X86 backend doesn't support MMX

[lattner]

Bugzilla Link	1222
Resolution	FIXED
Resolved on	Aug 19, 2007 21:09
Version	1.0
OS	All
Depends On	llvm/llvm-bugzilla-archive#1260

Extended Description

The X86 backend currently has no support for MMX registers. This can be seen when using MMX
intrinsic or when using MMX inline asm. Here's an example that uses inline asm:

void foo() {
const int Aligned64data_last_mult = 4;
asm {
movd mm1, [-1][eax][esi]
movd mm0, [-1][eax]
movd mm2, [-1][eax][esi2]
movq [edi][Aligned64data_last_mult], mm0
lea eax, [esi][ecx2]
movd mm4, [esi]
movd mm5, [esi][ecx]
movd [-4][edi], mm3
}
}

llvm-gcc -fasm-blocks -msse3 should successfully compile this, but doesn't due to lack of support
for mmx registers.

-Chris

[lattner]

assigned to @isanbard

[isanbard]

I'm giving the whole MMX thing a shot.

-bw

[isanbard]

We support MMX now. However, we still need to make asm blocks work.

-bw

[lattner]

Bill implemented MMX, and this patch fixes microsoft-style asm blocks with mmx:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070813/052983.html

-Chris

[lattner]

mentioned in issue llvm/llvm-bugzilla-archive#1260