http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55864
--- Comment #3 from beldmit at nm dot ru 2013-01-03 16:33:02 UTC ---
If this case is better to reproduce the problem or I completely do not
understand the asm usage?
The mmx registers and operations are used to speedup the 64-bit xor.
void _xor64_asm(const unsigned char * arg1, const unsigned char * arg2,
unsigned char* out)
{
__asm__(
"movq (%1), %%mm0;"
"movq (%2), %%mm1;"
"pxor %%mm0, %%mm1;"
"movl %0, %%ecx;"
"movq %%mm1, (%%ecx);"
"emms"
: "=ym"(out)
: "r"(arg1), "r"(arg2)
: "mm1", "mm2", "ecx");
}