[llvm-commits] [llvm] r161152 - in /llvm/trunk: include/llvm/Target/TargetInstrInfo.h lib/CodeGen/PeepholeOptimizer.cpp lib/Target/X86/X86InstrInfo.cpp lib/Target/X86/X86InstrInfo.h test/CodeGen/X86/2012-05-19-avx2-store.ll test/CodeGen/X86/break

[NAKAMURA Takumi]

>>  - It behaves as if avx were enabled without -avx, on avx-capable hosts.
>>    (expressions would match also to vpcmpeqd and vxorps. Is it your expected?)

Fixed in r161156. See also;
http://bb.pgr.jp/builders/cmake-llvm-x86_64-linux/builds/1214
It was not yours but had been there, though.

That said, this had been missing unexpected case on avx hosts.

>>  - Return value may be "%xmm0" instead of %xmm[0-9]+

Updated in r161155.

2012/8/2 Manman Ren <mren at apple.com>:
>>  - It is incompatible to win32 due to PR11031.
>>    Could you reconfirm to add -mtriple=i686-win32 instead of -march?

> I am re-writing this testing case to get rid of "grep".  Is this re-writing making win32 fail?
> I will look at PR11031 to figure out what is the problem.

Fixed in r161155.
You could reproduce to rewrite RUN line locally as below;

- ; RUN: llc < %s -march=x86 -mattr=+sse2 | FileCheck %s
+ ; RUN: llc < %s -mtriple=i686-win32 -mattr=+sse2 | FileCheck %s


...Takumi