visus
/
toolchain-gcc-arm-embedded


			
							<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
<html>
<!-- Copyright (C) 1988-2020 Free Software Foundation, Inc.

Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.3 or
any later version published by the Free Software Foundation; with the
Invariant Sections being "Funding Free Software", the Front-Cover
Texts being (a) (see below), and with the Back-Cover Texts being (b)
(see below).  A copy of the license is included in the section entitled
"GNU Free Documentation License".

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

A GNU Manual

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

You have freedom to copy and modify this GNU Manual, like GNU
     software.  Copies published by the Free Software Foundation raise
     funds for GNU development. -->
<!-- Created by GNU Texinfo 6.5, http://www.gnu.org/software/texinfo/ -->
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<title>Optimize Options (Using the GNU Compiler Collection (GCC))</title>

<meta name="description" content="Optimize Options (Using the GNU Compiler Collection (GCC))">
<meta name="keywords" content="Optimize Options (Using the GNU Compiler Collection (GCC))">
<meta name="resource-type" content="document">
<meta name="distribution" content="global">
<meta name="Generator" content="makeinfo">
<link href="index.html#Top" rel="start" title="Top">
<link href="Option-Index.html#Option-Index" rel="index" title="Option Index">
<link href="index.html#SEC_Contents" rel="contents" title="Table of Contents">
<link href="Invoking-GCC.html#Invoking-GCC" rel="up" title="Invoking GCC">
<link href="Instrumentation-Options.html#Instrumentation-Options" rel="next" title="Instrumentation Options">
<link href="Debugging-Options.html#Debugging-Options" rel="prev" title="Debugging Options">
<style type="text/css">
<!--
a.summary-letter {text-decoration: none}
blockquote.indentedblock {margin-right: 0em}
blockquote.smallindentedblock {margin-right: 0em; font-size: smaller}
blockquote.smallquotation {font-size: smaller}
div.display {margin-left: 3.2em}
div.example {margin-left: 3.2em}
div.lisp {margin-left: 3.2em}
div.smalldisplay {margin-left: 3.2em}
div.smallexample {margin-left: 3.2em}
div.smalllisp {margin-left: 3.2em}
kbd {font-style: oblique}
pre.display {font-family: inherit}
pre.format {font-family: inherit}
pre.menu-comment {font-family: serif}
pre.menu-preformatted {font-family: serif}
pre.smalldisplay {font-family: inherit; font-size: smaller}
pre.smallexample {font-size: smaller}
pre.smallformat {font-family: inherit; font-size: smaller}
pre.smalllisp {font-size: smaller}
span.nolinebreak {white-space: nowrap}
span.roman {font-family: initial; font-weight: normal}
span.sansserif {font-family: sans-serif; font-weight: normal}
ul.no-bullet {list-style: none}
-->
</style>


</head>

<body lang="en">
<a name="Optimize-Options"></a>
<div class="header">
<p>
Next: <a href="Instrumentation-Options.html#Instrumentation-Options" accesskey="n" rel="next">Instrumentation Options</a>, Previous: <a href="Debugging-Options.html#Debugging-Options" accesskey="p" rel="prev">Debugging Options</a>, Up: <a href="Invoking-GCC.html#Invoking-GCC" accesskey="u" rel="up">Invoking GCC</a> &nbsp; [<a href="index.html#SEC_Contents" title="Table of contents" rel="contents">Contents</a>][<a href="Option-Index.html#Option-Index" title="Index" rel="index">Index</a>]</p>
</div>
<hr>
<a name="Options-That-Control-Optimization"></a>
<h3 class="section">3.11 Options That Control Optimization</h3>
<a name="index-optimize-options"></a>
<a name="index-options_002c-optimization"></a>

<p>These options control various sorts of optimizations.
</p>
<p>Without any optimization option, the compiler&rsquo;s goal is to reduce the
cost of compilation and to make debugging produce the expected
results.  Statements are independent: if you stop the program with a
breakpoint between statements, you can then assign a new value to any
variable or change the program counter to any other statement in the
function and get exactly the results you expect from the source
code.
</p>
<p>Turning on optimization flags makes the compiler attempt to improve
the performance and/or code size at the expense of compilation time
and possibly the ability to debug the program.
</p>
<p>The compiler performs optimization based on the knowledge it has of the
program.  Compiling multiple files at once to a single output file mode allows
the compiler to use information gained from all of the files when compiling
each of them.
</p>
<p>Not all optimizations are controlled directly by a flag.  Only
optimizations that have a flag are listed in this section.
</p>
<p>Most optimizations are completely disabled at <samp>-O0</samp> or if an
<samp>-O</samp> level is not set on the command line, even if individual
optimization flags are specified.  Similarly, <samp>-Og</samp> suppresses
many optimization passes.
</p>
<p>Depending on the target and how GCC was configured, a slightly different
set of optimizations may be enabled at each <samp>-O</samp> level than
those listed here.  You can invoke GCC with <samp>-Q --help=optimizers</samp>
to find out the exact set of optimizations that are enabled at each level.
See <a href="Overall-Options.html#Overall-Options">Overall Options</a>, for examples.
</p>
<dl compact="compact">
<dt><code>-O</code></dt>
<dt><code>-O1</code></dt>
<dd><a name="index-O"></a>
<a name="index-O1"></a>
<p>Optimize.  Optimizing compilation takes somewhat more time, and a lot
more memory for a large function.
</p>
<p>With <samp>-O</samp>, the compiler tries to reduce code size and execution
time, without performing any optimizations that take a great deal of
compilation time.
</p>

<p><samp>-O</samp> turns on the following optimization flags:
</p>
<div class="smallexample">
<pre class="smallexample">-fauto-inc-dec 
-fbranch-count-reg 
-fcombine-stack-adjustments 
-fcompare-elim 
-fcprop-registers 
-fdce 
-fdefer-pop 
-fdelayed-branch 
-fdse 
-fforward-propagate 
-fguess-branch-probability 
-fif-conversion 
-fif-conversion2 
-finline-functions-called-once 
-fipa-profile 
-fipa-pure-const 
-fipa-reference 
-fipa-reference-addressable 
-fmerge-constants 
-fmove-loop-invariants 
-fomit-frame-pointer 
-freorder-blocks 
-fshrink-wrap 
-fshrink-wrap-separate 
-fsplit-wide-types 
-fssa-backprop 
-fssa-phiopt 
-ftree-bit-ccp 
-ftree-ccp 
-ftree-ch 
-ftree-coalesce-vars 
-ftree-copy-prop 
-ftree-dce 
-ftree-dominator-opts 
-ftree-dse 
-ftree-forwprop 
-ftree-fre 
-ftree-phiprop 
-ftree-pta 
-ftree-scev-cprop 
-ftree-sink 
-ftree-slsr 
-ftree-sra 
-ftree-ter 
-funit-at-a-time
</pre></div>

</dd>
<dt><code>-O2</code></dt>
<dd><a name="index-O2"></a>
<p>Optimize even more.  GCC performs nearly all supported optimizations
that do not involve a space-speed tradeoff.
As compared to <samp>-O</samp>, this option increases both compilation time
and the performance of the generated code.
</p>
<p><samp>-O2</samp> turns on all optimization flags specified by <samp>-O</samp>.  It
also turns on the following optimization flags:
</p>
<div class="smallexample">
<pre class="smallexample">-falign-functions  -falign-jumps 
-falign-labels  -falign-loops 
-fcaller-saves 
-fcode-hoisting 
-fcrossjumping 
-fcse-follow-jumps  -fcse-skip-blocks 
-fdelete-null-pointer-checks 
-fdevirtualize  -fdevirtualize-speculatively 
-fexpensive-optimizations 
-ffinite-loops 
-fgcse  -fgcse-lm  
-fhoist-adjacent-loads 
-finline-functions 
-finline-small-functions 
-findirect-inlining 
-fipa-bit-cp  -fipa-cp  -fipa-icf 
-fipa-ra  -fipa-sra  -fipa-vrp 
-fisolate-erroneous-paths-dereference 
-flra-remat 
-foptimize-sibling-calls 
-foptimize-strlen 
-fpartial-inlining 
-fpeephole2 
-freorder-blocks-algorithm=stc 
-freorder-blocks-and-partition  -freorder-functions 
-frerun-cse-after-loop  
-fschedule-insns  -fschedule-insns2 
-fsched-interblock  -fsched-spec 
-fstore-merging 
-fstrict-aliasing 
-fthread-jumps 
-ftree-builtin-call-dce 
-ftree-pre 
-ftree-switch-conversion  -ftree-tail-merge 
-ftree-vrp
</pre></div>

<p>Please note the warning under <samp>-fgcse</samp> about
invoking <samp>-O2</samp> on programs that use computed gotos.
</p>
</dd>
<dt><code>-O3</code></dt>
<dd><a name="index-O3"></a>
<p>Optimize yet more.  <samp>-O3</samp> turns on all optimizations specified
by <samp>-O2</samp> and also turns on the following optimization flags:
</p>
<div class="smallexample">
<pre class="smallexample">-fgcse-after-reload 
-fipa-cp-clone
-floop-interchange 
-floop-unroll-and-jam 
-fpeel-loops 
-fpredictive-commoning 
-fsplit-loops 
-fsplit-paths 
-ftree-loop-distribution 
-ftree-loop-vectorize 
-ftree-partial-pre 
-ftree-slp-vectorize 
-funswitch-loops 
-fvect-cost-model 
-fvect-cost-model=dynamic 
-fversion-loops-for-strides
</pre></div>

</dd>
<dt><code>-O0</code></dt>
<dd><a name="index-O0"></a>
<p>Reduce compilation time and make debugging produce the expected
results.  This is the default.
</p>
</dd>
<dt><code>-Os</code></dt>
<dd><a name="index-Os"></a>
<p>Optimize for size.  <samp>-Os</samp> enables all <samp>-O2</samp> optimizations 
except those that often increase code size:
</p>
<div class="smallexample">
<pre class="smallexample">-falign-functions  -falign-jumps 
-falign-labels  -falign-loops 
-fprefetch-loop-arrays  -freorder-blocks-algorithm=stc
</pre></div>

<p>It also enables <samp>-finline-functions</samp>, causes the compiler to tune for
code size rather than execution speed, and performs further optimizations
designed to reduce code size.
</p>
</dd>
<dt><code>-Ofast</code></dt>
<dd><a name="index-Ofast"></a>
<p>Disregard strict standards compliance.  <samp>-Ofast</samp> enables all
<samp>-O3</samp> optimizations.  It also enables optimizations that are not
valid for all standard-compliant programs.
It turns on <samp>-ffast-math</samp>, <samp>-fallow-store-data-races</samp>
and the Fortran-specific <samp>-fstack-arrays</samp>, unless
<samp>-fmax-stack-var-size</samp> is specified, and <samp>-fno-protect-parens</samp>.
</p>
</dd>
<dt><code>-Og</code></dt>
<dd><a name="index-Og"></a>
<p>Optimize debugging experience.  <samp>-Og</samp> should be the optimization
level of choice for the standard edit-compile-debug cycle, offering
a reasonable level of optimization while maintaining fast compilation
and a good debugging experience.  It is a better choice than <samp>-O0</samp>
for producing debuggable code because some compiler passes
that collect debug information are disabled at <samp>-O0</samp>.
</p>
<p>Like <samp>-O0</samp>, <samp>-Og</samp> completely disables a number of 
optimization passes so that individual options controlling them have
no effect.  Otherwise <samp>-Og</samp> enables all <samp>-O1</samp> 
optimization flags except for those that may interfere with debugging:
</p>
<div class="smallexample">
<pre class="smallexample">-fbranch-count-reg  -fdelayed-branch 
-fdse  -fif-conversion  -fif-conversion2  
-finline-functions-called-once 
-fmove-loop-invariants  -fssa-phiopt 
-ftree-bit-ccp  -ftree-dse  -ftree-pta  -ftree-sra
</pre></div>

</dd>
</dl>

<p>If you use multiple <samp>-O</samp> options, with or without level numbers,
the last such option is the one that is effective.
</p>
<p>Options of the form <samp>-f<var>flag</var></samp> specify machine-independent
flags.  Most flags have both positive and negative forms; the negative
form of <samp>-ffoo</samp> is <samp>-fno-foo</samp>.  In the table
below, only one of the forms is listed&mdash;the one you typically 
use.  You can figure out the other form by either removing &lsquo;<samp>no-</samp>&rsquo;
or adding it.
</p>
<p>The following options control specific optimizations.  They are either
activated by <samp>-O</samp> options or are related to ones that are.  You
can use the following flags in the rare cases when &ldquo;fine-tuning&rdquo; of
optimizations to be performed is desired.
</p>
<dl compact="compact">
<dt><code>-fno-defer-pop</code></dt>
<dd><a name="index-fno_002ddefer_002dpop"></a>
<a name="index-fdefer_002dpop"></a>
<p>For machines that must pop arguments after a function call, always pop 
the arguments as soon as each function returns.  
At levels <samp>-O1</samp> and higher, <samp>-fdefer-pop</samp> is the default;
this allows the compiler to let arguments accumulate on the stack for several
function calls and pop them all at once.
</p>
</dd>
<dt><code>-fforward-propagate</code></dt>
<dd><a name="index-fforward_002dpropagate"></a>
<p>Perform a forward propagation pass on RTL.  The pass tries to combine two
instructions and checks if the result can be simplified.  If loop unrolling
is active, two passes are performed and the second is scheduled after
loop unrolling.
</p>
<p>This option is enabled by default at optimization levels <samp>-O</samp>,
<samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-ffp-contract=<var>style</var></code></dt>
<dd><a name="index-ffp_002dcontract"></a>
<p><samp>-ffp-contract=off</samp> disables floating-point expression contraction.
<samp>-ffp-contract=fast</samp> enables floating-point expression contraction
such as forming of fused multiply-add operations if the target has
native support for them.
<samp>-ffp-contract=on</samp> enables floating-point expression contraction
if allowed by the language standard.  This is currently not implemented
and treated equal to <samp>-ffp-contract=off</samp>.
</p>
<p>The default is <samp>-ffp-contract=fast</samp>.
</p>
</dd>
<dt><code>-fomit-frame-pointer</code></dt>
<dd><a name="index-fomit_002dframe_002dpointer"></a>
<p>Omit the frame pointer in functions that don&rsquo;t need one.  This avoids the
instructions to save, set up and restore the frame pointer; on many targets
it also makes an extra register available.
</p>
<p>On some targets this flag has no effect because the standard calling sequence
always uses a frame pointer, so it cannot be omitted.
</p>
<p>Note that <samp>-fno-omit-frame-pointer</samp> doesn&rsquo;t guarantee the frame pointer
is used in all functions.  Several targets always omit the frame pointer in
leaf functions.
</p>
<p>Enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-foptimize-sibling-calls</code></dt>
<dd><a name="index-foptimize_002dsibling_002dcalls"></a>
<p>Optimize sibling and tail recursive calls.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-foptimize-strlen</code></dt>
<dd><a name="index-foptimize_002dstrlen"></a>
<p>Optimize various standard C string functions (e.g. <code>strlen</code>,
<code>strchr</code> or <code>strcpy</code>) and
their <code>_FORTIFY_SOURCE</code> counterparts into faster alternatives.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>.
</p>
</dd>
<dt><code>-fno-inline</code></dt>
<dd><a name="index-fno_002dinline"></a>
<a name="index-finline"></a>
<p>Do not expand any functions inline apart from those marked with
the <code>always_inline</code> attribute.  This is the default when not
optimizing.
</p>
<p>Single functions can be exempted from inlining by marking them
with the <code>noinline</code> attribute.
</p>
</dd>
<dt><code>-finline-small-functions</code></dt>
<dd><a name="index-finline_002dsmall_002dfunctions"></a>
<p>Integrate functions into their callers when their body is smaller than expected
function call code (so overall size of program gets smaller).  The compiler
heuristically decides which functions are simple enough to be worth integrating
in this way.  This inlining applies to all functions, even those not declared
inline.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-findirect-inlining</code></dt>
<dd><a name="index-findirect_002dinlining"></a>
<p>Inline also indirect calls that are discovered to be known at compile
time thanks to previous inlining.  This option has any effect only
when inlining itself is turned on by the <samp>-finline-functions</samp>
or <samp>-finline-small-functions</samp> options.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-finline-functions</code></dt>
<dd><a name="index-finline_002dfunctions"></a>
<p>Consider all functions for inlining, even if they are not declared inline.
The compiler heuristically decides which functions are worth integrating
in this way.
</p>
<p>If all calls to a given function are integrated, and the function is
declared <code>static</code>, then the function is normally not output as
assembler code in its own right.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.  Also enabled
by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-finline-functions-called-once</code></dt>
<dd><a name="index-finline_002dfunctions_002dcalled_002donce"></a>
<p>Consider all <code>static</code> functions called once for inlining into their
caller even if they are not marked <code>inline</code>.  If a call to a given
function is integrated, then the function is not output as assembler code
in its own right.
</p>
<p>Enabled at levels <samp>-O1</samp>, <samp>-O2</samp>, <samp>-O3</samp> and <samp>-Os</samp>,
but not <samp>-Og</samp>.
</p>
</dd>
<dt><code>-fearly-inlining</code></dt>
<dd><a name="index-fearly_002dinlining"></a>
<p>Inline functions marked by <code>always_inline</code> and functions whose body seems
smaller than the function call overhead early before doing
<samp>-fprofile-generate</samp> instrumentation and real inlining pass.  Doing so
makes profiling significantly cheaper and usually inlining faster on programs
having large chains of nested wrapper functions.
</p>
<p>Enabled by default.
</p>
</dd>
<dt><code>-fipa-sra</code></dt>
<dd><a name="index-fipa_002dsra"></a>
<p>Perform interprocedural scalar replacement of aggregates, removal of
unused parameters and replacement of parameters passed by reference
by parameters passed by value.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp> and <samp>-Os</samp>.
</p>
</dd>
<dt><code>-finline-limit=<var>n</var></code></dt>
<dd><a name="index-finline_002dlimit"></a>
<p>By default, GCC limits the size of functions that can be inlined.  This flag
allows coarse control of this limit.  <var>n</var> is the size of functions that
can be inlined in number of pseudo instructions.
</p>
<p>Inlining is actually controlled by a number of parameters, which may be
specified individually by using <samp>--param <var>name</var>=<var>value</var></samp>.
The <samp>-finline-limit=<var>n</var></samp> option sets some of these parameters
as follows:
</p>
<dl compact="compact">
<dt><code>max-inline-insns-single</code></dt>
<dd><p>is set to <var>n</var>/2.
</p></dd>
<dt><code>max-inline-insns-auto</code></dt>
<dd><p>is set to <var>n</var>/2.
</p></dd>
</dl>

<p>See below for a documentation of the individual
parameters controlling inlining and for the defaults of these parameters.
</p>
<p><em>Note:</em> there may be no value to <samp>-finline-limit</samp> that results
in default behavior.
</p>
<p><em>Note:</em> pseudo instruction represents, in this particular context, an
abstract measurement of function&rsquo;s size.  In no way does it represent a count
of assembly instructions and as such its exact meaning might change from one
release to an another.
</p>
</dd>
<dt><code>-fno-keep-inline-dllexport</code></dt>
<dd><a name="index-fno_002dkeep_002dinline_002ddllexport"></a>
<a name="index-fkeep_002dinline_002ddllexport"></a>
<p>This is a more fine-grained version of <samp>-fkeep-inline-functions</samp>,
which applies only to functions that are declared using the <code>dllexport</code>
attribute or declspec.  See <a href="Function-Attributes.html#Function-Attributes">Declaring Attributes of
Functions</a>.
</p>
</dd>
<dt><code>-fkeep-inline-functions</code></dt>
<dd><a name="index-fkeep_002dinline_002dfunctions"></a>
<p>In C, emit <code>static</code> functions that are declared <code>inline</code>
into the object file, even if the function has been inlined into all
of its callers.  This switch does not affect functions using the
<code>extern inline</code> extension in GNU C90.  In C++, emit any and all
inline functions into the object file.
</p>
</dd>
<dt><code>-fkeep-static-functions</code></dt>
<dd><a name="index-fkeep_002dstatic_002dfunctions"></a>
<p>Emit <code>static</code> functions into the object file, even if the function
is never used.
</p>
</dd>
<dt><code>-fkeep-static-consts</code></dt>
<dd><a name="index-fkeep_002dstatic_002dconsts"></a>
<p>Emit variables declared <code>static const</code> when optimization isn&rsquo;t turned
on, even if the variables aren&rsquo;t referenced.
</p>
<p>GCC enables this option by default.  If you want to force the compiler to
check if a variable is referenced, regardless of whether or not
optimization is turned on, use the <samp>-fno-keep-static-consts</samp> option.
</p>
</dd>
<dt><code>-fmerge-constants</code></dt>
<dd><a name="index-fmerge_002dconstants"></a>
<p>Attempt to merge identical constants (string constants and floating-point
constants) across compilation units.
</p>
<p>This option is the default for optimized compilation if the assembler and
linker support it.  Use <samp>-fno-merge-constants</samp> to inhibit this
behavior.
</p>
<p>Enabled at levels <samp>-O</samp>, <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fmerge-all-constants</code></dt>
<dd><a name="index-fmerge_002dall_002dconstants"></a>
<p>Attempt to merge identical constants and identical variables.
</p>
<p>This option implies <samp>-fmerge-constants</samp>.  In addition to
<samp>-fmerge-constants</samp> this considers e.g. even constant initialized
arrays or initialized constant variables with integral or floating-point
types.  Languages like C or C++ require each variable, including multiple
instances of the same variable in recursive calls, to have distinct locations,
so using this option results in non-conforming
behavior.
</p>
</dd>
<dt><code>-fmodulo-sched</code></dt>
<dd><a name="index-fmodulo_002dsched"></a>
<p>Perform swing modulo scheduling immediately before the first scheduling
pass.  This pass looks at innermost loops and reorders their
instructions by overlapping different iterations.
</p>
</dd>
<dt><code>-fmodulo-sched-allow-regmoves</code></dt>
<dd><a name="index-fmodulo_002dsched_002dallow_002dregmoves"></a>
<p>Perform more aggressive SMS-based modulo scheduling with register moves
allowed.  By setting this flag certain anti-dependences edges are
deleted, which triggers the generation of reg-moves based on the
life-range analysis.  This option is effective only with
<samp>-fmodulo-sched</samp> enabled.
</p>
</dd>
<dt><code>-fno-branch-count-reg</code></dt>
<dd><a name="index-fno_002dbranch_002dcount_002dreg"></a>
<a name="index-fbranch_002dcount_002dreg"></a>
<p>Disable the optimization pass that scans for opportunities to use 
&ldquo;decrement and branch&rdquo; instructions on a count register instead of
instruction sequences that decrement a register, compare it against zero, and
then branch based upon the result.  This option is only meaningful on
architectures that support such instructions, which include x86, PowerPC,
IA-64 and S/390.  Note that the <samp>-fno-branch-count-reg</samp> option
doesn&rsquo;t remove the decrement and branch instructions from the generated
instruction stream introduced by other optimization passes.
</p>
<p>The default is <samp>-fbranch-count-reg</samp> at <samp>-O1</samp> and higher,
except for <samp>-Og</samp>.
</p>
</dd>
<dt><code>-fno-function-cse</code></dt>
<dd><a name="index-fno_002dfunction_002dcse"></a>
<a name="index-ffunction_002dcse"></a>
<p>Do not put function addresses in registers; make each instruction that
calls a constant function contain the function&rsquo;s address explicitly.
</p>
<p>This option results in less efficient code, but some strange hacks
that alter the assembler output may be confused by the optimizations
performed when this option is not used.
</p>
<p>The default is <samp>-ffunction-cse</samp>
</p>
</dd>
<dt><code>-fno-zero-initialized-in-bss</code></dt>
<dd><a name="index-fno_002dzero_002dinitialized_002din_002dbss"></a>
<a name="index-fzero_002dinitialized_002din_002dbss"></a>
<p>If the target supports a BSS section, GCC by default puts variables that
are initialized to zero into BSS.  This can save space in the resulting
code.
</p>
<p>This option turns off this behavior because some programs explicitly
rely on variables going to the data section&mdash;e.g., so that the
resulting executable can find the beginning of that section and/or make
assumptions based on that.
</p>
<p>The default is <samp>-fzero-initialized-in-bss</samp>.
</p>
</dd>
<dt><code>-fthread-jumps</code></dt>
<dd><a name="index-fthread_002djumps"></a>
<p>Perform optimizations that check to see if a jump branches to a
location where another comparison subsumed by the first is found.  If
so, the first branch is redirected to either the destination of the
second branch or a point immediately following it, depending on whether
the condition is known to be true or false.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fsplit-wide-types</code></dt>
<dd><a name="index-fsplit_002dwide_002dtypes"></a>
<p>When using a type that occupies multiple registers, such as <code>long
long</code> on a 32-bit system, split the registers apart and allocate them
independently.  This normally generates better code for those types,
but may make debugging more difficult.
</p>
<p>Enabled at levels <samp>-O</samp>, <samp>-O2</samp>, <samp>-O3</samp>,
<samp>-Os</samp>.
</p>
</dd>
<dt><code>-fsplit-wide-types-early</code></dt>
<dd><a name="index-fsplit_002dwide_002dtypes_002dearly"></a>
<p>Fully split wide types early, instead of very late.
This option has no effect unless <samp>-fsplit-wide-types</samp> is turned on.
</p>
<p>This is the default on some targets.
</p>
</dd>
<dt><code>-fcse-follow-jumps</code></dt>
<dd><a name="index-fcse_002dfollow_002djumps"></a>
<p>In common subexpression elimination (CSE), scan through jump instructions
when the target of the jump is not reached by any other path.  For
example, when CSE encounters an <code>if</code> statement with an
<code>else</code> clause, CSE follows the jump when the condition
tested is false.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fcse-skip-blocks</code></dt>
<dd><a name="index-fcse_002dskip_002dblocks"></a>
<p>This is similar to <samp>-fcse-follow-jumps</samp>, but causes CSE to
follow jumps that conditionally skip over blocks.  When CSE
encounters a simple <code>if</code> statement with no else clause,
<samp>-fcse-skip-blocks</samp> causes CSE to follow the jump around the
body of the <code>if</code>.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-frerun-cse-after-loop</code></dt>
<dd><a name="index-frerun_002dcse_002dafter_002dloop"></a>
<p>Re-run common subexpression elimination after loop optimizations are
performed.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fgcse</code></dt>
<dd><a name="index-fgcse"></a>
<p>Perform a global common subexpression elimination pass.
This pass also performs global constant and copy propagation.
</p>
<p><em>Note:</em> When compiling a program using computed gotos, a GCC
extension, you may get better run-time performance if you disable
the global common subexpression elimination pass by adding
<samp>-fno-gcse</samp> to the command line.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fgcse-lm</code></dt>
<dd><a name="index-fgcse_002dlm"></a>
<p>When <samp>-fgcse-lm</samp> is enabled, global common subexpression elimination
attempts to move loads that are only killed by stores into themselves.  This
allows a loop containing a load/store sequence to be changed to a load outside
the loop, and a copy/store within the loop.
</p>
<p>Enabled by default when <samp>-fgcse</samp> is enabled.
</p>
</dd>
<dt><code>-fgcse-sm</code></dt>
<dd><a name="index-fgcse_002dsm"></a>
<p>When <samp>-fgcse-sm</samp> is enabled, a store motion pass is run after
global common subexpression elimination.  This pass attempts to move
stores out of loops.  When used in conjunction with <samp>-fgcse-lm</samp>,
loops containing a load/store sequence can be changed to a load before
the loop and a store after the loop.
</p>
<p>Not enabled at any optimization level.
</p>
</dd>
<dt><code>-fgcse-las</code></dt>
<dd><a name="index-fgcse_002dlas"></a>
<p>When <samp>-fgcse-las</samp> is enabled, the global common subexpression
elimination pass eliminates redundant loads that come after stores to the
same memory location (both partial and full redundancies).
</p>
<p>Not enabled at any optimization level.
</p>
</dd>
<dt><code>-fgcse-after-reload</code></dt>
<dd><a name="index-fgcse_002dafter_002dreload"></a>
<p>When <samp>-fgcse-after-reload</samp> is enabled, a redundant load elimination
pass is performed after reload.  The purpose of this pass is to clean up
redundant spilling.
</p>
<p>Enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-faggressive-loop-optimizations</code></dt>
<dd><a name="index-faggressive_002dloop_002doptimizations"></a>
<p>This option tells the loop optimizer to use language constraints to
derive bounds for the number of iterations of a loop.  This assumes that
loop code does not invoke undefined behavior by for example causing signed
integer overflows or out-of-bound array accesses.  The bounds for the
number of iterations of a loop are used to guide loop unrolling and peeling
and loop exit test optimizations.
This option is enabled by default.
</p>
</dd>
<dt><code>-funconstrained-commons</code></dt>
<dd><a name="index-funconstrained_002dcommons"></a>
<p>This option tells the compiler that variables declared in common blocks
(e.g. Fortran) may later be overridden with longer trailing arrays. This
prevents certain optimizations that depend on knowing the array bounds.
</p>
</dd>
<dt><code>-fcrossjumping</code></dt>
<dd><a name="index-fcrossjumping"></a>
<p>Perform cross-jumping transformation.
This transformation unifies equivalent code and saves code size.  The
resulting code may or may not perform better than without cross-jumping.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fauto-inc-dec</code></dt>
<dd><a name="index-fauto_002dinc_002ddec"></a>
<p>Combine increments or decrements of addresses with memory accesses.
This pass is always skipped on architectures that do not have
instructions to support this.  Enabled by default at <samp>-O</samp> and
higher on architectures that support this.
</p>
</dd>
<dt><code>-fdce</code></dt>
<dd><a name="index-fdce"></a>
<p>Perform dead code elimination (DCE) on RTL.
Enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fdse</code></dt>
<dd><a name="index-fdse"></a>
<p>Perform dead store elimination (DSE) on RTL.
Enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fif-conversion</code></dt>
<dd><a name="index-fif_002dconversion"></a>
<p>Attempt to transform conditional jumps into branch-less equivalents.  This
includes use of conditional moves, min, max, set flags and abs instructions, and
some tricks doable by standard arithmetics.  The use of conditional execution
on chips where it is available is controlled by <samp>-fif-conversion2</samp>.
</p>
<p>Enabled at levels <samp>-O</samp>, <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>, but
not with <samp>-Og</samp>.
</p>
</dd>
<dt><code>-fif-conversion2</code></dt>
<dd><a name="index-fif_002dconversion2"></a>
<p>Use conditional execution (where available) to transform conditional jumps into
branch-less equivalents.
</p>
<p>Enabled at levels <samp>-O</samp>, <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>, but
not with <samp>-Og</samp>.
</p>
</dd>
<dt><code>-fdeclone-ctor-dtor</code></dt>
<dd><a name="index-fdeclone_002dctor_002ddtor"></a>
<p>The C++ ABI requires multiple entry points for constructors and
destructors: one for a base subobject, one for a complete object, and
one for a virtual destructor that calls operator delete afterwards.
For a hierarchy with virtual bases, the base and complete variants are
clones, which means two copies of the function.  With this option, the
base and complete variants are changed to be thunks that call a common
implementation.
</p>
<p>Enabled by <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fdelete-null-pointer-checks</code></dt>
<dd><a name="index-fdelete_002dnull_002dpointer_002dchecks"></a>
<p>Assume that programs cannot safely dereference null pointers, and that
no code or data element resides at address zero.
This option enables simple constant
folding optimizations at all optimization levels.  In addition, other
optimization passes in GCC use this flag to control global dataflow
analyses that eliminate useless checks for null pointers; these assume
that a memory access to address zero always results in a trap, so
that if a pointer is checked after it has already been dereferenced,
it cannot be null.
</p>
<p>Note however that in some environments this assumption is not true.
Use <samp>-fno-delete-null-pointer-checks</samp> to disable this optimization
for programs that depend on that behavior.
</p>
<p>This option is enabled by default on most targets.  On Nios II ELF, it
defaults to off.  On AVR, CR16, and MSP430, this option is completely disabled.
</p>
<p>Passes that use the dataflow information
are enabled independently at different optimization levels.
</p>
</dd>
<dt><code>-fdevirtualize</code></dt>
<dd><a name="index-fdevirtualize"></a>
<p>Attempt to convert calls to virtual functions to direct calls.  This
is done both within a procedure and interprocedurally as part of
indirect inlining (<samp>-findirect-inlining</samp>) and interprocedural constant
propagation (<samp>-fipa-cp</samp>).
Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fdevirtualize-speculatively</code></dt>
<dd><a name="index-fdevirtualize_002dspeculatively"></a>
<p>Attempt to convert calls to virtual functions to speculative direct calls.
Based on the analysis of the type inheritance graph, determine for a given call
the set of likely targets. If the set is small, preferably of size 1, change
the call into a conditional deciding between direct and indirect calls.  The
speculative calls enable more optimizations, such as inlining.  When they seem
useless after further optimization, they are converted back into original form.
</p>
</dd>
<dt><code>-fdevirtualize-at-ltrans</code></dt>
<dd><a name="index-fdevirtualize_002dat_002dltrans"></a>
<p>Stream extra information needed for aggressive devirtualization when running
the link-time optimizer in local transformation mode.  
This option enables more devirtualization but
significantly increases the size of streamed data. For this reason it is
disabled by default.
</p>
</dd>
<dt><code>-fexpensive-optimizations</code></dt>
<dd><a name="index-fexpensive_002doptimizations"></a>
<p>Perform a number of minor optimizations that are relatively expensive.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-free</code></dt>
<dd><a name="index-free-1"></a>
<p>Attempt to remove redundant extension instructions.  This is especially
helpful for the x86-64 architecture, which implicitly zero-extends in 64-bit
registers after writing to their lower 32-bit half.
</p>
<p>Enabled for Alpha, AArch64 and x86 at levels <samp>-O2</samp>,
<samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fno-lifetime-dse</code></dt>
<dd><a name="index-fno_002dlifetime_002ddse"></a>
<a name="index-flifetime_002ddse"></a>
<p>In C++ the value of an object is only affected by changes within its
lifetime: when the constructor begins, the object has an indeterminate
value, and any changes during the lifetime of the object are dead when
the object is destroyed.  Normally dead store elimination will take
advantage of this; if your code relies on the value of the object
storage persisting beyond the lifetime of the object, you can use this
flag to disable this optimization.  To preserve stores before the
constructor starts (e.g. because your operator new clears the object
storage) but still treat the object as dead after the destructor, you
can use <samp>-flifetime-dse=1</samp>.  The default behavior can be
explicitly selected with <samp>-flifetime-dse=2</samp>.
<samp>-flifetime-dse=0</samp> is equivalent to <samp>-fno-lifetime-dse</samp>.
</p>
</dd>
<dt><code>-flive-range-shrinkage</code></dt>
<dd><a name="index-flive_002drange_002dshrinkage"></a>
<p>Attempt to decrease register pressure through register live range
shrinkage.  This is helpful for fast processors with small or moderate
size register sets.
</p>
</dd>
<dt><code>-fira-algorithm=<var>algorithm</var></code></dt>
<dd><a name="index-fira_002dalgorithm"></a>
<p>Use the specified coloring algorithm for the integrated register
allocator.  The <var>algorithm</var> argument can be &lsquo;<samp>priority</samp>&rsquo;, which
specifies Chow&rsquo;s priority coloring, or &lsquo;<samp>CB</samp>&rsquo;, which specifies
Chaitin-Briggs coloring.  Chaitin-Briggs coloring is not implemented
for all architectures, but for those targets that do support it, it is
the default because it generates better code.
</p>
</dd>
<dt><code>-fira-region=<var>region</var></code></dt>
<dd><a name="index-fira_002dregion"></a>
<p>Use specified regions for the integrated register allocator.  The
<var>region</var> argument should be one of the following:
</p>
<dl compact="compact">
<dt>&lsquo;<samp>all</samp>&rsquo;</dt>
<dd><p>Use all loops as register allocation regions.
This can give the best results for machines with a small and/or
irregular register set.
</p>
</dd>
<dt>&lsquo;<samp>mixed</samp>&rsquo;</dt>
<dd><p>Use all loops except for loops with small register pressure 
as the regions.  This value usually gives
the best results in most cases and for most architectures,
and is enabled by default when compiling with optimization for speed
(<samp>-O</samp>, <samp>-O2</samp>, &hellip;).
</p>
</dd>
<dt>&lsquo;<samp>one</samp>&rsquo;</dt>
<dd><p>Use all functions as a single region.  
This typically results in the smallest code size, and is enabled by default for
<samp>-Os</samp> or <samp>-O0</samp>.
</p>
</dd>
</dl>

</dd>
<dt><code>-fira-hoist-pressure</code></dt>
<dd><a name="index-fira_002dhoist_002dpressure"></a>
<p>Use IRA to evaluate register pressure in the code hoisting pass for
decisions to hoist expressions.  This option usually results in smaller
code, but it can slow the compiler down.
</p>
<p>This option is enabled at level <samp>-Os</samp> for all targets.
</p>
</dd>
<dt><code>-fira-loop-pressure</code></dt>
<dd><a name="index-fira_002dloop_002dpressure"></a>
<p>Use IRA to evaluate register pressure in loops for decisions to move
loop invariants.  This option usually results in generation
of faster and smaller code on machines with large register files (&gt;= 32
registers), but it can slow the compiler down.
</p>
<p>This option is enabled at level <samp>-O3</samp> for some targets.
</p>
</dd>
<dt><code>-fno-ira-share-save-slots</code></dt>
<dd><a name="index-fno_002dira_002dshare_002dsave_002dslots"></a>
<a name="index-fira_002dshare_002dsave_002dslots"></a>
<p>Disable sharing of stack slots used for saving call-used hard
registers living through a call.  Each hard register gets a
separate stack slot, and as a result function stack frames are
larger.
</p>
</dd>
<dt><code>-fno-ira-share-spill-slots</code></dt>
<dd><a name="index-fno_002dira_002dshare_002dspill_002dslots"></a>
<a name="index-fira_002dshare_002dspill_002dslots"></a>
<p>Disable sharing of stack slots allocated for pseudo-registers.  Each
pseudo-register that does not get a hard register gets a separate
stack slot, and as a result function stack frames are larger.
</p>
</dd>
<dt><code>-flra-remat</code></dt>
<dd><a name="index-flra_002dremat"></a>
<p>Enable CFG-sensitive rematerialization in LRA.  Instead of loading
values of spilled pseudos, LRA tries to rematerialize (recalculate)
values if it is profitable.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fdelayed-branch</code></dt>
<dd><a name="index-fdelayed_002dbranch"></a>
<p>If supported for the target machine, attempt to reorder instructions
to exploit instruction slots available after delayed branch
instructions.
</p>
<p>Enabled at levels <samp>-O</samp>, <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>,
but not at <samp>-Og</samp>.
</p>
</dd>
<dt><code>-fschedule-insns</code></dt>
<dd><a name="index-fschedule_002dinsns"></a>
<p>If supported for the target machine, attempt to reorder instructions to
eliminate execution stalls due to required data being unavailable.  This
helps machines that have slow floating point or memory load instructions
by allowing other instructions to be issued until the result of the load
or floating-point instruction is required.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>.
</p>
</dd>
<dt><code>-fschedule-insns2</code></dt>
<dd><a name="index-fschedule_002dinsns2"></a>
<p>Similar to <samp>-fschedule-insns</samp>, but requests an additional pass of
instruction scheduling after register allocation has been done.  This is
especially useful on machines with a relatively small number of
registers and where memory load instructions take more than one cycle.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fno-sched-interblock</code></dt>
<dd><a name="index-fno_002dsched_002dinterblock"></a>
<a name="index-fsched_002dinterblock"></a>
<p>Disable instruction scheduling across basic blocks, which
is normally enabled when scheduling before register allocation, i.e.
with <samp>-fschedule-insns</samp> or at <samp>-O2</samp> or higher.
</p>
</dd>
<dt><code>-fno-sched-spec</code></dt>
<dd><a name="index-fno_002dsched_002dspec"></a>
<a name="index-fsched_002dspec"></a>
<p>Disable speculative motion of non-load instructions, which
is normally enabled when scheduling before register allocation, i.e.
with <samp>-fschedule-insns</samp> or at <samp>-O2</samp> or higher.
</p>
</dd>
<dt><code>-fsched-pressure</code></dt>
<dd><a name="index-fsched_002dpressure"></a>
<p>Enable register pressure sensitive insn scheduling before register
allocation.  This only makes sense when scheduling before register
allocation is enabled, i.e. with <samp>-fschedule-insns</samp> or at
<samp>-O2</samp> or higher.  Usage of this option can improve the
generated code and decrease its size by preventing register pressure
increase above the number of available hard registers and subsequent
spills in register allocation.
</p>
</dd>
<dt><code>-fsched-spec-load</code></dt>
<dd><a name="index-fsched_002dspec_002dload"></a>
<p>Allow speculative motion of some load instructions.  This only makes
sense when scheduling before register allocation, i.e. with
<samp>-fschedule-insns</samp> or at <samp>-O2</samp> or higher.
</p>
</dd>
<dt><code>-fsched-spec-load-dangerous</code></dt>
<dd><a name="index-fsched_002dspec_002dload_002ddangerous"></a>
<p>Allow speculative motion of more load instructions.  This only makes
sense when scheduling before register allocation, i.e. with
<samp>-fschedule-insns</samp> or at <samp>-O2</samp> or higher.
</p>
</dd>
<dt><code>-fsched-stalled-insns</code></dt>
<dt><code>-fsched-stalled-insns=<var>n</var></code></dt>
<dd><a name="index-fsched_002dstalled_002dinsns"></a>
<p>Define how many insns (if any) can be moved prematurely from the queue
of stalled insns into the ready list during the second scheduling pass.
<samp>-fno-sched-stalled-insns</samp> means that no insns are moved
prematurely, <samp>-fsched-stalled-insns=0</samp> means there is no limit
on how many queued insns can be moved prematurely.
<samp>-fsched-stalled-insns</samp> without a value is equivalent to
<samp>-fsched-stalled-insns=1</samp>.
</p>
</dd>
<dt><code>-fsched-stalled-insns-dep</code></dt>
<dt><code>-fsched-stalled-insns-dep=<var>n</var></code></dt>
<dd><a name="index-fsched_002dstalled_002dinsns_002ddep"></a>
<p>Define how many insn groups (cycles) are examined for a dependency
on a stalled insn that is a candidate for premature removal from the queue
of stalled insns.  This has an effect only during the second scheduling pass,
and only if <samp>-fsched-stalled-insns</samp> is used.
<samp>-fno-sched-stalled-insns-dep</samp> is equivalent to
<samp>-fsched-stalled-insns-dep=0</samp>.
<samp>-fsched-stalled-insns-dep</samp> without a value is equivalent to
<samp>-fsched-stalled-insns-dep=1</samp>.
</p>
</dd>
<dt><code>-fsched2-use-superblocks</code></dt>
<dd><a name="index-fsched2_002duse_002dsuperblocks"></a>
<p>When scheduling after register allocation, use superblock scheduling.
This allows motion across basic block boundaries,
resulting in faster schedules.  This option is experimental, as not all machine
descriptions used by GCC model the CPU closely enough to avoid unreliable
results from the algorithm.
</p>
<p>This only makes sense when scheduling after register allocation, i.e. with
<samp>-fschedule-insns2</samp> or at <samp>-O2</samp> or higher.
</p>
</dd>
<dt><code>-fsched-group-heuristic</code></dt>
<dd><a name="index-fsched_002dgroup_002dheuristic"></a>
<p>Enable the group heuristic in the scheduler.  This heuristic favors
the instruction that belongs to a schedule group.  This is enabled
by default when scheduling is enabled, i.e. with <samp>-fschedule-insns</samp>
or <samp>-fschedule-insns2</samp> or at <samp>-O2</samp> or higher.
</p>
</dd>
<dt><code>-fsched-critical-path-heuristic</code></dt>
<dd><a name="index-fsched_002dcritical_002dpath_002dheuristic"></a>
<p>Enable the critical-path heuristic in the scheduler.  This heuristic favors
instructions on the critical path.  This is enabled by default when
scheduling is enabled, i.e. with <samp>-fschedule-insns</samp>
or <samp>-fschedule-insns2</samp> or at <samp>-O2</samp> or higher.
</p>
</dd>
<dt><code>-fsched-spec-insn-heuristic</code></dt>
<dd><a name="index-fsched_002dspec_002dinsn_002dheuristic"></a>
<p>Enable the speculative instruction heuristic in the scheduler.  This
heuristic favors speculative instructions with greater dependency weakness.
This is enabled by default when scheduling is enabled, i.e.
with <samp>-fschedule-insns</samp> or <samp>-fschedule-insns2</samp>
or at <samp>-O2</samp> or higher.
</p>
</dd>
<dt><code>-fsched-rank-heuristic</code></dt>
<dd><a name="index-fsched_002drank_002dheuristic"></a>
<p>Enable the rank heuristic in the scheduler.  This heuristic favors
the instruction belonging to a basic block with greater size or frequency.
This is enabled by default when scheduling is enabled, i.e.
with <samp>-fschedule-insns</samp> or <samp>-fschedule-insns2</samp> or
at <samp>-O2</samp> or higher.
</p>
</dd>
<dt><code>-fsched-last-insn-heuristic</code></dt>
<dd><a name="index-fsched_002dlast_002dinsn_002dheuristic"></a>
<p>Enable the last-instruction heuristic in the scheduler.  This heuristic
favors the instruction that is less dependent on the last instruction
scheduled.  This is enabled by default when scheduling is enabled,
i.e. with <samp>-fschedule-insns</samp> or <samp>-fschedule-insns2</samp> or
at <samp>-O2</samp> or higher.
</p>
</dd>
<dt><code>-fsched-dep-count-heuristic</code></dt>
<dd><a name="index-fsched_002ddep_002dcount_002dheuristic"></a>
<p>Enable the dependent-count heuristic in the scheduler.  This heuristic
favors the instruction that has more instructions depending on it.
This is enabled by default when scheduling is enabled, i.e.
with <samp>-fschedule-insns</samp> or <samp>-fschedule-insns2</samp> or
at <samp>-O2</samp> or higher.
</p>
</dd>
<dt><code>-freschedule-modulo-scheduled-loops</code></dt>
<dd><a name="index-freschedule_002dmodulo_002dscheduled_002dloops"></a>
<p>Modulo scheduling is performed before traditional scheduling.  If a loop
is modulo scheduled, later scheduling passes may change its schedule.  
Use this option to control that behavior.
</p>
</dd>
<dt><code>-fselective-scheduling</code></dt>
<dd><a name="index-fselective_002dscheduling"></a>
<p>Schedule instructions using selective scheduling algorithm.  Selective
scheduling runs instead of the first scheduler pass.
</p>
</dd>
<dt><code>-fselective-scheduling2</code></dt>
<dd><a name="index-fselective_002dscheduling2"></a>
<p>Schedule instructions using selective scheduling algorithm.  Selective
scheduling runs instead of the second scheduler pass.
</p>
</dd>
<dt><code>-fsel-sched-pipelining</code></dt>
<dd><a name="index-fsel_002dsched_002dpipelining"></a>
<p>Enable software pipelining of innermost loops during selective scheduling.
This option has no effect unless one of <samp>-fselective-scheduling</samp> or
<samp>-fselective-scheduling2</samp> is turned on.
</p>
</dd>
<dt><code>-fsel-sched-pipelining-outer-loops</code></dt>
<dd><a name="index-fsel_002dsched_002dpipelining_002douter_002dloops"></a>
<p>When pipelining loops during selective scheduling, also pipeline outer loops.
This option has no effect unless <samp>-fsel-sched-pipelining</samp> is turned on.
</p>
</dd>
<dt><code>-fsemantic-interposition</code></dt>
<dd><a name="index-fsemantic_002dinterposition"></a>
<p>Some object formats, like ELF, allow interposing of symbols by the 
dynamic linker.
This means that for symbols exported from the DSO, the compiler cannot perform
interprocedural propagation, inlining and other optimizations in anticipation
that the function or variable in question may change. While this feature is
useful, for example, to rewrite memory allocation functions by a debugging
implementation, it is expensive in the terms of code quality.
With <samp>-fno-semantic-interposition</samp> the compiler assumes that 
if interposition happens for functions the overwriting function will have 
precisely the same semantics (and side effects). 
Similarly if interposition happens
for variables, the constructor of the variable will be the same. The flag
has no effect for functions explicitly declared inline 
(where it is never allowed for interposition to change semantics) 
and for symbols explicitly declared weak.
</p>
</dd>
<dt><code>-fshrink-wrap</code></dt>
<dd><a name="index-fshrink_002dwrap"></a>
<p>Emit function prologues only before parts of the function that need it,
rather than at the top of the function.  This flag is enabled by default at
<samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fshrink-wrap-separate</code></dt>
<dd><a name="index-fshrink_002dwrap_002dseparate"></a>
<p>Shrink-wrap separate parts of the prologue and epilogue separately, so that
those parts are only executed when needed.
This option is on by default, but has no effect unless <samp>-fshrink-wrap</samp>
is also turned on and the target supports this.
</p>
</dd>
<dt><code>-fcaller-saves</code></dt>
<dd><a name="index-fcaller_002dsaves"></a>
<p>Enable allocation of values to registers that are clobbered by
function calls, by emitting extra instructions to save and restore the
registers around such calls.  Such allocation is done only when it
seems to result in better code.
</p>
<p>This option is always enabled by default on certain machines, usually
those which have no call-preserved registers to use instead.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fcombine-stack-adjustments</code></dt>
<dd><a name="index-fcombine_002dstack_002dadjustments"></a>
<p>Tracks stack adjustments (pushes and pops) and stack memory references
and then tries to find ways to combine them.
</p>
<p>Enabled by default at <samp>-O1</samp> and higher.
</p>
</dd>
<dt><code>-fipa-ra</code></dt>
<dd><a name="index-fipa_002dra"></a>
<p>Use caller save registers for allocation if those registers are not used by
any called function.  In that case it is not necessary to save and restore
them around calls.  This is only possible if called functions are part of
same compilation unit as current function and they are compiled before it.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>, however the option
is disabled if generated code will be instrumented for profiling
(<samp>-p</samp>, or <samp>-pg</samp>) or if callee&rsquo;s register usage cannot be known
exactly (this happens on targets that do not expose prologues
and epilogues in RTL).
</p>
</dd>
<dt><code>-fconserve-stack</code></dt>
<dd><a name="index-fconserve_002dstack"></a>
<p>Attempt to minimize stack usage.  The compiler attempts to use less
stack space, even if that makes the program slower.  This option
implies setting the <samp>large-stack-frame</samp> parameter to 100
and the <samp>large-stack-frame-growth</samp> parameter to 400.
</p>
</dd>
<dt><code>-ftree-reassoc</code></dt>
<dd><a name="index-ftree_002dreassoc"></a>
<p>Perform reassociation on trees.  This flag is enabled by default
at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fcode-hoisting</code></dt>
<dd><a name="index-fcode_002dhoisting"></a>
<p>Perform code hoisting.  Code hoisting tries to move the
evaluation of expressions executed on all paths to the function exit
as early as possible.  This is especially useful as a code size
optimization, but it often helps for code speed as well.
This flag is enabled by default at <samp>-O2</samp> and higher.
</p>
</dd>
<dt><code>-ftree-pre</code></dt>
<dd><a name="index-ftree_002dpre"></a>
<p>Perform partial redundancy elimination (PRE) on trees.  This flag is
enabled by default at <samp>-O2</samp> and <samp>-O3</samp>.
</p>
</dd>
<dt><code>-ftree-partial-pre</code></dt>
<dd><a name="index-ftree_002dpartial_002dpre"></a>
<p>Make partial redundancy elimination (PRE) more aggressive.  This flag is
enabled by default at <samp>-O3</samp>.
</p>
</dd>
<dt><code>-ftree-forwprop</code></dt>
<dd><a name="index-ftree_002dforwprop"></a>
<p>Perform forward propagation on trees.  This flag is enabled by default
at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-ftree-fre</code></dt>
<dd><a name="index-ftree_002dfre"></a>
<p>Perform full redundancy elimination (FRE) on trees.  The difference
between FRE and PRE is that FRE only considers expressions
that are computed on all paths leading to the redundant computation.
This analysis is faster than PRE, though it exposes fewer redundancies.
This flag is enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-ftree-phiprop</code></dt>
<dd><a name="index-ftree_002dphiprop"></a>
<p>Perform hoisting of loads from conditional pointers on trees.  This
pass is enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fhoist-adjacent-loads</code></dt>
<dd><a name="index-fhoist_002dadjacent_002dloads"></a>
<p>Speculatively hoist loads from both branches of an if-then-else if the
loads are from adjacent locations in the same structure and the target
architecture has a conditional move instruction.  This flag is enabled
by default at <samp>-O2</samp> and higher.
</p>
</dd>
<dt><code>-ftree-copy-prop</code></dt>
<dd><a name="index-ftree_002dcopy_002dprop"></a>
<p>Perform copy propagation on trees.  This pass eliminates unnecessary
copy operations.  This flag is enabled by default at <samp>-O</samp> and
higher.
</p>
</dd>
<dt><code>-fipa-pure-const</code></dt>
<dd><a name="index-fipa_002dpure_002dconst"></a>
<p>Discover which functions are pure or constant.
Enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fipa-reference</code></dt>
<dd><a name="index-fipa_002dreference"></a>
<p>Discover which static variables do not escape the
compilation unit.
Enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fipa-reference-addressable</code></dt>
<dd><a name="index-fipa_002dreference_002daddressable"></a>
<p>Discover read-only, write-only and non-addressable static variables.
Enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fipa-stack-alignment</code></dt>
<dd><a name="index-fipa_002dstack_002dalignment"></a>
<p>Reduce stack alignment on call sites if possible.
Enabled by default.
</p>
</dd>
<dt><code>-fipa-pta</code></dt>
<dd><a name="index-fipa_002dpta"></a>
<p>Perform interprocedural pointer analysis and interprocedural modification
and reference analysis.  This option can cause excessive memory and
compile-time usage on large compilation units.  It is not enabled by
default at any optimization level.
</p>
</dd>
<dt><code>-fipa-profile</code></dt>
<dd><a name="index-fipa_002dprofile"></a>
<p>Perform interprocedural profile propagation.  The functions called only from
cold functions are marked as cold. Also functions executed once (such as
<code>cold</code>, <code>noreturn</code>, static constructors or destructors) are identified. Cold
functions and loop less parts of functions executed once are then optimized for
size.
Enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fipa-cp</code></dt>
<dd><a name="index-fipa_002dcp"></a>
<p>Perform interprocedural constant propagation.
This optimization analyzes the program to determine when values passed
to functions are constants and then optimizes accordingly.
This optimization can substantially increase performance
if the application has constants passed to functions.
This flag is enabled by default at <samp>-O2</samp>, <samp>-Os</samp> and <samp>-O3</samp>.
It is also enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-fipa-cp-clone</code></dt>
<dd><a name="index-fipa_002dcp_002dclone"></a>
<p>Perform function cloning to make interprocedural constant propagation stronger.
When enabled, interprocedural constant propagation performs function cloning
when externally visible function can be called with constant arguments.
Because this optimization can create multiple copies of functions,
it may significantly increase code size
(see <samp>--param ipa-cp-unit-growth=<var>value</var></samp>).
This flag is enabled by default at <samp>-O3</samp>.
It is also enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-fipa-bit-cp</code></dt>
<dd><a name="index-fipa_002dbit_002dcp"></a>
<p>When enabled, perform interprocedural bitwise constant
propagation. This flag is enabled by default at <samp>-O2</samp> and
by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
It requires that <samp>-fipa-cp</samp> is enabled.  
</p>
</dd>
<dt><code>-fipa-vrp</code></dt>
<dd><a name="index-fipa_002dvrp"></a>
<p>When enabled, perform interprocedural propagation of value
ranges. This flag is enabled by default at <samp>-O2</samp>. It requires
that <samp>-fipa-cp</samp> is enabled.
</p>
</dd>
<dt><code>-fipa-icf</code></dt>
<dd><a name="index-fipa_002dicf"></a>
<p>Perform Identical Code Folding for functions and read-only variables.
The optimization reduces code size and may disturb unwind stacks by replacing
a function by equivalent one with a different name. The optimization works
more effectively with link-time optimization enabled.
</p>
<p>Although the behavior is similar to the Gold Linker&rsquo;s ICF optimization, GCC ICF
works on different levels and thus the optimizations are not same - there are
equivalences that are found only by GCC and equivalences found only by Gold.
</p>
<p>This flag is enabled by default at <samp>-O2</samp> and <samp>-Os</samp>.
</p>
</dd>
<dt><code>-flive-patching=<var>level</var></code></dt>
<dd><a name="index-flive_002dpatching"></a>
<p>Control GCC&rsquo;s optimizations to produce output suitable for live-patching.
</p>
<p>If the compiler&rsquo;s optimization uses a function&rsquo;s body or information extracted
from its body to optimize/change another function, the latter is called an
impacted function of the former.  If a function is patched, its impacted
functions should be patched too.
</p>
<p>The impacted functions are determined by the compiler&rsquo;s interprocedural
optimizations.  For example, a caller is impacted when inlining a function
into its caller,
cloning a function and changing its caller to call this new clone,
or extracting a function&rsquo;s pureness/constness information to optimize
its direct or indirect callers, etc.
</p>
<p>Usually, the more IPA optimizations enabled, the larger the number of
impacted functions for each function.  In order to control the number of
impacted functions and more easily compute the list of impacted function,
IPA optimizations can be partially enabled at two different levels.
</p>
<p>The <var>level</var> argument should be one of the following:
</p>
<dl compact="compact">
<dt>&lsquo;<samp>inline-clone</samp>&rsquo;</dt>
<dd>
<p>Only enable inlining and cloning optimizations, which includes inlining,
cloning, interprocedural scalar replacement of aggregates and partial inlining.
As a result, when patching a function, all its callers and its clones&rsquo;
callers are impacted, therefore need to be patched as well.
</p>
<p><samp>-flive-patching=inline-clone</samp> disables the following optimization flags:
</p><div class="smallexample">
<pre class="smallexample">-fwhole-program  -fipa-pta  -fipa-reference  -fipa-ra 
-fipa-icf  -fipa-icf-functions  -fipa-icf-variables 
-fipa-bit-cp  -fipa-vrp  -fipa-pure-const  -fipa-reference-addressable 
-fipa-stack-alignment
</pre></div>

</dd>
<dt>&lsquo;<samp>inline-only-static</samp>&rsquo;</dt>
<dd>
<p>Only enable inlining of static functions.
As a result, when patching a static function, all its callers are impacted
and so need to be patched as well.
</p>
<p>In addition to all the flags that <samp>-flive-patching=inline-clone</samp>
disables,
<samp>-flive-patching=inline-only-static</samp> disables the following additional
optimization flags:
</p><div class="smallexample">
<pre class="smallexample">-fipa-cp-clone  -fipa-sra  -fpartial-inlining  -fipa-cp
</pre></div>

</dd>
</dl>

<p>When <samp>-flive-patching</samp> is specified without any value, the default value
is <var>inline-clone</var>.
</p>
<p>This flag is disabled by default.
</p>
<p>Note that <samp>-flive-patching</samp> is not supported with link-time optimization
(<samp>-flto</samp>).
</p>
</dd>
<dt><code>-fisolate-erroneous-paths-dereference</code></dt>
<dd><a name="index-fisolate_002derroneous_002dpaths_002ddereference"></a>
<p>Detect paths that trigger erroneous or undefined behavior due to
dereferencing a null pointer.  Isolate those paths from the main control
flow and turn the statement with erroneous or undefined behavior into a trap.
This flag is enabled by default at <samp>-O2</samp> and higher and depends on
<samp>-fdelete-null-pointer-checks</samp> also being enabled.
</p>
</dd>
<dt><code>-fisolate-erroneous-paths-attribute</code></dt>
<dd><a name="index-fisolate_002derroneous_002dpaths_002dattribute"></a>
<p>Detect paths that trigger erroneous or undefined behavior due to a null value
being used in a way forbidden by a <code>returns_nonnull</code> or <code>nonnull</code>
attribute.  Isolate those paths from the main control flow and turn the
statement with erroneous or undefined behavior into a trap.  This is not
currently enabled, but may be enabled by <samp>-O2</samp> in the future.
</p>
</dd>
<dt><code>-ftree-sink</code></dt>
<dd><a name="index-ftree_002dsink"></a>
<p>Perform forward store motion on trees.  This flag is
enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-ftree-bit-ccp</code></dt>
<dd><a name="index-ftree_002dbit_002dccp"></a>
<p>Perform sparse conditional bit constant propagation on trees and propagate
pointer alignment information.
This pass only operates on local scalar variables and is enabled by default
at <samp>-O1</samp> and higher, except for <samp>-Og</samp>.
It requires that <samp>-ftree-ccp</samp> is enabled.
</p>
</dd>
<dt><code>-ftree-ccp</code></dt>
<dd><a name="index-ftree_002dccp"></a>
<p>Perform sparse conditional constant propagation (CCP) on trees.  This
pass only operates on local scalar variables and is enabled by default
at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fssa-backprop</code></dt>
<dd><a name="index-fssa_002dbackprop"></a>
<p>Propagate information about uses of a value up the definition chain
in order to simplify the definitions.  For example, this pass strips
sign operations if the sign of a value never matters.  The flag is
enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fssa-phiopt</code></dt>
<dd><a name="index-fssa_002dphiopt"></a>
<p>Perform pattern matching on SSA PHI nodes to optimize conditional
code.  This pass is enabled by default at <samp>-O1</samp> and higher,
except for <samp>-Og</samp>.
</p>
</dd>
<dt><code>-ftree-switch-conversion</code></dt>
<dd><a name="index-ftree_002dswitch_002dconversion"></a>
<p>Perform conversion of simple initializations in a switch to
initializations from a scalar array.  This flag is enabled by default
at <samp>-O2</samp> and higher.
</p>
</dd>
<dt><code>-ftree-tail-merge</code></dt>
<dd><a name="index-ftree_002dtail_002dmerge"></a>
<p>Look for identical code sequences.  When found, replace one with a jump to the
other.  This optimization is known as tail merging or cross jumping.  This flag
is enabled by default at <samp>-O2</samp> and higher.  The compilation time
in this pass can
be limited using <samp>max-tail-merge-comparisons</samp> parameter and
<samp>max-tail-merge-iterations</samp> parameter.
</p>
</dd>
<dt><code>-ftree-dce</code></dt>
<dd><a name="index-ftree_002ddce"></a>
<p>Perform dead code elimination (DCE) on trees.  This flag is enabled by
default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-ftree-builtin-call-dce</code></dt>
<dd><a name="index-ftree_002dbuiltin_002dcall_002ddce"></a>
<p>Perform conditional dead code elimination (DCE) for calls to built-in functions
that may set <code>errno</code> but are otherwise free of side effects.  This flag is
enabled by default at <samp>-O2</samp> and higher if <samp>-Os</samp> is not also
specified.
</p>
</dd>
<dt><code>-ffinite-loops</code></dt>
<dd><a name="index-ffinite_002dloops"></a>
<a name="index-fno_002dfinite_002dloops"></a>
<p>Assume that a loop with an exit will eventually take the exit and not loop
indefinitely.  This allows the compiler to remove loops that otherwise have
no side-effects, not considering eventual endless looping as such.
</p>
<p>This option is enabled by default at <samp>-O2</samp> for C++ with -std=c++11
or higher.
</p>
</dd>
<dt><code>-ftree-dominator-opts</code></dt>
<dd><a name="index-ftree_002ddominator_002dopts"></a>
<p>Perform a variety of simple scalar cleanups (constant/copy
propagation, redundancy elimination, range propagation and expression
simplification) based on a dominator tree traversal.  This also
performs jump threading (to reduce jumps to jumps). This flag is
enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-ftree-dse</code></dt>
<dd><a name="index-ftree_002ddse"></a>
<p>Perform dead store elimination (DSE) on trees.  A dead store is a store into
a memory location that is later overwritten by another store without
any intervening loads.  In this case the earlier store can be deleted.  This
flag is enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-ftree-ch</code></dt>
<dd><a name="index-ftree_002dch"></a>
<p>Perform loop header copying on trees.  This is beneficial since it increases
effectiveness of code motion optimizations.  It also saves one jump.  This flag
is enabled by default at <samp>-O</samp> and higher.  It is not enabled
for <samp>-Os</samp>, since it usually increases code size.
</p>
</dd>
<dt><code>-ftree-loop-optimize</code></dt>
<dd><a name="index-ftree_002dloop_002doptimize"></a>
<p>Perform loop optimizations on trees.  This flag is enabled by default
at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-ftree-loop-linear</code></dt>
<dt><code>-floop-strip-mine</code></dt>
<dt><code>-floop-block</code></dt>
<dd><a name="index-ftree_002dloop_002dlinear"></a>
<a name="index-floop_002dstrip_002dmine"></a>
<a name="index-floop_002dblock"></a>
<p>Perform loop nest optimizations.  Same as
<samp>-floop-nest-optimize</samp>.  To use this code transformation, GCC has
to be configured with <samp>--with-isl</samp> to enable the Graphite loop
transformation infrastructure.
</p>
</dd>
<dt><code>-fgraphite-identity</code></dt>
<dd><a name="index-fgraphite_002didentity"></a>
<p>Enable the identity transformation for graphite.  For every SCoP we generate
the polyhedral representation and transform it back to gimple.  Using
<samp>-fgraphite-identity</samp> we can check the costs or benefits of the
GIMPLE -&gt; GRAPHITE -&gt; GIMPLE transformation.  Some minimal optimizations
are also performed by the code generator isl, like index splitting and
dead code elimination in loops.
</p>
</dd>
<dt><code>-floop-nest-optimize</code></dt>
<dd><a name="index-floop_002dnest_002doptimize"></a>
<p>Enable the isl based loop nest optimizer.  This is a generic loop nest
optimizer based on the Pluto optimization algorithms.  It calculates a loop
structure optimized for data-locality and parallelism.  This option
is experimental.
</p>
</dd>
<dt><code>-floop-parallelize-all</code></dt>
<dd><a name="index-floop_002dparallelize_002dall"></a>
<p>Use the Graphite data dependence analysis to identify loops that can
be parallelized.  Parallelize all the loops that can be analyzed to
not contain loop carried dependences without checking that it is
profitable to parallelize the loops.
</p>
</dd>
<dt><code>-ftree-coalesce-vars</code></dt>
<dd><a name="index-ftree_002dcoalesce_002dvars"></a>
<p>While transforming the program out of the SSA representation, attempt to
reduce copying by coalescing versions of different user-defined
variables, instead of just compiler temporaries.  This may severely
limit the ability to debug an optimized program compiled with
<samp>-fno-var-tracking-assignments</samp>.  In the negated form, this flag
prevents SSA coalescing of user variables.  This option is enabled by
default if optimization is enabled, and it does very little otherwise.
</p>
</dd>
<dt><code>-ftree-loop-if-convert</code></dt>
<dd><a name="index-ftree_002dloop_002dif_002dconvert"></a>
<p>Attempt to transform conditional jumps in the innermost loops to
branch-less equivalents.  The intent is to remove control-flow from
the innermost loops in order to improve the ability of the
vectorization pass to handle these loops.  This is enabled by default
if vectorization is enabled.
</p>
</dd>
<dt><code>-ftree-loop-distribution</code></dt>
<dd><a name="index-ftree_002dloop_002ddistribution"></a>
<p>Perform loop distribution.  This flag can improve cache performance on
big loop bodies and allow further loop optimizations, like
parallelization or vectorization, to take place.  For example, the loop
</p><div class="smallexample">
<pre class="smallexample">DO I = 1, N
  A(I) = B(I) + C
  D(I) = E(I) * F
ENDDO
</pre></div>
<p>is transformed to
</p><div class="smallexample">
<pre class="smallexample">DO I = 1, N
   A(I) = B(I) + C
ENDDO
DO I = 1, N
   D(I) = E(I) * F
ENDDO
</pre></div>
<p>This flag is enabled by default at <samp>-O3</samp>.
It is also enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-ftree-loop-distribute-patterns</code></dt>
<dd><a name="index-ftree_002dloop_002ddistribute_002dpatterns"></a>
<p>Perform loop distribution of patterns that can be code generated with
calls to a library.  This flag is enabled by default at <samp>-O2</samp> and
higher, and by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
<p>This pass distributes the initialization loops and generates a call to
memset zero.  For example, the loop
</p><div class="smallexample">
<pre class="smallexample">DO I = 1, N
  A(I) = 0
  B(I) = A(I) + I
ENDDO
</pre></div>
<p>is transformed to
</p><div class="smallexample">
<pre class="smallexample">DO I = 1, N
   A(I) = 0
ENDDO
DO I = 1, N
   B(I) = A(I) + I
ENDDO
</pre></div>
<p>and the initialization loop is transformed into a call to memset zero.
This flag is enabled by default at <samp>-O3</samp>.
It is also enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-floop-interchange</code></dt>
<dd><a name="index-floop_002dinterchange"></a>
<p>Perform loop interchange outside of graphite.  This flag can improve cache
performance on loop nest and allow further loop optimizations, like
vectorization, to take place.  For example, the loop
</p><div class="smallexample">
<pre class="smallexample">for (int i = 0; i &lt; N; i++)
  for (int j = 0; j &lt; N; j++)
    for (int k = 0; k &lt; N; k++)
      c[i][j] = c[i][j] + a[i][k]*b[k][j];
</pre></div>
<p>is transformed to
</p><div class="smallexample">
<pre class="smallexample">for (int i = 0; i &lt; N; i++)
  for (int k = 0; k &lt; N; k++)
    for (int j = 0; j &lt; N; j++)
      c[i][j] = c[i][j] + a[i][k]*b[k][j];
</pre></div>
<p>This flag is enabled by default at <samp>-O3</samp>.
It is also enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-floop-unroll-and-jam</code></dt>
<dd><a name="index-floop_002dunroll_002dand_002djam"></a>
<p>Apply unroll and jam transformations on feasible loops.  In a loop
nest this unrolls the outer loop by some factor and fuses the resulting
multiple inner loops.  This flag is enabled by default at <samp>-O3</samp>.
It is also enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-ftree-loop-im</code></dt>
<dd><a name="index-ftree_002dloop_002dim"></a>
<p>Perform loop invariant motion on trees.  This pass moves only invariants that
are hard to handle at RTL level (function calls, operations that expand to
nontrivial sequences of insns).  With <samp>-funswitch-loops</samp> it also moves
operands of conditions that are invariant out of the loop, so that we can use
just trivial invariantness analysis in loop unswitching.  The pass also includes
store motion.
</p>
</dd>
<dt><code>-ftree-loop-ivcanon</code></dt>
<dd><a name="index-ftree_002dloop_002divcanon"></a>
<p>Create a canonical counter for number of iterations in loops for which
determining number of iterations requires complicated analysis.  Later
optimizations then may determine the number easily.  Useful especially
in connection with unrolling.
</p>
</dd>
<dt><code>-ftree-scev-cprop</code></dt>
<dd><a name="index-ftree_002dscev_002dcprop"></a>
<p>Perform final value replacement.  If a variable is modified in a loop
in such a way that its value when exiting the loop can be determined using
only its initial value and the number of loop iterations, replace uses of
the final value by such a computation, provided it is sufficiently cheap.
This reduces data dependencies and may allow further simplifications.
Enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-fivopts</code></dt>
<dd><a name="index-fivopts"></a>
<p>Perform induction variable optimizations (strength reduction, induction
variable merging and induction variable elimination) on trees.
</p>
</dd>
<dt><code>-ftree-parallelize-loops=n</code></dt>
<dd><a name="index-ftree_002dparallelize_002dloops"></a>
<p>Parallelize loops, i.e., split their iteration space to run in n threads.
This is only possible for loops whose iterations are independent
and can be arbitrarily reordered.  The optimization is only
profitable on multiprocessor machines, for loops that are CPU-intensive,
rather than constrained e.g. by memory bandwidth.  This option
implies <samp>-pthread</samp>, and thus is only supported on targets
that have support for <samp>-pthread</samp>.
</p>
</dd>
<dt><code>-ftree-pta</code></dt>
<dd><a name="index-ftree_002dpta"></a>
<p>Perform function-local points-to analysis on trees.  This flag is
enabled by default at <samp>-O1</samp> and higher, except for <samp>-Og</samp>.
</p>
</dd>
<dt><code>-ftree-sra</code></dt>
<dd><a name="index-ftree_002dsra"></a>
<p>Perform scalar replacement of aggregates.  This pass replaces structure
references with scalars to prevent committing structures to memory too
early.  This flag is enabled by default at <samp>-O1</samp> and higher,
except for <samp>-Og</samp>.
</p>
</dd>
<dt><code>-fstore-merging</code></dt>
<dd><a name="index-fstore_002dmerging"></a>
<p>Perform merging of narrow stores to consecutive memory addresses.  This pass
merges contiguous stores of immediate values narrower than a word into fewer
wider stores to reduce the number of instructions.  This is enabled by default
at <samp>-O2</samp> and higher as well as <samp>-Os</samp>.
</p>
</dd>
<dt><code>-ftree-ter</code></dt>
<dd><a name="index-ftree_002dter"></a>
<p>Perform temporary expression replacement during the SSA-&gt;normal phase.  Single
use/single def temporaries are replaced at their use location with their
defining expression.  This results in non-GIMPLE code, but gives the expanders
much more complex trees to work on resulting in better RTL generation.  This is
enabled by default at <samp>-O</samp> and higher.
</p>
</dd>
<dt><code>-ftree-slsr</code></dt>
<dd><a name="index-ftree_002dslsr"></a>
<p>Perform straight-line strength reduction on trees.  This recognizes related
expressions involving multiplications and replaces them by less expensive
calculations when possible.  This is enabled by default at <samp>-O</samp> and
higher.
</p>
</dd>
<dt><code>-ftree-vectorize</code></dt>
<dd><a name="index-ftree_002dvectorize"></a>
<p>Perform vectorization on trees. This flag enables <samp>-ftree-loop-vectorize</samp>
and <samp>-ftree-slp-vectorize</samp> if not explicitly specified.
</p>
</dd>
<dt><code>-ftree-loop-vectorize</code></dt>
<dd><a name="index-ftree_002dloop_002dvectorize"></a>
<p>Perform loop vectorization on trees. This flag is enabled by default at
<samp>-O3</samp> and by <samp>-ftree-vectorize</samp>, <samp>-fprofile-use</samp>,
and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-ftree-slp-vectorize</code></dt>
<dd><a name="index-ftree_002dslp_002dvectorize"></a>
<p>Perform basic block vectorization on trees. This flag is enabled by default at
<samp>-O3</samp> and by <samp>-ftree-vectorize</samp>, <samp>-fprofile-use</samp>,
and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-fvect-cost-model=<var>model</var></code></dt>
<dd><a name="index-fvect_002dcost_002dmodel"></a>
<p>Alter the cost model used for vectorization.  The <var>model</var> argument
should be one of &lsquo;<samp>unlimited</samp>&rsquo;, &lsquo;<samp>dynamic</samp>&rsquo; or &lsquo;<samp>cheap</samp>&rsquo;.
With the &lsquo;<samp>unlimited</samp>&rsquo; model the vectorized code-path is assumed
to be profitable while with the &lsquo;<samp>dynamic</samp>&rsquo; model a runtime check
guards the vectorized code-path to enable it only for iteration
counts that will likely execute faster than when executing the original
scalar loop.  The &lsquo;<samp>cheap</samp>&rsquo; model disables vectorization of
loops where doing so would be cost prohibitive for example due to
required runtime checks for data dependence or alignment but otherwise
is equal to the &lsquo;<samp>dynamic</samp>&rsquo; model.
The default cost model depends on other optimization flags and is
either &lsquo;<samp>dynamic</samp>&rsquo; or &lsquo;<samp>cheap</samp>&rsquo;.
</p>
</dd>
<dt><code>-fsimd-cost-model=<var>model</var></code></dt>
<dd><a name="index-fsimd_002dcost_002dmodel"></a>
<p>Alter the cost model used for vectorization of loops marked with the OpenMP
simd directive.  The <var>model</var> argument should be one of
&lsquo;<samp>unlimited</samp>&rsquo;, &lsquo;<samp>dynamic</samp>&rsquo;, &lsquo;<samp>cheap</samp>&rsquo;.  All values of <var>model</var>
have the same meaning as described in <samp>-fvect-cost-model</samp> and by
default a cost model defined with <samp>-fvect-cost-model</samp> is used.
</p>
</dd>
<dt><code>-ftree-vrp</code></dt>
<dd><a name="index-ftree_002dvrp"></a>
<p>Perform Value Range Propagation on trees.  This is similar to the
constant propagation pass, but instead of values, ranges of values are
propagated.  This allows the optimizers to remove unnecessary range
checks like array bound checks and null pointer checks.  This is
enabled by default at <samp>-O2</samp> and higher.  Null pointer check
elimination is only done if <samp>-fdelete-null-pointer-checks</samp> is
enabled.
</p>
</dd>
<dt><code>-fsplit-paths</code></dt>
<dd><a name="index-fsplit_002dpaths"></a>
<p>Split paths leading to loop backedges.  This can improve dead code
elimination and common subexpression elimination.  This is enabled by
default at <samp>-O3</samp> and above.
</p>
</dd>
<dt><code>-fsplit-ivs-in-unroller</code></dt>
<dd><a name="index-fsplit_002divs_002din_002dunroller"></a>
<p>Enables expression of values of induction variables in later iterations
of the unrolled loop using the value in the first iteration.  This breaks
long dependency chains, thus improving efficiency of the scheduling passes.
</p>
<p>A combination of <samp>-fweb</samp> and CSE is often sufficient to obtain the
same effect.  However, that is not reliable in cases where the loop body
is more complicated than a single basic block.  It also does not work at all
on some architectures due to restrictions in the CSE pass.
</p>
<p>This optimization is enabled by default.
</p>
</dd>
<dt><code>-fvariable-expansion-in-unroller</code></dt>
<dd><a name="index-fvariable_002dexpansion_002din_002dunroller"></a>
<p>With this option, the compiler creates multiple copies of some
local variables when unrolling a loop, which can result in superior code.
</p>
<p>This optimization is enabled by default for PowerPC targets, but disabled
by default otherwise.
</p>
</dd>
<dt><code>-fpartial-inlining</code></dt>
<dd><a name="index-fpartial_002dinlining"></a>
<p>Inline parts of functions.  This option has any effect only
when inlining itself is turned on by the <samp>-finline-functions</samp>
or <samp>-finline-small-functions</samp> options.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fpredictive-commoning</code></dt>
<dd><a name="index-fpredictive_002dcommoning"></a>
<p>Perform predictive commoning optimization, i.e., reusing computations
(especially memory loads and stores) performed in previous
iterations of loops.
</p>
<p>This option is enabled at level <samp>-O3</samp>.
It is also enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-fprefetch-loop-arrays</code></dt>
<dd><a name="index-fprefetch_002dloop_002darrays"></a>
<p>If supported by the target machine, generate instructions to prefetch
memory to improve the performance of loops that access large arrays.
</p>
<p>This option may generate better or worse code; results are highly
dependent on the structure of loops within the source code.
</p>
<p>Disabled at level <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fno-printf-return-value</code></dt>
<dd><a name="index-fno_002dprintf_002dreturn_002dvalue"></a>
<a name="index-fprintf_002dreturn_002dvalue"></a>
<p>Do not substitute constants for known return value of formatted output
functions such as <code>sprintf</code>, <code>snprintf</code>, <code>vsprintf</code>, and
<code>vsnprintf</code> (but not <code>printf</code> of <code>fprintf</code>).  This
transformation allows GCC to optimize or even eliminate branches based
on the known return value of these functions called with arguments that
are either constant, or whose values are known to be in a range that
makes determining the exact return value possible.  For example, when
<samp>-fprintf-return-value</samp> is in effect, both the branch and the
body of the <code>if</code> statement (but not the call to <code>snprint</code>)
can be optimized away when <code>i</code> is a 32-bit or smaller integer
because the return value is guaranteed to be at most 8.
</p>
<div class="smallexample">
<pre class="smallexample">char buf[9];
if (snprintf (buf, &quot;%08x&quot;, i) &gt;= sizeof buf)
  &hellip;
</pre></div>

<p>The <samp>-fprintf-return-value</samp> option relies on other optimizations
and yields best results with <samp>-O2</samp> and above.  It works in tandem
with the <samp>-Wformat-overflow</samp> and <samp>-Wformat-truncation</samp>
options.  The <samp>-fprintf-return-value</samp> option is enabled by default.
</p>
</dd>
<dt><code>-fno-peephole</code></dt>
<dt><code>-fno-peephole2</code></dt>
<dd><a name="index-fno_002dpeephole"></a>
<a name="index-fpeephole"></a>
<a name="index-fno_002dpeephole2"></a>
<a name="index-fpeephole2"></a>
<p>Disable any machine-specific peephole optimizations.  The difference
between <samp>-fno-peephole</samp> and <samp>-fno-peephole2</samp> is in how they
are implemented in the compiler; some targets use one, some use the
other, a few use both.
</p>
<p><samp>-fpeephole</samp> is enabled by default.
<samp>-fpeephole2</samp> enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fno-guess-branch-probability</code></dt>
<dd><a name="index-fno_002dguess_002dbranch_002dprobability"></a>
<a name="index-fguess_002dbranch_002dprobability"></a>
<p>Do not guess branch probabilities using heuristics.
</p>
<p>GCC uses heuristics to guess branch probabilities if they are
not provided by profiling feedback (<samp>-fprofile-arcs</samp>).  These
heuristics are based on the control flow graph.  If some branch probabilities
are specified by <code>__builtin_expect</code>, then the heuristics are
used to guess branch probabilities for the rest of the control flow graph,
taking the <code>__builtin_expect</code> info into account.  The interactions
between the heuristics and <code>__builtin_expect</code> can be complex, and in
some cases, it may be useful to disable the heuristics so that the effects
of <code>__builtin_expect</code> are easier to understand.
</p>
<p>It is also possible to specify expected probability of the expression
with <code>__builtin_expect_with_probability</code> built-in function.
</p>
<p>The default is <samp>-fguess-branch-probability</samp> at levels
<samp>-O</samp>, <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-freorder-blocks</code></dt>
<dd><a name="index-freorder_002dblocks"></a>
<p>Reorder basic blocks in the compiled function in order to reduce number of
taken branches and improve code locality.
</p>
<p>Enabled at levels <samp>-O</samp>, <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-freorder-blocks-algorithm=<var>algorithm</var></code></dt>
<dd><a name="index-freorder_002dblocks_002dalgorithm"></a>
<p>Use the specified algorithm for basic block reordering.  The
<var>algorithm</var> argument can be &lsquo;<samp>simple</samp>&rsquo;, which does not increase
code size (except sometimes due to secondary effects like alignment),
or &lsquo;<samp>stc</samp>&rsquo;, the &ldquo;software trace cache&rdquo; algorithm, which tries to
put all often executed code together, minimizing the number of branches
executed by making extra copies of code.
</p>
<p>The default is &lsquo;<samp>simple</samp>&rsquo; at levels <samp>-O</samp>, <samp>-Os</samp>, and
&lsquo;<samp>stc</samp>&rsquo; at levels <samp>-O2</samp>, <samp>-O3</samp>.
</p>
</dd>
<dt><code>-freorder-blocks-and-partition</code></dt>
<dd><a name="index-freorder_002dblocks_002dand_002dpartition"></a>
<p>In addition to reordering basic blocks in the compiled function, in order
to reduce number of taken branches, partitions hot and cold basic blocks
into separate sections of the assembly and <samp>.o</samp> files, to improve
paging and cache locality performance.
</p>
<p>This optimization is automatically turned off in the presence of
exception handling or unwind tables (on targets using setjump/longjump or target specific scheme), for linkonce sections, for functions with a user-defined
section attribute and on any architecture that does not support named
sections.  When <samp>-fsplit-stack</samp> is used this option is not
enabled by default (to avoid linker errors), but may be enabled
explicitly (if using a working linker).
</p>
<p>Enabled for x86 at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-freorder-functions</code></dt>
<dd><a name="index-freorder_002dfunctions"></a>
<p>Reorder functions in the object file in order to
improve code locality.  This is implemented by using special
subsections <code>.text.hot</code> for most frequently executed functions and
<code>.text.unlikely</code> for unlikely executed functions.  Reordering is done by
the linker so object file format must support named sections and linker must
place them in a reasonable way.
</p>
<p>This option isn&rsquo;t effective unless you either provide profile feedback
(see <samp>-fprofile-arcs</samp> for details) or manually annotate functions with 
<code>hot</code> or <code>cold</code> attributes (see <a href="Common-Function-Attributes.html#Common-Function-Attributes">Common Function Attributes</a>).
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fstrict-aliasing</code></dt>
<dd><a name="index-fstrict_002daliasing"></a>
<p>Allow the compiler to assume the strictest aliasing rules applicable to
the language being compiled.  For C (and C++), this activates
optimizations based on the type of expressions.  In particular, an
object of one type is assumed never to reside at the same address as an
object of a different type, unless the types are almost the same.  For
example, an <code>unsigned int</code> can alias an <code>int</code>, but not a
<code>void*</code> or a <code>double</code>.  A character type may alias any other
type.
</p>
<a name="Type_002dpunning"></a><p>Pay special attention to code like this:
</p><div class="smallexample">
<pre class="smallexample">union a_union {
  int i;
  double d;
};

int f() {
  union a_union t;
  t.d = 3.0;
  return t.i;
}
</pre></div>
<p>The practice of reading from a different union member than the one most
recently written to (called &ldquo;type-punning&rdquo;) is common.  Even with
<samp>-fstrict-aliasing</samp>, type-punning is allowed, provided the memory
is accessed through the union type.  So, the code above works as
expected.  See <a href="Structures-unions-enumerations-and-bit_002dfields-implementation.html#Structures-unions-enumerations-and-bit_002dfields-implementation">Structures unions enumerations and bit-fields implementation</a>.  However, this code might not:
</p><div class="smallexample">
<pre class="smallexample">int f() {
  union a_union t;
  int* ip;
  t.d = 3.0;
  ip = &amp;t.i;
  return *ip;
}
</pre></div>

<p>Similarly, access by taking the address, casting the resulting pointer
and dereferencing the result has undefined behavior, even if the cast
uses a union type, e.g.:
</p><div class="smallexample">
<pre class="smallexample">int f() {
  double d = 3.0;
  return ((union a_union *) &amp;d)-&gt;i;
}
</pre></div>

<p>The <samp>-fstrict-aliasing</samp> option is enabled at levels
<samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-falign-functions</code></dt>
<dt><code>-falign-functions=<var>n</var></code></dt>
<dt><code>-falign-functions=<var>n</var>:<var>m</var></code></dt>
<dt><code>-falign-functions=<var>n</var>:<var>m</var>:<var>n2</var></code></dt>
<dt><code>-falign-functions=<var>n</var>:<var>m</var>:<var>n2</var>:<var>m2</var></code></dt>
<dd><a name="index-falign_002dfunctions"></a>
<p>Align the start of functions to the next power-of-two greater than or
equal to <var>n</var>, skipping up to <var>m</var>-1 bytes.  This ensures that at
least the first <var>m</var> bytes of the function can be fetched by the CPU
without crossing an <var>n</var>-byte alignment boundary.
</p>
<p>If <var>m</var> is not specified, it defaults to <var>n</var>.
</p>
<p>Examples: <samp>-falign-functions=32</samp> aligns functions to the next
32-byte boundary, <samp>-falign-functions=24</samp> aligns to the next
32-byte boundary only if this can be done by skipping 23 bytes or less,
<samp>-falign-functions=32:7</samp> aligns to the next
32-byte boundary only if this can be done by skipping 6 bytes or less.
</p>
<p>The second pair of <var>n2</var>:<var>m2</var> values allows you to specify
a secondary alignment: <samp>-falign-functions=64:7:32:3</samp> aligns to
the next 64-byte boundary if this can be done by skipping 6 bytes or less,
otherwise aligns to the next 32-byte boundary if this can be done
by skipping 2 bytes or less.
If <var>m2</var> is not specified, it defaults to <var>n2</var>.
</p>
<p>Some assemblers only support this flag when <var>n</var> is a power of two;
in that case, it is rounded up.
</p>
<p><samp>-fno-align-functions</samp> and <samp>-falign-functions=1</samp> are
equivalent and mean that functions are not aligned.
</p>
<p>If <var>n</var> is not specified or is zero, use a machine-dependent default.
The maximum allowed <var>n</var> option value is 65536.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>.
</p>
</dd>
<dt><code>-flimit-function-alignment</code></dt>
<dd><p>If this option is enabled, the compiler tries to avoid unnecessarily
overaligning functions. It attempts to instruct the assembler to align
by the amount specified by <samp>-falign-functions</samp>, but not to
skip more bytes than the size of the function.
</p>
</dd>
<dt><code>-falign-labels</code></dt>
<dt><code>-falign-labels=<var>n</var></code></dt>
<dt><code>-falign-labels=<var>n</var>:<var>m</var></code></dt>
<dt><code>-falign-labels=<var>n</var>:<var>m</var>:<var>n2</var></code></dt>
<dt><code>-falign-labels=<var>n</var>:<var>m</var>:<var>n2</var>:<var>m2</var></code></dt>
<dd><a name="index-falign_002dlabels"></a>
<p>Align all branch targets to a power-of-two boundary.
</p>
<p>Parameters of this option are analogous to the <samp>-falign-functions</samp> option.
<samp>-fno-align-labels</samp> and <samp>-falign-labels=1</samp> are
equivalent and mean that labels are not aligned.
</p>
<p>If <samp>-falign-loops</samp> or <samp>-falign-jumps</samp> are applicable and
are greater than this value, then their values are used instead.
</p>
<p>If <var>n</var> is not specified or is zero, use a machine-dependent default
which is very likely to be &lsquo;<samp>1</samp>&rsquo;, meaning no alignment.
The maximum allowed <var>n</var> option value is 65536.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>.
</p>
</dd>
<dt><code>-falign-loops</code></dt>
<dt><code>-falign-loops=<var>n</var></code></dt>
<dt><code>-falign-loops=<var>n</var>:<var>m</var></code></dt>
<dt><code>-falign-loops=<var>n</var>:<var>m</var>:<var>n2</var></code></dt>
<dt><code>-falign-loops=<var>n</var>:<var>m</var>:<var>n2</var>:<var>m2</var></code></dt>
<dd><a name="index-falign_002dloops"></a>
<p>Align loops to a power-of-two boundary.  If the loops are executed
many times, this makes up for any execution of the dummy padding
instructions.
</p>
<p>If <samp>-falign-labels</samp> is greater than this value, then its value
is used instead.
</p>
<p>Parameters of this option are analogous to the <samp>-falign-functions</samp> option.
<samp>-fno-align-loops</samp> and <samp>-falign-loops=1</samp> are
equivalent and mean that loops are not aligned.
The maximum allowed <var>n</var> option value is 65536.
</p>
<p>If <var>n</var> is not specified or is zero, use a machine-dependent default.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>.
</p>
</dd>
<dt><code>-falign-jumps</code></dt>
<dt><code>-falign-jumps=<var>n</var></code></dt>
<dt><code>-falign-jumps=<var>n</var>:<var>m</var></code></dt>
<dt><code>-falign-jumps=<var>n</var>:<var>m</var>:<var>n2</var></code></dt>
<dt><code>-falign-jumps=<var>n</var>:<var>m</var>:<var>n2</var>:<var>m2</var></code></dt>
<dd><a name="index-falign_002djumps"></a>
<p>Align branch targets to a power-of-two boundary, for branch targets
where the targets can only be reached by jumping.  In this case,
no dummy operations need be executed.
</p>
<p>If <samp>-falign-labels</samp> is greater than this value, then its value
is used instead.
</p>
<p>Parameters of this option are analogous to the <samp>-falign-functions</samp> option.
<samp>-fno-align-jumps</samp> and <samp>-falign-jumps=1</samp> are
equivalent and mean that loops are not aligned.
</p>
<p>If <var>n</var> is not specified or is zero, use a machine-dependent default.
The maximum allowed <var>n</var> option value is 65536.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>.
</p>
</dd>
<dt><code>-fno-allocation-dce</code></dt>
<dd><a name="index-fno_002dallocation_002ddce"></a>
<p>Do not remove unused C++ allocations in dead code elimination.
</p>
</dd>
<dt><code>-fallow-store-data-races</code></dt>
<dd><a name="index-fallow_002dstore_002ddata_002draces"></a>
<p>Allow the compiler to introduce new data races on stores.
</p>
<p>Enabled at level <samp>-Ofast</samp>.
</p>
</dd>
<dt><code>-funit-at-a-time</code></dt>
<dd><a name="index-funit_002dat_002da_002dtime"></a>
<p>This option is left for compatibility reasons. <samp>-funit-at-a-time</samp>
has no effect, while <samp>-fno-unit-at-a-time</samp> implies
<samp>-fno-toplevel-reorder</samp> and <samp>-fno-section-anchors</samp>.
</p>
<p>Enabled by default.
</p>
</dd>
<dt><code>-fno-toplevel-reorder</code></dt>
<dd><a name="index-fno_002dtoplevel_002dreorder"></a>
<a name="index-ftoplevel_002dreorder"></a>
<p>Do not reorder top-level functions, variables, and <code>asm</code>
statements.  Output them in the same order that they appear in the
input file.  When this option is used, unreferenced static variables
are not removed.  This option is intended to support existing code
that relies on a particular ordering.  For new code, it is better to
use attributes when possible.
</p>
<p><samp>-ftoplevel-reorder</samp> is the default at <samp>-O1</samp> and higher, and
also at <samp>-O0</samp> if <samp>-fsection-anchors</samp> is explicitly requested.
Additionally <samp>-fno-toplevel-reorder</samp> implies
<samp>-fno-section-anchors</samp>.
</p>
</dd>
<dt><code>-fweb</code></dt>
<dd><a name="index-fweb"></a>
<p>Constructs webs as commonly used for register allocation purposes and assign
each web individual pseudo register.  This allows the register allocation pass
to operate on pseudos directly, but also strengthens several other optimization
passes, such as CSE, loop optimizer and trivial dead code remover.  It can,
however, make debugging impossible, since variables no longer stay in a
&ldquo;home register&rdquo;.
</p>
<p>Enabled by default with <samp>-funroll-loops</samp>.
</p>
</dd>
<dt><code>-fwhole-program</code></dt>
<dd><a name="index-fwhole_002dprogram"></a>
<p>Assume that the current compilation unit represents the whole program being
compiled.  All public functions and variables with the exception of <code>main</code>
and those merged by attribute <code>externally_visible</code> become static functions
and in effect are optimized more aggressively by interprocedural optimizers.
</p>
<p>This option should not be used in combination with <samp>-flto</samp>.
Instead relying on a linker plugin should provide safer and more precise
information.
</p>
</dd>
<dt><code>-flto[=<var>n</var>]</code></dt>
<dd><a name="index-flto"></a>
<p>This option runs the standard link-time optimizer.  When invoked
with source code, it generates GIMPLE (one of GCC&rsquo;s internal
representations) and writes it to special ELF sections in the object
file.  When the object files are linked together, all the function
bodies are read from these ELF sections and instantiated as if they
had been part of the same translation unit.
</p>
<p>To use the link-time optimizer, <samp>-flto</samp> and optimization
options should be specified at compile time and during the final link.
It is recommended that you compile all the files participating in the
same link with the same options and also specify those options at
link time.  
For example:
</p>
<div class="smallexample">
<pre class="smallexample">gcc -c -O2 -flto foo.c
gcc -c -O2 -flto bar.c
gcc -o myprog -flto -O2 foo.o bar.o
</pre></div>

<p>The first two invocations to GCC save a bytecode representation
of GIMPLE into special ELF sections inside <samp>foo.o</samp> and
<samp>bar.o</samp>.  The final invocation reads the GIMPLE bytecode from
<samp>foo.o</samp> and <samp>bar.o</samp>, merges the two files into a single
internal image, and compiles the result as usual.  Since both
<samp>foo.o</samp> and <samp>bar.o</samp> are merged into a single image, this
causes all the interprocedural analyses and optimizations in GCC to
work across the two files as if they were a single one.  This means,
for example, that the inliner is able to inline functions in
<samp>bar.o</samp> into functions in <samp>foo.o</samp> and vice-versa.
</p>
<p>Another (simpler) way to enable link-time optimization is:
</p>
<div class="smallexample">
<pre class="smallexample">gcc -o myprog -flto -O2 foo.c bar.c
</pre></div>

<p>The above generates bytecode for <samp>foo.c</samp> and <samp>bar.c</samp>,
merges them together into a single GIMPLE representation and optimizes
them as usual to produce <samp>myprog</samp>.
</p>
<p>The important thing to keep in mind is that to enable link-time
optimizations you need to use the GCC driver to perform the link step.
GCC automatically performs link-time optimization if any of the
objects involved were compiled with the <samp>-flto</samp> command-line option.  
You can always override
the automatic decision to do link-time optimization
by passing <samp>-fno-lto</samp> to the link command.
</p>
<p>To make whole program optimization effective, it is necessary to make
certain whole program assumptions.  The compiler needs to know
what functions and variables can be accessed by libraries and runtime
outside of the link-time optimized unit.  When supported by the linker,
the linker plugin (see <samp>-fuse-linker-plugin</samp>) passes information
to the compiler about used and externally visible symbols.  When
the linker plugin is not available, <samp>-fwhole-program</samp> should be
used to allow the compiler to make these assumptions, which leads
to more aggressive optimization decisions.
</p>
<p>When a file is compiled with <samp>-flto</samp> without
<samp>-fuse-linker-plugin</samp>, the generated object file is larger than
a regular object file because it contains GIMPLE bytecodes and the usual
final code (see <samp>-ffat-lto-objects</samp>.  This means that
object files with LTO information can be linked as normal object
files; if <samp>-fno-lto</samp> is passed to the linker, no
interprocedural optimizations are applied.  Note that when
<samp>-fno-fat-lto-objects</samp> is enabled the compile stage is faster
but you cannot perform a regular, non-LTO link on them.
</p>
<p>When producing the final binary, GCC only
applies link-time optimizations to those files that contain bytecode.
Therefore, you can mix and match object files and libraries with
GIMPLE bytecodes and final object code.  GCC automatically selects
which files to optimize in LTO mode and which files to link without
further processing.
</p>
<p>Generally, options specified at link time override those
specified at compile time, although in some cases GCC attempts to infer
link-time options from the settings used to compile the input files.
</p>
<p>If you do not specify an optimization level option <samp>-O</samp> at
link time, then GCC uses the highest optimization level 
used when compiling the object files.  Note that it is generally 
ineffective to specify an optimization level option only at link time and 
not at compile time, for two reasons.  First, compiling without 
optimization suppresses compiler passes that gather information 
needed for effective optimization at link time.  Second, some early
optimization passes can be performed only at compile time and 
not at link time.
</p>
<p>There are some code generation flags preserved by GCC when
generating bytecodes, as they need to be used during the final link.
Currently, the following options and their settings are taken from
the first object file that explicitly specifies them: 
<samp>-fPIC</samp>, <samp>-fpic</samp>, <samp>-fpie</samp>, <samp>-fcommon</samp>,
<samp>-fexceptions</samp>, <samp>-fnon-call-exceptions</samp>, <samp>-fgnu-tm</samp>
and all the <samp>-m</samp> target flags.
</p>
<p>Certain ABI-changing flags are required to match in all compilation units,
and trying to override this at link time with a conflicting value
is ignored.  This includes options such as <samp>-freg-struct-return</samp>
and <samp>-fpcc-struct-return</samp>. 
</p>
<p>Other options such as <samp>-ffp-contract</samp>, <samp>-fno-strict-overflow</samp>,
<samp>-fwrapv</samp>, <samp>-fno-trapv</samp> or <samp>-fno-strict-aliasing</samp>
are passed through to the link stage and merged conservatively for
conflicting translation units.  Specifically
<samp>-fno-strict-overflow</samp>, <samp>-fwrapv</samp> and <samp>-fno-trapv</samp> take
precedence; and for example <samp>-ffp-contract=off</samp> takes precedence
over <samp>-ffp-contract=fast</samp>.  You can override them at link time.
</p>
<p>Diagnostic options such as <samp>-Wstringop-overflow</samp> are passed
through to the link stage and their setting matches that of the
compile-step at function granularity.  Note that this matters only
for diagnostics emitted during optimization.  Note that code
transforms such as inlining can lead to warnings being enabled
or disabled for regions if code not consistent with the setting
at compile time.
</p>
<p>When you need to pass options to the assembler via <samp>-Wa</samp> or
<samp>-Xassembler</samp> make sure to either compile such translation
units with <samp>-fno-lto</samp> or consistently use the same assembler
options on all translation units.  You can alternatively also
specify assembler options at LTO link time.
</p>
<p>To enable debug info generation you need to supply <samp>-g</samp> at
compile time.  If any of the input files at link time were built
with debug info generation enabled the link will enable debug info
generation as well.  Any elaborate debug info settings
like the dwarf level <samp>-gdwarf-5</samp> need to be explicitly repeated
at the linker command line and mixing different settings in different
translation units is discouraged.
</p>
<p>If LTO encounters objects with C linkage declared with incompatible
types in separate translation units to be linked together (undefined
behavior according to ISO C99 6.2.7), a non-fatal diagnostic may be
issued.  The behavior is still undefined at run time.  Similar
diagnostics may be raised for other languages.
</p>
<p>Another feature of LTO is that it is possible to apply interprocedural
optimizations on files written in different languages:
</p>
<div class="smallexample">
<pre class="smallexample">gcc -c -flto foo.c
g++ -c -flto bar.cc
gfortran -c -flto baz.f90
g++ -o myprog -flto -O3 foo.o bar.o baz.o -lgfortran
</pre></div>

<p>Notice that the final link is done with <code>g++</code> to get the C++
runtime libraries and <samp>-lgfortran</samp> is added to get the Fortran
runtime libraries.  In general, when mixing languages in LTO mode, you
should use the same link command options as when mixing languages in a
regular (non-LTO) compilation.
</p>
<p>If object files containing GIMPLE bytecode are stored in a library archive, say
<samp>libfoo.a</samp>, it is possible to extract and use them in an LTO link if you
are using a linker with plugin support.  To create static libraries suitable
for LTO, use <code>gcc-ar</code> and <code>gcc-ranlib</code> instead of <code>ar</code>
and <code>ranlib</code>; 
to show the symbols of object files with GIMPLE bytecode, use
<code>gcc-nm</code>.  Those commands require that <code>ar</code>, <code>ranlib</code>
and <code>nm</code> have been compiled with plugin support.  At link time, use the
flag <samp>-fuse-linker-plugin</samp> to ensure that the library participates in
the LTO optimization process:
</p>
<div class="smallexample">
<pre class="smallexample">gcc -o myprog -O2 -flto -fuse-linker-plugin a.o b.o -lfoo
</pre></div>

<p>With the linker plugin enabled, the linker extracts the needed
GIMPLE files from <samp>libfoo.a</samp> and passes them on to the running GCC
to make them part of the aggregated GIMPLE image to be optimized.
</p>
<p>If you are not using a linker with plugin support and/or do not
enable the linker plugin, then the objects inside <samp>libfoo.a</samp>
are extracted and linked as usual, but they do not participate
in the LTO optimization process.  In order to make a static library suitable
for both LTO optimization and usual linkage, compile its object files with
<samp>-flto</samp> <samp>-ffat-lto-objects</samp>.
</p>
<p>Link-time optimizations do not require the presence of the whole program to
operate.  If the program does not require any symbols to be exported, it is
possible to combine <samp>-flto</samp> and <samp>-fwhole-program</samp> to allow
the interprocedural optimizers to use more aggressive assumptions which may
lead to improved optimization opportunities.
Use of <samp>-fwhole-program</samp> is not needed when linker plugin is
active (see <samp>-fuse-linker-plugin</samp>).
</p>
<p>The current implementation of LTO makes no
attempt to generate bytecode that is portable between different
types of hosts.  The bytecode files are versioned and there is a
strict version check, so bytecode files generated in one version of
GCC do not work with an older or newer version of GCC.
</p>
<p>Link-time optimization does not work well with generation of debugging
information on systems other than those using a combination of ELF and
DWARF.
</p>
<p>If you specify the optional <var>n</var>, the optimization and code
generation done at link time is executed in parallel using <var>n</var>
parallel jobs by utilizing an installed <code>make</code> program.  The
environment variable <code>MAKE</code> may be used to override the program
used.
</p>
<p>You can also specify <samp>-flto=jobserver</samp> to use GNU make&rsquo;s
job server mode to determine the number of parallel jobs. This
is useful when the Makefile calling GCC is already executing in parallel.
You must prepend a &lsquo;<samp>+</samp>&rsquo; to the command recipe in the parent Makefile
for this to work.  This option likely only works if <code>MAKE</code> is
GNU make.  Even without the option value, GCC tries to automatically
detect a running GNU make&rsquo;s job server.
</p>
<p>Use <samp>-flto=auto</samp> to use GNU make&rsquo;s job server, if available,
or otherwise fall back to autodetection of the number of CPU threads
present in your system.
</p>
</dd>
<dt><code>-flto-partition=<var>alg</var></code></dt>
<dd><a name="index-flto_002dpartition"></a>
<p>Specify the partitioning algorithm used by the link-time optimizer.
The value is either &lsquo;<samp>1to1</samp>&rsquo; to specify a partitioning mirroring
the original source files or &lsquo;<samp>balanced</samp>&rsquo; to specify partitioning
into equally sized chunks (whenever possible) or &lsquo;<samp>max</samp>&rsquo; to create
new partition for every symbol where possible.  Specifying &lsquo;<samp>none</samp>&rsquo;
as an algorithm disables partitioning and streaming completely. 
The default value is &lsquo;<samp>balanced</samp>&rsquo;. While &lsquo;<samp>1to1</samp>&rsquo; can be used
as an workaround for various code ordering issues, the &lsquo;<samp>max</samp>&rsquo;
partitioning is intended for internal testing only.
The value &lsquo;<samp>one</samp>&rsquo; specifies that exactly one partition should be
used while the value &lsquo;<samp>none</samp>&rsquo; bypasses partitioning and executes
the link-time optimization step directly from the WPA phase.
</p>
</dd>
<dt><code>-flto-compression-level=<var>n</var></code></dt>
<dd><a name="index-flto_002dcompression_002dlevel"></a>
<p>This option specifies the level of compression used for intermediate
language written to LTO object files, and is only meaningful in
conjunction with LTO mode (<samp>-flto</samp>).  Valid
values are 0 (no compression) to 9 (maximum compression).  Values
outside this range are clamped to either 0 or 9.  If the option is not
given, a default balanced compression setting is used.
</p>
</dd>
<dt><code>-fuse-linker-plugin</code></dt>
<dd><a name="index-fuse_002dlinker_002dplugin"></a>
<p>Enables the use of a linker plugin during link-time optimization.  This
option relies on plugin support in the linker, which is available in gold
or in GNU ld 2.21 or newer.
</p>
<p>This option enables the extraction of object files with GIMPLE bytecode out
of library archives. This improves the quality of optimization by exposing
more code to the link-time optimizer.  This information specifies what
symbols can be accessed externally (by non-LTO object or during dynamic
linking).  Resulting code quality improvements on binaries (and shared
libraries that use hidden visibility) are similar to <samp>-fwhole-program</samp>.
See <samp>-flto</samp> for a description of the effect of this flag and how to
use it.
</p>
<p>This option is enabled by default when LTO support in GCC is enabled
and GCC was configured for use with
a linker supporting plugins (GNU ld 2.21 or newer or gold).
</p>
</dd>
<dt><code>-ffat-lto-objects</code></dt>
<dd><a name="index-ffat_002dlto_002dobjects"></a>
<p>Fat LTO objects are object files that contain both the intermediate language
and the object code. This makes them usable for both LTO linking and normal
linking. This option is effective only when compiling with <samp>-flto</samp>
and is ignored at link time.
</p>
<p><samp>-fno-fat-lto-objects</samp> improves compilation time over plain LTO, but
requires the complete toolchain to be aware of LTO. It requires a linker with
linker plugin support for basic functionality.  Additionally,
<code>nm</code>, <code>ar</code> and <code>ranlib</code>
need to support linker plugins to allow a full-featured build environment
(capable of building static libraries etc).  GCC provides the <code>gcc-ar</code>,
<code>gcc-nm</code>, <code>gcc-ranlib</code> wrappers to pass the right options
to these tools. With non fat LTO makefiles need to be modified to use them.
</p>
<p>Note that modern binutils provide plugin auto-load mechanism.
Installing the linker plugin into <samp>$libdir/bfd-plugins</samp> has the same
effect as usage of the command wrappers (<code>gcc-ar</code>, <code>gcc-nm</code> and
<code>gcc-ranlib</code>).
</p>
<p>The default is <samp>-fno-fat-lto-objects</samp> on targets with linker plugin
support.
</p>
</dd>
<dt><code>-fcompare-elim</code></dt>
<dd><a name="index-fcompare_002delim"></a>
<p>After register allocation and post-register allocation instruction splitting,
identify arithmetic instructions that compute processor flags similar to a
comparison operation based on that arithmetic.  If possible, eliminate the
explicit comparison operation.
</p>
<p>This pass only applies to certain targets that cannot explicitly represent
the comparison operation before register allocation is complete.
</p>
<p>Enabled at levels <samp>-O</samp>, <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fcprop-registers</code></dt>
<dd><a name="index-fcprop_002dregisters"></a>
<p>After register allocation and post-register allocation instruction splitting,
perform a copy-propagation pass to try to reduce scheduling dependencies
and occasionally eliminate the copy.
</p>
<p>Enabled at levels <samp>-O</samp>, <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-fprofile-correction</code></dt>
<dd><a name="index-fprofile_002dcorrection"></a>
<p>Profiles collected using an instrumented binary for multi-threaded programs may
be inconsistent due to missed counter updates. When this option is specified,
GCC uses heuristics to correct or smooth out such inconsistencies. By
default, GCC emits an error message when an inconsistent profile is detected.
</p>
<p>This option is enabled by <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-fprofile-partial-training</code></dt>
<dd><a name="index-fprofile_002dpartial_002dtraining"></a>
<p>With <code>-fprofile-use</code> all portions of programs not executed during train
run are optimized agressively for size rather than speed.  In some cases it is
not practical to train all possible hot paths in the program. (For
example, program may contain functions specific for a given hardware and
trianing may not cover all hardware configurations program is run on.)  With
<code>-fprofile-partial-training</code> profile feedback will be ignored for all
functions not executed during the train run leading them to be optimized as if
they were compiled without profile feedback. This leads to better performance
when train run is not representative but also leads to significantly bigger
code.
</p>
</dd>
<dt><code>-fprofile-use</code></dt>
<dt><code>-fprofile-use=<var>path</var></code></dt>
<dd><a name="index-fprofile_002duse"></a>
<p>Enable profile feedback-directed optimizations, 
and the following optimizations, many of which
are generally profitable only with profile feedback available:
</p>
<div class="smallexample">
<pre class="smallexample">-fbranch-probabilities  -fprofile-values 
-funroll-loops  -fpeel-loops  -ftracer  -fvpt 
-finline-functions  -fipa-cp  -fipa-cp-clone  -fipa-bit-cp 
-fpredictive-commoning  -fsplit-loops  -funswitch-loops 
-fgcse-after-reload  -ftree-loop-vectorize  -ftree-slp-vectorize 
-fvect-cost-model=dynamic  -ftree-loop-distribute-patterns 
-fprofile-reorder-functions
</pre></div>

<p>Before you can use this option, you must first generate profiling information.
See <a href="Instrumentation-Options.html#Instrumentation-Options">Instrumentation Options</a>, for information about the
<samp>-fprofile-generate</samp> option.
</p>
<p>By default, GCC emits an error message if the feedback profiles do not
match the source code.  This error can be turned into a warning by using
<samp>-Wno-error=coverage-mismatch</samp>.  Note this may result in poorly
optimized code.  Additionally, by default, GCC also emits a warning message if
the feedback profiles do not exist (see <samp>-Wmissing-profile</samp>).
</p>
<p>If <var>path</var> is specified, GCC looks at the <var>path</var> to find
the profile feedback data files. See <samp>-fprofile-dir</samp>.
</p>
</dd>
<dt><code>-fauto-profile</code></dt>
<dt><code>-fauto-profile=<var>path</var></code></dt>
<dd><a name="index-fauto_002dprofile"></a>
<p>Enable sampling-based feedback-directed optimizations, 
and the following optimizations,
many of which are generally profitable only with profile feedback available:
</p>
<div class="smallexample">
<pre class="smallexample">-fbranch-probabilities  -fprofile-values 
-funroll-loops  -fpeel-loops  -ftracer  -fvpt 
-finline-functions  -fipa-cp  -fipa-cp-clone  -fipa-bit-cp 
-fpredictive-commoning  -fsplit-loops  -funswitch-loops 
-fgcse-after-reload  -ftree-loop-vectorize  -ftree-slp-vectorize 
-fvect-cost-model=dynamic  -ftree-loop-distribute-patterns 
-fprofile-correction
</pre></div>

<p><var>path</var> is the name of a file containing AutoFDO profile information.
If omitted, it defaults to <samp>fbdata.afdo</samp> in the current directory.
</p>
<p>Producing an AutoFDO profile data file requires running your program
with the <code>perf</code> utility on a supported GNU/Linux target system.
For more information, see <a href="https://perf.wiki.kernel.org/">https://perf.wiki.kernel.org/</a>.
</p>
<p>E.g.
</p><div class="smallexample">
<pre class="smallexample">perf record -e br_inst_retired:near_taken -b -o perf.data \
    -- your_program
</pre></div>

<p>Then use the <code>create_gcov</code> tool to convert the raw profile data
to a format that can be used by GCC.&nbsp; You must also supply the 
unstripped binary for your program to this tool.  
See <a href="https://github.com/google/autofdo">https://github.com/google/autofdo</a>.
</p>
<p>E.g.
</p><div class="smallexample">
<pre class="smallexample">create_gcov --binary=your_program.unstripped --profile=perf.data \
    --gcov=profile.afdo
</pre></div>
</dd>
</dl>

<p>The following options control compiler behavior regarding floating-point 
arithmetic.  These options trade off between speed and
correctness.  All must be specifically enabled.
</p>
<dl compact="compact">
<dt><code>-ffloat-store</code></dt>
<dd><a name="index-ffloat_002dstore"></a>
<p>Do not store floating-point variables in registers, and inhibit other
options that might change whether a floating-point value is taken from a
register or memory.
</p>
<a name="index-floating_002dpoint-precision"></a>
<p>This option prevents undesirable excess precision on machines such as
the 68000 where the floating registers (of the 68881) keep more
precision than a <code>double</code> is supposed to have.  Similarly for the
x86 architecture.  For most programs, the excess precision does only
good, but a few programs rely on the precise definition of IEEE floating
point.  Use <samp>-ffloat-store</samp> for such programs, after modifying
them to store all pertinent intermediate computations into variables.
</p>
</dd>
<dt><code>-fexcess-precision=<var>style</var></code></dt>
<dd><a name="index-fexcess_002dprecision"></a>
<p>This option allows further control over excess precision on machines
where floating-point operations occur in a format with more precision or
range than the IEEE standard and interchange floating-point types.  By
default, <samp>-fexcess-precision=fast</samp> is in effect; this means that
operations may be carried out in a wider precision than the types specified
in the source if that would result in faster code, and it is unpredictable
when rounding to the types specified in the source code takes place.
When compiling C, if <samp>-fexcess-precision=standard</samp> is specified then
excess precision follows the rules specified in ISO C99; in particular,
both casts and assignments cause values to be rounded to their
semantic types (whereas <samp>-ffloat-store</samp> only affects
assignments).  This option is enabled by default for C if a strict
conformance option such as <samp>-std=c99</samp> is used.
<samp>-ffast-math</samp> enables <samp>-fexcess-precision=fast</samp> by default
regardless of whether a strict conformance option is used.
</p>
<a name="index-mfpmath"></a>
<p><samp>-fexcess-precision=standard</samp> is not implemented for languages
other than C.  On the x86, it has no effect if <samp>-mfpmath=sse</samp>
or <samp>-mfpmath=sse+387</samp> is specified; in the former case, IEEE
semantics apply without excess precision, and in the latter, rounding
is unpredictable.
</p>
</dd>
<dt><code>-ffast-math</code></dt>
<dd><a name="index-ffast_002dmath"></a>
<p>Sets the options <samp>-fno-math-errno</samp>, <samp>-funsafe-math-optimizations</samp>,
<samp>-ffinite-math-only</samp>, <samp>-fno-rounding-math</samp>,
<samp>-fno-signaling-nans</samp>, <samp>-fcx-limited-range</samp> and
<samp>-fexcess-precision=fast</samp>.
</p>
<p>This option causes the preprocessor macro <code>__FAST_MATH__</code> to be defined.
</p>
<p>This option is not turned on by any <samp>-O</samp> option besides
<samp>-Ofast</samp> since it can result in incorrect output for programs
that depend on an exact implementation of IEEE or ISO rules/specifications
for math functions. It may, however, yield faster code for programs
that do not require the guarantees of these specifications.
</p>
</dd>
<dt><code>-fno-math-errno</code></dt>
<dd><a name="index-fno_002dmath_002derrno"></a>
<a name="index-fmath_002derrno"></a>
<p>Do not set <code>errno</code> after calling math functions that are executed
with a single instruction, e.g., <code>sqrt</code>.  A program that relies on
IEEE exceptions for math error handling may want to use this flag
for speed while maintaining IEEE arithmetic compatibility.
</p>
<p>This option is not turned on by any <samp>-O</samp> option since
it can result in incorrect output for programs that depend on
an exact implementation of IEEE or ISO rules/specifications for
math functions. It may, however, yield faster code for programs
that do not require the guarantees of these specifications.
</p>
<p>The default is <samp>-fmath-errno</samp>.
</p>
<p>On Darwin systems, the math library never sets <code>errno</code>.  There is
therefore no reason for the compiler to consider the possibility that
it might, and <samp>-fno-math-errno</samp> is the default.
</p>
</dd>
<dt><code>-funsafe-math-optimizations</code></dt>
<dd><a name="index-funsafe_002dmath_002doptimizations"></a>

<p>Allow optimizations for floating-point arithmetic that (a) assume
that arguments and results are valid and (b) may violate IEEE or
ANSI standards.  When used at link time, it may include libraries
or startup files that change the default FPU control word or other
similar optimizations.
</p>
<p>This option is not turned on by any <samp>-O</samp> option since
it can result in incorrect output for programs that depend on
an exact implementation of IEEE or ISO rules/specifications for
math functions. It may, however, yield faster code for programs
that do not require the guarantees of these specifications.
Enables <samp>-fno-signed-zeros</samp>, <samp>-fno-trapping-math</samp>,
<samp>-fassociative-math</samp> and <samp>-freciprocal-math</samp>.
</p>
<p>The default is <samp>-fno-unsafe-math-optimizations</samp>.
</p>
</dd>
<dt><code>-fassociative-math</code></dt>
<dd><a name="index-fassociative_002dmath"></a>

<p>Allow re-association of operands in series of floating-point operations.
This violates the ISO C and C++ language standard by possibly changing
computation result.  NOTE: re-ordering may change the sign of zero as
well as ignore NaNs and inhibit or create underflow or overflow (and
thus cannot be used on code that relies on rounding behavior like
<code>(x + 2**52) - 2**52</code>.  May also reorder floating-point comparisons
and thus may not be used when ordered comparisons are required.
This option requires that both <samp>-fno-signed-zeros</samp> and
<samp>-fno-trapping-math</samp> be in effect.  Moreover, it doesn&rsquo;t make
much sense with <samp>-frounding-math</samp>. For Fortran the option
is automatically enabled when both <samp>-fno-signed-zeros</samp> and
<samp>-fno-trapping-math</samp> are in effect.
</p>
<p>The default is <samp>-fno-associative-math</samp>.
</p>
</dd>
<dt><code>-freciprocal-math</code></dt>
<dd><a name="index-freciprocal_002dmath"></a>

<p>Allow the reciprocal of a value to be used instead of dividing by
the value if this enables optimizations.  For example <code>x / y</code>
can be replaced with <code>x * (1/y)</code>, which is useful if <code>(1/y)</code>
is subject to common subexpression elimination.  Note that this loses
precision and increases the number of flops operating on the value.
</p>
<p>The default is <samp>-fno-reciprocal-math</samp>.
</p>
</dd>
<dt><code>-ffinite-math-only</code></dt>
<dd><a name="index-ffinite_002dmath_002donly"></a>
<p>Allow optimizations for floating-point arithmetic that assume
that arguments and results are not NaNs or +-Infs.
</p>
<p>This option is not turned on by any <samp>-O</samp> option since
it can result in incorrect output for programs that depend on
an exact implementation of IEEE or ISO rules/specifications for
math functions. It may, however, yield faster code for programs
that do not require the guarantees of these specifications.
</p>
<p>The default is <samp>-fno-finite-math-only</samp>.
</p>
</dd>
<dt><code>-fno-signed-zeros</code></dt>
<dd><a name="index-fno_002dsigned_002dzeros"></a>
<a name="index-fsigned_002dzeros"></a>
<p>Allow optimizations for floating-point arithmetic that ignore the
signedness of zero.  IEEE arithmetic specifies the behavior of
distinct +0.0 and -0.0 values, which then prohibits simplification
of expressions such as x+0.0 or 0.0*x (even with <samp>-ffinite-math-only</samp>).
This option implies that the sign of a zero result isn&rsquo;t significant.
</p>
<p>The default is <samp>-fsigned-zeros</samp>.
</p>
</dd>
<dt><code>-fno-trapping-math</code></dt>
<dd><a name="index-fno_002dtrapping_002dmath"></a>
<a name="index-ftrapping_002dmath"></a>
<p>Compile code assuming that floating-point operations cannot generate
user-visible traps.  These traps include division by zero, overflow,
underflow, inexact result and invalid operation.  This option requires
that <samp>-fno-signaling-nans</samp> be in effect.  Setting this option may
allow faster code if one relies on &ldquo;non-stop&rdquo; IEEE arithmetic, for example.
</p>
<p>This option should never be turned on by any <samp>-O</samp> option since
it can result in incorrect output for programs that depend on
an exact implementation of IEEE or ISO rules/specifications for
math functions.
</p>
<p>The default is <samp>-ftrapping-math</samp>.
</p>
</dd>
<dt><code>-frounding-math</code></dt>
<dd><a name="index-frounding_002dmath"></a>
<p>Disable transformations and optimizations that assume default floating-point
rounding behavior.  This is round-to-zero for all floating point
to integer conversions, and round-to-nearest for all other arithmetic
truncations.  This option should be specified for programs that change
the FP rounding mode dynamically, or that may be executed with a
non-default rounding mode.  This option disables constant folding of
floating-point expressions at compile time (which may be affected by
rounding mode) and arithmetic transformations that are unsafe in the
presence of sign-dependent rounding modes.
</p>
<p>The default is <samp>-fno-rounding-math</samp>.
</p>
<p>This option is experimental and does not currently guarantee to
disable all GCC optimizations that are affected by rounding mode.
Future versions of GCC may provide finer control of this setting
using C99&rsquo;s <code>FENV_ACCESS</code> pragma.  This command-line option
will be used to specify the default state for <code>FENV_ACCESS</code>.
</p>
</dd>
<dt><code>-fsignaling-nans</code></dt>
<dd><a name="index-fsignaling_002dnans"></a>
<p>Compile code assuming that IEEE signaling NaNs may generate user-visible
traps during floating-point operations.  Setting this option disables
optimizations that may change the number of exceptions visible with
signaling NaNs.  This option implies <samp>-ftrapping-math</samp>.
</p>
<p>This option causes the preprocessor macro <code>__SUPPORT_SNAN__</code> to
be defined.
</p>
<p>The default is <samp>-fno-signaling-nans</samp>.
</p>
<p>This option is experimental and does not currently guarantee to
disable all GCC optimizations that affect signaling NaN behavior.
</p>
</dd>
<dt><code>-fno-fp-int-builtin-inexact</code></dt>
<dd><a name="index-fno_002dfp_002dint_002dbuiltin_002dinexact"></a>
<a name="index-ffp_002dint_002dbuiltin_002dinexact"></a>
<p>Do not allow the built-in functions <code>ceil</code>, <code>floor</code>,
<code>round</code> and <code>trunc</code>, and their <code>float</code> and <code>long
double</code> variants, to generate code that raises the &ldquo;inexact&rdquo;
floating-point exception for noninteger arguments.  ISO C99 and C11
allow these functions to raise the &ldquo;inexact&rdquo; exception, but ISO/IEC
TS 18661-1:2014, the C bindings to IEEE 754-2008, as integrated into
ISO C2X, does not allow these functions to do so.
</p>
<p>The default is <samp>-ffp-int-builtin-inexact</samp>, allowing the
exception to be raised, unless C2X or a later C standard is selected.
This option does nothing unless <samp>-ftrapping-math</samp> is in effect.
</p>
<p>Even if <samp>-fno-fp-int-builtin-inexact</samp> is used, if the functions
generate a call to a library function then the &ldquo;inexact&rdquo; exception
may be raised if the library implementation does not follow TS 18661.
</p>
</dd>
<dt><code>-fsingle-precision-constant</code></dt>
<dd><a name="index-fsingle_002dprecision_002dconstant"></a>
<p>Treat floating-point constants as single precision instead of
implicitly converting them to double-precision constants.
</p>
</dd>
<dt><code>-fcx-limited-range</code></dt>
<dd><a name="index-fcx_002dlimited_002drange"></a>
<p>When enabled, this option states that a range reduction step is not
needed when performing complex division.  Also, there is no checking
whether the result of a complex multiplication or division is <code>NaN
+ I*NaN</code>, with an attempt to rescue the situation in that case.  The
default is <samp>-fno-cx-limited-range</samp>, but is enabled by
<samp>-ffast-math</samp>.
</p>
<p>This option controls the default setting of the ISO C99
<code>CX_LIMITED_RANGE</code> pragma.  Nevertheless, the option applies to
all languages.
</p>
</dd>
<dt><code>-fcx-fortran-rules</code></dt>
<dd><a name="index-fcx_002dfortran_002drules"></a>
<p>Complex multiplication and division follow Fortran rules.  Range
reduction is done as part of complex division, but there is no checking
whether the result of a complex multiplication or division is <code>NaN
+ I*NaN</code>, with an attempt to rescue the situation in that case.
</p>
<p>The default is <samp>-fno-cx-fortran-rules</samp>.
</p>
</dd>
</dl>

<p>The following options control optimizations that may improve
performance, but are not enabled by any <samp>-O</samp> options.  This
section includes experimental options that may produce broken code.
</p>
<dl compact="compact">
<dt><code>-fbranch-probabilities</code></dt>
<dd><a name="index-fbranch_002dprobabilities"></a>
<p>After running a program compiled with <samp>-fprofile-arcs</samp>
(see <a href="Instrumentation-Options.html#Instrumentation-Options">Instrumentation Options</a>),
you can compile it a second time using
<samp>-fbranch-probabilities</samp>, to improve optimizations based on
the number of times each branch was taken.  When a program
compiled with <samp>-fprofile-arcs</samp> exits, it saves arc execution
counts to a file called <samp><var>sourcename</var>.gcda</samp> for each source
file.  The information in this data file is very dependent on the
structure of the generated code, so you must use the same source code
and the same optimization options for both compilations.
</p>
<p>With <samp>-fbranch-probabilities</samp>, GCC puts a
&lsquo;<samp>REG_BR_PROB</samp>&rsquo; note on each &lsquo;<samp>JUMP_INSN</samp>&rsquo; and &lsquo;<samp>CALL_INSN</samp>&rsquo;.
These can be used to improve optimization.  Currently, they are only
used in one place: in <samp>reorg.c</samp>, instead of guessing which path a
branch is most likely to take, the &lsquo;<samp>REG_BR_PROB</samp>&rsquo; values are used to
exactly determine which path is taken more often.
</p>
<p>Enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-fprofile-values</code></dt>
<dd><a name="index-fprofile_002dvalues"></a>
<p>If combined with <samp>-fprofile-arcs</samp>, it adds code so that some
data about values of expressions in the program is gathered.
</p>
<p>With <samp>-fbranch-probabilities</samp>, it reads back the data gathered
from profiling values of expressions for usage in optimizations.
</p>
<p>Enabled by <samp>-fprofile-generate</samp>, <samp>-fprofile-use</samp>, and
<samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-fprofile-reorder-functions</code></dt>
<dd><a name="index-fprofile_002dreorder_002dfunctions"></a>
<p>Function reordering based on profile instrumentation collects
first time of execution of a function and orders these functions
in ascending order.
</p>
<p>Enabled with <samp>-fprofile-use</samp>.
</p>
</dd>
<dt><code>-fvpt</code></dt>
<dd><a name="index-fvpt"></a>
<p>If combined with <samp>-fprofile-arcs</samp>, this option instructs the compiler
to add code to gather information about values of expressions.
</p>
<p>With <samp>-fbranch-probabilities</samp>, it reads back the data gathered
and actually performs the optimizations based on them.
Currently the optimizations include specialization of division operations
using the knowledge about the value of the denominator.
</p>
<p>Enabled with <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-frename-registers</code></dt>
<dd><a name="index-frename_002dregisters"></a>
<p>Attempt to avoid false dependencies in scheduled code by making use
of registers left over after register allocation.  This optimization
most benefits processors with lots of registers.  Depending on the
debug information format adopted by the target, however, it can
make debugging impossible, since variables no longer stay in
a &ldquo;home register&rdquo;.
</p>
<p>Enabled by default with <samp>-funroll-loops</samp>.
</p>
</dd>
<dt><code>-fschedule-fusion</code></dt>
<dd><a name="index-fschedule_002dfusion"></a>
<p>Performs a target dependent pass over the instruction stream to schedule
instructions of same type together because target machine can execute them
more efficiently if they are adjacent to each other in the instruction flow.
</p>
<p>Enabled at levels <samp>-O2</samp>, <samp>-O3</samp>, <samp>-Os</samp>.
</p>
</dd>
<dt><code>-ftracer</code></dt>
<dd><a name="index-ftracer"></a>
<p>Perform tail duplication to enlarge superblock size.  This transformation
simplifies the control flow of the function allowing other optimizations to do
a better job.
</p>
<p>Enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-funroll-loops</code></dt>
<dd><a name="index-funroll_002dloops"></a>
<p>Unroll loops whose number of iterations can be determined at compile time or
upon entry to the loop.  <samp>-funroll-loops</samp> implies
<samp>-frerun-cse-after-loop</samp>, <samp>-fweb</samp> and <samp>-frename-registers</samp>.
It also turns on complete loop peeling (i.e. complete removal of loops with
a small constant number of iterations).  This option makes code larger, and may
or may not make it run faster.
</p>
<p>Enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-funroll-all-loops</code></dt>
<dd><a name="index-funroll_002dall_002dloops"></a>
<p>Unroll all loops, even if their number of iterations is uncertain when
the loop is entered.  This usually makes programs run more slowly.
<samp>-funroll-all-loops</samp> implies the same options as
<samp>-funroll-loops</samp>.
</p>
</dd>
<dt><code>-fpeel-loops</code></dt>
<dd><a name="index-fpeel_002dloops"></a>
<p>Peels loops for which there is enough information that they do not
roll much (from profile feedback or static analysis).  It also turns on
complete loop peeling (i.e. complete removal of loops with small constant
number of iterations).
</p>
<p>Enabled by <samp>-O3</samp>, <samp>-fprofile-use</samp>, and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-fmove-loop-invariants</code></dt>
<dd><a name="index-fmove_002dloop_002dinvariants"></a>
<p>Enables the loop invariant motion pass in the RTL loop optimizer.  Enabled
at level <samp>-O1</samp> and higher, except for <samp>-Og</samp>.
</p>
</dd>
<dt><code>-fsplit-loops</code></dt>
<dd><a name="index-fsplit_002dloops"></a>
<p>Split a loop into two if it contains a condition that&rsquo;s always true
for one side of the iteration space and false for the other.
</p>
<p>Enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-funswitch-loops</code></dt>
<dd><a name="index-funswitch_002dloops"></a>
<p>Move branches with loop invariant conditions out of the loop, with duplicates
of the loop on both branches (modified according to result of the condition).
</p>
<p>Enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-fversion-loops-for-strides</code></dt>
<dd><a name="index-fversion_002dloops_002dfor_002dstrides"></a>
<p>If a loop iterates over an array with a variable stride, create another
version of the loop that assumes the stride is always one.  For example:
</p>
<div class="smallexample">
<pre class="smallexample">for (int i = 0; i &lt; n; ++i)
  x[i * stride] = &hellip;;
</pre></div>

<p>becomes:
</p>
<div class="smallexample">
<pre class="smallexample">if (stride == 1)
  for (int i = 0; i &lt; n; ++i)
    x[i] = &hellip;;
else
  for (int i = 0; i &lt; n; ++i)
    x[i * stride] = &hellip;;
</pre></div>

<p>This is particularly useful for assumed-shape arrays in Fortran where
(for example) it allows better vectorization assuming contiguous accesses.
This flag is enabled by default at <samp>-O3</samp>.
It is also enabled by <samp>-fprofile-use</samp> and <samp>-fauto-profile</samp>.
</p>
</dd>
<dt><code>-ffunction-sections</code></dt>
<dt><code>-fdata-sections</code></dt>
<dd><a name="index-ffunction_002dsections"></a>
<a name="index-fdata_002dsections"></a>
<p>Place each function or data item into its own section in the output
file if the target supports arbitrary sections.  The name of the
function or the name of the data item determines the section&rsquo;s name
in the output file.
</p>
<p>Use these options on systems where the linker can perform optimizations to
improve locality of reference in the instruction space.  Most systems using the
ELF object format have linkers with such optimizations.  On AIX, the linker
rearranges sections (CSECTs) based on the call graph.  The performance impact
varies.
</p>
<p>Together with a linker garbage collection (linker <samp>--gc-sections</samp>
option) these options may lead to smaller statically-linked executables (after
stripping).
</p>
<p>On ELF/DWARF systems these options do not degenerate the quality of the debug
information.  There could be issues with other object files/debug info formats.
</p>
<p>Only use these options when there are significant benefits from doing so.  When
you specify these options, the assembler and linker create larger object and
executable files and are also slower.  These options affect code generation.
They prevent optimizations by the compiler and assembler using relative
locations inside a translation unit since the locations are unknown until
link time.  An example of such an optimization is relaxing calls to short call
instructions.
</p>
</dd>
<dt><code>-fstdarg-opt</code></dt>
<dd><a name="index-fstdarg_002dopt"></a>
<p>Optimize the prologue of variadic argument functions with respect to usage of
those arguments.
</p>
</dd>
<dt><code>-fsection-anchors</code></dt>
<dd><a name="index-fsection_002danchors"></a>
<p>Try to reduce the number of symbolic address calculations by using
shared &ldquo;anchor&rdquo; symbols to address nearby objects.  This transformation
can help to reduce the number of GOT entries and GOT accesses on some
targets.
</p>
<p>For example, the implementation of the following function <code>foo</code>:
</p>
<div class="smallexample">
<pre class="smallexample">static int a, b, c;
int foo (void) { return a + b + c; }
</pre></div>

<p>usually calculates the addresses of all three variables, but if you
compile it with <samp>-fsection-anchors</samp>, it accesses the variables
from a common anchor point instead.  The effect is similar to the
following pseudocode (which isn&rsquo;t valid C):
</p>
<div class="smallexample">
<pre class="smallexample">int foo (void)
{
  register int *xr = &amp;x;
  return xr[&amp;a - &amp;x] + xr[&amp;b - &amp;x] + xr[&amp;c - &amp;x];
}
</pre></div>

<p>Not all targets support this option.
</p>
</dd>
<dt><code>--param <var>name</var>=<var>value</var></code></dt>
<dd><a name="index-param"></a>
<p>In some places, GCC uses various constants to control the amount of
optimization that is done.  For example, GCC does not inline functions
that contain more than a certain number of instructions.  You can
control some of these constants on the command line using the
<samp>--param</samp> option.
</p>
<p>The names of specific parameters, and the meaning of the values, are
tied to the internals of the compiler, and are subject to change
without notice in future releases.
</p>
<p>In order to get minimal, maximal and default value of a parameter,
one can use <samp>--help=param -Q</samp> options.
</p>
<p>In each case, the <var>value</var> is an integer.  The following choices
of <var>name</var> are recognized for all targets:
</p>
<dl compact="compact">
<dt><code>predictable-branch-outcome</code></dt>
<dd><p>When branch is predicted to be taken with probability lower than this threshold
(in percent), then it is considered well predictable.
</p>
</dd>
<dt><code>max-rtl-if-conversion-insns</code></dt>
<dd><p>RTL if-conversion tries to remove conditional branches around a block and
replace them with conditionally executed instructions.  This parameter
gives the maximum number of instructions in a block which should be
considered for if-conversion.  The compiler will
also use other heuristics to decide whether if-conversion is likely to be
profitable.
</p>
</dd>
<dt><code>max-rtl-if-conversion-predictable-cost</code></dt>
<dt><code>max-rtl-if-conversion-unpredictable-cost</code></dt>
<dd><p>RTL if-conversion will try to remove conditional branches around a block
and replace them with conditionally executed instructions.  These parameters
give the maximum permissible cost for the sequence that would be generated
by if-conversion depending on whether the branch is statically determined
to be predictable or not.  The units for this parameter are the same as
those for the GCC internal seq_cost metric.  The compiler will try to
provide a reasonable default for this parameter using the BRANCH_COST
target macro.
</p>
</dd>
<dt><code>max-crossjump-edges</code></dt>
<dd><p>The maximum number of incoming edges to consider for cross-jumping.
The algorithm used by <samp>-fcrossjumping</samp> is <em>O(N^2)</em> in
the number of edges incoming to each block.  Increasing values mean
more aggressive optimization, making the compilation time increase with
probably small improvement in executable size.
</p>
</dd>
<dt><code>min-crossjump-insns</code></dt>
<dd><p>The minimum number of instructions that must be matched at the end
of two blocks before cross-jumping is performed on them.  This
value is ignored in the case where all instructions in the block being
cross-jumped from are matched.
</p>
</dd>
<dt><code>max-grow-copy-bb-insns</code></dt>
<dd><p>The maximum code size expansion factor when copying basic blocks
instead of jumping.  The expansion is relative to a jump instruction.
</p>
</dd>
<dt><code>max-goto-duplication-insns</code></dt>
<dd><p>The maximum number of instructions to duplicate to a block that jumps
to a computed goto.  To avoid <em>O(N^2)</em> behavior in a number of
passes, GCC factors computed gotos early in the compilation process,
and unfactors them as late as possible.  Only computed jumps at the
end of a basic blocks with no more than max-goto-duplication-insns are
unfactored.
</p>
</dd>
<dt><code>max-delay-slot-insn-search</code></dt>
<dd><p>The maximum number of instructions to consider when looking for an
instruction to fill a delay slot.  If more than this arbitrary number of
instructions are searched, the time savings from filling the delay slot
are minimal, so stop searching.  Increasing values mean more
aggressive optimization, making the compilation time increase with probably
small improvement in execution time.
</p>
</dd>
<dt><code>max-delay-slot-live-search</code></dt>
<dd><p>When trying to fill delay slots, the maximum number of instructions to
consider when searching for a block with valid live register
information.  Increasing this arbitrarily chosen value means more
aggressive optimization, increasing the compilation time.  This parameter
should be removed when the delay slot code is rewritten to maintain the
control-flow graph.
</p>
</dd>
<dt><code>max-gcse-memory</code></dt>
<dd><p>The approximate maximum amount of memory that can be allocated in
order to perform the global common subexpression elimination
optimization.  If more memory than specified is required, the
optimization is not done.
</p>
</dd>
<dt><code>max-gcse-insertion-ratio</code></dt>
<dd><p>If the ratio of expression insertions to deletions is larger than this value
for any expression, then RTL PRE inserts or removes the expression and thus
leaves partially redundant computations in the instruction stream.
</p>
</dd>
<dt><code>max-pending-list-length</code></dt>
<dd><p>The maximum number of pending dependencies scheduling allows
before flushing the current state and starting over.  Large functions
with few branches or calls can create excessively large lists which
needlessly consume memory and resources.
</p>
</dd>
<dt><code>max-modulo-backtrack-attempts</code></dt>
<dd><p>The maximum number of backtrack attempts the scheduler should make
when modulo scheduling a loop.  Larger values can exponentially increase
compilation time.
</p>
</dd>
<dt><code>max-inline-insns-single</code></dt>
<dd><p>Several parameters control the tree inliner used in GCC.  This number sets the
maximum number of instructions (counted in GCC&rsquo;s internal representation) in a
single function that the tree inliner considers for inlining.  This only
affects functions declared inline and methods implemented in a class
declaration (C++). 
</p>

</dd>
<dt><code>max-inline-insns-auto</code></dt>
<dd><p>When you use <samp>-finline-functions</samp> (included in <samp>-O3</samp>),
a lot of functions that would otherwise not be considered for inlining
by the compiler are investigated.  To those functions, a different
(more restrictive) limit compared to functions declared inline can
be applied (<samp>--param max-inline-insns-auto</samp>).
</p>
</dd>
<dt><code>max-inline-insns-small</code></dt>
<dd><p>This is bound applied to calls which are considered relevant with
<samp>-finline-small-functions</samp>.
</p>
</dd>
<dt><code>max-inline-insns-size</code></dt>
<dd><p>This is bound applied to calls which are optimized for size. Small growth
may be desirable to anticipate optimization oppurtunities exposed by inlining.
</p>
</dd>
<dt><code>uninlined-function-insns</code></dt>
<dd><p>Number of instructions accounted by inliner for function overhead such as
function prologue and epilogue.
</p>
</dd>
<dt><code>uninlined-function-time</code></dt>
<dd><p>Extra time accounted by inliner for function overhead such as time needed to
execute function prologue and epilogue
</p>
</dd>
<dt><code>inline-heuristics-hint-percent</code></dt>
<dd><p>The scale (in percents) applied to <samp>inline-insns-single</samp>,
<samp>inline-insns-single-O2</samp>, <samp>inline-insns-auto</samp>
when inline heuristics hints that inlining is
very profitable (will enable later optimizations).
</p>
</dd>
<dt><code>uninlined-thunk-insns</code></dt>
<dt><code>uninlined-thunk-time</code></dt>
<dd><p>Same as <samp>--param uninlined-function-insns</samp> and
<samp>--param uninlined-function-time</samp> but applied to function thunks
</p>
</dd>
<dt><code>inline-min-speedup</code></dt>
<dd><p>When estimated performance improvement of caller + callee runtime exceeds this
threshold (in percent), the function can be inlined regardless of the limit on
<samp>--param max-inline-insns-single</samp> and <samp>--param
max-inline-insns-auto</samp>.
</p>
</dd>
<dt><code>large-function-insns</code></dt>
<dd><p>The limit specifying really large functions.  For functions larger than this
limit after inlining, inlining is constrained by
<samp>--param large-function-growth</samp>.  This parameter is useful primarily
to avoid extreme compilation time caused by non-linear algorithms used by the
back end.
</p>
</dd>
<dt><code>large-function-growth</code></dt>
<dd><p>Specifies maximal growth of large function caused by inlining in percents.
For example, parameter value 100 limits large function growth to 2.0 times
the original size.
</p>
</dd>
<dt><code>large-unit-insns</code></dt>
<dd><p>The limit specifying large translation unit.  Growth caused by inlining of
units larger than this limit is limited by <samp>--param inline-unit-growth</samp>.
For small units this might be too tight.
For example, consider a unit consisting of function A
that is inline and B that just calls A three times.  If B is small relative to
A, the growth of unit is 300\% and yet such inlining is very sane.  For very
large units consisting of small inlineable functions, however, the overall unit
growth limit is needed to avoid exponential explosion of code size.  Thus for
smaller units, the size is increased to <samp>--param large-unit-insns</samp>
before applying <samp>--param inline-unit-growth</samp>.
</p>
</dd>
<dt><code>inline-unit-growth</code></dt>
<dd><p>Specifies maximal overall growth of the compilation unit caused by inlining.
For example, parameter value 20 limits unit growth to 1.2 times the original
size. Cold functions (either marked cold via an attribute or by profile
feedback) are not accounted into the unit size.
</p>
</dd>
<dt><code>ipa-cp-unit-growth</code></dt>
<dd><p>Specifies maximal overall growth of the compilation unit caused by
interprocedural constant propagation.  For example, parameter value 10 limits
unit growth to 1.1 times the original size.
</p>
</dd>
<dt><code>large-stack-frame</code></dt>
<dd><p>The limit specifying large stack frames.  While inlining the algorithm is trying
to not grow past this limit too much.
</p>
</dd>
<dt><code>large-stack-frame-growth</code></dt>
<dd><p>Specifies maximal growth of large stack frames caused by inlining in percents.
For example, parameter value 1000 limits large stack frame growth to 11 times
the original size.
</p>
</dd>
<dt><code>max-inline-insns-recursive</code></dt>
<dt><code>max-inline-insns-recursive-auto</code></dt>
<dd><p>Specifies the maximum number of instructions an out-of-line copy of a
self-recursive inline
function can grow into by performing recursive inlining.
</p>
<p><samp>--param max-inline-insns-recursive</samp> applies to functions
declared inline.
For functions not declared inline, recursive inlining
happens only when <samp>-finline-functions</samp> (included in <samp>-O3</samp>) is
enabled; <samp>--param max-inline-insns-recursive-auto</samp> applies instead.
</p>
</dd>
<dt><code>max-inline-recursive-depth</code></dt>
<dt><code>max-inline-recursive-depth-auto</code></dt>
<dd><p>Specifies the maximum recursion depth used for recursive inlining.
</p>
<p><samp>--param max-inline-recursive-depth</samp> applies to functions
declared inline.  For functions not declared inline, recursive inlining
happens only when <samp>-finline-functions</samp> (included in <samp>-O3</samp>) is
enabled; <samp>--param max-inline-recursive-depth-auto</samp> applies instead.
</p>
</dd>
<dt><code>min-inline-recursive-probability</code></dt>
<dd><p>Recursive inlining is profitable only for function having deep recursion
in average and can hurt for function having little recursion depth by
increasing the prologue size or complexity of function body to other
optimizers.
</p>
<p>When profile feedback is available (see <samp>-fprofile-generate</samp>) the actual
recursion depth can be guessed from the probability that function recurses
via a given call expression.  This parameter limits inlining only to call
expressions whose probability exceeds the given threshold (in percents).
</p>
</dd>
<dt><code>early-inlining-insns</code></dt>
<dd><p>Specify growth that the early inliner can make.  In effect it increases
the amount of inlining for code having a large abstraction penalty.
</p>
</dd>
<dt><code>max-early-inliner-iterations</code></dt>
<dd><p>Limit of iterations of the early inliner.  This basically bounds
the number of nested indirect calls the early inliner can resolve.
Deeper chains are still handled by late inlining.
</p>
</dd>
<dt><code>comdat-sharing-probability</code></dt>
<dd><p>Probability (in percent) that C++ inline function with comdat visibility
are shared across multiple compilation units.
</p>
</dd>
<dt><code>profile-func-internal-id</code></dt>
<dd><p>A parameter to control whether to use function internal id in profile
database lookup. If the value is 0, the compiler uses an id that
is based on function assembler name and filename, which makes old profile
data more tolerant to source changes such as function reordering etc.
</p>
</dd>
<dt><code>min-vect-loop-bound</code></dt>
<dd><p>The minimum number of iterations under which loops are not vectorized
when <samp>-ftree-vectorize</samp> is used.  The number of iterations after
vectorization needs to be greater than the value specified by this option
to allow vectorization.
</p>
</dd>
<dt><code>gcse-cost-distance-ratio</code></dt>
<dd><p>Scaling factor in calculation of maximum distance an expression
can be moved by GCSE optimizations.  This is currently supported only in the
code hoisting pass.  The bigger the ratio, the more aggressive code hoisting
is with simple expressions, i.e., the expressions that have cost
less than <samp>gcse-unrestricted-cost</samp>.  Specifying 0 disables
hoisting of simple expressions.
</p>
</dd>
<dt><code>gcse-unrestricted-cost</code></dt>
<dd><p>Cost, roughly measured as the cost of a single typical machine
instruction, at which GCSE optimizations do not constrain
the distance an expression can travel.  This is currently
supported only in the code hoisting pass.  The lesser the cost,
the more aggressive code hoisting is.  Specifying 0 
allows all expressions to travel unrestricted distances.
</p>
</dd>
<dt><code>max-hoist-depth</code></dt>
<dd><p>The depth of search in the dominator tree for expressions to hoist.
This is used to avoid quadratic behavior in hoisting algorithm.
The value of 0 does not limit on the search, but may slow down compilation
of huge functions.
</p>
</dd>
<dt><code>max-tail-merge-comparisons</code></dt>
<dd><p>The maximum amount of similar bbs to compare a bb with.  This is used to
avoid quadratic behavior in tree tail merging.
</p>
</dd>
<dt><code>max-tail-merge-iterations</code></dt>
<dd><p>The maximum amount of iterations of the pass over the function.  This is used to
limit compilation time in tree tail merging.
</p>
</dd>
<dt><code>store-merging-allow-unaligned</code></dt>
<dd><p>Allow the store merging pass to introduce unaligned stores if it is legal to
do so.
</p>
</dd>
<dt><code>max-stores-to-merge</code></dt>
<dd><p>The maximum number of stores to attempt to merge into wider stores in the store
merging pass.
</p>
</dd>
<dt><code>max-unrolled-insns</code></dt>
<dd><p>The maximum number of instructions that a loop may have to be unrolled.
If a loop is unrolled, this parameter also determines how many times
the loop code is unrolled.
</p>
</dd>
<dt><code>max-average-unrolled-insns</code></dt>
<dd><p>The maximum number of instructions biased by probabilities of their execution
that a loop may have to be unrolled.  If a loop is unrolled,
this parameter also determines how many times the loop code is unrolled.
</p>
</dd>
<dt><code>max-unroll-times</code></dt>
<dd><p>The maximum number of unrollings of a single loop.
</p>
</dd>
<dt><code>max-peeled-insns</code></dt>
<dd><p>The maximum number of instructions that a loop may have to be peeled.
If a loop is peeled, this parameter also determines how many times
the loop code is peeled.
</p>
</dd>
<dt><code>max-peel-times</code></dt>
<dd><p>The maximum number of peelings of a single loop.
</p>
</dd>
<dt><code>max-peel-branches</code></dt>
<dd><p>The maximum number of branches on the hot path through the peeled sequence.
</p>
</dd>
<dt><code>max-completely-peeled-insns</code></dt>
<dd><p>The maximum number of insns of a completely peeled loop.
</p>
</dd>
<dt><code>max-completely-peel-times</code></dt>
<dd><p>The maximum number of iterations of a loop to be suitable for complete peeling.
</p>
</dd>
<dt><code>max-completely-peel-loop-nest-depth</code></dt>
<dd><p>The maximum depth of a loop nest suitable for complete peeling.
</p>
</dd>
<dt><code>max-unswitch-insns</code></dt>
<dd><p>The maximum number of insns of an unswitched loop.
</p>
</dd>
<dt><code>max-unswitch-level</code></dt>
<dd><p>The maximum number of branches unswitched in a single loop.
</p>
</dd>
<dt><code>lim-expensive</code></dt>
<dd><p>The minimum cost of an expensive expression in the loop invariant motion.
</p>
</dd>
<dt><code>min-loop-cond-split-prob</code></dt>
<dd><p>When FDO profile information is available, <samp>min-loop-cond-split-prob</samp>
specifies minimum threshold for probability of semi-invariant condition
statement to trigger loop split.
</p>
</dd>
<dt><code>iv-consider-all-candidates-bound</code></dt>
<dd><p>Bound on number of candidates for induction variables, below which
all candidates are considered for each use in induction variable
optimizations.  If there are more candidates than this,
only the most relevant ones are considered to avoid quadratic time complexity.
</p>
</dd>
<dt><code>iv-max-considered-uses</code></dt>
<dd><p>The induction variable optimizations give up on loops that contain more
induction variable uses.
</p>
</dd>
<dt><code>iv-always-prune-cand-set-bound</code></dt>
<dd><p>If the number of candidates in the set is smaller than this value,
always try to remove unnecessary ivs from the set
when adding a new one.
</p>
</dd>
<dt><code>avg-loop-niter</code></dt>
<dd><p>Average number of iterations of a loop.
</p>
</dd>
<dt><code>dse-max-object-size</code></dt>
<dd><p>Maximum size (in bytes) of objects tracked bytewise by dead store elimination.
Larger values may result in larger compilation times.
</p>
</dd>
<dt><code>dse-max-alias-queries-per-store</code></dt>
<dd><p>Maximum number of queries into the alias oracle per store.
Larger values result in larger compilation times and may result in more
removed dead stores.
</p>
</dd>
<dt><code>scev-max-expr-size</code></dt>
<dd><p>Bound on size of expressions used in the scalar evolutions analyzer.
Large expressions slow the analyzer.
</p>
</dd>
<dt><code>scev-max-expr-complexity</code></dt>
<dd><p>Bound on the complexity of the expressions in the scalar evolutions analyzer.
Complex expressions slow the analyzer.
</p>
</dd>
<dt><code>max-tree-if-conversion-phi-args</code></dt>
<dd><p>Maximum number of arguments in a PHI supported by TREE if conversion
unless the loop is marked with simd pragma.
</p>
</dd>
<dt><code>vect-max-version-for-alignment-checks</code></dt>
<dd><p>The maximum number of run-time checks that can be performed when
doing loop versioning for alignment in the vectorizer.
</p>
</dd>
<dt><code>vect-max-version-for-alias-checks</code></dt>
<dd><p>The maximum number of run-time checks that can be performed when
doing loop versioning for alias in the vectorizer.
</p>
</dd>
<dt><code>vect-max-peeling-for-alignment</code></dt>
<dd><p>The maximum number of loop peels to enhance access alignment
for vectorizer. Value -1 means no limit.
</p>
</dd>
<dt><code>max-iterations-to-track</code></dt>
<dd><p>The maximum number of iterations of a loop the brute-force algorithm
for analysis of the number of iterations of the loop tries to evaluate.
</p>
</dd>
<dt><code>hot-bb-count-fraction</code></dt>
<dd><p>The denominator n of fraction 1/n of the maximal execution count of a
basic block in the entire program that a basic block needs to at least
have in order to be considered hot.  The default is 10000, which means
that a basic block is considered hot if its execution count is greater
than 1/10000 of the maximal execution count.  0 means that it is never
considered hot.  Used in non-LTO mode.
</p>
</dd>
<dt><code>hot-bb-count-ws-permille</code></dt>
<dd><p>The number of most executed permilles, ranging from 0 to 1000, of the
profiled execution of the entire program to which the execution count
of a basic block must be part of in order to be considered hot.  The
default is 990, which means that a basic block is considered hot if
its execution count contributes to the upper 990 permilles, or 99.0%,
of the profiled execution of the entire program.  0 means that it is
never considered hot.  Used in LTO mode.
</p>
</dd>
<dt><code>hot-bb-frequency-fraction</code></dt>
<dd><p>The denominator n of fraction 1/n of the execution frequency of the
entry block of a function that a basic block of this function needs
to at least have in order to be considered hot.  The default is 1000,
which means that a basic block is considered hot in a function if it
is executed more frequently than 1/1000 of the frequency of the entry
block of the function.  0 means that it is never considered hot.
</p>
</dd>
<dt><code>unlikely-bb-count-fraction</code></dt>
<dd><p>The denominator n of fraction 1/n of the number of profiled runs of
the entire program below which the execution count of a basic block
must be in order for the basic block to be considered unlikely executed.
The default is 20, which means that a basic block is considered unlikely
executed if it is executed in fewer than 1/20, or 5%, of the runs of
the program.  0 means that it is always considered unlikely executed.
</p>
</dd>
<dt><code>max-predicted-iterations</code></dt>
<dd><p>The maximum number of loop iterations we predict statically.  This is useful
in cases where a function contains a single loop with known bound and
another loop with unknown bound.
The known number of iterations is predicted correctly, while
the unknown number of iterations average to roughly 10.  This means that the
loop without bounds appears artificially cold relative to the other one.
</p>
</dd>
<dt><code>builtin-expect-probability</code></dt>
<dd><p>Control the probability of the expression having the specified value. This
parameter takes a percentage (i.e. 0 ... 100) as input.
</p>
</dd>
<dt><code>builtin-string-cmp-inline-length</code></dt>
<dd><p>The maximum length of a constant string for a builtin string cmp call 
eligible for inlining.
</p>
</dd>
<dt><code>align-threshold</code></dt>
<dd>
<p>Select fraction of the maximal frequency of executions of a basic block in
a function to align the basic block.
</p>
</dd>
<dt><code>align-loop-iterations</code></dt>
<dd>
<p>A loop expected to iterate at least the selected number of iterations is
aligned.
</p>
</dd>
<dt><code>tracer-dynamic-coverage</code></dt>
<dt><code>tracer-dynamic-coverage-feedback</code></dt>
<dd>
<p>This value is used to limit superblock formation once the given percentage of
executed instructions is covered.  This limits unnecessary code size
expansion.
</p>
<p>The <samp>tracer-dynamic-coverage-feedback</samp> parameter
is used only when profile
feedback is available.  The real profiles (as opposed to statically estimated
ones) are much less balanced allowing the threshold to be larger value.
</p>
</dd>
<dt><code>tracer-max-code-growth</code></dt>
<dd><p>Stop tail duplication once code growth has reached given percentage.  This is
a rather artificial limit, as most of the duplicates are eliminated later in
cross jumping, so it may be set to much higher values than is the desired code
growth.
</p>
</dd>
<dt><code>tracer-min-branch-ratio</code></dt>
<dd>
<p>Stop reverse growth when the reverse probability of best edge is less than this
threshold (in percent).
</p>
</dd>
<dt><code>tracer-min-branch-probability</code></dt>
<dt><code>tracer-min-branch-probability-feedback</code></dt>
<dd>
<p>Stop forward growth if the best edge has probability lower than this
threshold.
</p>
<p>Similarly to <samp>tracer-dynamic-coverage</samp> two parameters are
provided.  <samp>tracer-min-branch-probability-feedback</samp> is used for
compilation with profile feedback and <samp>tracer-min-branch-probability</samp>
compilation without.  The value for compilation with profile feedback
needs to be more conservative (higher) in order to make tracer
effective.
</p>
</dd>
<dt><code>stack-clash-protection-guard-size</code></dt>
<dd><p>Specify the size of the operating system provided stack guard as
2 raised to <var>num</var> bytes.  Higher values may reduce the
number of explicit probes, but a value larger than the operating system
provided guard will leave code vulnerable to stack clash style attacks.
</p>
</dd>
<dt><code>stack-clash-protection-probe-interval</code></dt>
<dd><p>Stack clash protection involves probing stack space as it is allocated.  This
param controls the maximum distance between probes into the stack as 2 raised
to <var>num</var> bytes.  Higher values may reduce the number of explicit probes, but a value
larger than the operating system provided guard will leave code vulnerable to
stack clash style attacks.
</p>
</dd>
<dt><code>max-cse-path-length</code></dt>
<dd>
<p>The maximum number of basic blocks on path that CSE considers.
</p>
</dd>
<dt><code>max-cse-insns</code></dt>
<dd><p>The maximum number of instructions CSE processes before flushing.
</p>
</dd>
<dt><code>ggc-min-expand</code></dt>
<dd>
<p>GCC uses a garbage collector to manage its own memory allocation.  This
parameter specifies the minimum percentage by which the garbage
collector&rsquo;s heap should be allowed to expand between collections.
Tuning this may improve compilation speed; it has no effect on code
generation.
</p>
<p>The default is 30% + 70% * (RAM/1GB) with an upper bound of 100% when
RAM &gt;= 1GB.  If <code>getrlimit</code> is available, the notion of &ldquo;RAM&rdquo; is
the smallest of actual RAM and <code>RLIMIT_DATA</code> or <code>RLIMIT_AS</code>.  If
GCC is not able to calculate RAM on a particular platform, the lower
bound of 30% is used.  Setting this parameter and
<samp>ggc-min-heapsize</samp> to zero causes a full collection to occur at
every opportunity.  This is extremely slow, but can be useful for
debugging.
</p>
</dd>
<dt><code>ggc-min-heapsize</code></dt>
<dd>
<p>Minimum size of the garbage collector&rsquo;s heap before it begins bothering
to collect garbage.  The first collection occurs after the heap expands
by <samp>ggc-min-expand</samp>% beyond <samp>ggc-min-heapsize</samp>.  Again,
tuning this may improve compilation speed, and has no effect on code
generation.
</p>
<p>The default is the smaller of RAM/8, RLIMIT_RSS, or a limit that
tries to ensure that RLIMIT_DATA or RLIMIT_AS are not exceeded, but
with a lower bound of 4096 (four megabytes) and an upper bound of
131072 (128 megabytes).  If GCC is not able to calculate RAM on a
particular platform, the lower bound is used.  Setting this parameter
very large effectively disables garbage collection.  Setting this
parameter and <samp>ggc-min-expand</samp> to zero causes a full collection
to occur at every opportunity.
</p>
</dd>
<dt><code>max-reload-search-insns</code></dt>
<dd><p>The maximum number of instruction reload should look backward for equivalent
register.  Increasing values mean more aggressive optimization, making the
compilation time increase with probably slightly better performance.
</p>
</dd>
<dt><code>max-cselib-memory-locations</code></dt>
<dd><p>The maximum number of memory locations cselib should take into account.
Increasing values mean more aggressive optimization, making the compilation time
increase with probably slightly better performance.
</p>
</dd>
<dt><code>max-sched-ready-insns</code></dt>
<dd><p>The maximum number of instructions ready to be issued the scheduler should
consider at any given time during the first scheduling pass.  Increasing
values mean more thorough searches, making the compilation time increase
with probably little benefit.
</p>
</dd>
<dt><code>max-sched-region-blocks</code></dt>
<dd><p>The maximum number of blocks in a region to be considered for
interblock scheduling.
</p>
</dd>
<dt><code>max-pipeline-region-blocks</code></dt>
<dd><p>The maximum number of blocks in a region to be considered for
pipelining in the selective scheduler.
</p>
</dd>
<dt><code>max-sched-region-insns</code></dt>
<dd><p>The maximum number of insns in a region to be considered for
interblock scheduling.
</p>
</dd>
<dt><code>max-pipeline-region-insns</code></dt>
<dd><p>The maximum number of insns in a region to be considered for
pipelining in the selective scheduler.
</p>
</dd>
<dt><code>min-spec-prob</code></dt>
<dd><p>The minimum probability (in percents) of reaching a source block
for interblock speculative scheduling.
</p>
</dd>
<dt><code>max-sched-extend-regions-iters</code></dt>
<dd><p>The maximum number of iterations through CFG to extend regions.
A value of 0 disables region extensions.
</p>
</dd>
<dt><code>max-sched-insn-conflict-delay</code></dt>
<dd><p>The maximum conflict delay for an insn to be considered for speculative motion.
</p>
</dd>
<dt><code>sched-spec-prob-cutoff</code></dt>
<dd><p>The minimal probability of speculation success (in percents), so that
speculative insns are scheduled.
</p>
</dd>
<dt><code>sched-state-edge-prob-cutoff</code></dt>
<dd><p>The minimum probability an edge must have for the scheduler to save its
state across it.
</p>
</dd>
<dt><code>sched-mem-true-dep-cost</code></dt>
<dd><p>Minimal distance (in CPU cycles) between store and load targeting same
memory locations.
</p>
</dd>
<dt><code>selsched-max-lookahead</code></dt>
<dd><p>The maximum size of the lookahead window of selective scheduling.  It is a
depth of search for available instructions.
</p>
</dd>
<dt><code>selsched-max-sched-times</code></dt>
<dd><p>The maximum number of times that an instruction is scheduled during
selective scheduling.  This is the limit on the number of iterations
through which the instruction may be pipelined.
</p>
</dd>
<dt><code>selsched-insns-to-rename</code></dt>
<dd><p>The maximum number of best instructions in the ready list that are considered
for renaming in the selective scheduler.
</p>
</dd>
<dt><code>sms-min-sc</code></dt>
<dd><p>The minimum value of stage count that swing modulo scheduler
generates.
</p>
</dd>
<dt><code>max-last-value-rtl</code></dt>
<dd><p>The maximum size measured as number of RTLs that can be recorded in an expression
in combiner for a pseudo register as last known value of that register.
</p>
</dd>
<dt><code>max-combine-insns</code></dt>
<dd><p>The maximum number of instructions the RTL combiner tries to combine.
</p>
</dd>
<dt><code>integer-share-limit</code></dt>
<dd><p>Small integer constants can use a shared data structure, reducing the
compiler&rsquo;s memory usage and increasing its speed.  This sets the maximum
value of a shared integer constant.
</p>
</dd>
<dt><code>ssp-buffer-size</code></dt>
<dd><p>The minimum size of buffers (i.e. arrays) that receive stack smashing
protection when <samp>-fstack-protection</samp> is used.
</p>
</dd>
<dt><code>min-size-for-stack-sharing</code></dt>
<dd><p>The minimum size of variables taking part in stack slot sharing when not
optimizing.
</p>
</dd>
<dt><code>max-jump-thread-duplication-stmts</code></dt>
<dd><p>Maximum number of statements allowed in a block that needs to be
duplicated when threading jumps.
</p>
</dd>
<dt><code>max-fields-for-field-sensitive</code></dt>
<dd><p>Maximum number of fields in a structure treated in
a field sensitive manner during pointer analysis.
</p>
</dd>
<dt><code>prefetch-latency</code></dt>
<dd><p>Estimate on average number of instructions that are executed before
prefetch finishes.  The distance prefetched ahead is proportional
to this constant.  Increasing this number may also lead to less
streams being prefetched (see <samp>simultaneous-prefetches</samp>).
</p>
</dd>
<dt><code>simultaneous-prefetches</code></dt>
<dd><p>Maximum number of prefetches that can run at the same time.
</p>
</dd>
<dt><code>l1-cache-line-size</code></dt>
<dd><p>The size of cache line in L1 data cache, in bytes.
</p>
</dd>
<dt><code>l1-cache-size</code></dt>
<dd><p>The size of L1 data cache, in kilobytes.
</p>
</dd>
<dt><code>l2-cache-size</code></dt>
<dd><p>The size of L2 data cache, in kilobytes.
</p>
</dd>
<dt><code>prefetch-dynamic-strides</code></dt>
<dd><p>Whether the loop array prefetch pass should issue software prefetch hints
for strides that are non-constant.  In some cases this may be
beneficial, though the fact the stride is non-constant may make it
hard to predict when there is clear benefit to issuing these hints.
</p>
<p>Set to 1 if the prefetch hints should be issued for non-constant
strides.  Set to 0 if prefetch hints should be issued only for strides that
are known to be constant and below <samp>prefetch-minimum-stride</samp>.
</p>
</dd>
<dt><code>prefetch-minimum-stride</code></dt>
<dd><p>Minimum constant stride, in bytes, to start using prefetch hints for.  If
the stride is less than this threshold, prefetch hints will not be issued.
</p>
<p>This setting is useful for processors that have hardware prefetchers, in
which case there may be conflicts between the hardware prefetchers and
the software prefetchers.  If the hardware prefetchers have a maximum
stride they can handle, it should be used here to improve the use of
software prefetchers.
</p>
<p>A value of -1 means we don&rsquo;t have a threshold and therefore
prefetch hints can be issued for any constant stride.
</p>
<p>This setting is only useful for strides that are known and constant.
</p>
</dd>
<dt><code>loop-interchange-max-num-stmts</code></dt>
<dd><p>The maximum number of stmts in a loop to be interchanged.
</p>
</dd>
<dt><code>loop-interchange-stride-ratio</code></dt>
<dd><p>The minimum ratio between stride of two loops for interchange to be profitable.
</p>
</dd>
<dt><code>min-insn-to-prefetch-ratio</code></dt>
<dd><p>The minimum ratio between the number of instructions and the
number of prefetches to enable prefetching in a loop.
</p>
</dd>
<dt><code>prefetch-min-insn-to-mem-ratio</code></dt>
<dd><p>The minimum ratio between the number of instructions and the
number of memory references to enable prefetching in a loop.
</p>
</dd>
<dt><code>use-canonical-types</code></dt>
<dd><p>Whether the compiler should use the &ldquo;canonical&rdquo; type system.
Should always be 1, which uses a more efficient internal
mechanism for comparing types in C++ and Objective-C++.  However, if
bugs in the canonical type system are causing compilation failures,
set this value to 0 to disable canonical types.
</p>
</dd>
<dt><code>switch-conversion-max-branch-ratio</code></dt>
<dd><p>Switch initialization conversion refuses to create arrays that are
bigger than <samp>switch-conversion-max-branch-ratio</samp> times the number of
branches in the switch.
</p>
</dd>
<dt><code>max-partial-antic-length</code></dt>
<dd><p>Maximum length of the partial antic set computed during the tree
partial redundancy elimination optimization (<samp>-ftree-pre</samp>) when
optimizing at <samp>-O3</samp> and above.  For some sorts of source code
the enhanced partial redundancy elimination optimization can run away,
consuming all of the memory available on the host machine.  This
parameter sets a limit on the length of the sets that are computed,
which prevents the runaway behavior.  Setting a value of 0 for
this parameter allows an unlimited set length.
</p>
</dd>
<dt><code>rpo-vn-max-loop-depth</code></dt>
<dd><p>Maximum loop depth that is value-numbered optimistically.
When the limit hits the innermost
<var>rpo-vn-max-loop-depth</var> loops and the outermost loop in the
loop nest are value-numbered optimistically and the remaining ones not.
</p>
</dd>
<dt><code>sccvn-max-alias-queries-per-access</code></dt>
<dd><p>Maximum number of alias-oracle queries we perform when looking for
redundancies for loads and stores.  If this limit is hit the search
is aborted and the load or store is not considered redundant.  The
number of queries is algorithmically limited to the number of
stores on all paths from the load to the function entry.
</p>
</dd>
<dt><code>ira-max-loops-num</code></dt>
<dd><p>IRA uses regional register allocation by default.  If a function
contains more loops than the number given by this parameter, only at most
the given number of the most frequently-executed loops form regions
for regional register allocation.
</p>
</dd>
<dt><code>ira-max-conflict-table-size</code></dt>
<dd><p>Although IRA uses a sophisticated algorithm to compress the conflict
table, the table can still require excessive amounts of memory for
huge functions.  If the conflict table for a function could be more
than the size in MB given by this parameter, the register allocator
instead uses a faster, simpler, and lower-quality
algorithm that does not require building a pseudo-register conflict table.  
</p>
</dd>
<dt><code>ira-loop-reserved-regs</code></dt>
<dd><p>IRA can be used to evaluate more accurate register pressure in loops
for decisions to move loop invariants (see <samp>-O3</samp>).  The number
of available registers reserved for some other purposes is given
by this parameter.  Default of the parameter
is the best found from numerous experiments.
</p>
</dd>
<dt><code>lra-inheritance-ebb-probability-cutoff</code></dt>
<dd><p>LRA tries to reuse values reloaded in registers in subsequent insns.
This optimization is called inheritance.  EBB is used as a region to
do this optimization.  The parameter defines a minimal fall-through
edge probability in percentage used to add BB to inheritance EBB in
LRA.  The default value was chosen
from numerous runs of SPEC2000 on x86-64.
</p>
</dd>
<dt><code>loop-invariant-max-bbs-in-loop</code></dt>
<dd><p>Loop invariant motion can be very expensive, both in compilation time and
in amount of needed compile-time memory, with very large loops.  Loops
with more basic blocks than this parameter won&rsquo;t have loop invariant
motion optimization performed on them.
</p>
</dd>
<dt><code>loop-max-datarefs-for-datadeps</code></dt>
<dd><p>Building data dependencies is expensive for very large loops.  This
parameter limits the number of data references in loops that are
considered for data dependence analysis.  These large loops are no
handled by the optimizations using loop data dependencies.
</p>
</dd>
<dt><code>max-vartrack-size</code></dt>
<dd><p>Sets a maximum number of hash table slots to use during variable
tracking dataflow analysis of any function.  If this limit is exceeded
with variable tracking at assignments enabled, analysis for that
function is retried without it, after removing all debug insns from
the function.  If the limit is exceeded even without debug insns, var
tracking analysis is completely disabled for the function.  Setting
the parameter to zero makes it unlimited.
</p>
</dd>
<dt><code>max-vartrack-expr-depth</code></dt>
<dd><p>Sets a maximum number of recursion levels when attempting to map
variable names or debug temporaries to value expressions.  This trades
compilation time for more complete debug information.  If this is set too
low, value expressions that are available and could be represented in
debug information may end up not being used; setting this higher may
enable the compiler to find more complex debug expressions, but compile
time and memory use may grow.
</p>
</dd>
<dt><code>max-debug-marker-count</code></dt>
<dd><p>Sets a threshold on the number of debug markers (e.g. begin stmt
markers) to avoid complexity explosion at inlining or expanding to RTL.
If a function has more such gimple stmts than the set limit, such stmts
will be dropped from the inlined copy of a function, and from its RTL
expansion.
</p>
</dd>
<dt><code>min-nondebug-insn-uid</code></dt>
<dd><p>Use uids starting at this parameter for nondebug insns.  The range below
the parameter is reserved exclusively for debug insns created by
<samp>-fvar-tracking-assignments</samp>, but debug insns may get
(non-overlapping) uids above it if the reserved range is exhausted.
</p>
</dd>
<dt><code>ipa-sra-ptr-growth-factor</code></dt>
<dd><p>IPA-SRA replaces a pointer to an aggregate with one or more new
parameters only when their cumulative size is less or equal to
<samp>ipa-sra-ptr-growth-factor</samp> times the size of the original
pointer parameter.
</p>
</dd>
<dt><code>ipa-sra-max-replacements</code></dt>
<dd><p>Maximum pieces of an aggregate that IPA-SRA tracks.  As a
consequence, it is also the maximum number of replacements of a formal
parameter.
</p>
</dd>
<dt><code>sra-max-scalarization-size-Ospeed</code></dt>
<dt><code>sra-max-scalarization-size-Osize</code></dt>
<dd><p>The two Scalar Reduction of Aggregates passes (SRA and IPA-SRA) aim to
replace scalar parts of aggregates with uses of independent scalar
variables.  These parameters control the maximum size, in storage units,
of aggregate which is considered for replacement when compiling for
speed
(<samp>sra-max-scalarization-size-Ospeed</samp>) or size
(<samp>sra-max-scalarization-size-Osize</samp>) respectively.
</p>
</dd>
<dt><code>sra-max-propagations</code></dt>
<dd><p>The maximum number of artificial accesses that Scalar Replacement of
Aggregates (SRA) will track, per one local variable, in order to
facilitate copy propagation.
</p>
</dd>
<dt><code>tm-max-aggregate-size</code></dt>
<dd><p>When making copies of thread-local variables in a transaction, this
parameter specifies the size in bytes after which variables are
saved with the logging functions as opposed to save/restore code
sequence pairs.  This option only applies when using
<samp>-fgnu-tm</samp>.
</p>
</dd>
<dt><code>graphite-max-nb-scop-params</code></dt>
<dd><p>To avoid exponential effects in the Graphite loop transforms, the
number of parameters in a Static Control Part (SCoP) is bounded.
A value of zero can be used to lift
the bound.  A variable whose value is unknown at compilation time and
defined outside a SCoP is a parameter of the SCoP.
</p>
</dd>
<dt><code>loop-block-tile-size</code></dt>
<dd><p>Loop blocking or strip mining transforms, enabled with
<samp>-floop-block</samp> or <samp>-floop-strip-mine</samp>, strip mine each
loop in the loop nest by a given number of iterations.  The strip
length can be changed using the <samp>loop-block-tile-size</samp>
parameter.
</p>
</dd>
<dt><code>ipa-cp-value-list-size</code></dt>
<dd><p>IPA-CP attempts to track all possible values and types passed to a function&rsquo;s
parameter in order to propagate them and perform devirtualization.
<samp>ipa-cp-value-list-size</samp> is the maximum number of values and types it
stores per one formal parameter of a function.
</p>
</dd>
<dt><code>ipa-cp-eval-threshold</code></dt>
<dd><p>IPA-CP calculates its own score of cloning profitability heuristics
and performs those cloning opportunities with scores that exceed
<samp>ipa-cp-eval-threshold</samp>.
</p>
</dd>
<dt><code>ipa-cp-max-recursive-depth</code></dt>
<dd><p>Maximum depth of recursive cloning for self-recursive function.
</p>
</dd>
<dt><code>ipa-cp-min-recursive-probability</code></dt>
<dd><p>Recursive cloning only when the probability of call being executed exceeds
the parameter.
</p>
</dd>
<dt><code>ipa-cp-recursion-penalty</code></dt>
<dd><p>Percentage penalty the recursive functions will receive when they
are evaluated for cloning.
</p>
</dd>
<dt><code>ipa-cp-single-call-penalty</code></dt>
<dd><p>Percentage penalty functions containing a single call to another
function will receive when they are evaluated for cloning.
</p>
</dd>
<dt><code>ipa-max-agg-items</code></dt>
<dd><p>IPA-CP is also capable to propagate a number of scalar values passed
in an aggregate. <samp>ipa-max-agg-items</samp> controls the maximum
number of such values per one parameter.
</p>
</dd>
<dt><code>ipa-cp-loop-hint-bonus</code></dt>
<dd><p>When IPA-CP determines that a cloning candidate would make the number
of iterations of a loop known, it adds a bonus of
<samp>ipa-cp-loop-hint-bonus</samp> to the profitability score of
the candidate.
</p>
</dd>
<dt><code>ipa-max-aa-steps</code></dt>
<dd><p>During its analysis of function bodies, IPA-CP employs alias analysis
in order to track values pointed to by function parameters.  In order
not spend too much time analyzing huge functions, it gives up and
consider all memory clobbered after examining
<samp>ipa-max-aa-steps</samp> statements modifying memory.
</p>
</dd>
<dt><code>ipa-max-switch-predicate-bounds</code></dt>
<dd><p>Maximal number of boundary endpoints of case ranges of switch statement.
For switch exceeding this limit, IPA-CP will not construct cloning cost
predicate, which is used to estimate cloning benefit, for default case
of the switch statement.
</p>
</dd>
<dt><code>ipa-max-param-expr-ops</code></dt>
<dd><p>IPA-CP will analyze conditional statement that references some function
parameter to estimate benefit for cloning upon certain constant value.
But if number of operations in a parameter expression exceeds
<samp>ipa-max-param-expr-ops</samp>, the expression is treated as complicated
one, and is not handled by IPA analysis.
</p>
</dd>
<dt><code>lto-partitions</code></dt>
<dd><p>Specify desired number of partitions produced during WHOPR compilation.
The number of partitions should exceed the number of CPUs used for compilation.
</p>
</dd>
<dt><code>lto-min-partition</code></dt>
<dd><p>Size of minimal partition for WHOPR (in estimated instructions).
This prevents expenses of splitting very small programs into too many
partitions.
</p>
</dd>
<dt><code>lto-max-partition</code></dt>
<dd><p>Size of max partition for WHOPR (in estimated instructions).
to provide an upper bound for individual size of partition.
Meant to be used only with balanced partitioning.
</p>
</dd>
<dt><code>lto-max-streaming-parallelism</code></dt>
<dd><p>Maximal number of parallel processes used for LTO streaming.
</p>
</dd>
<dt><code>cxx-max-namespaces-for-diagnostic-help</code></dt>
<dd><p>The maximum number of namespaces to consult for suggestions when C++
name lookup fails for an identifier.
</p>
</dd>
<dt><code>sink-frequency-threshold</code></dt>
<dd><p>The maximum relative execution frequency (in percents) of the target block
relative to a statement&rsquo;s original block to allow statement sinking of a
statement.  Larger numbers result in more aggressive statement sinking.
A small positive adjustment is applied for
statements with memory operands as those are even more profitable so sink.
</p>
</dd>
<dt><code>max-stores-to-sink</code></dt>
<dd><p>The maximum number of conditional store pairs that can be sunk.  Set to 0
if either vectorization (<samp>-ftree-vectorize</samp>) or if-conversion
(<samp>-ftree-loop-if-convert</samp>) is disabled.
</p>
</dd>
<dt><code>case-values-threshold</code></dt>
<dd><p>The smallest number of different values for which it is best to use a
jump-table instead of a tree of conditional branches.  If the value is
0, use the default for the machine.
</p>
</dd>
<dt><code>jump-table-max-growth-ratio-for-size</code></dt>
<dd><p>The maximum code size growth ratio when expanding
into a jump table (in percent).  The parameter is used when
optimizing for size.
</p>
</dd>
<dt><code>jump-table-max-growth-ratio-for-speed</code></dt>
<dd><p>The maximum code size growth ratio when expanding
into a jump table (in percent).  The parameter is used when
optimizing for speed.
</p>
</dd>
<dt><code>tree-reassoc-width</code></dt>
<dd><p>Set the maximum number of instructions executed in parallel in
reassociated tree. This parameter overrides target dependent
heuristics used by default if has non zero value.
</p>
</dd>
<dt><code>sched-pressure-algorithm</code></dt>
<dd><p>Choose between the two available implementations of
<samp>-fsched-pressure</samp>.  Algorithm 1 is the original implementation
and is the more likely to prevent instructions from being reordered.
Algorithm 2 was designed to be a compromise between the relatively
conservative approach taken by algorithm 1 and the rather aggressive
approach taken by the default scheduler.  It relies more heavily on
having a regular register file and accurate register pressure classes.
See <samp>haifa-sched.c</samp> in the GCC sources for more details.
</p>
<p>The default choice depends on the target.
</p>
</dd>
<dt><code>max-slsr-cand-scan</code></dt>
<dd><p>Set the maximum number of existing candidates that are considered when
seeking a basis for a new straight-line strength reduction candidate.
</p>
</dd>
<dt><code>asan-globals</code></dt>
<dd><p>Enable buffer overflow detection for global objects.  This kind
of protection is enabled by default if you are using
<samp>-fsanitize=address</samp> option.
To disable global objects protection use <samp>--param asan-globals=0</samp>.
</p>
</dd>
<dt><code>asan-stack</code></dt>
<dd><p>Enable buffer overflow detection for stack objects.  This kind of
protection is enabled by default when using <samp>-fsanitize=address</samp>.
To disable stack protection use <samp>--param asan-stack=0</samp> option.
</p>
</dd>
<dt><code>asan-instrument-reads</code></dt>
<dd><p>Enable buffer overflow detection for memory reads.  This kind of
protection is enabled by default when using <samp>-fsanitize=address</samp>.
To disable memory reads protection use
<samp>--param asan-instrument-reads=0</samp>.
</p>
</dd>
<dt><code>asan-instrument-writes</code></dt>
<dd><p>Enable buffer overflow detection for memory writes.  This kind of
protection is enabled by default when using <samp>-fsanitize=address</samp>.
To disable memory writes protection use
<samp>--param asan-instrument-writes=0</samp> option.
</p>
</dd>
<dt><code>asan-memintrin</code></dt>
<dd><p>Enable detection for built-in functions.  This kind of protection
is enabled by default when using <samp>-fsanitize=address</samp>.
To disable built-in functions protection use
<samp>--param asan-memintrin=0</samp>.
</p>
</dd>
<dt><code>asan-use-after-return</code></dt>
<dd><p>Enable detection of use-after-return.  This kind of protection
is enabled by default when using the <samp>-fsanitize=address</samp> option.
To disable it use <samp>--param asan-use-after-return=0</samp>.
</p>
<p>Note: By default the check is disabled at run time.  To enable it,
add <code>detect_stack_use_after_return=1</code> to the environment variable
<code>ASAN_OPTIONS</code>.
</p>
</dd>
<dt><code>asan-instrumentation-with-call-threshold</code></dt>
<dd><p>If number of memory accesses in function being instrumented
is greater or equal to this number, use callbacks instead of inline checks.
E.g. to disable inline code use
<samp>--param asan-instrumentation-with-call-threshold=0</samp>.
</p>
</dd>
<dt><code>use-after-scope-direct-emission-threshold</code></dt>
<dd><p>If the size of a local variable in bytes is smaller or equal to this
number, directly poison (or unpoison) shadow memory instead of using
run-time callbacks.
</p>
</dd>
<dt><code>max-fsm-thread-path-insns</code></dt>
<dd><p>Maximum number of instructions to copy when duplicating blocks on a
finite state automaton jump thread path.
</p>
</dd>
<dt><code>max-fsm-thread-length</code></dt>
<dd><p>Maximum number of basic blocks on a finite state automaton jump thread
path.
</p>
</dd>
<dt><code>max-fsm-thread-paths</code></dt>
<dd><p>Maximum number of new jump thread paths to create for a finite state
automaton.
</p>
</dd>
<dt><code>parloops-chunk-size</code></dt>
<dd><p>Chunk size of omp schedule for loops parallelized by parloops.
</p>
</dd>
<dt><code>parloops-schedule</code></dt>
<dd><p>Schedule type of omp schedule for loops parallelized by parloops (static,
dynamic, guided, auto, runtime).
</p>
</dd>
<dt><code>parloops-min-per-thread</code></dt>
<dd><p>The minimum number of iterations per thread of an innermost parallelized
loop for which the parallelized variant is preferred over the single threaded
one.  Note that for a parallelized loop nest the
minimum number of iterations of the outermost loop per thread is two.
</p>
</dd>
<dt><code>max-ssa-name-query-depth</code></dt>
<dd><p>Maximum depth of recursion when querying properties of SSA names in things
like fold routines.  One level of recursion corresponds to following a
use-def chain.
</p>
</dd>
<dt><code>hsa-gen-debug-stores</code></dt>
<dd><p>Enable emission of special debug stores within HSA kernels which are
then read and reported by libgomp plugin.  Generation of these stores
is disabled by default, use <samp>--param hsa-gen-debug-stores=1</samp> to
enable it.
</p>
</dd>
<dt><code>max-speculative-devirt-maydefs</code></dt>
<dd><p>The maximum number of may-defs we analyze when looking for a must-def
specifying the dynamic type of an object that invokes a virtual call
we may be able to devirtualize speculatively.
</p>
</dd>
<dt><code>max-vrp-switch-assertions</code></dt>
<dd><p>The maximum number of assertions to add along the default edge of a switch
statement during VRP.
</p>
</dd>
<dt><code>unroll-jam-min-percent</code></dt>
<dd><p>The minimum percentage of memory references that must be optimized
away for the unroll-and-jam transformation to be considered profitable.
</p>
</dd>
<dt><code>unroll-jam-max-unroll</code></dt>
<dd><p>The maximum number of times the outer loop should be unrolled by
the unroll-and-jam transformation.
</p>
</dd>
<dt><code>max-rtl-if-conversion-unpredictable-cost</code></dt>
<dd><p>Maximum permissible cost for the sequence that would be generated
by the RTL if-conversion pass for a branch that is considered unpredictable.
</p>
</dd>
<dt><code>max-variable-expansions-in-unroller</code></dt>
<dd><p>If <samp>-fvariable-expansion-in-unroller</samp> is used, the maximum number
of times that an individual variable will be expanded during loop unrolling.
</p>
</dd>
<dt><code>tracer-min-branch-probability-feedback</code></dt>
<dd><p>Stop forward growth if the probability of best edge is less than
this threshold (in percent). Used when profile feedback is available.
</p>
</dd>
<dt><code>partial-inlining-entry-probability</code></dt>
<dd><p>Maximum probability of the entry BB of split region
(in percent relative to entry BB of the function)
to make partial inlining happen.
</p>
</dd>
<dt><code>max-tracked-strlens</code></dt>
<dd><p>Maximum number of strings for which strlen optimization pass will
track string lengths.
</p>
</dd>
<dt><code>gcse-after-reload-partial-fraction</code></dt>
<dd><p>The threshold ratio for performing partial redundancy
elimination after reload.
</p>
</dd>
<dt><code>gcse-after-reload-critical-fraction</code></dt>
<dd><p>The threshold ratio of critical edges execution count that
permit performing redundancy elimination after reload.
</p>
</dd>
<dt><code>max-loop-header-insns</code></dt>
<dd><p>The maximum number of insns in loop header duplicated
by the copy loop headers pass.
</p>
</dd>
<dt><code>vect-epilogues-nomask</code></dt>
<dd><p>Enable loop epilogue vectorization using smaller vector size.
</p>
</dd>
<dt><code>slp-max-insns-in-bb</code></dt>
<dd><p>Maximum number of instructions in basic block to be
considered for SLP vectorization.
</p>
</dd>
<dt><code>avoid-fma-max-bits</code></dt>
<dd><p>Maximum number of bits for which we avoid creating FMAs.
</p>
</dd>
<dt><code>sms-loop-average-count-threshold</code></dt>
<dd><p>A threshold on the average loop count considered by the swing modulo scheduler.
</p>
</dd>
<dt><code>sms-dfa-history</code></dt>
<dd><p>The number of cycles the swing modulo scheduler considers when checking
conflicts using DFA.
</p>
</dd>
<dt><code>max-inline-insns-recursive-auto</code></dt>
<dd><p>The maximum number of instructions non-inline function
can grow to via recursive inlining.
</p>
</dd>
<dt><code>graphite-allow-codegen-errors</code></dt>
<dd><p>Whether codegen errors should be ICEs when <samp>-fchecking</samp>.
</p>
</dd>
<dt><code>sms-max-ii-factor</code></dt>
<dd><p>A factor for tuning the upper bound that swing modulo scheduler
uses for scheduling a loop.
</p>
</dd>
<dt><code>lra-max-considered-reload-pseudos</code></dt>
<dd><p>The max number of reload pseudos which are considered during
spilling a non-reload pseudo.
</p>
</dd>
<dt><code>max-pow-sqrt-depth</code></dt>
<dd><p>Maximum depth of sqrt chains to use when synthesizing exponentiation
by a real constant.
</p>
</dd>
<dt><code>max-dse-active-local-stores</code></dt>
<dd><p>Maximum number of active local stores in RTL dead store elimination.
</p>
</dd>
<dt><code>asan-instrument-allocas</code></dt>
<dd><p>Enable asan allocas/VLAs protection.
</p>
</dd>
<dt><code>max-iterations-computation-cost</code></dt>
<dd><p>Bound on the cost of an expression to compute the number of iterations.
</p>
</dd>
<dt><code>max-isl-operations</code></dt>
<dd><p>Maximum number of isl operations, 0 means unlimited.
</p>
</dd>
<dt><code>graphite-max-arrays-per-scop</code></dt>
<dd><p>Maximum number of arrays per scop.
</p>
</dd>
<dt><code>max-vartrack-reverse-op-size</code></dt>
<dd><p>Max. size of loc list for which reverse ops should be added.
</p>
</dd>
<dt><code>tracer-dynamic-coverage-feedback</code></dt>
<dd><p>The percentage of function, weighted by execution frequency,
that must be covered by trace formation.
Used when profile feedback is available.
</p>
</dd>
<dt><code>max-inline-recursive-depth-auto</code></dt>
<dd><p>The maximum depth of recursive inlining for non-inline functions.
</p>
</dd>
<dt><code>fsm-scale-path-stmts</code></dt>
<dd><p>Scale factor to apply to the number of statements in a threading path
when comparing to the number of (scaled) blocks.
</p>
</dd>
<dt><code>fsm-maximum-phi-arguments</code></dt>
<dd><p>Maximum number of arguments a PHI may have before the FSM threader
will not try to thread through its block.
</p>
</dd>
<dt><code>uninit-control-dep-attempts</code></dt>
<dd><p>Maximum number of nested calls to search for control dependencies
during uninitialized variable analysis.
</p>
</dd>
<dt><code>sra-max-scalarization-size-Osize</code></dt>
<dd><p>Maximum size, in storage units, of an aggregate
which should be considered for scalarization when compiling for size.
</p>
</dd>
<dt><code>fsm-scale-path-blocks</code></dt>
<dd><p>Scale factor to apply to the number of blocks in a threading path
when comparing to the number of (scaled) statements.
</p>
</dd>
<dt><code>sched-autopref-queue-depth</code></dt>
<dd><p>Hardware autoprefetcher scheduler model control flag.
Number of lookahead cycles the model looks into; at &rsquo;
&rsquo; only enable instruction sorting heuristic.
</p>
</dd>
<dt><code>loop-versioning-max-inner-insns</code></dt>
<dd><p>The maximum number of instructions that an inner loop can have
before the loop versioning pass considers it too big to copy.
</p>
</dd>
<dt><code>loop-versioning-max-outer-insns</code></dt>
<dd><p>The maximum number of instructions that an outer loop can have
before the loop versioning pass considers it too big to copy,
discounting any instructions in inner loops that directly benefit
from versioning.
</p>
</dd>
<dt><code>ssa-name-def-chain-limit</code></dt>
<dd><p>The maximum number of SSA_NAME assignments to follow in determining
a property of a variable such as its value.  This limits the number
of iterations or recursive calls GCC performs when optimizing certain
statements or when determining their validity prior to issuing
diagnostics.
</p>
</dd>
<dt><code>store-merging-max-size</code></dt>
<dd><p>Maximum size of a single store merging region in bytes.
</p>
</dd>
<dt><code>hash-table-verification-limit</code></dt>
<dd><p>The number of elements for which hash table verification is done
for each searched element.
</p>
</dd>
<dt><code>max-find-base-term-values</code></dt>
<dd><p>Maximum number of VALUEs handled during a single find_base_term call.
</p>
</dd>
<dt><code>analyzer-max-enodes-per-program-point</code></dt>
<dd><p>The maximum number of exploded nodes per program point within
the analyzer, before terminating analysis of that point.
</p>
</dd>
<dt><code>analyzer-min-snodes-for-call-summary</code></dt>
<dd><p>The minimum number of supernodes within a function for the
analyzer to consider summarizing its effects at call sites.
</p>
</dd>
<dt><code>analyzer-max-recursion-depth</code></dt>
<dd><p>The maximum number of times a callsite can appear in a call stack
within the analyzer, before terminating analysis of a call that would
recurse deeper.
</p>
</dd>
<dt><code>gimple-fe-computed-hot-bb-threshold</code></dt>
<dd><p>The number of executions of a basic block which is considered hot.
The parameter is used only in GIMPLE FE.
</p>
</dd>
<dt><code>analyzer-bb-explosion-factor</code></dt>
<dd><p>The maximum number of &rsquo;after supernode&rsquo; exploded nodes within the analyzer
per supernode, before terminating analysis.
</p>
</dd>
</dl>

<p>The following choices of <var>name</var> are available on AArch64 targets:
</p>
<dl compact="compact">
<dt><code>aarch64-sve-compare-costs</code></dt>
<dd><p>When vectorizing for SVE, consider using &ldquo;unpacked&rdquo; vectors for
smaller elements and use the cost model to pick the cheapest approach.
Also use the cost model to choose between SVE and Advanced SIMD vectorization.
</p>
<p>Using unpacked vectors includes storing smaller elements in larger
containers and accessing elements with extending loads and truncating
stores.
</p>
</dd>
<dt><code>aarch64-float-recp-precision</code></dt>
<dd><p>The number of Newton iterations for calculating the reciprocal for float type.
The precision of division is proportional to this param when division
approximation is enabled.  The default value is 1.
</p>
</dd>
<dt><code>aarch64-double-recp-precision</code></dt>
<dd><p>The number of Newton iterations for calculating the reciprocal for double type.
The precision of division is propotional to this param when division
approximation is enabled.  The default value is 2.
</p>
</dd>
</dl>

</dd>
</dl>

<hr>
<div class="header">
<p>
Next: <a href="Instrumentation-Options.html#Instrumentation-Options" accesskey="n" rel="next">Instrumentation Options</a>, Previous: <a href="Debugging-Options.html#Debugging-Options" accesskey="p" rel="prev">Debugging Options</a>, Up: <a href="Invoking-GCC.html#Invoking-GCC" accesskey="u" rel="up">Invoking GCC</a> &nbsp; [<a href="index.html#SEC_Contents" title="Table of contents" rel="contents">Contents</a>][<a href="Option-Index.html#Option-Index" title="Index" rel="index">Index</a>]</p>
</div>


</body>
</html>