         (line  163)

  54046. * ior and attributes:                    Expressions.        (line   50)

  54047. * ior, canonicalization of:              Insn Canonicalizations.

  54048.                                                              (line   67)

  54049. * iorM3 instruction pattern:             Standard Names.     (line  442)

  54050. * IPA passes:                            IPA passes.         (line    6)

  54051. * IRA_HARD_REGNO_ADD_COST_MULTIPLIER:    Allocation Order.   (line   44)

  54052. * is_a:                                  Machine Modes.      (line  347)

  54053. * IS_ASM_LOGICAL_LINE_SEPARATOR:         Data Output.        (line  129)

  54054. * is_gimple_addressable:                 Logical Operators.  (line  113)

  54055. * is_gimple_asm_val:                     Logical Operators.  (line  117)

  54056. * is_gimple_assign:                      Logical Operators.  (line  149)

  54057. * is_gimple_call:                        Logical Operators.  (line  152)

  54058. * is_gimple_call_addr:                   Logical Operators.  (line  120)

  54059. * is_gimple_constant:                    Logical Operators.  (line  128)

  54060. * is_gimple_debug:                       Logical Operators.  (line  155)

  54061. * is_gimple_ip_invariant:                Logical Operators.  (line  137)

  54062. * is_gimple_ip_invariant_address:        Logical Operators.  (line  142)

  54063. * is_gimple_mem_ref_addr:                Logical Operators.  (line  124)

  54064. * is_gimple_min_invariant:               Logical Operators.  (line  131)

  54065. * is_gimple_omp:                         Logical Operators.  (line  166)

  54066. * is_gimple_val:                         Logical Operators.  (line  107)

  54067. * iterators in .md files:                Iterators.          (line    6)

  54068. * IV analysis on GIMPLE:                 Scalar evolutions.  (line    6)

  54069. * IV analysis on RTL:                    loop-iv.            (line    6)

  54070. * JMP_BUF_SIZE:                          Exception Region Output.

  54071.                                                              (line   83)

  54072. * jump:                                  Flags.              (line  295)

  54073. * jump instruction pattern:              Standard Names.     (line 1704)

  54074. * jump instruction patterns:             Jump Patterns.      (line    6)

  54075. * jump instructions and set:             Side Effects.       (line   56)

  54076. * jump, in call_insn:                    Flags.              (line  175)

  54077. * jump, in insn:                         Flags.              (line  171)

  54078. * jump, in mem:                          Flags.              (line   59)

  54079. * Jumps:                                 Jumps.              (line    6)

  54080. * JUMP_ALIGN:                            Alignment Output.   (line    8)

  54081. * jump_insn:                             Insns.              (line   73)

  54082. * jump_insn and /f:                      Flags.              (line  135)

  54083. * jump_insn and /j:                      Flags.              (line   10)

  54084. * jump_insn and /s:                      Flags.              (line   38)

  54085. * jump_insn and /s <1>:                  Flags.              (line  162)

  54086. * jump_insn and /u:                      Flags.              (line   28)

  54087. * jump_insn and /v:                      Flags.              (line   33)

  54088. * JUMP_LABEL:                            Insns.              (line   80)

  54089. * JUMP_TABLES_IN_TEXT_SECTION:           Sections.           (line  155)

  54090. * jump_table_data:                       Insns.              (line  166)

  54091. * jump_table_data and /s:                Flags.              (line  162)

  54092. * jump_table_data and /v:                Flags.              (line   33)

  54093. * LABEL_ALIGN:                           Alignment Output.   (line   42)

  54094. * LABEL_ALIGN_AFTER_BARRIER:             Alignment Output.   (line   21)

  54095. * LABEL_ALTERNATE_NAME:                  Edges.              (line  180)

  54096. * LABEL_ALT_ENTRY_P:                     Insns.              (line  146)

  54097. * LABEL_DECL:                            Declarations.       (line    6)

  54098. * LABEL_KIND:                            Insns.              (line  146)

  54099. * LABEL_NUSES:                           Insns.              (line  142)

  54100. * LABEL_PRESERVE_P:                      Flags.              (line   48)

  54101. * label_ref:                             Constants.          (line  199)

  54102. * label_ref and /v:                      Flags.              (line   54)

  54103. * label_ref, RTL sharing:                Sharing.            (line   38)

  54104. * LABEL_REF_NONLOCAL_P:                  Flags.              (line   54)

  54105. * language-dependent trees:              Language-dependent trees.

  54106.                                                              (line    6)

  54107. * language-independent intermediate representation: Parsing pass.

  54108.                                                              (line   13)

  54109. * lang_hooks.gimplify_expr:              Gimplification pass.

  54110.                                                              (line   18)

  54111. * lang_hooks.parse_file:                 Parsing pass.       (line    6)

  54112. * large return values:                   Aggregate Return.   (line    6)

  54113. * LAST_STACK_REG:                        Stack Registers.    (line   30)

  54114. * LAST_VIRTUAL_REGISTER:                 Regs and Memory.    (line   51)

  54115. * late IPA passes:                       Late IPA passes.    (line    6)

  54116. * lceilMN2:                              Standard Names.     (line 1137)

  54117. * LCSSA:                                 LCSSA.              (line    6)

  54118. * LDD_SUFFIX:                            Macros for Initialization.

  54119.                                                              (line  121)

  54120. * ldexpM3 instruction pattern:           Standard Names.     (line  922)

  54121. * LD_FINI_SWITCH:                        Macros for Initialization.

  54122.                                                              (line   28)

  54123. * LD_INIT_SWITCH:                        Macros for Initialization.

  54124.                                                              (line   24)

  54125. * le:                                    Comparisons.        (line   76)

  54126. * le and attributes:                     Expressions.        (line   83)

  54127. * leaf functions:                        Leaf Functions.     (line    6)

  54128. * leaf_function_p:                       Standard Names.     (line 1788)

  54129. * LEAF_REGISTERS:                        Leaf Functions.     (line   23)

  54130. * LEAF_REG_REMAP:                        Leaf Functions.     (line   37)

  54131. * left rotate:                           Arithmetic.         (line  195)

  54132. * left shift:                            Arithmetic.         (line  173)

  54133. * LEGITIMATE_PIC_OPERAND_P:              PIC.                (line   31)

  54134. * LEGITIMIZE_RELOAD_ADDRESS:             Addressing Modes.   (line  150)

  54135. * length:                                GTY Options.        (line   47)

  54136. * less than:                             Comparisons.        (line   68)

  54137. * less than or equal:                    Comparisons.        (line   76)

  54138. * leu:                                   Comparisons.        (line   76)

  54139. * leu and attributes:                    Expressions.        (line   83)

  54140. * LE_EXPR:                               Unary and Binary Expressions.

  54141.                                                              (line    6)

  54142. * lfloorMN2:                             Standard Names.     (line 1132)

  54143. * LIB2FUNCS_EXTRA:                       Target Fragment.    (line   11)

  54144. * LIBCALL_VALUE:                         Scalar Return.      (line   56)

  54145. * libgcc.a:                              Library Calls.      (line    6)

  54146. * LIBGCC2_CFLAGS:                        Target Fragment.    (line    8)

  54147. * LIBGCC2_GNU_PREFIX:                    Type Layout.        (line  102)

  54148. * LIBGCC2_UNWIND_ATTRIBUTE:              Misc.               (line 1047)

  54149. * LIBGCC_SPEC:                           Driver.             (line  115)

  54150. * library subroutine names:              Library Calls.      (line    6)

  54151. * LIBRARY_PATH_ENV:                      Misc.               (line  543)

  54152. * LIB_SPEC:                              Driver.             (line  107)

  54153. * LIMIT_RELOAD_CLASS:                    Register Classes.   (line  296)

  54154. * LINK_COMMAND_SPEC:                     Driver.             (line  240)

  54155. * LINK_EH_SPEC:                          Driver.             (line  142)

  54156. * LINK_GCC_C_SEQUENCE_SPEC:              Driver.             (line  232)

  54157. * LINK_LIBGCC_SPECIAL_1:                 Driver.             (line  227)

  54158. * LINK_SPEC:                             Driver.             (line  100)

  54159. * list:                                  Containers.         (line    6)

  54160. * Liveness representation:               Liveness information.

  54161.                                                              (line    6)

  54162. * load address instruction:              Simple Constraints. (line  162)

  54163. * LOAD_EXTEND_OP:                        Misc.               (line   80)

  54164. * load_multiple instruction pattern:     Standard Names.     (line  136)

  54165. * Local Register Allocator (LRA):        RTL passes.         (line  187)

  54166. * LOCAL_ALIGNMENT:                       Storage Layout.     (line  284)

  54167. * LOCAL_CLASS_P:                         Classes.            (line   70)

  54168. * LOCAL_DECL_ALIGNMENT:                  Storage Layout.     (line  321)

  54169. * LOCAL_INCLUDE_DIR:                     Driver.             (line  316)

  54170. * LOCAL_LABEL_PREFIX:                    Instruction Output. (line  151)

  54171. * LOCAL_REGNO:                           Register Basics.    (line  143)

  54172. * location information:                  Guidelines for Diagnostics.

  54173.                                                              (line  159)

  54174. * log10M2 instruction pattern:           Standard Names.     (line 1026)

  54175. * log1pM2 instruction pattern:           Standard Names.     (line 1016)

  54176. * log2M2 instruction pattern:            Standard Names.     (line 1033)

  54177. * logbM2 instruction pattern:            Standard Names.     (line 1040)

  54178. * Logical Operators:                     Logical Operators.  (line    6)

  54179. * logical-and, bitwise:                  Arithmetic.         (line  158)

  54180. * LOGICAL_OP_NON_SHORT_CIRCUIT:          Costs.              (line  294)

  54181. * logM2 instruction pattern:             Standard Names.     (line 1009)

  54182. * LOG_LINKS:                             Insns.              (line  337)

  54183. * longjmp and automatic variables:       Interface.          (line   52)

  54184. * LONG_ACCUM_TYPE_SIZE:                  Type Layout.        (line   92)

  54185. * LONG_DOUBLE_TYPE_SIZE:                 Type Layout.        (line   57)

  54186. * LONG_FRACT_TYPE_SIZE:                  Type Layout.        (line   72)

  54187. * LONG_LONG_ACCUM_TYPE_SIZE:             Type Layout.        (line   97)

  54188. * LONG_LONG_FRACT_TYPE_SIZE:             Type Layout.        (line   77)

  54189. * LONG_LONG_TYPE_SIZE:                   Type Layout.        (line   32)

  54190. * LONG_TYPE_SIZE:                        Type Layout.        (line   21)

  54191. * Loop analysis:                         Loop representation.

  54192.                                                              (line    6)

  54193. * Loop manipulation:                     Loop manipulation.  (line    6)

  54194. * Loop querying:                         Loop querying.      (line    6)

  54195. * Loop representation:                   Loop representation.

  54196.                                                              (line    6)

  54197. * Loop-closed SSA form:                  LCSSA.              (line    6)

  54198. * looping instruction patterns:          Looping Patterns.   (line    6)

  54199. * LOOP_ALIGN:                            Alignment Output.   (line   29)

  54200. * LOOP_EXPR:                             Unary and Binary Expressions.

  54201.                                                              (line    6)

  54202. * lowering, language-dependent intermediate representation: Parsing pass.

  54203.                                                              (line   13)

  54204. * lo_sum:                                Arithmetic.         (line   25)

  54205. * lrintMN2:                              Standard Names.     (line 1122)

  54206. * lroundMN2:                             Standard Names.     (line 1127)

  54207. * lshiftrt:                              Arithmetic.         (line  190)

  54208. * lshiftrt and attributes:               Expressions.        (line   83)

  54209. * LSHIFT_EXPR:                           Unary and Binary Expressions.

  54210.                                                              (line    6)

  54211. * lshrM3 instruction pattern:            Standard Names.     (line  840)

  54212. * lt:                                    Comparisons.        (line   68)

  54213. * lt and attributes:                     Expressions.        (line   83)

  54214. * LTGT_EXPR:                             Unary and Binary Expressions.

  54215.                                                              (line    6)

  54216. * lto:                                   LTO.                (line    6)

  54217. * ltrans:                                LTO.                (line    6)

  54218. * ltu:                                   Comparisons.        (line   68)

  54219. * LT_EXPR:                               Unary and Binary Expressions.

  54220.                                                              (line    6)

  54221. * m in constraint:                       Simple Constraints. (line   17)

  54222. * machine attributes:                    Target Attributes.  (line    6)

  54223. * machine description macros:            Target Macros.      (line    6)

  54224. * machine descriptions:                  Machine Desc.       (line    6)

  54225. * machine mode conversions:              Conversions.        (line    6)

  54226. * machine mode wrapper classes:          Machine Modes.      (line  286)

  54227. * machine modes:                         Machine Modes.      (line    6)

  54228. * machine specific constraints:          Machine Constraints.

  54229.                                                              (line    6)

  54230. * machine-independent predicates:        Machine-Independent Predicates.

  54231.                                                              (line    6)

  54232. * machine_mode:                          Machine Modes.      (line    6)

  54233. * MACH_DEP_SECTION_ASM_FLAG:             Sections.           (line  120)

  54234. * macros, target description:            Target Macros.      (line    6)

  54235. * maddMN4 instruction pattern:           Standard Names.     (line  761)

  54236. * makefile fragment:                     Fragments.          (line    6)

  54237. * makefile targets:                      Makefile.           (line    6)

  54238. * MAKE_DECL_ONE_ONLY:                    Label Output.       (line  281)

  54239. * make_safe_from:                        Expander Definitions.

  54240.                                                              (line  151)

  54241. * MALLOC_ABI_ALIGNMENT:                  Storage Layout.     (line  190)

  54242. * Manipulating GIMPLE statements:        Manipulating GIMPLE statements.

  54243.                                                              (line    6)

  54244. * marking roots:                         GGC Roots.          (line    6)

  54245. * maskloadMN instruction pattern:        Standard Names.     (line  396)

  54246. * maskstoreMN instruction pattern:       Standard Names.     (line  403)

  54247. * mask_fold_left_plus_M instruction pattern: Standard Names. (line  564)

  54248. * mask_gather_loadMN instruction pattern: Standard Names.    (line  249)

  54249. * MASK_RETURN_ADDR:                      Exception Region Output.

  54250.                                                              (line   35)

  54251. * mask_scatter_storeMN instruction pattern: Standard Names.  (line  272)

  54252. * Match and Simplify:                    Match and Simplify. (line    6)

  54253. * matching constraint:                   Simple Constraints. (line  140)

  54254. * matching operands:                     Output Template.    (line   49)

  54255. * match_dup:                             RTL Template.       (line   73)

  54256. * match_dup <1>:                         define_peephole2.   (line   28)

  54257. * match_dup and attributes:              Insn Lengths.       (line   16)

  54258. * match_operand:                         RTL Template.       (line   16)

  54259. * match_operand and attributes:          Expressions.        (line   55)

  54260. * match_operator:                        RTL Template.       (line   95)

  54261. * match_op_dup:                          RTL Template.       (line  163)

  54262. * match_parallel:                        RTL Template.       (line  172)

  54263. * match_par_dup:                         RTL Template.       (line  219)

  54264. * match_scratch:                         RTL Template.       (line   58)

  54265. * match_scratch <1>:                     define_peephole2.   (line   28)

  54266. * match_test and attributes:             Expressions.        (line   64)

  54267. * math library:                          Soft float library routines.

  54268.                                                              (line    6)

  54269. * math, in RTL:                          Arithmetic.         (line    6)

  54270. * matherr:                               Library Calls.      (line   59)

  54271. * MATH_LIBRARY:                          Misc.               (line  536)

  54272. * maxM3 instruction pattern:             Standard Names.     (line  503)

  54273. * MAX_BITSIZE_MODE_ANY_INT:              Machine Modes.      (line  444)

  54274. * MAX_BITSIZE_MODE_ANY_MODE:             Machine Modes.      (line  450)

  54275. * MAX_BITS_PER_WORD:                     Storage Layout.     (line   54)

  54276. * MAX_CONDITIONAL_EXECUTE:               Misc.               (line  558)

  54277. * MAX_FIXED_MODE_SIZE:                   Storage Layout.     (line  466)

  54278. * MAX_MOVE_MAX:                          Misc.               (line  127)

  54279. * MAX_OFILE_ALIGNMENT:                   Storage Layout.     (line  228)

  54280. * MAX_REGS_PER_ADDRESS:                  Addressing Modes.   (line   42)

  54281. * MAX_STACK_ALIGNMENT:                   Storage Layout.     (line  222)

  54282. * maybe_undef:                           GTY Options.        (line  141)

  54283. * may_trap_p, tree_could_trap_p:         Edges.              (line  114)

  54284. * mcount:                                Profiling.          (line   12)

  54285. * MD_EXEC_PREFIX:                        Driver.             (line  271)

  54286. * MD_FALLBACK_FRAME_STATE_FOR:           Exception Handling. (line   93)

  54287. * MD_HANDLE_UNWABI:                      Exception Handling. (line  112)

  54288. * MD_STARTFILE_PREFIX:                   Driver.             (line  299)

  54289. * MD_STARTFILE_PREFIX_1:                 Driver.             (line  304)

  54290. * mem:                                   Regs and Memory.    (line  396)

  54291. * mem and /c:                            Flags.              (line   70)

  54292. * mem and /f:                            Flags.              (line   74)

  54293. * mem and /j:                            Flags.              (line   59)

  54294. * mem and /u:                            Flags.              (line   78)

  54295. * mem and /v:                            Flags.              (line   65)

  54296. * mem, RTL sharing:                      Sharing.            (line   43)

  54297. * memory model:                          Memory model.       (line    6)

  54298. * memory reference, nonoffsettable:      Simple Constraints. (line  254)

  54299. * memory references in constraints:      Simple Constraints. (line   17)

  54300. * memory_barrier instruction pattern:    Standard Names.     (line 2172)

  54301. * memory_blockage instruction pattern:   Standard Names.     (line 2163)

  54302. * MEMORY_MOVE_COST:                      Costs.              (line   53)

  54303. * memory_operand:                        Machine-Independent Predicates.

  54304.                                                              (line   57)

  54305. * MEM_ADDR_SPACE:                        Special Accessors.  (line   48)

  54306. * MEM_ALIAS_SET:                         Special Accessors.  (line    9)

  54307. * MEM_ALIGN:                             Special Accessors.  (line   45)

  54308. * MEM_EXPR:                              Special Accessors.  (line   19)

  54309. * MEM_KEEP_ALIAS_SET_P:                  Flags.              (line   59)

  54310. * MEM_NOTRAP_P:                          Flags.              (line   70)

  54311. * MEM_OFFSET:                            Special Accessors.  (line   31)

  54312. * MEM_OFFSET_KNOWN_P:                    Special Accessors.  (line   27)

  54313. * MEM_POINTER:                           Flags.              (line   74)

  54314. * MEM_READONLY_P:                        Flags.              (line   78)

  54315. * MEM_REF:                               Storage References. (line    6)

  54316. * MEM_SIZE:                              Special Accessors.  (line   39)

  54317. * MEM_SIZE_KNOWN_P:                      Special Accessors.  (line   35)

  54318. * mem_thread_fence instruction pattern:  Standard Names.     (line 2462)

  54319. * MEM_VOLATILE_P:                        Flags.              (line   65)

  54320. * METHOD_TYPE:                           Types.              (line    6)

  54321. * MINIMUM_ALIGNMENT:                     Storage Layout.     (line  334)

  54322. * MINIMUM_ATOMIC_ALIGNMENT:              Storage Layout.     (line  198)

  54323. * minM3 instruction pattern:             Standard Names.     (line  503)

  54324. * minus:                                 Arithmetic.         (line   38)

  54325. * minus and attributes:                  Expressions.        (line   83)

  54326. * minus, canonicalization of:            Insn Canonicalizations.

  54327.                                                              (line   27)

  54328. * MINUS_EXPR:                            Unary and Binary Expressions.

  54329.                                                              (line    6)

  54330. * MIN_UNITS_PER_WORD:                    Storage Layout.     (line   64)

  54331. * MIPS coprocessor-definition macros:    MIPS Coprocessors.  (line    6)

  54332. * miscellaneous register hooks:          Miscellaneous Register Hooks.

  54333.                                                              (line    6)

  54334. * mnemonic attribute:                    Mnemonic Attribute. (line    6)

  54335. * mod:                                   Arithmetic.         (line  136)

  54336. * mod and attributes:                    Expressions.        (line   83)

  54337. * mode classes:                          Machine Modes.      (line  226)

  54338. * mode iterators in .md files:           Mode Iterators.     (line    6)

  54339. * mode switching:                        Mode Switching.     (line    6)

  54340. * MODE_ACCUM:                            Machine Modes.      (line  256)

  54341. * MODE_BASE_REG_CLASS:                   Register Classes.   (line  116)

  54342. * MODE_BASE_REG_REG_CLASS:               Register Classes.   (line  122)

  54343. * MODE_CC:                               Machine Modes.      (line  271)

  54344. * MODE_CC <1>:                           MODE_CC Condition Codes.

  54345.                                                              (line    6)

  54346. * MODE_CODE_BASE_REG_CLASS:              Register Classes.   (line  129)

  54347. * MODE_COMPLEX_FLOAT:                    Machine Modes.      (line  267)

  54348. * MODE_COMPLEX_INT:                      Machine Modes.      (line  264)

  54349. * MODE_DECIMAL_FLOAT:                    Machine Modes.      (line  244)

  54350. * MODE_FLOAT:                            Machine Modes.      (line  240)

  54351. * MODE_FRACT:                            Machine Modes.      (line  248)

  54352. * MODE_INT:                              Machine Modes.      (line  232)

  54353. * MODE_PARTIAL_INT:                      Machine Modes.      (line  236)

  54354. * MODE_POINTER_BOUNDS:                   Machine Modes.      (line  276)

  54355. * MODE_RANDOM:                           Machine Modes.      (line  281)

  54356. * MODE_UACCUM:                           Machine Modes.      (line  260)

  54357. * MODE_UFRACT:                           Machine Modes.      (line  252)

  54358. * modifiers in constraints:              Modifiers.          (line    6)

  54359. * MODIFY_EXPR:                           Unary and Binary Expressions.

  54360.                                                              (line    6)

  54361. * modM3 instruction pattern:             Standard Names.     (line  442)

  54362. * modulo scheduling:                     RTL passes.         (line  123)

  54363. * MOVE_MAX:                              Misc.               (line  122)

  54364. * MOVE_MAX_PIECES:                       Costs.              (line  210)

  54365. * MOVE_RATIO:                            Costs.              (line  149)

  54366. * movM instruction pattern:              Standard Names.     (line   11)

  54367. * movmemM instruction pattern:           Standard Names.     (line 1281)

  54368. * movmisalignM instruction pattern:      Standard Names.     (line  125)

  54369. * movMODEcc instruction pattern:         Standard Names.     (line 1576)

  54370. * movstr instruction pattern:            Standard Names.     (line 1317)

  54371. * movstrictM instruction pattern:        Standard Names.     (line  119)

  54372. * msubMN4 instruction pattern:           Standard Names.     (line  784)

  54373. * mulhisi3 instruction pattern:          Standard Names.     (line  737)

  54374. * mulM3 instruction pattern:             Standard Names.     (line  442)

  54375. * mulqihi3 instruction pattern:          Standard Names.     (line  741)

  54376. * mulsidi3 instruction pattern:          Standard Names.     (line  741)

  54377. * mult:                                  Arithmetic.         (line   93)

  54378. * mult and attributes:                   Expressions.        (line   83)

  54379. * mult, canonicalization of:             Insn Canonicalizations.

  54380.                                                              (line   27)

  54381. * mult, canonicalization of <1>:         Insn Canonicalizations.

  54382.                                                              (line  107)

  54383. * MULTIARCH_DIRNAME:                     Target Fragment.    (line  173)

  54384. * MULTILIB_DEFAULTS:                     Driver.             (line  256)

  54385. * MULTILIB_DIRNAMES:                     Target Fragment.    (line   44)

  54386. * MULTILIB_EXCEPTIONS:                   Target Fragment.    (line   70)

  54387. * MULTILIB_EXTRA_OPTS:                   Target Fragment.    (line  135)

  54388. * MULTILIB_MATCHES:                      Target Fragment.    (line   63)

  54389. * MULTILIB_OPTIONS:                      Target Fragment.    (line   24)

  54390. * MULTILIB_OSDIRNAMES:                   Target Fragment.    (line  142)

  54391. * MULTILIB_REQUIRED:                     Target Fragment.    (line   82)

  54392. * MULTILIB_REUSE:                        Target Fragment.    (line  103)

  54393. * multiple alternative constraints:      Multi-Alternative.  (line    6)

  54394. * MULTIPLE_SYMBOL_SPACES:                Misc.               (line  515)

  54395. * multiplication:                        Arithmetic.         (line   93)

  54396. * multiplication with signed saturation: Arithmetic.         (line   93)

  54397. * multiplication with unsigned saturation: Arithmetic.       (line   93)

  54398. * MULT_EXPR:                             Unary and Binary Expressions.

  54399.                                                              (line    6)

  54400. * MULT_HIGHPART_EXPR:                    Unary and Binary Expressions.

  54401.                                                              (line    6)

  54402. * mulvM4 instruction pattern:            Standard Names.     (line  458)

  54403. * n in constraint:                       Simple Constraints. (line   73)

  54404. * name:                                  Identifiers.        (line    6)

  54405. * named address spaces:                  Named Address Spaces.

  54406.                                                              (line    6)

  54407. * named patterns and conditions:         Patterns.           (line   61)

  54408. * names, pattern:                        Standard Names.     (line    6)

  54409. * namespace, scope:                      Namespaces.         (line    6)

  54410. * NAMESPACE_DECL:                        Declarations.       (line    6)

  54411. * NAMESPACE_DECL <1>:                    Namespaces.         (line    6)

  54412. * NATIVE_SYSTEM_HEADER_COMPONENT:        Driver.             (line  326)

  54413. * ne:                                    Comparisons.        (line   56)

  54414. * ne and attributes:                     Expressions.        (line   83)

  54415. * nearbyintM2 instruction pattern:       Standard Names.     (line 1106)

  54416. * neg:                                   Arithmetic.         (line   82)

  54417. * neg and attributes:                    Expressions.        (line   83)

  54418. * neg, canonicalization of:              Insn Canonicalizations.

  54419.                                                              (line   27)

  54420. * NEGATE_EXPR:                           Unary and Binary Expressions.

  54421.                                                              (line    6)

  54422. * negation:                              Arithmetic.         (line   82)

  54423. * negation with signed saturation:       Arithmetic.         (line   82)

  54424. * negation with unsigned saturation:     Arithmetic.         (line   82)

  54425. * negM2 instruction pattern:             Standard Names.     (line  872)

  54426. * negMODEcc instruction pattern:         Standard Names.     (line 1645)

  54427. * negvM3 instruction pattern:            Standard Names.     (line  875)

  54428. * nested functions, support for:         Trampolines.        (line    6)

  54429. * nested_ptr:                            GTY Options.        (line  149)

  54430. * next_bb, prev_bb, FOR_EACH_BB, FOR_ALL_BB: Basic Blocks.   (line   25)

  54431. * NEXT_INSN:                             Insns.              (line   30)

  54432. * NEXT_OBJC_RUNTIME:                     Library Calls.      (line   86)

  54433. * NE_EXPR:                               Unary and Binary Expressions.

  54434.                                                              (line    6)

  54435. * nil:                                   RTL Objects.        (line   73)

  54436. * NM_FLAGS:                              Macros for Initialization.

  54437.                                                              (line  110)

  54438. * nondeterministic finite state automaton: Processor pipeline description.

  54439.                                                              (line  304)

  54440. * nonimmediate_operand:                  Machine-Independent Predicates.

  54441.                                                              (line  100)

  54442. * nonlocal goto handler:                 Edges.              (line  171)

  54443. * nonlocal_goto instruction pattern:     Standard Names.     (line 1998)

  54444. * nonlocal_goto_receiver instruction pattern: Standard Names.

  54445.                                                              (line 2015)

  54446. * nonmemory_operand:                     Machine-Independent Predicates.

  54447.                                                              (line   96)

  54448. * nonoffsettable memory reference:       Simple Constraints. (line  254)

  54449. * NON_LVALUE_EXPR:                       Unary and Binary Expressions.

  54450.                                                              (line    6)

  54451. * nop instruction pattern:               Standard Names.     (line 1821)

  54452. * NOP_EXPR:                              Unary and Binary Expressions.

  54453.                                                              (line    6)

  54454. * normal predicates:                     Predicates.         (line   31)

  54455. * not:                                   Arithmetic.         (line  154)

  54456. * not and attributes:                    Expressions.        (line   50)

  54457. * not equal:                             Comparisons.        (line   56)

  54458. * not, canonicalization of:              Insn Canonicalizations.

  54459.                                                              (line   27)

  54460. * note:                                  Insns.              (line  183)

  54461. * note and /i:                           Flags.              (line   48)

  54462. * note and /v:                           Flags.              (line   33)

  54463. * NOTE_INSN_BASIC_BLOCK:                 Basic Blocks.       (line   50)

  54464. * NOTE_INSN_BASIC_BLOCK <1>:             Basic Blocks.       (line   52)

  54465. * NOTE_INSN_BEGIN_STMT:                  Insns.              (line  233)

  54466. * NOTE_INSN_BLOCK_BEG:                   Insns.              (line  208)

  54467. * NOTE_INSN_BLOCK_END:                   Insns.              (line  208)

  54468. * NOTE_INSN_DELETED:                     Insns.              (line  198)

  54469. * NOTE_INSN_DELETED_LABEL:               Insns.              (line  203)

  54470. * NOTE_INSN_EH_REGION_BEG:               Insns.              (line  214)

  54471. * NOTE_INSN_EH_REGION_END:               Insns.              (line  214)

  54472. * NOTE_INSN_FUNCTION_BEG:                Insns.              (line  221)

  54473. * NOTE_INSN_INLINE_ENTRY:                Insns.              (line  238)

  54474. * NOTE_INSN_VAR_LOCATION:                Insns.              (line  225)

  54475. * NOTE_LINE_NUMBER:                      Insns.              (line  183)

  54476. * NOTE_SOURCE_FILE:                      Insns.              (line  183)

  54477. * NOTE_VAR_LOCATION:                     Insns.              (line  225)

  54478. * NOTICE_UPDATE_CC:                      CC0 Condition Codes.

  54479.                                                              (line   30)

  54480. * notMODEcc instruction pattern:         Standard Names.     (line 1652)

  54481. * NO_DBX_BNSYM_ENSYM:                    DBX Hooks.          (line   25)

  54482. * NO_DBX_FUNCTION_END:                   DBX Hooks.          (line   19)

  54483. * NO_DBX_GCC_MARKER:                     File Names and DBX. (line   27)

  54484. * NO_DBX_MAIN_SOURCE_DIRECTORY:          File Names and DBX. (line   22)

  54485. * NO_DOLLAR_IN_LABEL:                    Label Output.       (line   64)

  54486. * NO_DOT_IN_LABEL:                       Label Output.       (line   70)

  54487. * NO_FUNCTION_CSE:                       Costs.              (line  289)

  54488. * NO_PROFILE_COUNTERS:                   Profiling.          (line   27)

  54489. * NO_REGS:                               Register Classes.   (line   17)

  54490. * Number of iterations analysis:         Number of iterations.

  54491.                                                              (line    6)

  54492. * NUM_MACHINE_MODES:                     Machine Modes.      (line  383)

  54493. * NUM_MODES_FOR_MODE_SWITCHING:          Mode Switching.     (line   30)

  54494. * NUM_POLY_INT_COEFFS:                   Overview of poly_int.

  54495.                                                              (line   24)

  54496. * N_REG_CLASSES:                         Register Classes.   (line   81)

  54497. * o in constraint:                       Simple Constraints. (line   23)

  54498. * OACC_CACHE:                            OpenACC.            (line    6)

  54499. * OACC_DATA:                             OpenACC.            (line    6)

  54500. * OACC_DECLARE:                          OpenACC.            (line    6)

  54501. * OACC_ENTER_DATA:                       OpenACC.            (line    6)

  54502. * OACC_EXIT_DATA:                        OpenACC.            (line    6)

  54503. * OACC_HOST_DATA:                        OpenACC.            (line    6)

  54504. * OACC_KERNELS:                          OpenACC.            (line    6)

  54505. * OACC_LOOP:                             OpenACC.            (line    6)

  54506. * OACC_PARALLEL:                         OpenACC.            (line    6)

  54507. * OACC_SERIAL:                           OpenACC.            (line    6)

  54508. * OACC_UPDATE:                           OpenACC.            (line    6)

  54509. * OBJC_GEN_METHOD_LABEL:                 Label Output.       (line  482)

  54510. * OBJC_JBLEN:                            Misc.               (line 1042)

  54511. * OBJECT_FORMAT_COFF:                    Macros for Initialization.

  54512.                                                              (line   96)

  54513. * offsettable address:                   Simple Constraints. (line   23)

  54514. * OFFSET_TYPE:                           Types.              (line    6)

  54515. * OImode:                                Machine Modes.      (line   51)

  54516. * OMP_ATOMIC:                            OpenMP.             (line    6)

  54517. * OMP_CLAUSE:                            OpenMP.             (line    6)

  54518. * OMP_CONTINUE:                          OpenMP.             (line    6)

  54519. * OMP_CRITICAL:                          OpenMP.             (line    6)

  54520. * OMP_FOR:                               OpenMP.             (line    6)

  54521. * OMP_MASTER:                            OpenMP.             (line    6)

  54522. * OMP_ORDERED:                           OpenMP.             (line    6)

  54523. * OMP_PARALLEL:                          OpenMP.             (line    6)

  54524. * OMP_RETURN:                            OpenMP.             (line    6)

  54525. * OMP_SECTION:                           OpenMP.             (line    6)

  54526. * OMP_SECTIONS:                          OpenMP.             (line    6)

  54527. * OMP_SINGLE:                            OpenMP.             (line    6)

  54528. * one_cmplM2 instruction pattern:        Standard Names.     (line 1241)

  54529. * operand access:                        Accessors.          (line    6)

  54530. * Operand Access Routines:               SSA Operands.       (line  116)

  54531. * operand constraints:                   Constraints.        (line    6)

  54532. * Operand Iterators:                     SSA Operands.       (line  116)

  54533. * operand predicates:                    Predicates.         (line    6)

  54534. * operand substitution:                  Output Template.    (line    6)

  54535. * Operands:                              Operands.           (line    6)

  54536. * operands:                              SSA Operands.       (line    6)

  54537. * operands <1>:                          Patterns.           (line   67)

  54538. * operator predicates:                   Predicates.         (line    6)

  54539. * optc-gen.awk:                          Options.            (line    6)

  54540. * OPTGROUP_ALL:                          Optimization groups.

  54541.                                                              (line   28)

  54542. * OPTGROUP_INLINE:                       Optimization groups.

  54543.                                                              (line   15)

  54544. * OPTGROUP_IPA:                          Optimization groups.

  54545.                                                              (line    9)

  54546. * OPTGROUP_LOOP:                         Optimization groups.

  54547.                                                              (line   12)

  54548. * OPTGROUP_OMP:                          Optimization groups.

  54549.                                                              (line   18)

  54550. * OPTGROUP_OTHER:                        Optimization groups.

  54551.                                                              (line   24)

  54552. * OPTGROUP_VEC:                          Optimization groups.

  54553.                                                              (line   21)

  54554. * optimization dumps:                    Optimization info.  (line    6)

  54555. * optimization groups:                   Optimization groups.

  54556.                                                              (line    6)

  54557. * optimization info file names:          Dump files and streams.

  54558.                                                              (line    6)

  54559. * Optimization infrastructure for GIMPLE: Tree SSA.          (line    6)

  54560. * OPTIMIZE_MODE_SWITCHING:               Mode Switching.     (line    8)

  54561. * option specification files:            Options.            (line    6)

  54562. * optional hardware or system features:  Run-time Target.    (line   59)

  54563. * options, directory search:             Including Patterns. (line   47)

  54564. * options, guidelines for:               Guidelines for Options.

  54565.                                                              (line    6)

  54566. * OPTION_DEFAULT_SPECS:                  Driver.             (line   25)

  54567. * opt_mode:                              Machine Modes.      (line  322)

  54568. * order of register allocation:          Allocation Order.   (line    6)

  54569. * ordered_comparison_operator:           Machine-Independent Predicates.

  54570.                                                              (line  115)

  54571. * ORDERED_EXPR:                          Unary and Binary Expressions.

  54572.                                                              (line    6)

  54573. * Ordering of Patterns:                  Pattern Ordering.   (line    6)

  54574. * ORIGINAL_REGNO:                        Special Accessors.  (line   53)

  54575. * other register constraints:            Simple Constraints. (line  171)

  54576. * outgoing_args_size:                    Stack Arguments.    (line   48)

  54577. * OUTGOING_REGNO:                        Register Basics.    (line  136)

  54578. * OUTGOING_REG_PARM_STACK_SPACE:         Stack Arguments.    (line   79)

  54579. * output of assembler code:              File Framework.     (line    6)

  54580. * output statements:                     Output Statement.   (line    6)

  54581. * output templates:                      Output Template.    (line    6)

  54582. * output_asm_insn:                       Output Statement.   (line   52)

  54583. * OUTPUT_QUOTED_STRING:                  File Framework.     (line  105)

  54584. * OVERLAPPING_REGISTER_NAMES:            Instruction Output. (line   20)

  54585. * OVERLOAD:                              Functions for C++.  (line    6)

  54586. * OVERRIDE_ABI_FORMAT:                   Register Arguments. (line  150)

  54587. * OVL_CURRENT:                           Functions for C++.  (line    6)

  54588. * OVL_NEXT:                              Functions for C++.  (line    6)

  54589. * p in constraint:                       Simple Constraints. (line  162)

  54590. * PAD_VARARGS_DOWN:                      Register Arguments. (line  230)

  54591. * parallel:                              Side Effects.       (line  210)

  54592. * parameters, c++ abi:                   C++ ABI.            (line    6)

  54593. * parameters, d abi:                     D Language and ABI. (line    6)

  54594. * parameters, miscellaneous:             Misc.               (line    6)

  54595. * parameters, precompiled headers:       PCH Target.         (line    6)

  54596. * parity:                                Arithmetic.         (line  242)

  54597. * parityM2 instruction pattern:          Standard Names.     (line 1228)

  54598. * PARM_BOUNDARY:                         Storage Layout.     (line  150)

  54599. * PARM_DECL:                             Declarations.       (line    6)

  54600. * PARSE_LDD_OUTPUT:                      Macros for Initialization.

  54601.                                                              (line  125)

  54602. * pass dumps:                            Passes.             (line    6)

  54603. * passes and files of the compiler:      Passes.             (line    6)

  54604. * passing arguments:                     Interface.          (line   36)

  54605. * pass_duplicate_computed_gotos:         Edges.              (line  161)

  54606. * PATH_SEPARATOR:                        Filesystem.         (line   31)

  54607. * PATTERN:                               Insns.              (line  307)

  54608. * pattern conditions:                    Patterns.           (line   55)

  54609. * pattern names:                         Standard Names.     (line    6)

  54610. * Pattern Ordering:                      Pattern Ordering.   (line    6)

  54611. * patterns:                              Patterns.           (line    6)

  54612. * pc:                                    Regs and Memory.    (line  383)

  54613. * pc and attributes:                     Insn Lengths.       (line   20)

  54614. * pc, RTL sharing:                       Sharing.            (line   28)

  54615. * PCC_BITFIELD_TYPE_MATTERS:             Storage Layout.     (line  360)

  54616. * PCC_STATIC_STRUCT_RETURN:              Aggregate Return.   (line   64)

  54617. * PC_REGNUM:                             Register Basics.    (line  150)

  54618. * pc_rtx:                                Regs and Memory.    (line  388)

  54619. * PDImode:                               Machine Modes.      (line   40)

  54620. * peephole optimization, RTL representation: Side Effects.   (line  244)

  54621. * peephole optimizer definitions:        Peephole Definitions.

  54622.                                                              (line    6)

  54623. * per-function data:                     Per-Function Data.  (line    6)

  54624. * percent sign:                          Output Template.    (line    6)

  54625. * PHI nodes:                             SSA.                (line   31)

  54626. * PIC:                                   PIC.                (line    6)

  54627. * PIC_OFFSET_TABLE_REGNUM:               PIC.                (line   15)

  54628. * PIC_OFFSET_TABLE_REG_CALL_CLOBBERED:   PIC.                (line   25)

  54629. * pipeline hazard recognizer:            Processor pipeline description.

  54630.                                                              (line    6)

  54631. * pipeline hazard recognizer <1>:        Processor pipeline description.

  54632.                                                              (line   53)

  54633. * Plugins:                               Plugins.            (line    6)

  54634. * plus:                                  Arithmetic.         (line   14)

  54635. * plus and attributes:                   Expressions.        (line   83)

  54636. * plus, canonicalization of:             Insn Canonicalizations.

  54637.                                                              (line   27)

  54638. * PLUS_EXPR:                             Unary and Binary Expressions.

  54639.                                                              (line    6)

  54640. * Pmode:                                 Misc.               (line  363)

  54641. * pmode_register_operand:                Machine-Independent Predicates.

  54642.                                                              (line   34)

  54643. * pointer:                               Types.              (line    6)

  54644. * POINTERS_EXTEND_UNSIGNED:              Storage Layout.     (line   76)

  54645. * POINTER_DIFF_EXPR:                     Unary and Binary Expressions.

  54646.                                                              (line    6)

  54647. * POINTER_PLUS_EXPR:                     Unary and Binary Expressions.

  54648.                                                              (line    6)

  54649. * POINTER_SIZE:                          Storage Layout.     (line   70)

  54650. * POINTER_TYPE:                          Types.              (line    6)

  54651. * polynomial integers:                   poly_int.           (line    6)

  54652. * poly_int:                              poly_int.           (line    6)

  54653. * poly_int, invariant range:             Overview of poly_int.

  54654.                                                              (line   31)

  54655. * poly_int, main typedefs:               Overview of poly_int.

  54656.                                                              (line   46)

  54657. * poly_int, runtime value:               Overview of poly_int.

  54658.                                                              (line    6)

  54659. * poly_int, template parameters:         Overview of poly_int.

  54660.                                                              (line   24)

  54661. * poly_int, use in target-independent code: Consequences of using poly_int.

  54662.                                                              (line   32)

  54663. * poly_int, use in target-specific code: Consequences of using poly_int.

  54664.                                                              (line   40)

  54665. * POLY_INT_CST:                          Constant expressions.

  54666.                                                              (line    6)

  54667. * popcount:                              Arithmetic.         (line  238)

  54668. * popcountM2 instruction pattern:        Standard Names.     (line 1216)

  54669. * pops_args:                             Function Entry.     (line  111)

  54670. * pop_operand:                           Machine-Independent Predicates.

  54671.                                                              (line   87)

  54672. * portability:                           Portability.        (line    6)

  54673. * position independent code:             PIC.                (line    6)

  54674. * POSTDECREMENT_EXPR:                    Unary and Binary Expressions.

  54675.                                                              (line    6)

  54676. * POSTINCREMENT_EXPR:                    Unary and Binary Expressions.

  54677.                                                              (line    6)

  54678. * post_dec:                              Incdec.             (line   25)

  54679. * post_inc:                              Incdec.             (line   30)

  54680. * POST_LINK_SPEC:                        Driver.             (line  236)

  54681. * post_modify:                           Incdec.             (line   33)

  54682. * post_order_compute, inverted_post_order_compute, walk_dominator_tree: Basic Blocks.

  54683.                                                              (line   34)

  54684. * POWI_MAX_MULTS:                        Misc.               (line  927)

  54685. * powM3 instruction pattern:             Standard Names.     (line 1054)

  54686. * pragma:                                Misc.               (line  420)

  54687. * PREDECREMENT_EXPR:                     Unary and Binary Expressions.

  54688.                                                              (line    6)

  54689. * predefined macros:                     Run-time Target.    (line    6)

  54690. * predicates:                            Predicates.         (line    6)

  54691. * predicates and machine modes:          Predicates.         (line   31)

  54692. * predication:                           Conditional Execution.

  54693.                                                              (line    6)

  54694. * predict.def:                           Profile information.

  54695.                                                              (line   24)

  54696. * PREFERRED_DEBUGGING_TYPE:              All Debuggers.      (line   40)

  54697. * PREFERRED_RELOAD_CLASS:                Register Classes.   (line  249)

  54698. * PREFERRED_STACK_BOUNDARY:              Storage Layout.     (line  164)

  54699. * prefetch:                              Side Effects.       (line  324)

  54700. * prefetch and /v:                       Flags.              (line   92)

  54701. * prefetch instruction pattern:          Standard Names.     (line 2140)

  54702. * PREFETCH_SCHEDULE_BARRIER_P:           Flags.              (line   92)

  54703. * PREINCREMENT_EXPR:                     Unary and Binary Expressions.

  54704.                                                              (line    6)

  54705. * presence_set:                          Processor pipeline description.

  54706.                                                              (line  223)

  54707. * preserving SSA form:                   SSA.                (line   74)

  54708. * pretend_args_size:                     Function Entry.     (line  117)

  54709. * prev_active_insn:                      define_peephole.    (line   60)

  54710. * PREV_INSN:                             Insns.              (line   26)

  54711. * pre_dec:                               Incdec.             (line    8)

  54712. * PRE_GCC3_DWARF_FRAME_REGISTERS:        Frame Registers.    (line  126)

  54713. * pre_inc:                               Incdec.             (line   22)

  54714. * pre_modify:                            Incdec.             (line   52)

  54715. * PRINT_OPERAND:                         Instruction Output. (line   95)

  54716. * PRINT_OPERAND_ADDRESS:                 Instruction Output. (line  122)

  54717. * PRINT_OPERAND_PUNCT_VALID_P:           Instruction Output. (line  115)

  54718. * probe_stack instruction pattern:       Standard Names.     (line 1990)

  54719. * probe_stack_address instruction pattern: Standard Names.   (line 1983)

  54720. * processor functional units:            Processor pipeline description.

  54721.                                                              (line    6)

  54722. * processor functional units <1>:        Processor pipeline description.

  54723.                                                              (line   68)

  54724. * processor pipeline description:        Processor pipeline description.

  54725.                                                              (line    6)

  54726. * product:                               Arithmetic.         (line   93)

  54727. * profile feedback:                      Profile information.

  54728.                                                              (line   14)

  54729. * profile representation:                Profile information.

  54730.                                                              (line    6)

  54731. * PROFILE_BEFORE_PROLOGUE:               Profiling.          (line   34)

  54732. * PROFILE_HOOK:                          Profiling.          (line   22)

  54733. * profiling, code generation:            Profiling.          (line    6)

  54734. * program counter:                       Regs and Memory.    (line  384)

  54735. * prologue:                              Function Entry.     (line    6)

  54736. * prologue instruction pattern:          Standard Names.     (line 2079)

  54737. * PROMOTE_MODE:                          Storage Layout.     (line   87)

  54738. * pseudo registers:                      Regs and Memory.    (line    9)

  54739. * PSImode:                               Machine Modes.      (line   32)

  54740. * PTRDIFF_TYPE:                          Type Layout.        (line  157)

  54741. * purge_dead_edges:                      Edges.              (line  103)

  54742. * purge_dead_edges <1>:                  Maintaining the CFG.

  54743.                                                              (line   81)

  54744. * push address instruction:              Simple Constraints. (line  162)

  54745. * pushM1 instruction pattern:            Standard Names.     (line  429)

  54746. * PUSH_ARGS:                             Stack Arguments.    (line   17)

  54747. * PUSH_ARGS_REVERSED:                    Stack Arguments.    (line   25)

  54748. * push_operand:                          Machine-Independent Predicates.

  54749.                                                              (line   80)

  54750. * push_reload:                           Addressing Modes.   (line  176)

  54751. * PUSH_ROUNDING:                         Stack Arguments.    (line   31)

  54752. * PUT_CODE:                              RTL Objects.        (line   47)

  54753. * PUT_MODE:                              Machine Modes.      (line  380)

  54754. * PUT_REG_NOTE_KIND:                     Insns.              (line  369)

  54755. * QCmode:                                Machine Modes.      (line  199)

  54756. * QFmode:                                Machine Modes.      (line   57)

  54757. * QImode:                                Machine Modes.      (line   25)

  54758. * QImode, in insn:                       Insns.              (line  291)

  54759. * QQmode:                                Machine Modes.      (line  106)

  54760. * qualified type:                        Types.              (line    6)

  54761. * qualified type <1>:                    Types for C++.      (line    6)

  54762. * querying function unit reservations:   Processor pipeline description.

  54763.                                                              (line   90)

  54764. * question mark:                         Multi-Alternative.  (line   42)

  54765. * quotient:                              Arithmetic.         (line  116)

  54766. * r in constraint:                       Simple Constraints. (line   64)

  54767. * RDIV_EXPR:                             Unary and Binary Expressions.

  54768.                                                              (line    6)

  54769. * READONLY_DATA_SECTION_ASM_OP:          Sections.           (line   62)

  54770. * real operands:                         SSA Operands.       (line    6)

  54771. * REALPART_EXPR:                         Unary and Binary Expressions.

  54772.                                                              (line    6)

  54773. * REAL_CST:                              Constant expressions.

  54774.                                                              (line    6)

  54775. * REAL_LIBGCC_SPEC:                      Driver.             (line  124)

  54776. * REAL_NM_FILE_NAME:                     Macros for Initialization.

  54777.                                                              (line  105)

  54778. * REAL_TYPE:                             Types.              (line    6)

  54779. * REAL_VALUE_ABS:                        Floating Point.     (line   58)

  54780. * REAL_VALUE_ATOF:                       Floating Point.     (line   39)

  54781. * REAL_VALUE_FIX:                        Floating Point.     (line   31)

  54782. * REAL_VALUE_ISINF:                      Floating Point.     (line   49)

  54783. * REAL_VALUE_ISNAN:                      Floating Point.     (line   52)

  54784. * REAL_VALUE_NEGATE:                     Floating Point.     (line   55)

  54785. * REAL_VALUE_NEGATIVE:                   Floating Point.     (line   46)

  54786. * REAL_VALUE_TO_TARGET_DECIMAL128:       Data Output.        (line  153)

  54787. * REAL_VALUE_TO_TARGET_DECIMAL32:        Data Output.        (line  151)

  54788. * REAL_VALUE_TO_TARGET_DECIMAL64:        Data Output.        (line  152)

  54789. * REAL_VALUE_TO_TARGET_DOUBLE:           Data Output.        (line  149)

  54790. * REAL_VALUE_TO_TARGET_LONG_DOUBLE:      Data Output.        (line  150)

  54791. * REAL_VALUE_TO_TARGET_SINGLE:           Data Output.        (line  148)

  54792. * REAL_VALUE_TYPE:                       Floating Point.     (line   25)

  54793. * REAL_VALUE_UNSIGNED_FIX:               Floating Point.     (line   34)

  54794. * recognizing insns:                     RTL Template.       (line    6)

  54795. * recog_data.operand:                    Instruction Output. (line   54)

  54796. * RECORD_TYPE:                           Types.              (line    6)

  54797. * RECORD_TYPE <1>:                       Classes.            (line    6)

  54798. * redirect_edge_and_branch:              Profile information.

  54799.                                                              (line   71)

  54800. * redirect_edge_and_branch, redirect_jump: Maintaining the CFG.

  54801.                                                              (line   89)

  54802. * reduc_and_scal_M instruction pattern:  Standard Names.     (line  535)

  54803. * reduc_ior_scal_M instruction pattern:  Standard Names.     (line  536)

  54804. * reduc_plus_scal_M instruction pattern: Standard Names.     (line  530)

  54805. * reduc_smax_scal_M instruction pattern: Standard Names.     (line  520)

  54806. * reduc_smin_scal_M instruction pattern: Standard Names.     (line  520)

  54807. * reduc_umax_scal_M instruction pattern: Standard Names.     (line  525)

  54808. * reduc_umin_scal_M instruction pattern: Standard Names.     (line  525)

  54809. * reduc_xor_scal_M instruction pattern:  Standard Names.     (line  537)

  54810. * reference:                             Types.              (line    6)

  54811. * REFERENCE_TYPE:                        Types.              (line    6)

  54812. * reg:                                   Regs and Memory.    (line    9)

  54813. * reg and /f:                            Flags.              (line  102)

  54814. * reg and /i:                            Flags.              (line   97)

  54815. * reg and /v:                            Flags.              (line  106)

  54816. * reg, RTL sharing:                      Sharing.            (line   17)

  54817. * register allocation order:             Allocation Order.   (line    6)

  54818. * register class definitions:            Register Classes.   (line    6)

  54819. * register class preference constraints: Class Preferences.  (line    6)

  54820. * register pairs:                        Values in Registers.

  54821.                                                              (line   65)

  54822. * Register Transfer Language (RTL):      RTL.                (line    6)

  54823. * register usage:                        Registers.          (line    6)

  54824. * registers arguments:                   Register Arguments. (line    6)

  54825. * registers in constraints:              Simple Constraints. (line   64)

  54826. * REGISTER_MOVE_COST:                    Costs.              (line    9)

  54827. * REGISTER_NAMES:                        Instruction Output. (line    8)

  54828. * register_operand:                      Machine-Independent Predicates.

  54829.                                                              (line   29)

  54830. * REGISTER_PREFIX:                       Instruction Output. (line  150)

  54831. * REGISTER_TARGET_PRAGMAS:               Misc.               (line  420)

  54832. * REGMODE_NATURAL_SIZE:                  Regs and Memory.    (line  191)

  54833. * REGMODE_NATURAL_SIZE <1>:              Regs and Memory.    (line  268)

  54834. * REGMODE_NATURAL_SIZE <2>:              Values in Registers.

  54835.                                                              (line   46)

  54836. * REGNO_MODE_CODE_OK_FOR_BASE_P:         Register Classes.   (line  172)

  54837. * REGNO_MODE_OK_FOR_BASE_P:              Register Classes.   (line  150)

  54838. * REGNO_MODE_OK_FOR_REG_BASE_P:          Register Classes.   (line  160)

  54839. * REGNO_OK_FOR_BASE_P:                   Register Classes.   (line  146)

  54840. * REGNO_OK_FOR_INDEX_P:                  Register Classes.   (line  186)

  54841. * REGNO_REG_CLASS:                       Register Classes.   (line  105)

  54842. * regs_ever_live:                        Function Entry.     (line   29)

  54843. * regular expressions:                   Processor pipeline description.

  54844.                                                              (line    6)

  54845. * regular expressions <1>:               Processor pipeline description.

  54846.                                                              (line  105)

  54847. * regular IPA passes:                    Regular IPA passes. (line    6)

  54848. * REG_ALLOC_ORDER:                       Allocation Order.   (line    8)

  54849. * REG_BR_PRED:                           Insns.              (line  541)

  54850. * REG_BR_PROB:                           Insns.              (line  533)

  54851. * REG_BR_PROB_BASE, BB_FREQ_BASE, count: Profile information.

  54852.                                                              (line   82)

  54853. * REG_BR_PROB_BASE, EDGE_FREQUENCY:      Profile information.

  54854.                                                              (line   52)

  54855. * REG_CALL_NOCF_CHECK:                   Insns.              (line  557)

  54856. * REG_CC_SETTER:                         Insns.              (line  505)

  54857. * REG_CC_USER:                           Insns.              (line  505)

  54858. * reg_class_contents:                    Register Basics.    (line  102)

  54859. * REG_CLASS_CONTENTS:                    Register Classes.   (line   91)

  54860. * reg_class_for_constraint:              C Constraint Interface.

  54861.                                                              (line   48)

  54862. * REG_CLASS_NAMES:                       Register Classes.   (line   86)

  54863. * REG_DEAD:                              Insns.              (line  380)

  54864. * REG_DEAD, REG_UNUSED:                  Liveness information.

  54865.                                                              (line   32)

  54866. * REG_DEP_ANTI:                          Insns.              (line  527)

  54867. * REG_DEP_OUTPUT:                        Insns.              (line  523)

  54868. * REG_DEP_TRUE:                          Insns.              (line  520)

  54869. * REG_EH_REGION, EDGE_ABNORMAL_CALL:     Edges.              (line  109)

  54870. * REG_EQUAL:                             Insns.              (line  434)

  54871. * REG_EQUIV:                             Insns.              (line  434)

  54872. * REG_EXPR:                              Special Accessors.  (line   58)

  54873. * REG_FRAME_RELATED_EXPR:                Insns.              (line  547)

  54874. * REG_FUNCTION_VALUE_P:                  Flags.              (line   97)

  54875. * REG_INC:                               Insns.              (line  396)

  54876. * reg_label and /v:                      Flags.              (line   54)

  54877. * REG_LABEL_OPERAND:                     Insns.              (line  410)

  54878. * REG_LABEL_TARGET:                      Insns.              (line  419)

  54879. * reg_names:                             Register Basics.    (line  102)

  54880. * reg_names <1>:                         Instruction Output. (line  107)

  54881. * REG_NONNEG:                            Insns.              (line  402)

  54882. * REG_NOTES:                             Insns.              (line  344)

  54883. * REG_NOTE_KIND:                         Insns.              (line  369)

  54884. * REG_OFFSET:                            Special Accessors.  (line   62)

  54885. * REG_OK_STRICT:                         Addressing Modes.   (line   99)

  54886. * REG_PARM_STACK_SPACE:                  Stack Arguments.    (line   58)

  54887. * REG_PARM_STACK_SPACE, and TARGET_FUNCTION_ARG: Register Arguments.

  54888.                                                              (line   49)

  54889. * REG_POINTER:                           Flags.              (line  102)

  54890. * REG_SETJMP:                            Insns.              (line  428)

  54891. * REG_UNUSED:                            Insns.              (line  389)

  54892. * REG_USERVAR_P:                         Flags.              (line  106)

  54893. * REG_VALUE_IN_UNWIND_CONTEXT:           Frame Registers.    (line  156)

  54894. * REG_WORDS_BIG_ENDIAN:                  Storage Layout.     (line   35)

  54895. * relative costs:                        Costs.              (line    6)

  54896. * RELATIVE_PREFIX_NOT_LINKDIR:           Driver.             (line  266)

  54897. * reloading:                             RTL passes.         (line  170)

  54898. * reload_completed:                      Standard Names.     (line 1788)

  54899. * reload_in instruction pattern:         Standard Names.     (line   98)

  54900. * reload_in_progress:                    Standard Names.     (line   57)

  54901. * reload_out instruction pattern:        Standard Names.     (line   98)

  54902. * remainder:                             Arithmetic.         (line  136)

  54903. * remainderM3 instruction pattern:       Standard Names.     (line  908)

  54904. * reorder:                               GTY Options.        (line  175)

  54905. * representation of RTL:                 RTL.                (line    6)

  54906. * reservation delays:                    Processor pipeline description.

  54907.                                                              (line    6)

  54908. * restore_stack_block instruction pattern: Standard Names.   (line 1904)

  54909. * restore_stack_function instruction pattern: Standard Names.

  54910.                                                              (line 1904)

  54911. * restore_stack_nonlocal instruction pattern: Standard Names.

  54912.                                                              (line 1904)

  54913. * rest_of_decl_compilation:              Parsing pass.       (line   51)

  54914. * rest_of_type_compilation:              Parsing pass.       (line   51)

  54915. * RESULT_DECL:                           Declarations.       (line    6)

  54916. * return:                                Side Effects.       (line   72)

  54917. * return instruction pattern:            Standard Names.     (line 1762)

  54918. * return values in registers:            Scalar Return.      (line    6)

  54919. * returning aggregate values:            Aggregate Return.   (line    6)

  54920. * returning structures and unions:       Interface.          (line   10)

  54921. * RETURN_ADDRESS_POINTER_REGNUM:         Frame Registers.    (line   64)

  54922. * RETURN_ADDR_IN_PREVIOUS_FRAME:         Frame Layout.       (line  127)

  54923. * RETURN_ADDR_OFFSET:                    Exception Handling. (line   59)

  54924. * RETURN_ADDR_RTX:                       Frame Layout.       (line  116)

  54925. * RETURN_EXPR:                           Statements for C++. (line    6)

  54926. * RETURN_STMT:                           Statements for C++. (line    6)

  54927. * return_val:                            Flags.              (line  283)

  54928. * return_val, in call_insn:              Flags.              (line  120)

  54929. * return_val, in reg:                    Flags.              (line   97)

  54930. * return_val, in symbol_ref:             Flags.              (line  216)

  54931. * reverse probability:                   Profile information.

  54932.                                                              (line   66)

  54933. * REVERSE_CONDITION:                     MODE_CC Condition Codes.

  54934.                                                              (line   92)

  54935. * REVERSIBLE_CC_MODE:                    MODE_CC Condition Codes.

  54936.                                                              (line   77)

  54937. * right rotate:                          Arithmetic.         (line  195)

  54938. * right shift:                           Arithmetic.         (line  190)

  54939. * rintM2 instruction pattern:            Standard Names.     (line 1114)

  54940. * RISC:                                  Processor pipeline description.

  54941.                                                              (line    6)

  54942. * RISC <1>:                              Processor pipeline description.

  54943.                                                              (line  223)

  54944. * roots, marking:                        GGC Roots.          (line    6)

  54945. * rotate:                                Arithmetic.         (line  195)

  54946. * rotate <1>:                            Arithmetic.         (line  195)

  54947. * rotatert:                              Arithmetic.         (line  195)

  54948. * rotlM3 instruction pattern:            Standard Names.     (line  840)

  54949. * rotrM3 instruction pattern:            Standard Names.     (line  840)

  54950. * roundM2 instruction pattern:           Standard Names.     (line 1087)

  54951. * ROUND_DIV_EXPR:                        Unary and Binary Expressions.

  54952.                                                              (line    6)

  54953. * ROUND_MOD_EXPR:                        Unary and Binary Expressions.

  54954.                                                              (line    6)

  54955. * ROUND_TYPE_ALIGN:                      Storage Layout.     (line  457)

  54956. * RSHIFT_EXPR:                           Unary and Binary Expressions.

  54957.                                                              (line    6)

  54958. * rsqrtM2 instruction pattern:           Standard Names.     (line  888)

  54959. * RTL addition:                          Arithmetic.         (line   14)

  54960. * RTL addition with signed saturation:   Arithmetic.         (line   14)

  54961. * RTL addition with unsigned saturation: Arithmetic.         (line   14)

  54962. * RTL classes:                           RTL Classes.        (line    6)

  54963. * RTL comparison:                        Arithmetic.         (line   46)

  54964. * RTL comparison operations:             Comparisons.        (line    6)

  54965. * RTL constant expression types:         Constants.          (line    6)

  54966. * RTL constants:                         Constants.          (line    6)

  54967. * RTL declarations:                      RTL Declarations.   (line    6)

  54968. * RTL difference:                        Arithmetic.         (line   38)

  54969. * RTL expression:                        RTL Objects.        (line    6)

  54970. * RTL expressions for arithmetic:        Arithmetic.         (line    6)

  54971. * RTL format:                            RTL Classes.        (line   73)

  54972. * RTL format characters:                 RTL Classes.        (line   78)

  54973. * RTL function-call insns:               Calls.              (line    6)

  54974. * RTL insn template:                     RTL Template.       (line    6)

  54975. * RTL integers:                          RTL Objects.        (line    6)

  54976. * RTL memory expressions:                Regs and Memory.    (line    6)

  54977. * RTL object types:                      RTL Objects.        (line    6)

  54978. * RTL postdecrement:                     Incdec.             (line    6)

  54979. * RTL postincrement:                     Incdec.             (line    6)

  54980. * RTL predecrement:                      Incdec.             (line    6)

  54981. * RTL preincrement:                      Incdec.             (line    6)

  54982. * RTL register expressions:              Regs and Memory.    (line    6)

  54983. * RTL representation:                    RTL.                (line    6)

  54984. * RTL side effect expressions:           Side Effects.       (line    6)

  54985. * RTL strings:                           RTL Objects.        (line    6)

  54986. * RTL structure sharing assumptions:     Sharing.            (line    6)

  54987. * RTL subtraction:                       Arithmetic.         (line   38)

  54988. * RTL subtraction with signed saturation: Arithmetic.        (line   38)

  54989. * RTL subtraction with unsigned saturation: Arithmetic.      (line   38)

  54990. * RTL sum:                               Arithmetic.         (line   14)

  54991. * RTL vectors:                           RTL Objects.        (line    6)

  54992. * RTL_CONST_CALL_P:                      Flags.              (line  115)

  54993. * RTL_CONST_OR_PURE_CALL_P:              Flags.              (line  125)

  54994. * RTL_LOOPING_CONST_OR_PURE_CALL_P:      Flags.              (line  129)

  54995. * RTL_PURE_CALL_P:                       Flags.              (line  120)

  54996. * RTX (See RTL):                         RTL Objects.        (line    6)

  54997. * RTX codes, classes of:                 RTL Classes.        (line    6)

  54998. * RTX_FRAME_RELATED_P:                   Flags.              (line  135)

  54999. * run-time conventions:                  Interface.          (line    6)

  55000. * run-time target specification:         Run-time Target.    (line    6)

  55001. * s in constraint:                       Simple Constraints. (line  100)

  55002. * SAD_EXPR:                              Vectors.            (line    6)

  55003. * same_type_p:                           Types.              (line   86)

  55004. * SAmode:                                Machine Modes.      (line  150)

  55005. * satfractMN2 instruction pattern:       Standard Names.     (line 1464)

  55006. * satfractunsMN2 instruction pattern:    Standard Names.     (line 1477)

  55007. * satisfies_constraint_M:                C Constraint Interface.

  55008.                                                              (line   36)

  55009. * sat_fract:                             Conversions.        (line   90)

  55010. * SAVE_EXPR:                             Unary and Binary Expressions.

  55011.                                                              (line    6)

  55012. * save_stack_block instruction pattern:  Standard Names.     (line 1904)

  55013. * save_stack_function instruction pattern: Standard Names.   (line 1904)

  55014. * save_stack_nonlocal instruction pattern: Standard Names.   (line 1904)

  55015. * SBSS_SECTION_ASM_OP:                   Sections.           (line   75)

  55016. * Scalar evolutions:                     Scalar evolutions.  (line    6)

  55017. * scalars, returned as values:           Scalar Return.      (line    6)

  55018. * scalar_float_mode:                     Machine Modes.      (line  293)

  55019. * scalar_int_mode:                       Machine Modes.      (line  290)

  55020. * scalar_mode:                           Machine Modes.      (line  296)

  55021. * scalbM3 instruction pattern:           Standard Names.     (line  915)

  55022. * scatter_storeMN instruction pattern:   Standard Names.     (line  255)

  55023. * SCHED_GROUP_P:                         Flags.              (line  162)

  55024. * SCmode:                                Machine Modes.      (line  199)

  55025. * scratch:                               Regs and Memory.    (line  320)

  55026. * scratch operands:                      Regs and Memory.    (line  320)

  55027. * scratch, RTL sharing:                  Sharing.            (line   38)

  55028. * scratch_operand:                       Machine-Independent Predicates.

  55029.                                                              (line   49)

  55030. * SDATA_SECTION_ASM_OP:                  Sections.           (line   57)

  55031. * sdiv_pow2M3 instruction pattern:       Standard Names.     (line  614)

  55032. * sdiv_pow2M3 instruction pattern <1>:   Standard Names.     (line  615)

  55033. * SDmode:                                Machine Modes.      (line   88)

  55034. * sdot_prodM instruction pattern:        Standard Names.     (line  569)

  55035. * search options:                        Including Patterns. (line   47)

  55036. * SECONDARY_INPUT_RELOAD_CLASS:          Register Classes.   (line  391)

  55037. * SECONDARY_MEMORY_NEEDED_RTX:           Register Classes.   (line  457)

  55038. * SECONDARY_OUTPUT_RELOAD_CLASS:         Register Classes.   (line  392)

  55039. * SECONDARY_RELOAD_CLASS:                Register Classes.   (line  390)

  55040. * SELECT_CC_MODE:                        MODE_CC Condition Codes.

  55041.                                                              (line    6)

  55042. * sequence:                              Side Effects.       (line  259)

  55043. * Sequence iterators:                    Sequence iterators. (line    6)

  55044. * set:                                   Side Effects.       (line   15)

  55045. * set and /f:                            Flags.              (line  135)

  55046. * setmemM instruction pattern:           Standard Names.     (line 1328)

  55047. * SETUP_FRAME_ADDRESSES:                 Frame Layout.       (line   94)

  55048. * SET_ASM_OP:                            Label Output.       (line  451)

  55049. * SET_ASM_OP <1>:                        Label Output.       (line  462)

  55050. * set_attr:                              Tagging Insns.      (line   31)

  55051. * set_attr_alternative:                  Tagging Insns.      (line   49)

  55052. * set_bb_seq:                            GIMPLE sequences.   (line   75)

  55053. * SET_DEST:                              Side Effects.       (line   69)

  55054. * SET_IS_RETURN_P:                       Flags.              (line  171)

  55055. * SET_LABEL_KIND:                        Insns.              (line  146)

  55056. * set_optab_libfunc:                     Library Calls.      (line   15)

  55057. * SET_RATIO:                             Costs.              (line  237)

  55058. * SET_SRC:                               Side Effects.       (line   69)

  55059. * set_thread_pointerMODE instruction pattern: Standard Names.

  55060.                                                              (line 2477)

  55061. * SET_TYPE_STRUCTURAL_EQUALITY:          Types.              (line    6)

  55062. * SET_TYPE_STRUCTURAL_EQUALITY <1>:      Types.              (line   81)

  55063. * SFmode:                                Machine Modes.      (line   69)

  55064. * sharing of RTL components:             Sharing.            (line    6)

  55065. * shift:                                 Arithmetic.         (line  173)

  55066. * SHIFT_COUNT_TRUNCATED:                 Misc.               (line  134)

  55067. * SHLIB_SUFFIX:                          Macros for Initialization.

  55068.                                                              (line  133)

  55069. * SHORT_ACCUM_TYPE_SIZE:                 Type Layout.        (line   82)

  55070. * SHORT_FRACT_TYPE_SIZE:                 Type Layout.        (line   62)

  55071. * SHORT_IMMEDIATES_SIGN_EXTEND:          Misc.               (line  108)

  55072. * SHORT_TYPE_SIZE:                       Type Layout.        (line   15)

  55073. * shrink-wrapping separate components:   Shrink-wrapping separate components.

  55074.                                                              (line    6)

  55075. * sibcall_epilogue instruction pattern:  Standard Names.     (line 2111)

  55076. * sibling call:                          Edges.              (line  121)

  55077. * SIBLING_CALL_P:                        Flags.              (line  175)

  55078. * signal-to-noise ratio (metaphorical usage for diagnostics): Guidelines for Diagnostics.

  55079.                                                              (line   39)

  55080. * signed division:                       Arithmetic.         (line  116)

  55081. * signed division with signed saturation: Arithmetic.        (line  116)

  55082. * signed maximum:                        Arithmetic.         (line  141)

  55083. * signed minimum:                        Arithmetic.         (line  141)

  55084. * significandM2 instruction pattern:     Standard Names.     (line 1047)

  55085. * sign_extend:                           Conversions.        (line   23)

  55086. * sign_extract:                          Bit-Fields.         (line    8)

  55087. * sign_extract, canonicalization of:     Insn Canonicalizations.

  55088.                                                              (line  103)

  55089. * SIG_ATOMIC_TYPE:                       Type Layout.        (line  208)

  55090. * SImode:                                Machine Modes.      (line   37)

  55091. * simple constraints:                    Simple Constraints. (line    6)

  55092. * simple_return:                         Side Effects.       (line   86)

  55093. * simple_return instruction pattern:     Standard Names.     (line 1777)

  55094. * sincosM3 instruction pattern:          Standard Names.     (line  943)

  55095. * sinM2 instruction pattern:             Standard Names.     (line  937)

  55096. * SIZETYPE:                              Type Layout.        (line  147)

  55097. * SIZE_ASM_OP:                           Label Output.       (line   33)

  55098. * SIZE_TYPE:                             Type Layout.        (line  131)

  55099. * skip:                                  GTY Options.        (line   76)

  55100. * SLOW_BYTE_ACCESS:                      Costs.              (line  117)

  55101. * small IPA passes:                      Small IPA passes.   (line    6)

  55102. * smax:                                  Arithmetic.         (line  141)

  55103. * smin:                                  Arithmetic.         (line  141)

  55104. * sms, swing, software pipelining:       RTL passes.         (line  123)

  55105. * smulhrsM3 instruction pattern:         Standard Names.     (line  604)

  55106. * smulhsM3 instruction pattern:          Standard Names.     (line  594)

  55107. * smulM3_highpart instruction pattern:   Standard Names.     (line  753)

  55108. * soft float library:                    Soft float library routines.

  55109.                                                              (line    6)

  55110. * source code, location information:     Guidelines for Diagnostics.

  55111.                                                              (line  159)

  55112. * special:                               GTY Options.        (line  238)

  55113. * special predicates:                    Predicates.         (line   31)

  55114. * SPECS:                                 Target Fragment.    (line  194)

  55115. * speculation_barrier instruction pattern: Standard Names.   (line 2178)

  55116. * speed of instructions:                 Costs.              (line    6)

  55117. * splitting instructions:                Insn Splitting.     (line    6)

  55118. * split_block:                           Maintaining the CFG.

  55119.                                                              (line   96)

  55120. * SQmode:                                Machine Modes.      (line  114)

  55121. * sqrt:                                  Arithmetic.         (line  206)

  55122. * sqrtM2 instruction pattern:            Standard Names.     (line  882)

  55123. * square root:                           Arithmetic.         (line  206)

  55124. * SSA:                                   SSA.                (line    6)

  55125. * ssaddM3 instruction pattern:           Standard Names.     (line  442)

  55126. * ssadM instruction pattern:             Standard Names.     (line  578)

  55127. * ssashlM3 instruction pattern:          Standard Names.     (line  828)

  55128. * SSA_NAME_DEF_STMT:                     SSA.                (line  184)

  55129. * SSA_NAME_VERSION:                      SSA.                (line  189)

  55130. * ssdivM3 instruction pattern:           Standard Names.     (line  442)

  55131. * ssmaddMN4 instruction pattern:         Standard Names.     (line  776)

  55132. * ssmsubMN4 instruction pattern:         Standard Names.     (line  800)

  55133. * ssmulM3 instruction pattern:           Standard Names.     (line  442)

  55134. * ssnegM2 instruction pattern:           Standard Names.     (line  872)

  55135. * sssubM3 instruction pattern:           Standard Names.     (line  442)

  55136. * ss_abs:                                Arithmetic.         (line  200)

  55137. * ss_ashift:                             Arithmetic.         (line  173)

  55138. * ss_div:                                Arithmetic.         (line  116)

  55139. * ss_minus:                              Arithmetic.         (line   38)

  55140. * ss_mult:                               Arithmetic.         (line   93)

  55141. * ss_neg:                                Arithmetic.         (line   82)

  55142. * ss_plus:                               Arithmetic.         (line   14)

  55143. * ss_truncate:                           Conversions.        (line   43)

  55144. * stack arguments:                       Stack Arguments.    (line    6)

  55145. * stack frame layout:                    Frame Layout.       (line    6)

  55146. * stack smashing protection:             Stack Smashing Protection.

  55147.                                                              (line    6)

  55148. * STACK_ALIGNMENT_NEEDED:                Frame Layout.       (line   41)

  55149. * STACK_BOUNDARY:                        Storage Layout.     (line  156)

  55150. * STACK_CHECK_BUILTIN:                   Stack Checking.     (line   31)

  55151. * STACK_CHECK_FIXED_FRAME_SIZE:          Stack Checking.     (line   83)

  55152. * STACK_CHECK_MAX_FRAME_SIZE:            Stack Checking.     (line   74)

  55153. * STACK_CHECK_MAX_VAR_SIZE:              Stack Checking.     (line   90)

  55154. * STACK_CHECK_MOVING_SP:                 Stack Checking.     (line   53)

  55155. * STACK_CHECK_PROBE_INTERVAL_EXP:        Stack Checking.     (line   45)

  55156. * STACK_CHECK_PROTECT:                   Stack Checking.     (line   62)

  55157. * STACK_CHECK_STATIC_BUILTIN:            Stack Checking.     (line   38)

  55158. * STACK_DYNAMIC_OFFSET:                  Frame Layout.       (line   67)

  55159. * STACK_DYNAMIC_OFFSET and virtual registers: Regs and Memory.

  55160.                                                              (line   83)

  55161. * STACK_GROWS_DOWNWARD:                  Frame Layout.       (line    8)

  55162. * STACK_PARMS_IN_REG_PARM_AREA:          Stack Arguments.    (line   89)

  55163. * STACK_POINTER_OFFSET:                  Frame Layout.       (line   51)

  55164. * STACK_POINTER_OFFSET and virtual registers: Regs and Memory.

  55165.                                                              (line   93)

  55166. * STACK_POINTER_REGNUM:                  Frame Registers.    (line    8)

  55167. * STACK_POINTER_REGNUM and virtual registers: Regs and Memory.

  55168.                                                              (line   83)

  55169. * stack_pointer_rtx:                     Frame Registers.    (line  104)

  55170. * stack_protect_combined_set instruction pattern: Standard Names.

  55171.                                                              (line 2487)

  55172. * stack_protect_combined_test instruction pattern: Standard Names.

  55173.                                                              (line 2517)

  55174. * stack_protect_set instruction pattern: Standard Names.     (line 2503)

  55175. * stack_protect_test instruction pattern: Standard Names.    (line 2534)

  55176. * STACK_PUSH_CODE:                       Frame Layout.       (line   12)

  55177. * STACK_REGS:                            Stack Registers.    (line   19)

  55178. * STACK_REG_COVER_CLASS:                 Stack Registers.    (line   22)

  55179. * STACK_SAVEAREA_MODE:                   Storage Layout.     (line  473)

  55180. * STACK_SIZE_MODE:                       Storage Layout.     (line  484)

  55181. * STACK_SLOT_ALIGNMENT:                  Storage Layout.     (line  305)

  55182. * standard pattern names:                Standard Names.     (line    6)

  55183. * STANDARD_STARTFILE_PREFIX:             Driver.             (line  278)

  55184. * STANDARD_STARTFILE_PREFIX_1:           Driver.             (line  285)

  55185. * STANDARD_STARTFILE_PREFIX_2:           Driver.             (line  292)

  55186. * STARTFILE_SPEC:                        Driver.             (line  147)

  55187. * Statement and operand traversals:      Statement and operand traversals.

  55188.                                                              (line    6)

  55189. * Statement Sequences:                   Statement Sequences.

  55190.                                                              (line    6)

  55191. * Statements:                            Statements.         (line    6)

  55192. * statements:                            Function Properties.

  55193.                                                              (line    6)

  55194. * statements <1>:                        Statements for C++. (line    6)

  55195. * static analysis:                       Static Analyzer.    (line    6)

  55196. * static analyzer:                       Static Analyzer.    (line    6)

  55197. * static analyzer, debugging:            Debugging the Analyzer.

  55198.                                                              (line    5)

  55199. * static analyzer, internals:            Analyzer Internals. (line    5)

  55200. * Static profile estimation:             Profile information.

  55201.                                                              (line   24)

  55202. * static single assignment:              SSA.                (line    6)

  55203. * STATIC_CHAIN_INCOMING_REGNUM:          Frame Registers.    (line   77)

  55204. * STATIC_CHAIN_REGNUM:                   Frame Registers.    (line   76)

  55205. * stdarg.h and register arguments:       Register Arguments. (line   44)

  55206. * STDC_0_IN_SYSTEM_HEADERS:              Misc.               (line  384)

  55207. * STMT_EXPR:                             Unary and Binary Expressions.

  55208.                                                              (line    6)

  55209. * STMT_IS_FULL_EXPR_P:                   Statements for C++. (line   22)

  55210. * storage layout:                        Storage Layout.     (line    6)

  55211. * STORE_FLAG_VALUE:                      Misc.               (line  235)

  55212. * STORE_MAX_PIECES:                      Costs.              (line  215)

  55213. * store_multiple instruction pattern:    Standard Names.     (line  159)

  55214. * strcpy:                                Storage Layout.     (line  258)

  55215. * STRICT_ALIGNMENT:                      Storage Layout.     (line  355)

  55216. * strict_low_part:                       RTL Declarations.   (line    9)

  55217. * strict_memory_address_p:               Addressing Modes.   (line  186)

  55218. * STRING_CST:                            Constant expressions.

  55219.                                                              (line    6)

  55220. * STRING_POOL_ADDRESS_P:                 Flags.              (line  179)

  55221. * strlenM instruction pattern:           Standard Names.     (line 1399)

  55222. * structure value address:               Aggregate Return.   (line    6)

  55223. * structures, returning:                 Interface.          (line   10)

  55224. * STRUCTURE_SIZE_BOUNDARY:               Storage Layout.     (line  347)

  55225. * subM3 instruction pattern:             Standard Names.     (line  442)

  55226. * SUBOBJECT:                             Statements for C++. (line    6)

  55227. * SUBOBJECT_CLEANUP:                     Statements for C++. (line    6)

  55228. * subreg:                                Regs and Memory.    (line   97)

  55229. * subreg and /s:                         Flags.              (line  201)

  55230. * subreg and /u:                         Flags.              (line  194)

  55231. * subreg and /u and /v:                  Flags.              (line  184)

  55232. * subreg, in strict_low_part:            RTL Declarations.   (line    9)

  55233. * SUBREG_BYTE:                           Regs and Memory.    (line  311)

  55234. * SUBREG_PROMOTED_UNSIGNED_P:            Flags.              (line  184)

  55235. * SUBREG_PROMOTED_UNSIGNED_SET:          Flags.              (line  194)

  55236. * SUBREG_PROMOTED_VAR_P:                 Flags.              (line  201)

  55237. * SUBREG_REG:                            Regs and Memory.    (line  311)

  55238. * subst iterators in .md files:          Subst Iterators.    (line    6)

  55239. * subvM4 instruction pattern:            Standard Names.     (line  458)

  55240. * SUCCESS_EXIT_CODE:                     Host Misc.          (line   12)

  55241. * support for nested functions:          Trampolines.        (line    6)

  55242. * SUPPORTS_INIT_PRIORITY:                Macros for Initialization.

  55243.                                                              (line   57)

  55244. * SUPPORTS_ONE_ONLY:                     Label Output.       (line  290)

  55245. * SUPPORTS_WEAK:                         Label Output.       (line  264)

  55246. * SWITCHABLE_TARGET:                     Run-time Target.    (line  160)

  55247. * SWITCH_BODY:                           Statements for C++. (line    6)

  55248. * SWITCH_COND:                           Statements for C++. (line    6)

  55249. * SWITCH_STMT:                           Statements for C++. (line    6)

  55250. * symbolic label:                        Sharing.            (line   20)

  55251. * SYMBOL_FLAG_ANCHOR:                    Special Accessors.  (line  117)

  55252. * SYMBOL_FLAG_EXTERNAL:                  Special Accessors.  (line   99)

  55253. * SYMBOL_FLAG_FUNCTION:                  Special Accessors.  (line   92)

  55254. * SYMBOL_FLAG_HAS_BLOCK_INFO:            Special Accessors.  (line  113)

  55255. * SYMBOL_FLAG_LOCAL:                     Special Accessors.  (line   95)

  55256. * SYMBOL_FLAG_SMALL:                     Special Accessors.  (line  104)

  55257. * SYMBOL_FLAG_TLS_SHIFT:                 Special Accessors.  (line  108)

  55258. * symbol_ref:                            Constants.          (line  189)

  55259. * symbol_ref and /f:                     Flags.              (line  179)

  55260. * symbol_ref and /i:                     Flags.              (line  216)

  55261. * symbol_ref and /u:                     Flags.              (line   19)

  55262. * symbol_ref and /v:                     Flags.              (line  220)

  55263. * symbol_ref, RTL sharing:               Sharing.            (line   20)

  55264. * SYMBOL_REF_ANCHOR_P:                   Special Accessors.  (line  117)

  55265. * SYMBOL_REF_BLOCK:                      Special Accessors.  (line  130)

  55266. * SYMBOL_REF_BLOCK_OFFSET:               Special Accessors.  (line  135)

  55267. * SYMBOL_REF_CONSTANT:                   Special Accessors.  (line   78)

  55268. * SYMBOL_REF_DATA:                       Special Accessors.  (line   82)

  55269. * SYMBOL_REF_DECL:                       Special Accessors.  (line   67)

  55270. * SYMBOL_REF_EXTERNAL_P:                 Special Accessors.  (line   99)

  55271. * SYMBOL_REF_FLAG:                       Flags.              (line  220)

  55272. * SYMBOL_REF_FLAG, in TARGET_ENCODE_SECTION_INFO: Sections.  (line  289)

  55273. * SYMBOL_REF_FLAGS:                      Special Accessors.  (line   86)

  55274. * SYMBOL_REF_FUNCTION_P:                 Special Accessors.  (line   92)

  55275. * SYMBOL_REF_HAS_BLOCK_INFO_P:           Special Accessors.  (line  113)

  55276. * SYMBOL_REF_LOCAL_P:                    Special Accessors.  (line   95)

  55277. * SYMBOL_REF_SMALL_P:                    Special Accessors.  (line  104)

  55278. * SYMBOL_REF_TLS_MODEL:                  Special Accessors.  (line  108)

  55279. * SYMBOL_REF_USED:                       Flags.              (line  211)

  55280. * SYMBOL_REF_WEAK:                       Flags.              (line  216)

  55281. * sync_addMODE instruction pattern:      Standard Names.     (line 2232)

  55282. * sync_andMODE instruction pattern:      Standard Names.     (line 2232)

  55283. * sync_compare_and_swapMODE instruction pattern: Standard Names.

  55284.                                                              (line 2192)

  55285. * sync_iorMODE instruction pattern:      Standard Names.     (line 2232)

  55286. * sync_lock_releaseMODE instruction pattern: Standard Names. (line 2297)

  55287. * sync_lock_test_and_setMODE instruction pattern: Standard Names.

  55288.                                                              (line 2271)

  55289. * sync_nandMODE instruction pattern:     Standard Names.     (line 2232)

  55290. * sync_new_addMODE instruction pattern:  Standard Names.     (line 2264)

  55291. * sync_new_andMODE instruction pattern:  Standard Names.     (line 2264)

  55292. * sync_new_iorMODE instruction pattern:  Standard Names.     (line 2264)

  55293. * sync_new_nandMODE instruction pattern: Standard Names.     (line 2264)

  55294. * sync_new_subMODE instruction pattern:  Standard Names.     (line 2264)

  55295. * sync_new_xorMODE instruction pattern:  Standard Names.     (line 2264)

  55296. * sync_old_addMODE instruction pattern:  Standard Names.     (line 2247)

  55297. * sync_old_andMODE instruction pattern:  Standard Names.     (line 2247)

  55298. * sync_old_iorMODE instruction pattern:  Standard Names.     (line 2247)

  55299. * sync_old_nandMODE instruction pattern: Standard Names.     (line 2247)

  55300. * sync_old_subMODE instruction pattern:  Standard Names.     (line 2247)

  55301. * sync_old_xorMODE instruction pattern:  Standard Names.     (line 2247)

  55302. * sync_subMODE instruction pattern:      Standard Names.     (line 2232)

  55303. * sync_xorMODE instruction pattern:      Standard Names.     (line 2232)

  55304. * SYSROOT_HEADERS_SUFFIX_SPEC:           Driver.             (line  176)

  55305. * SYSROOT_SUFFIX_SPEC:                   Driver.             (line  171)

  55306. * SYSTEM_IMPLICIT_EXTERN_C:              Misc.               (line  415)

  55307. * t-TARGET:                              Target Fragment.    (line    6)

  55308. * table jump:                            Basic Blocks.       (line   67)

  55309. * tablejump instruction pattern:         Standard Names.     (line 1850)

  55310. * tag:                                   GTY Options.        (line   90)

  55311. * tagging insns:                         Tagging Insns.      (line    6)

  55312. * tail calls:                            Tail Calls.         (line    6)

  55313. * TAmode:                                Machine Modes.      (line  158)

  55314. * tanM2 instruction pattern:             Standard Names.     (line  954)

  55315. * target attributes:                     Target Attributes.  (line    6)

  55316. * target description macros:             Target Macros.      (line    6)

  55317. * target functions:                      Target Structure.   (line    6)

  55318. * target hooks:                          Target Structure.   (line    6)

  55319. * target makefile fragment:              Target Fragment.    (line    6)

  55320. * target specifications:                 Run-time Target.    (line    6)

  55321. * targetm:                               Target Structure.   (line    6)

  55322. * targets, makefile:                     Makefile.           (line    6)

  55323. * TARGET_ABSOLUTE_BIGGEST_ALIGNMENT:     Storage Layout.     (line  185)

  55324. * TARGET_ADDITIONAL_ALLOCNO_CLASS_P:     Register Classes.   (line  639)

  55325. * TARGET_ADDRESS_COST:                   Costs.              (line  344)

  55326. * TARGET_ADDR_SPACE_ADDRESS_MODE:        Named Address Spaces.

  55327.                                                              (line   42)

  55328. * TARGET_ADDR_SPACE_CONVERT:             Named Address Spaces.

  55329.                                                              (line   89)

  55330. * TARGET_ADDR_SPACE_DEBUG:               Named Address Spaces.

  55331.                                                              (line   99)

  55332. * TARGET_ADDR_SPACE_DIAGNOSE_USAGE:      Named Address Spaces.

  55333.                                                              (line  103)

  55334. * TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P: Named Address Spaces.

  55335.                                                              (line   59)

  55336. * TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS:  Named Address Spaces.

  55337.                                                              (line   67)

  55338. * TARGET_ADDR_SPACE_POINTER_MODE:        Named Address Spaces.

  55339.                                                              (line   36)

  55340. * TARGET_ADDR_SPACE_SUBSET_P:            Named Address Spaces.

  55341.                                                              (line   74)

  55342. * TARGET_ADDR_SPACE_VALID_POINTER_MODE:  Named Address Spaces.

  55343.                                                              (line   48)

  55344. * TARGET_ADDR_SPACE_ZERO_ADDRESS_VALID:  Named Address Spaces.

  55345.                                                              (line   83)

  55346. * TARGET_ALIGN_ANON_BITFIELD:            Storage Layout.     (line  432)

  55347. * TARGET_ALLOCATE_INITIAL_VALUE:         Misc.               (line  807)

  55348. * TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS:  Misc.               (line 1064)

  55349. * TARGET_ALWAYS_STRIP_DOTDOT:            Driver.             (line  250)

  55350. * TARGET_ARG_PARTIAL_BYTES:              Register Arguments. (line   92)

  55351. * TARGET_ARM_EABI_UNWINDER:              Exception Region Output.

  55352.                                                              (line  133)

  55353. * TARGET_ARRAY_MODE:                     Register Arguments. (line  373)

  55354. * TARGET_ARRAY_MODE_SUPPORTED_P:         Register Arguments. (line  388)

  55355. * TARGET_ASAN_SHADOW_OFFSET:             Misc.               (line 1092)

  55356. * TARGET_ASM_ALIGNED_DI_OP:              Data Output.        (line   11)

  55357. * TARGET_ASM_ALIGNED_HI_OP:              Data Output.        (line    7)

  55358. * TARGET_ASM_ALIGNED_PDI_OP:             Data Output.        (line   10)

  55359. * TARGET_ASM_ALIGNED_PSI_OP:             Data Output.        (line    8)

  55360. * TARGET_ASM_ALIGNED_PTI_OP:             Data Output.        (line   12)

  55361. * TARGET_ASM_ALIGNED_SI_OP:              Data Output.        (line    9)

  55362. * TARGET_ASM_ALIGNED_TI_OP:              Data Output.        (line   13)

  55363. * TARGET_ASM_ASSEMBLE_UNDEFINED_DECL:    Label Output.       (line  231)

  55364. * TARGET_ASM_ASSEMBLE_VISIBILITY:        Label Output.       (line  301)

  55365. * TARGET_ASM_BYTE_OP:                    Data Output.        (line    6)

  55366. * TARGET_ASM_CAN_OUTPUT_MI_THUNK:        Function Entry.     (line  209)

  55367. * TARGET_ASM_CLOSE_PAREN:                Data Output.        (line  139)

  55368. * TARGET_ASM_CODE_END:                   File Framework.     (line   57)

  55369. * TARGET_ASM_CONSTRUCTOR:                Macros for Initialization.

  55370.                                                              (line   68)

  55371. * TARGET_ASM_DECLARE_CONSTANT_NAME:      Label Output.       (line  177)

  55372. * TARGET_ASM_DECL_END:                   Data Output.        (line   44)

  55373. * TARGET_ASM_DESTRUCTOR:                 Macros for Initialization.

  55374.                                                              (line   82)

  55375. * TARGET_ASM_ELF_FLAGS_NUMERIC:          File Framework.     (line  120)

  55376. * TARGET_ASM_EMIT_EXCEPT_PERSONALITY:    Dispatch Tables.    (line   89)

  55377. * TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL:    Dispatch Tables.    (line   82)

  55378. * TARGET_ASM_EMIT_UNWIND_LABEL:          Dispatch Tables.    (line   70)

  55379. * TARGET_ASM_EXTERNAL_LIBCALL:           Label Output.       (line  337)

  55380. * TARGET_ASM_FILE_END:                   File Framework.     (line   35)

  55381. * TARGET_ASM_FILE_START:                 File Framework.     (line    8)

  55382. * TARGET_ASM_FILE_START_APP_OFF:         File Framework.     (line   16)

  55383. * TARGET_ASM_FILE_START_FILE_DIRECTIVE:  File Framework.     (line   29)

  55384. * TARGET_ASM_FINAL_POSTSCAN_INSN:        Instruction Output. (line   82)

  55385. * TARGET_ASM_FUNCTION_BEGIN_EPILOGUE:    Function Entry.     (line   67)

  55386. * TARGET_ASM_FUNCTION_END_PROLOGUE:      Function Entry.     (line   61)

  55387. * TARGET_ASM_FUNCTION_EPILOGUE:          Function Entry.     (line   73)

  55388. * TARGET_ASM_FUNCTION_PROLOGUE:          Function Entry.     (line   18)

  55389. * TARGET_ASM_FUNCTION_RODATA_SECTION:    Sections.           (line  225)

  55390. * TARGET_ASM_FUNCTION_SECTION:           File Framework.     (line  132)

  55391. * TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS: File Framework.

  55392.                                                              (line  142)

  55393. * TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC: Sections.           (line  184)

  55394. * TARGET_ASM_GLOBALIZE_DECL_NAME:        Label Output.       (line  222)

  55395. * TARGET_ASM_GLOBALIZE_LABEL:            Label Output.       (line  213)

  55396. * TARGET_ASM_INIT_SECTIONS:              Sections.           (line  164)

  55397. * TARGET_ASM_INTEGER:                    Data Output.        (line   31)

  55398. * TARGET_ASM_INTERNAL_LABEL:             Label Output.       (line  380)

  55399. * TARGET_ASM_LTO_END:                    File Framework.     (line   52)

  55400. * TARGET_ASM_LTO_START:                  File Framework.     (line   47)

  55401. * TARGET_ASM_MARK_DECL_PRESERVED:        Label Output.       (line  343)

  55402. * TARGET_ASM_MERGEABLE_RODATA_PREFIX:    Sections.           (line  233)

  55403. * TARGET_ASM_NAMED_SECTION:              File Framework.     (line  112)

  55404. * TARGET_ASM_OPEN_PAREN:                 Data Output.        (line  138)

  55405. * TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA:    Data Output.        (line   48)

  55406. * TARGET_ASM_OUTPUT_ANCHOR:              Anchored Addresses. (line   42)

  55407. * TARGET_ASM_OUTPUT_DWARF_DTPREL:        DWARF.              (line  121)

  55408. * TARGET_ASM_OUTPUT_IDENT:               File Framework.     (line   99)

  55409. * TARGET_ASM_OUTPUT_MI_THUNK:            Function Entry.     (line  167)

  55410. * TARGET_ASM_OUTPUT_SOURCE_FILENAME:     File Framework.     (line   91)

  55411. * TARGET_ASM_POST_CFI_STARTPROC:         Dispatch Tables.    (line   61)

  55412. * TARGET_ASM_PRINT_PATCHABLE_FUNCTION_ENTRY: Function Entry. (line    9)

  55413. * TARGET_ASM_RECORD_GCC_SWITCHES:        File Framework.     (line  173)

  55414. * TARGET_ASM_RECORD_GCC_SWITCHES_SECTION: File Framework.    (line  218)

  55415. * TARGET_ASM_RELOC_RW_MASK:              Sections.           (line  173)

  55416. * TARGET_ASM_SELECT_RTX_SECTION:         Sections.           (line  242)

  55417. * TARGET_ASM_SELECT_SECTION:             Sections.           (line  191)

  55418. * TARGET_ASM_TM_CLONE_TABLE_SECTION:     Sections.           (line  238)

  55419. * TARGET_ASM_TRAMPOLINE_TEMPLATE:        Trampolines.        (line   81)

  55420. * TARGET_ASM_TTYPE:                      Exception Region Output.

  55421.                                                              (line  127)

  55422. * TARGET_ASM_UNALIGNED_DI_OP:            Data Output.        (line   18)

  55423. * TARGET_ASM_UNALIGNED_HI_OP:            Data Output.        (line   14)

  55424. * TARGET_ASM_UNALIGNED_PDI_OP:           Data Output.        (line   17)

  55425. * TARGET_ASM_UNALIGNED_PSI_OP:           Data Output.        (line   15)

  55426. * TARGET_ASM_UNALIGNED_PTI_OP:           Data Output.        (line   19)

  55427. * TARGET_ASM_UNALIGNED_SI_OP:            Data Output.        (line   16)

  55428. * TARGET_ASM_UNALIGNED_TI_OP:            Data Output.        (line   20)

  55429. * TARGET_ASM_UNIQUE_SECTION:             Sections.           (line  213)

  55430. * TARGET_ASM_UNWIND_EMIT:                Dispatch Tables.    (line   96)

  55431. * TARGET_ASM_UNWIND_EMIT_BEFORE_INSN:    Dispatch Tables.    (line  102)

  55432. * TARGET_ATOMIC_ALIGN_FOR_MODE:          Misc.               (line 1111)

  55433. * TARGET_ATOMIC_ASSIGN_EXPAND_FENV:      Misc.               (line 1117)

  55434. * TARGET_ATOMIC_TEST_AND_SET_TRUEVAL:    Misc.               (line 1102)

  55435. * TARGET_ATTRIBUTE_TABLE:                Target Attributes.  (line   10)

  55436. * TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P:   Target Attributes.  (line   17)

  55437. * TARGET_BINDS_LOCAL_P:                  Sections.           (line  320)

  55438. * TARGET_BUILD_BUILTIN_VA_LIST:          Register Arguments. (line  281)

  55439. * TARGET_BUILTIN_DECL:                   Misc.               (line  637)

  55440. * TARGET_BUILTIN_RECIPROCAL:             Addressing Modes.   (line  261)

  55441. * TARGET_BUILTIN_SETJMP_FRAME_VALUE:     Frame Layout.       (line  101)

  55442. * TARGET_CALLEE_COPIES:                  Register Arguments. (line  124)

  55443. * TARGET_CALL_ARGS:                      Varargs.            (line  123)

  55444. * TARGET_CALL_FUSAGE_CONTAINS_NON_CALLEE_CLOBBERS: Miscellaneous Register Hooks.

  55445.                                                              (line    6)

  55446. * TARGET_CANNOT_FORCE_CONST_MEM:         Addressing Modes.   (line  234)

  55447. * TARGET_CANNOT_MODIFY_JUMPS_P:          Misc.               (line  871)

  55448. * TARGET_CANNOT_SUBSTITUTE_MEM_EQUIV_P:  Register Classes.   (line  610)

  55449. * TARGET_CANONICALIZE_COMPARISON:        MODE_CC Condition Codes.

  55450.                                                              (line   55)

  55451. * TARGET_CANONICAL_VA_LIST_TYPE:         Register Arguments. (line  302)

  55452. * TARGET_CAN_CHANGE_MODE_CLASS:          Register Classes.   (line  543)

  55453. * TARGET_CAN_CHANGE_MODE_CLASS and subreg semantics: Regs and Memory.

  55454.                                                              (line  294)

  55455. * TARGET_CAN_ELIMINATE:                  Elimination.        (line   58)

  55456. * TARGET_CAN_FOLLOW_JUMP:                Misc.               (line  793)

  55457. * TARGET_CAN_INLINE_P:                   Target Attributes.  (line  173)

  55458. * TARGET_CAN_USE_DOLOOP_P:               Misc.               (line  757)

  55459. * TARGET_CASE_VALUES_THRESHOLD:          Misc.               (line   46)

  55460. * TARGET_CC_MODES_COMPATIBLE:            MODE_CC Condition Codes.

  55461.                                                              (line  120)

  55462. * TARGET_CHECK_BUILTIN_CALL:             Misc.               (line  669)

  55463. * TARGET_CHECK_PCH_TARGET_FLAGS:         PCH Target.         (line   26)

  55464. * TARGET_CHECK_STRING_OBJECT_FORMAT_ARG: Run-time Target.    (line  119)

  55465. * TARGET_CLASS_LIKELY_SPILLED_P:         Register Classes.   (line  499)

  55466. * TARGET_CLASS_MAX_NREGS:                Register Classes.   (line  515)

  55467. * TARGET_COMMUTATIVE_P:                  Misc.               (line  800)

  55468. * TARGET_COMPARE_BY_PIECES_BRANCH_RATIO: Costs.              (line  200)

  55469. * TARGET_COMPARE_VERSION_PRIORITY:       Misc.               (line  701)

  55470. * TARGET_COMPATIBLE_VECTOR_TYPES_P:      Register Arguments. (line  350)

  55471. * TARGET_COMPUTE_FRAME_LAYOUT:           Elimination.        (line   74)

  55472. * TARGET_COMPUTE_PRESSURE_CLASSES:       Register Classes.   (line  655)

  55473. * TARGET_COMP_TYPE_ATTRIBUTES:           Target Attributes.  (line   25)

  55474. * TARGET_CONDITIONAL_REGISTER_USAGE:     Register Basics.    (line  102)

  55475. * TARGET_CONSTANT_ALIGNMENT:             Storage Layout.     (line  271)

  55476. * TARGET_CONST_ANCHOR:                   Misc.               (line 1075)

  55477. * TARGET_CONST_NOT_OK_FOR_DEBUG_P:       Addressing Modes.   (line  230)

  55478. * TARGET_CONVERT_TO_TYPE:                Misc.               (line 1022)

  55479. * TARGET_CPU_CPP_BUILTINS:               Run-time Target.    (line    8)

  55480. * TARGET_CSTORE_MODE:                    Register Classes.   (line  647)

  55481. * TARGET_CUSTOM_FUNCTION_DESCRIPTORS:    Trampolines.        (line   39)

  55482. * TARGET_CXX_ADJUST_CLASS_AT_DEFINITION: C++ ABI.            (line   86)

  55483. * TARGET_CXX_CDTOR_RETURNS_THIS:         C++ ABI.            (line   37)

  55484. * TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT:   C++ ABI.            (line   61)

  55485. * TARGET_CXX_COOKIE_HAS_SIZE:            C++ ABI.            (line   24)

  55486. * TARGET_CXX_DECL_MANGLING_CONTEXT:      C++ ABI.            (line   92)

  55487. * TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY: C++ ABI.       (line   52)

  55488. * TARGET_CXX_GET_COOKIE_SIZE:            C++ ABI.            (line   17)

  55489. * TARGET_CXX_GUARD_MASK_BIT:             C++ ABI.            (line   11)

  55490. * TARGET_CXX_GUARD_TYPE:                 C++ ABI.            (line    6)

  55491. * TARGET_CXX_IMPLICIT_EXTERN_C:          Misc.               (line  407)

  55492. * TARGET_CXX_IMPORT_EXPORT_CLASS:        C++ ABI.            (line   28)

  55493. * TARGET_CXX_KEY_METHOD_MAY_BE_INLINE:   C++ ABI.            (line   42)

  55494. * TARGET_CXX_LIBRARY_RTTI_COMDAT:        C++ ABI.            (line   68)

  55495. * TARGET_CXX_USE_AEABI_ATEXIT:           C++ ABI.            (line   73)

  55496. * TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT:  C++ ABI.            (line   79)

  55497. * TARGET_C_EXCESS_PRECISION:             Storage Layout.     (line  109)

  55498. * TARGET_C_PREINCLUDE:                   Misc.               (line  395)

  55499. * TARGET_DEBUG_UNWIND_INFO:              DWARF.              (line   32)

  55500. * TARGET_DECIMAL_FLOAT_SUPPORTED_P:      Storage Layout.     (line  537)

  55501. * TARGET_DECLSPEC:                       Target Attributes.  (line   72)

  55502. * TARGET_DEFAULT_PACK_STRUCT:            Misc.               (line  479)

  55503. * TARGET_DEFAULT_SHORT_ENUMS:            Type Layout.        (line  123)

  55504. * TARGET_DEFAULT_TARGET_FLAGS:           Run-time Target.    (line   55)

  55505. * TARGET_DEFERRED_OUTPUT_DEFS:           Label Output.       (line  465)

  55506. * TARGET_DELAY_SCHED2:                   DWARF.              (line   77)

  55507. * TARGET_DELAY_VARTRACK:                 DWARF.              (line   81)

  55508. * TARGET_DELEGITIMIZE_ADDRESS:           Addressing Modes.   (line  221)

  55509. * TARGET_DIFFERENT_ADDR_DISPLACEMENT_P:  Register Classes.   (line  603)

  55510. * TARGET_DLLIMPORT_DECL_ATTRIBUTES:      Target Attributes.  (line   55)

  55511. * TARGET_DOLOOP_COST_FOR_ADDRESS:        Misc.               (line  746)

  55512. * TARGET_DOLOOP_COST_FOR_GENERIC:        Misc.               (line  735)

  55513. * TARGET_DWARF_CALLING_CONVENTION:       DWARF.              (line   12)

  55514. * TARGET_DWARF_FRAME_REG_MODE:           Exception Region Output.

  55515.                                                              (line  113)

  55516. * TARGET_DWARF_HANDLE_FRAME_UNSPEC:      Frame Layout.       (line  165)

  55517. * TARGET_DWARF_POLY_INDETERMINATE_VALUE: Frame Layout.       (line  177)

  55518. * TARGET_DWARF_REGISTER_SPAN:            Exception Region Output.

  55519.                                                              (line  104)

  55520. * TARGET_D_CPU_VERSIONS:                 D Language and ABI. (line    6)

  55521. * TARGET_D_CRITSEC_SIZE:                 D Language and ABI. (line   17)

  55522. * TARGET_D_OS_VERSIONS:                  D Language and ABI. (line   13)

  55523. * TARGET_EDOM:                           Library Calls.      (line   59)

  55524. * TARGET_EMPTY_RECORD_P:                 Aggregate Return.   (line   86)

  55525. * TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS:  Emulated TLS.       (line   67)

  55526. * TARGET_EMUTLS_GET_ADDRESS:             Emulated TLS.       (line   18)

  55527. * TARGET_EMUTLS_REGISTER_COMMON:         Emulated TLS.       (line   23)

  55528. * TARGET_EMUTLS_TMPL_PREFIX:             Emulated TLS.       (line   44)

  55529. * TARGET_EMUTLS_TMPL_SECTION:            Emulated TLS.       (line   35)

  55530. * TARGET_EMUTLS_VAR_ALIGN_FIXED:         Emulated TLS.       (line   62)

  55531. * TARGET_EMUTLS_VAR_FIELDS:              Emulated TLS.       (line   48)

  55532. * TARGET_EMUTLS_VAR_INIT:                Emulated TLS.       (line   55)

  55533. * TARGET_EMUTLS_VAR_PREFIX:              Emulated TLS.       (line   40)

  55534. * TARGET_EMUTLS_VAR_SECTION:             Emulated TLS.       (line   30)

  55535. * TARGET_ENCODE_SECTION_INFO:            Sections.           (line  263)

  55536. * TARGET_ENCODE_SECTION_INFO and address validation: Addressing Modes.

  55537.                                                              (line   82)

  55538. * TARGET_ENCODE_SECTION_INFO usage:      Instruction Output. (line  127)

  55539. * TARGET_END_CALL_ARGS:                  Varargs.            (line  137)

  55540. * TARGET_ENUM_VA_LIST_P:                 Register Arguments. (line  285)

  55541. * TARGET_ESTIMATED_POLY_VALUE:           Costs.              (line  425)

  55542. * TARGET_EXCEPT_UNWIND_INFO:             Exception Region Output.

  55543.                                                              (line   46)

  55544. * TARGET_EXECUTABLE_SUFFIX:              Misc.               (line  858)

  55545. * TARGET_EXPAND_BUILTIN:                 Misc.               (line  647)

  55546. * TARGET_EXPAND_BUILTIN_SAVEREGS:        Varargs.            (line   64)

  55547. * TARGET_EXPAND_DIVMOD_LIBFUNC:          Scheduling.         (line  461)

  55548. * TARGET_EXPAND_TO_RTL_HOOK:             Storage Layout.     (line  543)

  55549. * TARGET_EXPR:                           Unary and Binary Expressions.

  55550.                                                              (line    6)

  55551. * TARGET_EXTRA_INCLUDES:                 Misc.               (line  937)

  55552. * TARGET_EXTRA_LIVE_ON_ENTRY:            Tail Calls.         (line   20)

  55553. * TARGET_EXTRA_PRE_INCLUDES:             Misc.               (line  944)

  55554. * TARGET_FIXED_CONDITION_CODE_REGS:      MODE_CC Condition Codes.

  55555.                                                              (line  105)

  55556. * TARGET_FIXED_POINT_SUPPORTED_P:        Storage Layout.     (line  540)

  55557. * target_flags:                          Run-time Target.    (line   51)

  55558. * TARGET_FLAGS_REGNUM:                   MODE_CC Condition Codes.

  55559.                                                              (line  133)

  55560. * TARGET_FLOATN_BUILTIN_P:               Register Arguments. (line  438)

  55561. * TARGET_FLOATN_MODE:                    Register Arguments. (line  420)

  55562. * TARGET_FLOAT_EXCEPTIONS_ROUNDING_SUPPORTED_P: Run-time Target.

  55563.                                                              (line  179)

  55564. * TARGET_FNTYPE_ABI:                     Register Basics.    (line   58)

  55565. * TARGET_FN_ABI_VA_LIST:                 Register Arguments. (line  297)

  55566. * TARGET_FOLD_BUILTIN:                   Misc.               (line  684)

  55567. * TARGET_FORMAT_TYPES:                   Misc.               (line  965)

  55568. * TARGET_FRAME_POINTER_REQUIRED:         Elimination.        (line    8)

  55569. * TARGET_FUNCTION_ARG:                   Register Arguments. (line   10)

  55570. * TARGET_FUNCTION_ARG_ADVANCE:           Register Arguments. (line  195)

  55571. * TARGET_FUNCTION_ARG_BOUNDARY:          Register Arguments. (line  248)

  55572. * TARGET_FUNCTION_ARG_OFFSET:            Register Arguments. (line  206)

  55573. * TARGET_FUNCTION_ARG_PADDING:           Register Arguments. (line  214)

  55574. * TARGET_FUNCTION_ARG_ROUND_BOUNDARY:    Register Arguments. (line  254)

  55575. * TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P: Target Attributes.  (line  101)

  55576. * TARGET_FUNCTION_INCOMING_ARG:          Register Arguments. (line   64)

  55577. * TARGET_FUNCTION_OK_FOR_SIBCALL:        Tail Calls.         (line    6)

  55578. * TARGET_FUNCTION_VALUE:                 Scalar Return.      (line    9)

  55579. * TARGET_FUNCTION_VALUE_REGNO_P:         Scalar Return.      (line   96)

  55580. * TARGET_GENERATE_VERSION_DISPATCHER_BODY: Misc.             (line  717)

  55581. * TARGET_GEN_CCMP_FIRST:                 Misc.               (line  890)

  55582. * TARGET_GEN_CCMP_NEXT:                  Misc.               (line  901)

  55583. * TARGET_GET_DRAP_RTX:                   Misc.               (line 1058)

  55584. * TARGET_GET_FUNCTION_VERSIONS_DISPATCHER: Misc.             (line  710)

  55585. * TARGET_GET_MULTILIB_ABI_NAME:          Register Basics.    (line   99)

  55586. * TARGET_GET_PCH_VALIDITY:               PCH Target.         (line    6)

  55587. * TARGET_GET_RAW_ARG_MODE:               Aggregate Return.   (line   81)

  55588. * TARGET_GET_RAW_RESULT_MODE:            Aggregate Return.   (line   76)

  55589. * TARGET_GET_VALID_OPTION_VALUES:        Stack Smashing Protection.

  55590.                                                              (line   39)

  55591. * TARGET_GIMPLE_FOLD_BUILTIN:            Misc.               (line  694)

  55592. * TARGET_GIMPLIFY_VA_ARG_EXPR:           Register Arguments. (line  307)

  55593. * TARGET_GOACC_DIM_LIMIT:                Addressing Modes.   (line  540)

  55594. * TARGET_GOACC_FORK_JOIN:                Addressing Modes.   (line  544)

  55595. * TARGET_GOACC_REDUCTION:                Addressing Modes.   (line  555)

  55596. * TARGET_GOACC_VALIDATE_DIMS:            Addressing Modes.   (line  527)

  55597. * TARGET_HANDLE_C_OPTION:                Run-time Target.    (line   73)

  55598. * TARGET_HANDLE_GENERIC_ATTRIBUTE:       Target Attributes.  (line   93)

  55599. * TARGET_HANDLE_OPTION:                  Run-time Target.    (line   59)

  55600. * TARGET_HARD_REGNO_CALL_PART_CLOBBERED: Register Basics.    (line   76)

  55601. * TARGET_HARD_REGNO_MODE_OK:             Values in Registers.

  55602.                                                              (line   54)

  55603. * TARGET_HARD_REGNO_NREGS:               Values in Registers.

  55604.                                                              (line   10)

  55605. * TARGET_HARD_REGNO_SCRATCH_OK:          Values in Registers.

  55606.                                                              (line  139)

  55607. * TARGET_HAS_IFUNC_P:                    Misc.               (line 1106)

  55608. * TARGET_HAS_NO_HW_DIVIDE:               Library Calls.      (line   52)

  55609. * TARGET_HAVE_CONDITIONAL_EXECUTION:     Misc.               (line  884)

  55610. * TARGET_HAVE_COUNT_REG_DECR_P:          Misc.               (line  731)

  55611. * TARGET_HAVE_CTORS_DTORS:               Macros for Initialization.

  55612.                                                              (line   63)

  55613. * TARGET_HAVE_NAMED_SECTIONS:            File Framework.     (line  150)

  55614. * TARGET_HAVE_SPECULATION_SAFE_VALUE:    Misc.               (line 1189)

  55615. * TARGET_HAVE_SRODATA_SECTION:           Sections.           (line  309)

  55616. * TARGET_HAVE_SWITCHABLE_BSS_SECTIONS:   File Framework.     (line  155)

  55617. * TARGET_HAVE_TLS:                       Sections.           (line  329)

  55618. * TARGET_INIT_BUILTINS:                  Misc.               (line  621)

  55619. * TARGET_INIT_DWARF_REG_SIZES_EXTRA:     Exception Region Output.

  55620.                                                              (line  119)

  55621. * TARGET_INIT_LIBFUNCS:                  Library Calls.      (line   15)

  55622. * TARGET_INIT_PIC_REG:                   Register Arguments. (line   88)

  55623. * TARGET_INSERT_ATTRIBUTES:              Target Attributes.  (line   80)

  55624. * TARGET_INSN_CALLEE_ABI:                Register Basics.    (line   65)

  55625. * TARGET_INSN_COST:                      Costs.              (line  380)

  55626. * TARGET_INSTANTIATE_DECLS:              Storage Layout.     (line  551)

  55627. * TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN: Misc.              (line  989)

  55628. * TARGET_INVALID_BINARY_OP:              Misc.               (line 1008)

  55629. * TARGET_INVALID_CONVERSION:             Misc.               (line  995)

  55630. * TARGET_INVALID_UNARY_OP:               Misc.               (line 1001)

  55631. * TARGET_INVALID_WITHIN_DOLOOP:          Misc.               (line  774)

  55632. * TARGET_IN_SMALL_DATA_P:                Sections.           (line  305)

  55633. * TARGET_IRA_CHANGE_PSEUDO_ALLOCNO_CLASS: Register Classes.  (line  570)

  55634. * TARGET_KEEP_LEAF_WHEN_PROFILED:        Profiling.          (line   39)

  55635. * TARGET_LEGITIMATE_ADDRESS_P:           Addressing Modes.   (line   48)

  55636. * TARGET_LEGITIMATE_COMBINED_INSN:       Misc.               (line  788)

  55637. * TARGET_LEGITIMATE_CONSTANT_P:          Addressing Modes.   (line  213)

  55638. * TARGET_LEGITIMIZE_ADDRESS:             Addressing Modes.   (line  129)

  55639. * TARGET_LEGITIMIZE_ADDRESS_DISPLACEMENT: Register Classes.  (line  618)

  55640. * TARGET_LIBCALL_VALUE:                  Scalar Return.      (line   65)

  55641. * TARGET_LIBC_HAS_FAST_FUNCTION:         Library Calls.      (line   82)

  55642. * TARGET_LIBC_HAS_FUNCTION:              Library Calls.      (line   77)

  55643. * TARGET_LIBFUNC_GNU_PREFIX:             Library Calls.      (line   24)

  55644. * TARGET_LIBGCC_CMP_RETURN_MODE:         Storage Layout.     (line  493)

  55645. * TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P: Register Arguments.

  55646.                                                              (line  412)

  55647. * TARGET_LIBGCC_SDATA_SECTION:           Sections.           (line  136)

  55648. * TARGET_LIBGCC_SHIFT_COUNT_MODE:        Storage Layout.     (line  499)

  55649. * TARGET_LIB_INT_CMP_BIASED:             Library Calls.      (line   42)

  55650. * TARGET_LOAD_BOUNDS_FOR_ARG:            Varargs.            (line  153)

  55651. * TARGET_LOAD_RETURNED_BOUNDS:           Varargs.            (line  172)

  55652. * TARGET_LOOP_UNROLL_ADJUST:             Misc.               (line  918)

  55653. * TARGET_LRA_P:                          Register Classes.   (line  577)

  55654. * TARGET_MACHINE_DEPENDENT_REORG:        Misc.               (line  606)

  55655. * TARGET_MANGLE_ASSEMBLER_NAME:          Label Output.       (line  356)

  55656. * TARGET_MANGLE_DECL_ASSEMBLER_NAME:     Sections.           (line  253)

  55657. * TARGET_MANGLE_TYPE:                    Storage Layout.     (line  555)

  55658. * TARGET_MAX_ANCHOR_OFFSET:              Anchored Addresses. (line   38)

  55659. * TARGET_MAX_NOCE_IFCVT_SEQ_COST:        Costs.              (line  390)

  55660. * TARGET_MD_ASM_ADJUST:                  Misc.               (line  524)

  55661. * TARGET_MEMBER_TYPE_FORCES_BLK:         Storage Layout.     (line  445)

  55662. * TARGET_MEMMODEL_CHECK:                 Misc.               (line 1097)

  55663. * TARGET_MEMORY_MOVE_COST:               Costs.              (line   79)

  55664. * TARGET_MEM_CONSTRAINT:                 Addressing Modes.   (line  107)

  55665. * TARGET_MEM_REF:                        Storage References. (line    6)

  55666. * TARGET_MERGE_DECL_ATTRIBUTES:          Target Attributes.  (line   45)

  55667. * TARGET_MERGE_TYPE_ATTRIBUTES:          Target Attributes.  (line   37)

  55668. * TARGET_MIN_ANCHOR_OFFSET:              Anchored Addresses. (line   32)

  55669. * TARGET_MIN_ARITHMETIC_PRECISION:       Misc.               (line   63)

  55670. * TARGET_MIN_DIVISIONS_FOR_RECIP_MUL:    Misc.               (line  112)

  55671. * TARGET_MODES_TIEABLE_P:                Values in Registers.

  55672.                                                              (line  123)

  55673. * TARGET_MODE_AFTER:                     Mode Switching.     (line   57)

  55674. * TARGET_MODE_DEPENDENT_ADDRESS_P:       Addressing Modes.   (line  196)

  55675. * TARGET_MODE_EMIT:                      Mode Switching.     (line   42)

  55676. * TARGET_MODE_ENTRY:                     Mode Switching.     (line   64)

  55677. * TARGET_MODE_EXIT:                      Mode Switching.     (line   71)

  55678. * TARGET_MODE_NEEDED:                    Mode Switching.     (line   50)

  55679. * TARGET_MODE_PRIORITY:                  Mode Switching.     (line   78)

  55680. * TARGET_MODE_REP_EXTENDED:              Misc.               (line  197)

  55681. * TARGET_MS_BITFIELD_LAYOUT_P:           Storage Layout.     (line  509)

  55682. * TARGET_MUST_PASS_IN_STACK:             Register Arguments. (line   57)

  55683. * TARGET_MUST_PASS_IN_STACK, and TARGET_FUNCTION_ARG: Register Arguments.

  55684.                                                              (line   49)

  55685. * TARGET_NARROW_VOLATILE_BITFIELD:       Storage Layout.     (line  438)

  55686. * TARGET_NOCE_CONVERSION_PROFITABLE_P:   Costs.              (line  409)

  55687. * TARGET_NO_REGISTER_ALLOCATION:         DWARF.              (line   85)

  55688. * TARGET_NO_SPECULATION_IN_DELAY_SLOTS_P: Costs.             (line  415)

  55689. * TARGET_N_FORMAT_TYPES:                 Misc.               (line  970)

  55690. * TARGET_OBJC_CONSTRUCT_STRING_OBJECT:   Run-time Target.    (line   88)

  55691. * TARGET_OBJC_DECLARE_CLASS_DEFINITION:  Run-time Target.    (line  109)

  55692. * TARGET_OBJC_DECLARE_UNRESOLVED_CLASS_REFERENCE: Run-time Target.

  55693.                                                              (line  104)

  55694. * TARGET_OBJECT_SUFFIX:                  Misc.               (line  853)

  55695. * TARGET_OBJFMT_CPP_BUILTINS:            Run-time Target.    (line   45)

  55696. * TARGET_OFFLOAD_OPTIONS:                Misc.               (line 1140)

  55697. * TARGET_OMIT_STRUCT_RETURN_REG:         Scalar Return.      (line  117)

  55698. * TARGET_OMP_DEVICE_KIND_ARCH_ISA:       Addressing Modes.   (line  519)

  55699. * TARGET_OPTAB_SUPPORTED_P:              Costs.              (line  299)

  55700. * TARGET_OPTF:                           Misc.               (line  952)

  55701. * TARGET_OPTION_FUNCTION_VERSIONS:       Target Attributes.  (line  165)

  55702. * TARGET_OPTION_INIT_STRUCT:             Run-time Target.    (line  156)

  55703. * TARGET_OPTION_OPTIMIZATION_TABLE:      Run-time Target.    (line  142)

  55704. * TARGET_OPTION_OVERRIDE:                Target Attributes.  (line  152)

  55705. * TARGET_OPTION_POST_STREAM_IN:          Target Attributes.  (line  133)

  55706. * TARGET_OPTION_PRAGMA_PARSE:            Target Attributes.  (line  145)

  55707. * TARGET_OPTION_PRINT:                   Target Attributes.  (line  139)

  55708. * TARGET_OPTION_RESTORE:                 Target Attributes.  (line  127)

  55709. * TARGET_OPTION_SAVE:                    Target Attributes.  (line  120)

  55710. * TARGET_OPTION_VALID_ATTRIBUTE_P:       Target Attributes.  (line  108)

  55711. * TARGET_OS_CPP_BUILTINS:                Run-time Target.    (line   41)

  55712. * TARGET_OVERRIDES_FORMAT_ATTRIBUTES:    Misc.               (line  974)

  55713. * TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT: Misc.            (line  980)

  55714. * TARGET_OVERRIDES_FORMAT_INIT:          Misc.               (line  984)

  55715. * TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE:  Run-time Target.    (line  126)

  55716. * TARGET_PASS_BY_REFERENCE:              Register Arguments. (line  112)

  55717. * TARGET_PCH_VALID_P:                    PCH Target.         (line   11)

  55718. * TARGET_POSIX_IO:                       Misc.               (line  550)

  55719. * TARGET_PREDICT_DOLOOP_P:               Misc.               (line  724)

  55720. * TARGET_PREFERRED_ELSE_VALUE:           Addressing Modes.   (line  563)

  55721. * TARGET_PREFERRED_OUTPUT_RELOAD_CLASS:  Register Classes.   (line  284)

  55722. * TARGET_PREFERRED_RELOAD_CLASS:         Register Classes.   (line  213)

  55723. * TARGET_PREFERRED_RENAME_CLASS:         Register Classes.   (line  201)

  55724. * TARGET_PREPARE_PCH_SAVE:               PCH Target.         (line   34)

  55725. * TARGET_PRETEND_OUTGOING_VARARGS_NAMED: Varargs.            (line  144)

  55726. * TARGET_PROFILE_BEFORE_PROLOGUE:        Sections.           (line  313)

  55727. * TARGET_PROMOTED_TYPE:                  Misc.               (line 1014)

  55728. * TARGET_PROMOTE_FUNCTION_MODE:          Storage Layout.     (line  126)

  55729. * TARGET_PROMOTE_PROTOTYPES:             Stack Arguments.    (line   10)

  55730. * TARGET_PTRMEMFUNC_VBIT_LOCATION:       Type Layout.        (line  250)

  55731. * TARGET_RECORD_OFFLOAD_SYMBOL:          Misc.               (line 1135)

  55732. * TARGET_REF_MAY_ALIAS_ERRNO:            Register Arguments. (line  318)

  55733. * TARGET_REGISTER_MOVE_COST:             Costs.              (line   31)

  55734. * TARGET_REGISTER_PRIORITY:              Register Classes.   (line  582)

  55735. * TARGET_REGISTER_USAGE_LEVELING_P:      Register Classes.   (line  593)

  55736. * TARGET_RELAYOUT_FUNCTION:              Target Attributes.  (line  180)

  55737. * TARGET_RESET_LOCATION_VIEW:            DWARF.              (line   57)

  55738. * TARGET_RESOLVE_OVERLOADED_BUILTIN:     Misc.               (line  658)

  55739. * TARGET_RETURN_IN_MEMORY:               Aggregate Return.   (line   15)

  55740. * TARGET_RETURN_IN_MSB:                  Scalar Return.      (line  124)

  55741. * TARGET_RETURN_POPS_ARGS:               Stack Arguments.    (line   98)

  55742. * TARGET_RTX_COSTS:                      Costs.              (line  313)

  55743. * TARGET_RUN_TARGET_SELFTESTS:           Misc.               (line 1226)

  55744. * TARGET_SCALAR_MODE_SUPPORTED_P:        Register Arguments. (line  334)

  55745. * TARGET_SCHED_ADJUST_COST:              Scheduling.         (line   35)

  55746. * TARGET_SCHED_ADJUST_PRIORITY:          Scheduling.         (line   50)

  55747. * TARGET_SCHED_ALLOC_SCHED_CONTEXT:      Scheduling.         (line  294)

  55748. * TARGET_SCHED_CAN_SPECULATE_INSN:       Scheduling.         (line  354)

  55749. * TARGET_SCHED_CLEAR_SCHED_CONTEXT:      Scheduling.         (line  309)

  55750. * TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK: Scheduling.     (line  101)

  55751. * TARGET_SCHED_DFA_NEW_CYCLE:            Scheduling.         (line  255)

  55752. * TARGET_SCHED_DFA_POST_ADVANCE_CYCLE:   Scheduling.         (line  172)

  55753. * TARGET_SCHED_DFA_POST_CYCLE_INSN:      Scheduling.         (line  156)

  55754. * TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE:    Scheduling.         (line  165)

  55755. * TARGET_SCHED_DFA_PRE_CYCLE_INSN:       Scheduling.         (line  144)

  55756. * TARGET_SCHED_DISPATCH:                 Scheduling.         (line  370)

  55757. * TARGET_SCHED_DISPATCH_DO:              Scheduling.         (line  375)

  55758. * TARGET_SCHED_EXPOSED_PIPELINE:         Scheduling.         (line  379)

  55759. * TARGET_SCHED_FINISH:                   Scheduling.         (line  122)

  55760. * TARGET_SCHED_FINISH_GLOBAL:            Scheduling.         (line  137)

  55761. * TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BACKTRACK: Scheduling.  (line  235)

  55762. * TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BEGIN: Scheduling.      (line  223)

  55763. * TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD: Scheduling.

  55764.                                                              (line  179)

  55765. * TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD: Scheduling.

  55766.                                                              (line  207)

  55767. * TARGET_SCHED_FIRST_CYCLE_MULTIPASS_END: Scheduling.        (line  240)

  55768. * TARGET_SCHED_FIRST_CYCLE_MULTIPASS_FINI: Scheduling.       (line  250)

  55769. * TARGET_SCHED_FIRST_CYCLE_MULTIPASS_INIT: Scheduling.       (line  245)

  55770. * TARGET_SCHED_FIRST_CYCLE_MULTIPASS_ISSUE: Scheduling.      (line  229)

  55771. * TARGET_SCHED_FREE_SCHED_CONTEXT:       Scheduling.         (line  313)

  55772. * TARGET_SCHED_FUSION_PRIORITY:          Scheduling.         (line  389)

  55773. * TARGET_SCHED_GEN_SPEC_CHECK:           Scheduling.         (line  335)

  55774. * TARGET_SCHED_H_I_D_EXTENDED:           Scheduling.         (line  289)

  55775. * TARGET_SCHED_INIT:                     Scheduling.         (line  111)

  55776. * TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN: Scheduling.         (line  161)

  55777. * TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN:  Scheduling.         (line  153)

  55778. * TARGET_SCHED_INIT_GLOBAL:              Scheduling.         (line  129)

  55779. * TARGET_SCHED_INIT_SCHED_CONTEXT:       Scheduling.         (line  298)

  55780. * TARGET_SCHED_ISSUE_RATE:               Scheduling.         (line   11)

  55781. * TARGET_SCHED_IS_COSTLY_DEPENDENCE:     Scheduling.         (line  267)

  55782. * TARGET_SCHED_MACRO_FUSION_P:           Scheduling.         (line   87)

  55783. * TARGET_SCHED_MACRO_FUSION_PAIR_P:      Scheduling.         (line   91)

  55784. * TARGET_SCHED_NEEDS_BLOCK_P:            Scheduling.         (line  328)

  55785. * TARGET_SCHED_REASSOCIATION_WIDTH:      Scheduling.         (line  384)

  55786. * TARGET_SCHED_REORDER:                  Scheduling.         (line   58)

  55787. * TARGET_SCHED_REORDER2:                 Scheduling.         (line   75)

  55788. * TARGET_SCHED_SET_SCHED_CONTEXT:        Scheduling.         (line  305)

  55789. * TARGET_SCHED_SET_SCHED_FLAGS:          Scheduling.         (line  347)

  55790. * TARGET_SCHED_SMS_RES_MII:              Scheduling.         (line  361)

  55791. * TARGET_SCHED_SPECULATE_INSN:           Scheduling.         (line  316)

  55792. * TARGET_SCHED_VARIABLE_ISSUE:           Scheduling.         (line   22)

  55793. * TARGET_SECONDARY_MEMORY_NEEDED:        Register Classes.   (line  447)

  55794. * TARGET_SECONDARY_MEMORY_NEEDED_MODE:   Register Classes.   (line  466)

  55795. * TARGET_SECONDARY_RELOAD:               Register Classes.   (line  312)

  55796. * TARGET_SECTION_TYPE_FLAGS:             File Framework.     (line  160)

  55797. * TARGET_SELECT_EARLY_REMAT_MODES:       Register Classes.   (line  488)

  55798. * TARGET_SETJMP_PRESERVES_NONVOLATILE_REGS_P: Misc.          (line  223)

  55799. * TARGET_SETUP_INCOMING_VARARGS:         Varargs.            (line   71)

  55800. * TARGET_SET_CURRENT_FUNCTION:           Misc.               (line  835)

  55801. * TARGET_SET_DEFAULT_TYPE_ATTRIBUTES:    Target Attributes.  (line   33)

  55802. * TARGET_SET_UP_BY_PROLOGUE:             Tail Calls.         (line   29)

  55803. * TARGET_SHIFT_TRUNCATION_MASK:          Misc.               (line  160)

  55804. * TARGET_SHRINK_WRAP_COMPONENTS_FOR_BB:  Shrink-wrapping separate components.

  55805.                                                              (line   36)

  55806. * TARGET_SHRINK_WRAP_DISQUALIFY_COMPONENTS: Shrink-wrapping separate components.

  55807.                                                              (line   43)

  55808. * TARGET_SHRINK_WRAP_EMIT_EPILOGUE_COMPONENTS: Shrink-wrapping separate components.

  55809.                                                              (line   54)

  55810. * TARGET_SHRINK_WRAP_EMIT_PROLOGUE_COMPONENTS: Shrink-wrapping separate components.

  55811.                                                              (line   50)

  55812. * TARGET_SHRINK_WRAP_GET_SEPARATE_COMPONENTS: Shrink-wrapping separate components.

  55813.                                                              (line   27)

  55814. * TARGET_SHRINK_WRAP_SET_HANDLED_COMPONENTS: Shrink-wrapping separate components.

  55815.                                                              (line   58)

  55816. * TARGET_SIMD_CLONE_ADJUST:              Addressing Modes.   (line  506)

  55817. * TARGET_SIMD_CLONE_COMPUTE_VECSIZE_AND_SIMDLEN: Addressing Modes.

  55818.                                                              (line  498)

  55819. * TARGET_SIMD_CLONE_USABLE:              Addressing Modes.   (line  510)

  55820. * TARGET_SIMT_VF:                        Addressing Modes.   (line  516)

  55821. * TARGET_SLOW_UNALIGNED_ACCESS:          Costs.              (line  132)

  55822. * TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P: Register Arguments.

  55823.                                                              (line  448)

  55824. * TARGET_SPECULATION_SAFE_VALUE:         Misc.               (line 1208)

  55825. * TARGET_SPILL_CLASS:                    Register Classes.   (line  632)

  55826. * TARGET_SPLIT_COMPLEX_ARG:              Register Arguments. (line  269)

  55827. * TARGET_STACK_CLASH_PROTECTION_ALLOCA_PROBE_RANGE: Stack Checking.

  55828.                                                              (line   97)

  55829. * TARGET_STACK_PROTECT_FAIL:             Stack Smashing Protection.

  55830.                                                              (line   16)

  55831. * TARGET_STACK_PROTECT_GUARD:            Stack Smashing Protection.

  55832.                                                              (line    6)

  55833. * TARGET_STACK_PROTECT_RUNTIME_ENABLED_P: Stack Smashing Protection.

  55834.                                                              (line   25)

  55835. * TARGET_STARTING_FRAME_OFFSET:          Frame Layout.       (line   34)

  55836. * TARGET_STARTING_FRAME_OFFSET and virtual registers: Regs and Memory.

  55837.                                                              (line   74)

  55838. * TARGET_STATIC_CHAIN:                   Frame Registers.    (line   90)

  55839. * TARGET_STATIC_RTX_ALIGNMENT:           Storage Layout.     (line  243)

  55840. * TARGET_STORE_BOUNDS_FOR_ARG:           Varargs.            (line  163)

  55841. * TARGET_STORE_RETURNED_BOUNDS:          Varargs.            (line  177)

  55842. * TARGET_STRICT_ARGUMENT_NAMING:         Varargs.            (line  107)

  55843. * TARGET_STRING_OBJECT_REF_TYPE_P:       Run-time Target.    (line  114)

  55844. * TARGET_STRIP_NAME_ENCODING:            Sections.           (line  300)

  55845. * TARGET_STRUCT_VALUE_RTX:               Aggregate Return.   (line   44)

  55846. * TARGET_SUPPORTS_SPLIT_STACK:           Stack Smashing Protection.

  55847.                                                              (line   30)

  55848. * TARGET_SUPPORTS_WEAK:                  Label Output.       (line  272)

  55849. * TARGET_SUPPORTS_WIDE_INT:              Misc.               (line 1148)

  55850. * TARGET_TERMINATE_DW2_EH_FRAME_INFO:    Exception Region Output.

  55851.                                                              (line   98)

  55852. * TARGET_TRAMPOLINE_ADJUST_ADDRESS:      Trampolines.        (line  127)

  55853. * TARGET_TRAMPOLINE_INIT:                Trampolines.        (line  107)

  55854. * TARGET_TRANSLATE_MODE_ATTRIBUTE:       Register Arguments. (line  325)

  55855. * TARGET_TRULY_NOOP_TRUNCATION:          Misc.               (line  184)

  55856. * TARGET_UNSPEC_MAY_TRAP_P:              Misc.               (line  826)

  55857. * TARGET_UNWIND_TABLES_DEFAULT:          Exception Region Output.

  55858.                                                              (line   73)

  55859. * TARGET_UNWIND_WORD_MODE:               Storage Layout.     (line  505)

  55860. * TARGET_UPDATE_STACK_BOUNDARY:          Misc.               (line 1054)

  55861. * TARGET_USES_WEAK_UNWIND_INFO:          Exception Handling. (line  123)

  55862. * TARGET_USE_ANCHORS_FOR_SYMBOL_P:       Anchored Addresses. (line   53)

  55863. * TARGET_USE_BLOCKS_FOR_CONSTANT_P:      Addressing Modes.   (line  248)

  55864. * TARGET_USE_BLOCKS_FOR_DECL_P:          Addressing Modes.   (line  255)

  55865. * TARGET_USE_BY_PIECES_INFRASTRUCTURE_P: Costs.              (line  165)

  55866. * TARGET_USE_PSEUDO_PIC_REG:             Register Arguments. (line   84)

  55867. * TARGET_VALID_DLLIMPORT_ATTRIBUTE_P:    Target Attributes.  (line   66)

  55868. * TARGET_VALID_POINTER_MODE:             Register Arguments. (line  313)

  55869. * TARGET_VECTORIZE_ADD_STMT_COST:        Addressing Modes.   (line  461)

  55870. * TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_MODES: Addressing Modes.

  55871.                                                              (line  376)

  55872. * TARGET_VECTORIZE_BUILTIN_GATHER:       Addressing Modes.   (line  484)

  55873. * TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD: Addressing Modes.  (line  266)

  55874. * TARGET_VECTORIZE_BUILTIN_MD_VECTORIZED_FUNCTION: Addressing Modes.

  55875.                                                              (line  344)

  55876. * TARGET_VECTORIZE_BUILTIN_SCATTER:      Addressing Modes.   (line  491)

  55877. * TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST: Addressing Modes.

  55878.                                                              (line  292)

  55879. * TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION: Addressing Modes.

  55880.                                                              (line  336)

  55881. * TARGET_VECTORIZE_DESTROY_COST_DATA:    Addressing Modes.   (line  479)

  55882. * TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE: Addressing Modes.

  55883.                                                              (line  445)

  55884. * TARGET_VECTORIZE_FINISH_COST:          Addressing Modes.   (line  472)

  55885. * TARGET_VECTORIZE_GET_MASK_MODE:        Addressing Modes.   (line  433)

  55886. * TARGET_VECTORIZE_INIT_COST:            Addressing Modes.   (line  452)

  55887. * TARGET_VECTORIZE_PREFERRED_SIMD_MODE:  Addressing Modes.   (line  361)

  55888. * TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT: Addressing Modes.

  55889.                                                              (line  298)

  55890. * TARGET_VECTORIZE_RELATED_MODE:         Addressing Modes.   (line  407)

  55891. * TARGET_VECTORIZE_SPLIT_REDUCTION:      Addressing Modes.   (line  368)

  55892. * TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT: Addressing Modes.

  55893.                                                              (line  351)

  55894. * TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE: Addressing Modes.

  55895.                                                              (line  310)

  55896. * TARGET_VECTORIZE_VEC_PERM_CONST:       Addressing Modes.   (line  316)

  55897. * TARGET_VECTOR_ALIGNMENT:               Storage Layout.     (line  298)

  55898. * TARGET_VECTOR_MODE_SUPPORTED_P:        Register Arguments. (line  345)

  55899. * TARGET_VERIFY_TYPE_CONTEXT:            Misc.               (line 1029)

  55900. * TARGET_VTABLE_DATA_ENTRY_DISTANCE:     Type Layout.        (line  303)

  55901. * TARGET_VTABLE_ENTRY_ALIGN:             Type Layout.        (line  297)

  55902. * TARGET_VTABLE_USES_DESCRIPTORS:        Type Layout.        (line  286)

  55903. * TARGET_WANT_DEBUG_PUB_SECTIONS:        DWARF.              (line   72)

  55904. * TARGET_WARN_FUNC_RETURN:               Tail Calls.         (line   35)

  55905. * TARGET_WARN_PARAMETER_PASSING_ABI:     Aggregate Return.   (line   90)

  55906. * TARGET_WEAK_NOT_IN_ARCHIVE_TOC:        Label Output.       (line  308)

  55907. * TCmode:                                Machine Modes.      (line  199)

  55908. * TDmode:                                Machine Modes.      (line   97)

  55909. * TEMPLATE_DECL:                         Declarations.       (line    6)

  55910. * Temporaries:                           Temporaries.        (line    6)

  55911. * termination routines:                  Initialization.     (line    6)

  55912. * testing constraints:                   C Constraint Interface.

  55913.                                                              (line    6)

  55914. * TEXT_SECTION_ASM_OP:                   Sections.           (line   37)

  55915. * TFmode:                                Machine Modes.      (line  101)

  55916. * The Language:                          The Language.       (line    6)

  55917. * THEN_CLAUSE:                           Statements for C++. (line    6)

  55918. * THREAD_MODEL_SPEC:                     Driver.             (line  162)

  55919. * THROW_EXPR:                            Unary and Binary Expressions.

  55920.                                                              (line    6)

  55921. * THUNK_DECL:                            Declarations.       (line    6)

  55922. * THUNK_DELTA:                           Declarations.       (line    6)

  55923. * TImode:                                Machine Modes.      (line   48)

  55924. * TImode, in insn:                       Insns.              (line  291)

  55925. * TLS_COMMON_ASM_OP:                     Sections.           (line   80)

  55926. * TLS_SECTION_ASM_FLAG:                  Sections.           (line   85)

  55927. * tm.h macros:                           Target Macros.      (line    6)

  55928. * TQFmode:                               Machine Modes.      (line   65)

  55929. * TQmode:                                Machine Modes.      (line  122)

  55930. * trampolines for nested functions:      Trampolines.        (line    6)

  55931. * TRAMPOLINE_ALIGNMENT:                  Trampolines.        (line  101)

  55932. * TRAMPOLINE_SECTION:                    Trampolines.        (line   92)

  55933. * TRAMPOLINE_SIZE:                       Trampolines.        (line   97)

  55934. * TRANSFER_FROM_TRAMPOLINE:              Trampolines.        (line  163)

  55935. * trap instruction pattern:              Standard Names.     (line 2121)

  55936. * tree:                                  Tree overview.      (line    6)

  55937. * tree <1>:                              Macros and Functions.

  55938.                                                              (line    6)

  55939. * Tree SSA:                              Tree SSA.           (line    6)

  55940. * TREE_CHAIN:                            Macros and Functions.

  55941.                                                              (line    6)

  55942. * TREE_CODE:                             Tree overview.      (line    6)

  55943. * tree_fits_shwi_p:                      Constant expressions.

  55944.                                                              (line    6)

  55945. * tree_fits_uhwi_p:                      Constant expressions.

  55946.                                                              (line    6)

  55947. * TREE_INT_CST_ELT:                      Constant expressions.

  55948.                                                              (line    6)

  55949. * tree_int_cst_equal:                    Constant expressions.

  55950.                                                              (line    6)

  55951. * TREE_INT_CST_LOW:                      Constant expressions.

  55952.                                                              (line    6)

  55953. * tree_int_cst_lt:                       Constant expressions.

  55954.                                                              (line    6)

  55955. * TREE_INT_CST_NUNITS:                   Constant expressions.

  55956.                                                              (line    6)

  55957. * TREE_LIST:                             Containers.         (line    6)

  55958. * TREE_OPERAND:                          Expression trees.   (line    6)

  55959. * TREE_PUBLIC:                           Function Basics.    (line    6)

  55960. * TREE_PUBLIC <1>:                       Function Properties.

  55961.                                                              (line   28)

  55962. * TREE_PURPOSE:                          Containers.         (line    6)

  55963. * TREE_READONLY:                         Function Properties.

  55964.                                                              (line   37)

  55965. * tree_size:                             Macros and Functions.

  55966.                                                              (line   13)

  55967. * TREE_STATIC:                           Function Properties.

  55968.                                                              (line   31)

  55969. * TREE_STRING_LENGTH:                    Constant expressions.

  55970.                                                              (line    6)

  55971. * TREE_STRING_POINTER:                   Constant expressions.

  55972.                                                              (line    6)

  55973. * TREE_THIS_VOLATILE:                    Function Properties.

  55974.                                                              (line   34)

  55975. * tree_to_shwi:                          Constant expressions.

  55976.                                                              (line    6)

  55977. * tree_to_uhwi:                          Constant expressions.

  55978.                                                              (line    6)

  55979. * TREE_TYPE:                             Macros and Functions.

  55980.                                                              (line    6)

  55981. * TREE_TYPE <1>:                         Types.              (line    6)

  55982. * TREE_TYPE <2>:                         Working with declarations.

  55983.                                                              (line   11)

  55984. * TREE_TYPE <3>:                         Expression trees.   (line    6)

  55985. * TREE_TYPE <4>:                         Expression trees.   (line   17)

  55986. * TREE_TYPE <5>:                         Function Basics.    (line   47)

  55987. * TREE_TYPE <6>:                         Types for C++.      (line    6)

  55988. * TREE_VALUE:                            Containers.         (line    6)

  55989. * TREE_VEC:                              Containers.         (line    6)

  55990. * TREE_VEC_ELT:                          Containers.         (line    6)

  55991. * TREE_VEC_LENGTH:                       Containers.         (line    6)

  55992. * true positive:                         Guidelines for Diagnostics.

  55993.                                                              (line   39)

  55994. * truncate:                              Conversions.        (line   38)

  55995. * truncMN2 instruction pattern:          Standard Names.     (line 1442)

  55996. * TRUNC_DIV_EXPR:                        Unary and Binary Expressions.

  55997.                                                              (line    6)

  55998. * TRUNC_MOD_EXPR:                        Unary and Binary Expressions.

  55999.                                                              (line    6)

  56000. * TRUTH_ANDIF_EXPR:                      Unary and Binary Expressions.

  56001.                                                              (line    6)

  56002. * TRUTH_AND_EXPR:                        Unary and Binary Expressions.

  56003.                                                              (line    6)

  56004. * TRUTH_NOT_EXPR:                        Unary and Binary Expressions.

  56005.                                                              (line    6)

  56006. * TRUTH_ORIF_EXPR:                       Unary and Binary Expressions.

  56007.                                                              (line    6)

  56008. * TRUTH_OR_EXPR:                         Unary and Binary Expressions.

  56009.                                                              (line    6)

  56010. * TRUTH_XOR_EXPR:                        Unary and Binary Expressions.

  56011.                                                              (line    6)

  56012. * TRY_BLOCK:                             Statements for C++. (line    6)

  56013. * TRY_HANDLERS:                          Statements for C++. (line    6)

  56014. * TRY_STMTS:                             Statements for C++. (line    6)

  56015. * Tuple specific accessors:              Tuple specific accessors.

  56016.                                                              (line    6)

  56017. * tuples:                                Tuple representation.

  56018.                                                              (line    6)

  56019. * type:                                  Types.              (line    6)

  56020. * type declaration:                      Declarations.       (line    6)

  56021. * TYPENAME_TYPE:                         Types for C++.      (line    6)

  56022. * TYPENAME_TYPE_FULLNAME:                Types.              (line    6)

  56023. * TYPENAME_TYPE_FULLNAME <1>:            Types for C++.      (line    6)

  56024. * TYPEOF_TYPE:                           Types for C++.      (line    6)

  56025. * TYPE_ALIGN:                            Types.              (line    6)

  56026. * TYPE_ALIGN <1>:                        Types.              (line   30)

  56027. * TYPE_ALIGN <2>:                        Types for C++.      (line    6)

  56028. * TYPE_ALIGN <3>:                        Types for C++.      (line   44)

  56029. * TYPE_ARG_TYPES:                        Types.              (line    6)

  56030. * TYPE_ARG_TYPES <1>:                    Types for C++.      (line    6)

  56031. * TYPE_ASM_OP:                           Label Output.       (line   76)

  56032. * TYPE_ATTRIBUTES:                       Attributes.         (line   24)

  56033. * TYPE_BINFO:                            Classes.            (line    6)

  56034. * TYPE_BUILT_IN:                         Types for C++.      (line   66)

  56035. * TYPE_CANONICAL:                        Types.              (line    6)

  56036. * TYPE_CANONICAL <1>:                    Types.              (line   41)

  56037. * TYPE_CONTEXT:                          Types.              (line    6)

  56038. * TYPE_CONTEXT <1>:                      Types for C++.      (line    6)

  56039. * TYPE_DECL:                             Declarations.       (line    6)

  56040. * TYPE_FIELDS:                           Types.              (line    6)

  56041. * TYPE_FIELDS <1>:                       Types for C++.      (line    6)

  56042. * TYPE_FIELDS <2>:                       Classes.            (line    6)

  56043. * TYPE_HAS_ARRAY_NEW_OPERATOR:           Classes.            (line   93)

  56044. * TYPE_HAS_DEFAULT_CONSTRUCTOR:          Classes.            (line   78)

  56045. * TYPE_HAS_MUTABLE_P:                    Classes.            (line   83)

  56046. * TYPE_HAS_NEW_OPERATOR:                 Classes.            (line   90)

  56047. * TYPE_MAIN_VARIANT:                     Types.              (line    6)

  56048. * TYPE_MAIN_VARIANT <1>:                 Types.              (line   19)

  56049. * TYPE_MAIN_VARIANT <2>:                 Types for C++.      (line    6)

  56050. * TYPE_MAX_VALUE:                        Types.              (line    6)

  56051. * TYPE_METHOD_BASETYPE:                  Types.              (line    6)

  56052. * TYPE_METHOD_BASETYPE <1>:              Types for C++.      (line    6)

  56053. * TYPE_MIN_VALUE:                        Types.              (line    6)

  56054. * TYPE_NAME:                             Types.              (line    6)

  56055. * TYPE_NAME <1>:                         Types.              (line   33)

  56056. * TYPE_NAME <2>:                         Types for C++.      (line    6)

  56057. * TYPE_NAME <3>:                         Types for C++.      (line   47)

  56058. * TYPE_NOTHROW_P:                        Functions for C++.  (line  154)

  56059. * TYPE_OFFSET_BASETYPE:                  Types.              (line    6)

  56060. * TYPE_OFFSET_BASETYPE <1>:              Types for C++.      (line    6)

  56061. * TYPE_OPERAND_FMT:                      Label Output.       (line   87)

  56062. * TYPE_OVERLOADS_ARRAY_REF:              Classes.            (line  101)

  56063. * TYPE_OVERLOADS_ARROW:                  Classes.            (line  104)

  56064. * TYPE_OVERLOADS_CALL_EXPR:              Classes.            (line   97)

  56065. * TYPE_POLYMORPHIC_P:                    Classes.            (line   74)

  56066. * TYPE_PRECISION:                        Types.              (line    6)

  56067. * TYPE_PRECISION <1>:                    Types for C++.      (line    6)

  56068. * TYPE_PTRDATAMEM_P:                     Types for C++.      (line    6)

  56069. * TYPE_PTRDATAMEM_P <1>:                 Types for C++.      (line   69)

  56070. * TYPE_PTRFN_P:                          Types for C++.      (line   76)

  56071. * TYPE_PTROBV_P:                         Types for C++.      (line    6)

  56072. * TYPE_PTROB_P:                          Types for C++.      (line   79)

  56073. * TYPE_PTR_P:                            Types for C++.      (line   72)

  56074. * TYPE_QUAL_CONST:                       Types.              (line    6)

  56075. * TYPE_QUAL_CONST <1>:                   Types for C++.      (line    6)

  56076. * TYPE_QUAL_RESTRICT:                    Types.              (line    6)

  56077. * TYPE_QUAL_RESTRICT <1>:                Types for C++.      (line    6)

  56078. * TYPE_QUAL_VOLATILE:                    Types.              (line    6)

  56079. * TYPE_QUAL_VOLATILE <1>:                Types for C++.      (line    6)

  56080. * TYPE_RAISES_EXCEPTIONS:                Functions for C++.  (line  149)

  56081. * TYPE_SIZE:                             Types.              (line    6)

  56082. * TYPE_SIZE <1>:                         Types.              (line   25)

  56083. * TYPE_SIZE <2>:                         Types for C++.      (line    6)

  56084. * TYPE_SIZE <3>:                         Types for C++.      (line   39)

  56085. * TYPE_STRUCTURAL_EQUALITY_P:            Types.              (line    6)

  56086. * TYPE_STRUCTURAL_EQUALITY_P <1>:        Types.              (line   77)

  56087. * TYPE_UNQUALIFIED:                      Types.              (line    6)

  56088. * TYPE_UNQUALIFIED <1>:                  Types for C++.      (line    6)

  56089. * TYPE_VFIELD:                           Classes.            (line    6)

  56090. * uaddvM4 instruction pattern:           Standard Names.     (line  461)

  56091. * uavgM3_ceil instruction pattern:       Standard Names.     (line  860)

  56092. * uavgM3_floor instruction pattern:      Standard Names.     (line  848)

  56093. * UDAmode:                               Machine Modes.      (line  170)

  56094. * udiv:                                  Arithmetic.         (line  130)

  56095. * udivM3 instruction pattern:            Standard Names.     (line  442)

  56096. * udivmodM4 instruction pattern:         Standard Names.     (line  825)

  56097. * udot_prodM instruction pattern:        Standard Names.     (line  570)

  56098. * UDQmode:                               Machine Modes.      (line  138)

  56099. * UHAmode:                               Machine Modes.      (line  162)

  56100. * UHQmode:                               Machine Modes.      (line  130)

  56101. * UINT16_TYPE:                           Type Layout.        (line  214)

  56102. * UINT32_TYPE:                           Type Layout.        (line  215)

  56103. * UINT64_TYPE:                           Type Layout.        (line  216)

  56104. * UINT8_TYPE:                            Type Layout.        (line  213)

  56105. * UINTMAX_TYPE:                          Type Layout.        (line  197)

  56106. * UINTPTR_TYPE:                          Type Layout.        (line  234)

  56107. * UINT_FAST16_TYPE:                      Type Layout.        (line  230)

  56108. * UINT_FAST32_TYPE:                      Type Layout.        (line  231)

  56109. * UINT_FAST64_TYPE:                      Type Layout.        (line  232)

  56110. * UINT_FAST8_TYPE:                       Type Layout.        (line  229)

  56111. * UINT_LEAST16_TYPE:                     Type Layout.        (line  222)

  56112. * UINT_LEAST32_TYPE:                     Type Layout.        (line  223)

  56113. * UINT_LEAST64_TYPE:                     Type Layout.        (line  224)

  56114. * UINT_LEAST8_TYPE:                      Type Layout.        (line  221)

  56115. * umaddMN4 instruction pattern:          Standard Names.     (line  772)

  56116. * umax:                                  Arithmetic.         (line  149)

  56117. * umaxM3 instruction pattern:            Standard Names.     (line  442)

  56118. * umin:                                  Arithmetic.         (line  149)

  56119. * uminM3 instruction pattern:            Standard Names.     (line  442)

  56120. * umod:                                  Arithmetic.         (line  136)

  56121. * umodM3 instruction pattern:            Standard Names.     (line  442)

  56122. * umsubMN4 instruction pattern:          Standard Names.     (line  796)

  56123. * umulhisi3 instruction pattern:         Standard Names.     (line  744)

  56124. * umulhrsM3 instruction pattern:         Standard Names.     (line  605)

  56125. * umulhsM3 instruction pattern:          Standard Names.     (line  595)

  56126. * umulM3_highpart instruction pattern:   Standard Names.     (line  758)

  56127. * umulqihi3 instruction pattern:         Standard Names.     (line  744)

  56128. * umulsidi3 instruction pattern:         Standard Names.     (line  744)

  56129. * umulvM4 instruction pattern:           Standard Names.     (line  466)

  56130. * unchanging:                            Flags.              (line  307)

  56131. * unchanging, in call_insn:              Flags.              (line  115)

  56132. * unchanging, in jump_insn, call_insn and insn: Flags.       (line   28)

  56133. * unchanging, in mem:                    Flags.              (line   78)

  56134. * unchanging, in subreg:                 Flags.              (line  184)

  56135. * unchanging, in subreg <1>:             Flags.              (line  194)

  56136. * unchanging, in symbol_ref:             Flags.              (line   19)

  56137. * UNEQ_EXPR:                             Unary and Binary Expressions.

  56138.                                                              (line    6)

  56139. * UNGE_EXPR:                             Unary and Binary Expressions.

  56140.                                                              (line    6)

  56141. * UNGT_EXPR:                             Unary and Binary Expressions.

  56142.                                                              (line    6)

  56143. * unions, returning:                     Interface.          (line   10)

  56144. * UNION_TYPE:                            Types.              (line    6)

  56145. * UNION_TYPE <1>:                        Classes.            (line    6)

  56146. * UNITS_PER_WORD:                        Storage Layout.     (line   60)

  56147. * UNKNOWN_TYPE:                          Types.              (line    6)

  56148. * UNKNOWN_TYPE <1>:                      Types for C++.      (line    6)

  56149. * UNLE_EXPR:                             Unary and Binary Expressions.

  56150.                                                              (line    6)

  56151. * UNLIKELY_EXECUTED_TEXT_SECTION_NAME:   Sections.           (line   48)

  56152. * UNLT_EXPR:                             Unary and Binary Expressions.

  56153.                                                              (line    6)

  56154. * UNORDERED_EXPR:                        Unary and Binary Expressions.

  56155.                                                              (line    6)

  56156. * unshare_all_rtl:                       Sharing.            (line   61)

  56157. * unsigned division:                     Arithmetic.         (line  130)

  56158. * unsigned division with unsigned saturation: Arithmetic.    (line  130)

  56159. * unsigned greater than:                 Comparisons.        (line   64)

  56160. * unsigned greater than <1>:             Comparisons.        (line   72)

  56161. * unsigned less than:                    Comparisons.        (line   68)

  56162. * unsigned less than <1>:                Comparisons.        (line   76)

  56163. * unsigned minimum and maximum:          Arithmetic.         (line  149)

  56164. * unsigned_fix:                          Conversions.        (line   77)

  56165. * unsigned_float:                        Conversions.        (line   62)

  56166. * unsigned_fract_convert:                Conversions.        (line   97)

  56167. * unsigned_sat_fract:                    Conversions.        (line  103)

  56168. * unspec:                                Side Effects.       (line  299)

  56169. * unspec <1>:                            Constant Definitions.

  56170.                                                              (line  111)

  56171. * unspec_volatile:                       Side Effects.       (line  299)

  56172. * unspec_volatile <1>:                   Constant Definitions.

  56173.                                                              (line   99)

  56174. * untyped_call instruction pattern:      Standard Names.     (line 1747)

  56175. * untyped_return instruction pattern:    Standard Names.     (line 1810)

  56176. * UPDATE_PATH_HOST_CANONICALIZE (PATH):  Filesystem.         (line   59)

  56177. * update_ssa:                            SSA.                (line   74)

  56178. * update_stmt:                           Manipulating GIMPLE statements.

  56179.                                                              (line  140)

  56180. * update_stmt <1>:                       SSA Operands.       (line    6)

  56181. * update_stmt_if_modified:               Manipulating GIMPLE statements.

  56182.                                                              (line  143)

  56183. * UQQmode:                               Machine Modes.      (line  126)

  56184. * usaddM3 instruction pattern:           Standard Names.     (line  442)

  56185. * usadM instruction pattern:             Standard Names.     (line  579)

  56186. * USAmode:                               Machine Modes.      (line  166)

  56187. * usashlM3 instruction pattern:          Standard Names.     (line  828)

  56188. * usdivM3 instruction pattern:           Standard Names.     (line  442)

  56189. * use:                                   Side Effects.       (line  168)

  56190. * used:                                  Flags.              (line  325)

  56191. * used, in symbol_ref:                   Flags.              (line  211)

  56192. * user:                                  GTY Options.        (line  245)

  56193. * user experience guidelines:            User Experience Guidelines.

  56194.                                                              (line    6)

  56195. * user gc:                               User GC.            (line    6)

  56196. * USER_LABEL_PREFIX:                     Instruction Output. (line  152)

  56197. * USE_C_ALLOCA:                          Host Misc.          (line   19)

  56198. * USE_LD_AS_NEEDED:                      Driver.             (line  135)

  56199. * USE_LOAD_POST_DECREMENT:               Costs.              (line  254)

  56200. * USE_LOAD_POST_INCREMENT:               Costs.              (line  249)

  56201. * USE_LOAD_PRE_DECREMENT:                Costs.              (line  264)

  56202. * USE_LOAD_PRE_INCREMENT:                Costs.              (line  259)

  56203. * USE_SELECT_SECTION_FOR_FUNCTIONS:      Sections.           (line  205)

  56204. * USE_STORE_POST_DECREMENT:              Costs.              (line  274)

  56205. * USE_STORE_POST_INCREMENT:              Costs.              (line  269)

  56206. * USE_STORE_PRE_DECREMENT:               Costs.              (line  284)

  56207. * USE_STORE_PRE_INCREMENT:               Costs.              (line  279)

  56208. * USING_STMT:                            Statements for C++. (line    6)

  56209. * usmaddMN4 instruction pattern:         Standard Names.     (line  780)

  56210. * usmsubMN4 instruction pattern:         Standard Names.     (line  804)

  56211. * usmulhisi3 instruction pattern:        Standard Names.     (line  748)

  56212. * usmulM3 instruction pattern:           Standard Names.     (line  442)

  56213. * usmulqihi3 instruction pattern:        Standard Names.     (line  748)

  56214. * usmulsidi3 instruction pattern:        Standard Names.     (line  748)

  56215. * usnegM2 instruction pattern:           Standard Names.     (line  872)

  56216. * USQmode:                               Machine Modes.      (line  134)

  56217. * ussubM3 instruction pattern:           Standard Names.     (line  442)

  56218. * usubvM4 instruction pattern:           Standard Names.     (line  466)

  56219. * us_ashift:                             Arithmetic.         (line  173)

  56220. * us_minus:                              Arithmetic.         (line   38)

  56221. * us_mult:                               Arithmetic.         (line   93)

  56222. * us_neg:                                Arithmetic.         (line   82)

  56223. * us_plus:                               Arithmetic.         (line   14)

  56224. * us_truncate:                           Conversions.        (line   48)

  56225. * UTAmode:                               Machine Modes.      (line  174)

  56226. * UTQmode:                               Machine Modes.      (line  142)

  56227. * V in constraint:                       Simple Constraints. (line   43)

  56228. * values, returned by functions:         Scalar Return.      (line    6)

  56229. * varargs implementation:                Varargs.            (line    6)

  56230. * variable:                              Declarations.       (line    6)

  56231. * Variable Location Debug Information in RTL: Debug Information.

  56232.                                                              (line    6)

  56233. * VAR_DECL:                              Declarations.       (line    6)

  56234. * var_location:                          Debug Information.  (line   14)

  56235. * vashlM3 instruction pattern:           Standard Names.     (line  844)

  56236. * vashrM3 instruction pattern:           Standard Names.     (line  844)

  56237. * VA_ARG_EXPR:                           Unary and Binary Expressions.

  56238.                                                              (line    6)

  56239. * vcondeqMN instruction pattern:         Standard Names.     (line  385)

  56240. * vcondMN instruction pattern:           Standard Names.     (line  372)

  56241. * vconduMN instruction pattern:          Standard Names.     (line  382)

  56242. * vcond_mask_MN instruction pattern:     Standard Names.     (line  392)

  56243. * vector:                                Containers.         (line    6)

  56244. * vector operations:                     Vector Operations.  (line    6)

  56245. * VECTOR_CST:                            Constant expressions.

  56246.                                                              (line    6)

  56247. * VECTOR_STORE_FLAG_VALUE:               Misc.               (line  327)

  56248. * vec_cmpeqMN instruction pattern:       Standard Names.     (line  365)

  56249. * vec_cmpMN instruction pattern:         Standard Names.     (line  355)

  56250. * vec_cmpuMN instruction pattern:        Standard Names.     (line  362)

  56251. * vec_concat:                            Vector Operations.  (line   29)

  56252. * VEC_COND_EXPR:                         Vectors.            (line    6)

  56253. * vec_duplicate:                         Vector Operations.  (line   34)

  56254. * vec_duplicateM instruction pattern:    Standard Names.     (line  298)

  56255. * VEC_DUPLICATE_EXPR:                    Vectors.            (line    6)

  56256. * vec_extractMN instruction pattern:     Standard Names.     (line  282)

  56257. * vec_initMN instruction pattern:        Standard Names.     (line  291)

  56258. * vec_load_lanesMN instruction pattern:  Standard Names.     (line  165)

  56259. * VEC_LSHIFT_EXPR:                       Vectors.            (line    6)

  56260. * vec_mask_load_lanesMN instruction pattern: Standard Names. (line  189)

  56261. * vec_mask_store_lanesMN instruction pattern: Standard Names.

  56262.                                                              (line  219)

  56263. * vec_merge:                             Vector Operations.  (line   11)

  56264. * vec_packs_float_M instruction pattern: Standard Names.     (line  670)

  56265. * vec_packu_float_M instruction pattern: Standard Names.     (line  670)

  56266. * VEC_PACK_FIX_TRUNC_EXPR:               Vectors.            (line    6)

  56267. * VEC_PACK_FLOAT_EXPR:                   Vectors.            (line    6)

  56268. * VEC_PACK_SAT_EXPR:                     Vectors.            (line    6)

  56269. * vec_pack_sbool_trunc_M instruction pattern: Standard Names.

  56270.                                                              (line  647)

  56271. * vec_pack_sfix_trunc_M instruction pattern: Standard Names. (line  663)

  56272. * vec_pack_ssat_M instruction pattern:   Standard Names.     (line  656)

  56273. * VEC_PACK_TRUNC_EXPR:                   Vectors.            (line    6)

  56274. * vec_pack_trunc_M instruction pattern:  Standard Names.     (line  640)

  56275. * vec_pack_ufix_trunc_M instruction pattern: Standard Names. (line  663)

  56276. * vec_pack_usat_M instruction pattern:   Standard Names.     (line  656)

  56277. * vec_permM instruction pattern:         Standard Names.     (line  410)

  56278. * vec_permM instruction pattern <1>:     Addressing Modes.   (line  330)

  56279. * VEC_RSHIFT_EXPR:                       Vectors.            (line    6)

  56280. * vec_select:                            Vector Operations.  (line   19)

  56281. * vec_series:                            Vector Operations.  (line   41)

  56282. * vec_seriesM instruction pattern:       Standard Names.     (line  308)

  56283. * VEC_SERIES_EXPR:                       Vectors.            (line    6)

  56284. * vec_setM instruction pattern:          Standard Names.     (line  277)

  56285. * vec_shl_insert_M instruction pattern:  Standard Names.     (line  621)

  56286. * vec_shl_M instruction pattern:         Standard Names.     (line  628)

  56287. * vec_shr_M instruction pattern:         Standard Names.     (line  634)

  56288. * vec_store_lanesMN instruction pattern: Standard Names.     (line  206)

  56289. * vec_unpacks_float_hi_M instruction pattern: Standard Names.

  56290.                                                              (line  700)

  56291. * vec_unpacks_float_lo_M instruction pattern: Standard Names.

  56292.                                                              (line  700)

  56293. * vec_unpacks_hi_M instruction pattern:  Standard Names.     (line  677)

  56294. * vec_unpacks_lo_M instruction pattern:  Standard Names.     (line  677)

  56295. * vec_unpacks_sbool_hi_M instruction pattern: Standard Names.

  56296.                                                              (line  691)

  56297. * vec_unpacks_sbool_lo_M instruction pattern: Standard Names.

  56298.                                                              (line  691)

  56299. * vec_unpacku_float_hi_M instruction pattern: Standard Names.

  56300.                                                              (line  700)

  56301. * vec_unpacku_float_lo_M instruction pattern: Standard Names.

  56302.                                                              (line  700)

  56303. * vec_unpacku_hi_M instruction pattern:  Standard Names.     (line  684)

  56304. * vec_unpacku_lo_M instruction pattern:  Standard Names.     (line  684)

  56305. * VEC_UNPACK_FIX_TRUNC_HI_EXPR:          Vectors.            (line    6)

  56306. * VEC_UNPACK_FIX_TRUNC_LO_EXPR:          Vectors.            (line    6)

  56307. * VEC_UNPACK_FLOAT_HI_EXPR:              Vectors.            (line    6)

  56308. * VEC_UNPACK_FLOAT_LO_EXPR:              Vectors.            (line    6)

  56309. * VEC_UNPACK_HI_EXPR:                    Vectors.            (line    6)

  56310. * VEC_UNPACK_LO_EXPR:                    Vectors.            (line    6)

  56311. * vec_unpack_sfix_trunc_hi_M instruction pattern: Standard Names.

  56312.                                                              (line  709)

  56313. * vec_unpack_sfix_trunc_lo_M instruction pattern: Standard Names.

  56314.                                                              (line  709)

  56315. * vec_unpack_ufix_trunc_hi_M instruction pattern: Standard Names.

  56316.                                                              (line  709)

  56317. * vec_unpack_ufix_trunc_lo_M instruction pattern: Standard Names.

  56318.                                                              (line  709)

  56319. * VEC_WIDEN_MULT_HI_EXPR:                Vectors.            (line    6)

  56320. * VEC_WIDEN_MULT_LO_EXPR:                Vectors.            (line    6)

  56321. * vec_widen_smult_even_M instruction pattern: Standard Names.

  56322.                                                              (line  719)

  56323. * vec_widen_smult_hi_M instruction pattern: Standard Names.  (line  719)

  56324. * vec_widen_smult_lo_M instruction pattern: Standard Names.  (line  719)

  56325. * vec_widen_smult_odd_M instruction pattern: Standard Names. (line  719)

  56326. * vec_widen_sshiftl_hi_M instruction pattern: Standard Names.

  56327.                                                              (line  730)

  56328. * vec_widen_sshiftl_lo_M instruction pattern: Standard Names.

  56329.                                                              (line  730)

  56330. * vec_widen_umult_even_M instruction pattern: Standard Names.

  56331.                                                              (line  719)

  56332. * vec_widen_umult_hi_M instruction pattern: Standard Names.  (line  719)

  56333. * vec_widen_umult_lo_M instruction pattern: Standard Names.  (line  719)

  56334. * vec_widen_umult_odd_M instruction pattern: Standard Names. (line  719)

  56335. * vec_widen_ushiftl_hi_M instruction pattern: Standard Names.

  56336.                                                              (line  730)

  56337. * vec_widen_ushiftl_lo_M instruction pattern: Standard Names.

  56338.                                                              (line  730)

  56339. * verify_flow_info:                      Maintaining the CFG.

  56340.                                                              (line  116)

  56341. * virtual operands:                      SSA Operands.       (line    6)

  56342. * VIRTUAL_INCOMING_ARGS_REGNUM:          Regs and Memory.    (line   59)

  56343. * VIRTUAL_OUTGOING_ARGS_REGNUM:          Regs and Memory.    (line   87)

  56344. * VIRTUAL_STACK_DYNAMIC_REGNUM:          Regs and Memory.    (line   78)

  56345. * VIRTUAL_STACK_VARS_REGNUM:             Regs and Memory.    (line   69)

  56346. * VLIW:                                  Processor pipeline description.

  56347.                                                              (line    6)

  56348. * VLIW <1>:                              Processor pipeline description.

  56349.                                                              (line  223)

  56350. * vlshrM3 instruction pattern:           Standard Names.     (line  844)

  56351. * VMS:                                   Filesystem.         (line   37)

  56352. * VMS_DEBUGGING_INFO:                    VMS Debug.          (line    8)

  56353. * VOIDmode:                              Machine Modes.      (line  192)

  56354. * VOID_TYPE:                             Types.              (line    6)

  56355. * volatil:                               Flags.              (line  339)

  56356. * volatil, in insn, call_insn, jump_insn, code_label, jump_table_data, barrier, and note: Flags.

  56357.                                                              (line   33)

  56358. * volatil, in label_ref and reg_label:   Flags.              (line   54)

  56359. * volatil, in mem, asm_operands, and asm_input: Flags.       (line   65)

  56360. * volatil, in reg:                       Flags.              (line  106)

  56361. * volatil, in subreg:                    Flags.              (line  184)

  56362. * volatil, in subreg <1>:                Flags.              (line  194)

  56363. * volatil, in symbol_ref:                Flags.              (line  220)

  56364. * volatile memory references:            Flags.              (line  340)

  56365. * volatile, in prefetch:                 Flags.              (line   92)

  56366. * voting between constraint alternatives: Class Preferences. (line    6)

  56367. * vrotlM3 instruction pattern:           Standard Names.     (line  844)

  56368. * vrotrM3 instruction pattern:           Standard Names.     (line  844)

  56369. * walk_dominator_tree:                   SSA.                (line  195)

  56370. * walk_gimple_op:                        Statement and operand traversals.

  56371.                                                              (line   30)

  56372. * walk_gimple_seq:                       Statement and operand traversals.

  56373.                                                              (line   47)

  56374. * walk_gimple_stmt:                      Statement and operand traversals.

  56375.                                                              (line   10)

  56376. * WCHAR_TYPE:                            Type Layout.        (line  165)

  56377. * WCHAR_TYPE_SIZE:                       Type Layout.        (line  173)

  56378. * which_alternative:                     Output Statement.   (line   58)

  56379. * WHILE_BODY:                            Statements for C++. (line    6)

  56380. * WHILE_COND:                            Statements for C++. (line    6)

  56381. * WHILE_STMT:                            Statements for C++. (line    6)

  56382. * while_ultMN instruction pattern:       Standard Names.     (line  320)

  56383. * whopr:                                 LTO.                (line    6)

  56384. * widen_ssumM3 instruction pattern:      Standard Names.     (line  587)

  56385. * widen_usumM3 instruction pattern:      Standard Names.     (line  588)

  56386. * WIDEST_HARDWARE_FP_SIZE:               Type Layout.        (line  110)

  56387. * window_save instruction pattern:       Standard Names.     (line 2092)

  56388. * WINT_TYPE:                             Type Layout.        (line  178)

  56389. * WORDS_BIG_ENDIAN:                      Storage Layout.     (line   28)

  56390. * WORDS_BIG_ENDIAN, effect on subreg:    Regs and Memory.    (line  225)

  56391. * word_mode:                             Machine Modes.      (line  458)

  56392. * WORD_REGISTER_OPERATIONS:              Misc.               (line   53)

  56393. * wpa:                                   LTO.                (line    6)

  56394. * X in constraint:                       Simple Constraints. (line  122)

  56395. * x-HOST:                                Host Fragment.      (line    6)

  56396. * XCmode:                                Machine Modes.      (line  199)

  56397. * XCOFF_DEBUGGING_INFO:                  DBX Options.        (line   12)

  56398. * XEXP:                                  Accessors.          (line    6)

  56399. * XFmode:                                Machine Modes.      (line   82)

  56400. * XImode:                                Machine Modes.      (line   54)

  56401. * XINT:                                  Accessors.          (line    6)

  56402. * xm-MACHINE.h:                          Filesystem.         (line    6)

  56403. * xm-MACHINE.h <1>:                      Host Misc.          (line    6)

  56404. * xor:                                   Arithmetic.         (line  168)

  56405. * xor, canonicalization of:              Insn Canonicalizations.

  56406.                                                              (line   94)

  56407. * xorM3 instruction pattern:             Standard Names.     (line  442)

  56408. * xorsignM3 instruction pattern:         Standard Names.     (line 1149)

  56409. * XSTR:                                  Accessors.          (line    6)

  56410. * XVEC:                                  Accessors.          (line   38)

  56411. * XVECEXP:                               Accessors.          (line   45)

  56412. * XVECLEN:                               Accessors.          (line   41)

  56413. * XWINT:                                 Accessors.          (line    6)

  56414. * zero_extend:                           Conversions.        (line   28)

  56415. * zero_extendMN2 instruction pattern:    Standard Names.     (line 1452)

  56416. * zero_extract:                          Bit-Fields.         (line   30)

  56417. * zero_extract, canonicalization of:     Insn Canonicalizations.

  56418.                                                              (line  103)

  56419. 

  56420. 

  56421. 

  56422. Tag Table:

  56423. Node: Top1789

  56424. Node: Contributing5203

  56425. Node: Portability5933

  56426. Node: Interface7721

  56427. Node: Libgcc10762

  56428. Node: Integer library routines12589

  56429. Node: Soft float library routines19557

  56430. Node: Decimal float library routines31495

  56431. Node: Fixed-point fractional library routines47253

  56432. Node: Exception handling routines147649

  56433. Node: Miscellaneous routines148756

  56434. Node: Languages150876

  56435. Node: Source Tree152423

  56436. Node: Configure Terms153005

  56437. Node: Top Level155961

  56438. Node: gcc Directory159546

  56439. Node: Subdirectories160498

  56440. Node: Configuration162666

  56441. Node: Config Fragments163386

  56442. Node: System Config164611

  56443. Node: Configuration Files165547

  56444. Node: Build168163

  56445. Node: Makefile168575

  56446. Ref: Makefile-Footnote-1175350

  56447. Ref: Makefile-Footnote-2175497

  56448. Node: Library Files175571

  56449. Node: Headers176133

  56450. Node: Documentation178216

  56451. Node: Texinfo Manuals179075

  56452. Node: Man Page Generation181404

  56453. Node: Miscellaneous Docs183317

  56454. Node: Front End184704

  56455. Node: Front End Directory188383

  56456. Node: Front End Config189699

  56457. Node: Front End Makefile192535

  56458. Node: Back End196303

  56459. Node: Testsuites201189

  56460. Node: Test Idioms202178

  56461. Node: Test Directives205576

  56462. Node: Directives206103

  56463. Node: Selectors217476

  56464. Node: Effective-Target Keywords218832

  56465. Ref: arm_fp_ok230783

  56466. Ref: arm_fp_dp_ok230950

  56467. Ref: arm_neon_ok232062

  56468. Ref: arm_neon_ok_no_float_abi232231

  56469. Ref: arm_neonv2_ok232398

  56470. Ref: arm_fp16_ok232565

  56471. Ref: arm_neon_fp16_ok232907

  56472. Ref: arm_vfp3_ok233839

  56473. Ref: arm_arch_v8a_hard_ok233982

  56474. Ref: arm_v8_1a_neon_ok234732

  56475. Ref: arm_v8_2a_fp16_scalar_ok235160

  56476. Ref: arm_v8_2a_fp16_neon_ok235611

  56477. Ref: arm_v8_2a_dotprod_neon_ok236086

  56478. Ref: arm_fp16fml_neon_ok236506

  56479. Ref: arm_coproc1_ok239006

  56480. Ref: arm_coproc2_ok239132

  56481. Ref: arm_coproc3_ok239360

  56482. Ref: arm_simd32_ok239727

  56483. Ref: arm_qbit_ok239905

  56484. Ref: arm_softfp_ok240097

  56485. Ref: arm_hard_ok240170

  56486. Ref: stack_size_et252078

  56487. Node: Add Options254471

  56488. Ref: arm_fp16_ieee255709

  56489. Ref: arm_fp16_alternative255964

  56490. Ref: stack_size_ao258359

  56491. Node: Require Support258721

  56492. Node: Final Actions261623

  56493. Node: Ada Tests270463

  56494. Node: C Tests271626

  56495. Node: LTO Testing275998

  56496. Node: gcov Testing277641

  56497. Node: profopt Testing280611

  56498. Node: compat Testing282326

  56499. Node: Torture Tests286566

  56500. Node: GIMPLE Tests288200

  56501. Node: RTL Tests289442

  56502. Node: Options290748

  56503. Node: Option file format291189

  56504. Node: Option properties298178

  56505. Node: Passes314341

  56506. Node: Parsing pass315214

  56507. Node: Gimplification pass318742

  56508. Node: Pass manager320575

  56509. Node: IPA passes322416

  56510. Node: Small IPA passes323309

  56511. Node: Regular IPA passes326746

  56512. Node: Late IPA passes331808

  56513. Node: Tree SSA passes332767

  56514. Node: RTL passes354307

  56515. Node: Optimization info366571

  56516. Node: Dump setup367390

  56517. Node: Optimization groups368519

  56518. Node: Dump files and streams369498

  56519. Node: Dump output verbosity370696

  56520. Node: Dump types371752

  56521. Node: Dump examples374094

  56522. Node: poly_int375575

  56523. Node: Overview of poly_int377055

  56524. Node: Consequences of using poly_int379659

  56525. Node: Comparisons involving poly_int381294

  56526. Node: Comparison functions for poly_int382932

  56527. Node: Properties of the poly_int comparisons384139

  56528. Node: Comparing potentially-unordered poly_ints386581

  56529. Node: Comparing ordered poly_ints387492

  56530. Node: Checking for a poly_int marker value389516

  56531. Node: Range checks on poly_ints390365

  56532. Node: Sorting poly_ints393019

  56533. Node: Arithmetic on poly_ints393792

  56534. Node: Using poly_int with C++ arithmetic operators394593

  56535. Node: wi arithmetic on poly_ints396124

  56536. Node: Division of poly_ints396976

  56537. Node: Other poly_int arithmetic398483

  56538. Node: Alignment of poly_ints399889

  56539. Node: Computing bounds on poly_ints403166

  56540. Node: Converting poly_ints404555

  56541. Node: Miscellaneous poly_int routines408102

  56542. Node: Guidelines for using poly_int408742

  56543. Node: GENERIC413674

  56544. Node: Deficiencies415496

  56545. Node: Tree overview415737

  56546. Node: Macros and Functions419861

  56547. Node: Identifiers420686

  56548. Node: Containers422295

  56549. Node: Types423452

  56550. Node: Declarations435526

  56551. Node: Working with declarations436021

  56552. Node: Internal structure441625

  56553. Node: Current structure hierarchy442009

  56554. Node: Adding new DECL node types444102

  56555. Node: Attributes448386

  56556. Node: Expression trees449630

  56557. Node: Constant expressions451384

  56558. Node: Storage References457476

  56559. Node: Unary and Binary Expressions460995

  56560. Node: Vectors481897

  56561. Node: Statements488801

  56562. Node: Basic Statements489333

  56563. Node: Blocks493960

  56564. Node: Statement Sequences495661

  56565. Node: Empty Statements495994

  56566. Node: Jumps496568

  56567. Node: Cleanups497221

  56568. Node: OpenMP499262

  56569. Node: OpenACC505107

  56570. Node: Functions506224

  56571. Node: Function Basics506695

  56572. Node: Function Properties510379

  56573. Node: Language-dependent trees513160

  56574. Node: C and C++ Trees514047

  56575. Node: Types for C++516932

  56576. Node: Namespaces521902

  56577. Node: Classes525008

  56578. Node: Functions for C++529916

  56579. Node: Statements for C++536167

  56580. Node: C++ Expressions544220

  56581. Node: GIMPLE545725

  56582. Node: Tuple representation549390

  56583. Node: Class hierarchy of GIMPLE statements556350

  56584. Node: GIMPLE instruction set561338

  56585. Node: GIMPLE Exception Handling562970

  56586. Node: Temporaries564882

  56587. Ref: Temporaries-Footnote-1566200

  56588. Node: Operands566265

  56589. Node: Compound Expressions567026

  56590. Node: Compound Lvalues567260

  56591. Node: Conditional Expressions568022

  56592. Node: Logical Operators568681

  56593. Node: Manipulating GIMPLE statements576028

  56594. Node: Tuple specific accessors581964

  56595. Node: GIMPLE_ASM582743

  56596. Node: GIMPLE_ASSIGN585126

  56597. Node: GIMPLE_BIND589830

  56598. Node: GIMPLE_CALL591644

  56599. Node: GIMPLE_CATCH595787

  56600. Node: GIMPLE_COND596937

  56601. Node: GIMPLE_DEBUG599732

  56602. Node: GIMPLE_EH_FILTER604330

  56603. Node: GIMPLE_LABEL605893

  56604. Node: GIMPLE_GOTO606506

  56605. Node: GIMPLE_NOP607029

  56606. Node: GIMPLE_OMP_ATOMIC_LOAD607391

  56607. Node: GIMPLE_OMP_ATOMIC_STORE608387

  56608. Node: GIMPLE_OMP_CONTINUE609086

  56609. Node: GIMPLE_OMP_CRITICAL610565

  56610. Node: GIMPLE_OMP_FOR611559

  56611. Node: GIMPLE_OMP_MASTER614975

  56612. Node: GIMPLE_OMP_ORDERED615353

  56613. Node: GIMPLE_OMP_PARALLEL615747

  56614. Node: GIMPLE_OMP_RETURN618516

  56615. Node: GIMPLE_OMP_SECTION619161

  56616. Node: GIMPLE_OMP_SECTIONS619821

  56617. Node: GIMPLE_OMP_SINGLE621431

  56618. Node: GIMPLE_PHI622377

  56619. Node: GIMPLE_RESX623656

  56620. Node: GIMPLE_RETURN624375

  56621. Node: GIMPLE_SWITCH624949

  56622. Node: GIMPLE_TRY626824

  56623. Node: GIMPLE_WITH_CLEANUP_EXPR628596

  56624. Node: GIMPLE sequences629475

  56625. Node: Sequence iterators632681

  56626. Node: Adding a new GIMPLE statement code641138

  56627. Node: Statement and operand traversals642483

  56628. Node: Tree SSA645075

  56629. Node: Annotations646863

  56630. Node: SSA Operands647268

  56631. Node: SSA661347

  56632. Node: Alias analysis671053

  56633. Node: Memory model674827

  56634. Node: RTL676186

  56635. Node: RTL Objects678374

  56636. Node: RTL Classes682258

  56637. Node: Accessors687557

  56638. Node: Special Accessors689730

  56639. Node: Flags695517

  56640. Node: Machine Modes710780

  56641. Node: Constants728186

  56642. Node: Regs and Memory739575

  56643. Node: Arithmetic758827

  56644. Node: Comparisons769002

  56645. Node: Bit-Fields773294

  56646. Node: Vector Operations774845

  56647. Node: Conversions776949

  56648. Node: RTL Declarations781447

  56649. Node: Side Effects782291

  56650. Node: Incdec799300

  56651. Node: Assembler802636

  56652. Node: Debug Information804181

  56653. Node: Insns806108

  56654. Node: Calls833964

  56655. Node: Sharing836557

  56656. Node: Reading RTL839752

  56657. Node: Control Flow840743

  56658. Node: Basic Blocks842511

  56659. Node: Edges847965

  56660. Node: Profile information856584

  56661. Node: Maintaining the CFG861268

  56662. Node: Liveness information867036

  56663. Node: Loop Analysis and Representation869162

  56664. Node: Loop representation870198

  56665. Node: Loop querying877564

  56666. Node: Loop manipulation880385

  56667. Node: LCSSA882721

  56668. Node: Scalar evolutions884790

  56669. Node: loop-iv888034

  56670. Node: Number of iterations889956

  56671. Node: Dependency analysis894037

  56672. Node: Machine Desc900388

  56673. Node: Overview902951

  56674. Node: Patterns904991

  56675. Node: Example909958

  56676. Node: RTL Template911419

  56677. Node: Output Template922075

  56678. Node: Output Statement926256

  56679. Node: Predicates930595

  56680. Node: Machine-Independent Predicates933513

  56681. Node: Defining Predicates938457

  56682. Node: Constraints944420

  56683. Node: Simple Constraints945889

  56684. Node: Multi-Alternative958729

  56685. Node: Class Preferences961938

  56686. Node: Modifiers962830

  56687. Node: Machine Constraints967563

  56688. Node: Disable Insn Alternatives1029667

  56689. Node: Define Constraints1033159

  56690. Node: C Constraint Interface1040553

  56691. Node: Standard Names1043680

  56692. Ref: shift patterns1079924

  56693. Ref: prologue instruction pattern1134704

  56694. Ref: window_save instruction pattern1135197

  56695. Ref: epilogue instruction pattern1135474

  56696. Node: Pattern Ordering1157450

  56697. Node: Dependent Patterns1158686

  56698. Node: Jump Patterns1160306

  56699. Ref: Jump Patterns-Footnote-11162451

  56700. Node: Looping Patterns1162499

  56701. Node: Insn Canonicalizations1168138

  56702. Node: Expander Definitions1173345

  56703. Node: Insn Splitting1181559

  56704. Node: Including Patterns1196589

  56705. Node: Peephole Definitions1198373

  56706. Node: define_peephole1199626

  56707. Node: define_peephole21205956

  56708. Node: Insn Attributes1210045

  56709. Node: Defining Attributes1211227

  56710. Ref: define_enum_attr1214719

  56711. Node: Expressions1215755

  56712. Node: Tagging Insns1222505

  56713. Node: Attr Example1226858

  56714. Node: Insn Lengths1229231

  56715. Node: Constant Attributes1232639

  56716. Node: Mnemonic Attribute1233815

  56717. Node: Delay Slots1235334

  56718. Node: Processor pipeline description1238557

  56719. Ref: Processor pipeline description-Footnote-11257369

  56720. Node: Conditional Execution1257693

  56721. Node: Define Subst1261176

  56722. Node: Define Subst Example1263211

  56723. Node: Define Subst Pattern Matching1266206

  56724. Node: Define Subst Output Template1267432

  56725. Node: Constant Definitions1269755

  56726. Ref: define_enum1273537

  56727. Node: Iterators1274025

  56728. Node: Mode Iterators1274670

  56729. Node: Defining Mode Iterators1275648

  56730. Node: Substitutions1277142

  56731. Node: Examples1279384

  56732. Node: Code Iterators1280832

  56733. Node: Int Iterators1283962

  56734. Node: Subst Iterators1286408

  56735. Node: Parameterized Names1288128

  56736. Node: Target Macros1292146

  56737. Node: Target Structure1295209

  56738. Node: Driver1297701

  56739. Node: Run-time Target1316671

  56740. Node: Per-Function Data1326204

  56741. Node: Storage Layout1328968

  56742. Node: Type Layout1356467

  56743. Node: Registers1369808

  56744. Node: Register Basics1370782

  56745. Node: Allocation Order1378344

  56746. Node: Values in Registers1380828

  56747. Node: Leaf Functions1388304

  56748. Node: Stack Registers1391163

  56749. Node: Register Classes1392435

  56750. Node: Stack and Calling1427217

  56751. Node: Frame Layout1427823

  56752. Node: Exception Handling1439658

  56753. Node: Stack Checking1445868

  56754. Node: Frame Registers1451493

  56755. Node: Elimination1460044

  56756. Node: Stack Arguments1463900

  56757. Node: Register Arguments1471096

  56758. Node: Scalar Return1495957

  56759. Node: Aggregate Return1502413

  56760. Node: Caller Saves1506967

  56761. Node: Function Entry1507709

  56762. Node: Profiling1519261

  56763. Node: Tail Calls1521371

  56764. Node: Shrink-wrapping separate components1523281

  56765. Node: Stack Smashing Protection1526322

  56766. Node: Miscellaneous Register Hooks1528811

  56767. Node: Varargs1529676

  56768. Node: Trampolines1539076

  56769. Node: Library Calls1547899

  56770. Node: Addressing Modes1552789

  56771. Node: Anchored Addresses1581796

  56772. Node: Condition Code1584439

  56773. Node: CC0 Condition Codes1586766

  56774. Node: MODE_CC Condition Codes1590012

  56775. Node: Costs1596838

  56776. Node: Scheduling1618209

  56777. Node: Sections1642131

  56778. Node: PIC1658350

  56779. Node: Assembler Format1660409

  56780. Node: File Framework1661547

  56781. Ref: TARGET_HAVE_SWITCHABLE_BSS_SECTIONS1669146

  56782. Node: Data Output1672416

  56783. Node: Uninitialized Data1680704

  56784. Node: Label Output1685718

  56785. Node: Initialization1710329

  56786. Node: Macros for Initialization1716290

  56787. Node: Instruction Output1723009

  56788. Node: Dispatch Tables1733638

  56789. Node: Exception Region Output1738467

  56790. Node: Alignment Output1745549

  56791. Node: Debugging Info1749236

  56792. Node: All Debuggers1749890

  56793. Node: DBX Options1752662

  56794. Node: DBX Hooks1758100

  56795. Node: File Names and DBX1759409

  56796. Node: DWARF1761513

  56797. Node: VMS Debug1767328

  56798. Node: Floating Point1767907

  56799. Node: Mode Switching1770662

  56800. Node: Target Attributes1775099

  56801. Node: Emulated TLS1784515

  56802. Node: MIPS Coprocessors1787905

  56803. Node: PCH Target1789064

  56804. Node: C++ ABI1790906

  56805. Node: D Language and ABI1795698

  56806. Node: Named Address Spaces1796743

  56807. Node: Misc1802670

  56808. Ref: TARGET_SHIFT_TRUNCATION_MASK1810541

  56809. Node: Host Config1865875

  56810. Node: Host Common1866944

  56811. Node: Filesystem1869318

  56812. Node: Host Misc1873433

  56813. Node: Fragments1875882

  56814. Node: Target Fragment1877077

  56815. Node: Host Fragment1887889

  56816. Node: Collect21888129

  56817. Node: Header Dirs1890765

  56818. Node: Type Information1892188

  56819. Node: GTY Options1895464

  56820. Node: Inheritance and GTY1906722

  56821. Ref: Inheritance and GTY-Footnote-11908287

  56822. Node: User GC1908557

  56823. Node: GGC Roots1912296

  56824. Node: Files1913009

  56825. Node: Invoking the garbage collector1915716

  56826. Node: Troubleshooting1917221

  56827. Node: Plugins1918296

  56828. Node: Plugins loading1919425

  56829. Node: Plugin API1920524

  56830. Node: Plugins pass1928250

  56831. Node: Plugins GC1930221

  56832. Node: Plugins description1931938

  56833. Node: Plugins attr1932474

  56834. Node: Plugins recording1934754

  56835. Node: Plugins gate1935604

  56836. Node: Plugins tracking1936195

  56837. Node: Plugins building1936783

  56838. Node: LTO1940284

  56839. Node: LTO Overview1941156

  56840. Node: LTO object file layout1946983

  56841. Node: IPA1951613

  56842. Node: WHOPR1960663

  56843. Node: Internal flags1965223

  56844. Node: Match and Simplify1966634

  56845. Node: GIMPLE API1967596

  56846. Node: The Language1970391

  56847. Node: Static Analyzer1982366

  56848. Node: Analyzer Internals1982631

  56849. Node: Debugging the Analyzer1998275

  56850. Node: User Experience Guidelines2000886

  56851. Node: Guidelines for Diagnostics2001822

  56852. Ref: input_location_example2008883

  56853. Node: Guidelines for Options2018566

  56854. Node: Funding2018743

  56855. Node: GNU Project2021250

  56856. Node: Copying2021899

  56857. Node: GNU Free Documentation License2059410

  56858. Node: Contributors2084531

  56859. Node: Option Index2125507

  56860. Node: Concept Index2126384

  56861. 

  56862. End Tag Table