Only use memset.S with -Os, C lib is faster with -O

teensy3/memset.S

  */
 //#include <asm.h>
 //#include <arch/arm/cores.h>
+#if defined (__OPTIMIZE_SIZE__)
 #if defined (__ARM_ARCH_7M__) || defined (__ARM_ARCH_7EM__)
 .global	memset
 .text
     pop     { r0, pc }
 
 #endif
+#endif