[llvm-commits] [llvm] r131493 - in /llvm/trunk: lib/Transforms/InstCombine/InstCombineCalls.cpp lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp test/CodeGen/X86/2011-05-17-pmovzxwd.ll

Chris Lattner clattner at apple.com
Wed May 18 14:27:13 PDT 2011


Thanks Stuart!

-Chris

On May 18, 2011, at 10:07 AM, Stuart Hastings wrote:

> 
> On May 17, 2011, at 9:33 PM, Chris Lattner wrote:
> 
>> 
>> On May 17, 2011, at 3:13 PM, Stuart Hastings wrote:
>> 
>>> Author: stuart
>>> Date: Tue May 17 17:13:31 2011
>>> New Revision: 131493
>>> 
>>> URL: http://llvm.org/viewvc/llvm-project?rev=131493&view=rev
>>> Log:
>>> X86 pmovsx/pmovzx ignore the upper half of their inputs.
>>> rdar://problem/6945110
>> 
>> Hi Stuart,
>> 
>>> +  case Intrinsic::x86_sse41_pmovzxdq: {
>>> +    unsigned VWidth =
>>> +      cast<VectorType>(II->getArgOperand(0)->getType())->getNumElements();
>>> +    unsigned LowHalfElts = VWidth / 2;
>>> +    APInt InputDemandedElts(VWidth, 0);
>>> +    InputDemandedElts = InputDemandedElts.getBitsSet(VWidth, 0, LowHalfElts);
>> 
>> getBitsSet is a static method.  Please do this instead:
>> 
>> APInt InputDemandedElts(APInt::getBitsSet(VWidth, 0, LowHalfElts));
> 
> Done in 131538.
> 
>>> +++ llvm/trunk/test/CodeGen/X86/2011-05-17-pmovzxwd.ll Tue May 17 17:13:31 2011
>>> @@ -0,0 +1,15 @@
>>> +; RUN: opt -instcombine -S < %s | FileCheck %s
>>> +; <rdar://problem/6945110>
>>> +
>>> +define <4 x i32> @kernel3_vertical(<4 x i16> * %src, <8 x i16> * %foo) nounwind {
>>> +entry:
>>> +	%tmp = load <4 x i16>* %src
>>> +	%tmp1 = load <8 x i16>* %foo
>>> +; CHECK: shufflevector
>>> +	%tmp2 = shufflevector <4 x i16> %tmp, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
>>> +; CHECK-NOT: shufflevector
>>> +	%tmp3 = shufflevector <8 x i16> %tmp1, <8 x i16> %tmp2, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 4, i32 5, i32 6, i32 7>
>>> +	%0 = call <4 x i32> @llvm.x86.sse41.pmovzxwd(<8 x i16> %tmp3)
>>> +	ret <4 x i32> %0
>>> +}
>>> +declare <4 x i32> @llvm.x86.sse41.pmovzxwd(<8 x i16>) nounwind readnone
>> 
>> Please make the CHECK lines more specific so that it is more clear what you're doing here, add a comment explaining what is going on (to the test)
>> and merge the test into an extant .ll file.
> 
> Merged into vec_shuffle-36.ll in 131539.
> 
>> Thanks,
> 
> Thanks for the review,
> 
> stuart




More information about the llvm-commits mailing list