diff options
author | Simon Pilgrim <llvm-dev@redking.me.uk> | 2015-09-17 20:32:45 +0000 |
---|---|---|
committer | Simon Pilgrim <llvm-dev@redking.me.uk> | 2015-09-17 20:32:45 +0000 |
commit | 61116ddc7bd7953fd90455927f8ca60e3fd713c2 (patch) | |
tree | 48c002732fb9db84da254dfdce31bd886dea1585 /llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp | |
parent | b83278687dcab59646e43e6f6907d8cef7585692 (diff) | |
download | llvm-61116ddc7bd7953fd90455927f8ca60e3fd713c2.zip llvm-61116ddc7bd7953fd90455927f8ca60e3fd713c2.tar.gz llvm-61116ddc7bd7953fd90455927f8ca60e3fd713c2.tar.bz2 |
[InstCombine] Added vector demanded bits support for SSE4A EXTRQ/INSERTQ instructions
The SSE4A instructions EXTRQ/INSERTQ only use the lower 64-bits (or less) for many of their input vector operands and all of them have undefined upper 64-bits results.
Differential Revision: http://reviews.llvm.org/D12680
llvm-svn: 247934
Diffstat (limited to 'llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp')
-rw-r--r-- | llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp | 11 |
1 files changed, 10 insertions, 1 deletions
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp index 142e071..c68a031 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp @@ -412,7 +412,7 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask, Value *LHS, *RHS; if (matchSelectPattern(I, LHS, RHS).Flavor != SPF_UNKNOWN) return nullptr; - + if (SimplifyDemandedBits(I->getOperandUse(2), DemandedMask, RHSKnownZero, RHSKnownOne, Depth + 1) || SimplifyDemandedBits(I->getOperandUse(1), DemandedMask, LHSKnownZero, @@ -1237,6 +1237,15 @@ Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts, // like undef&0. The result is known zero, not undef. UndefElts &= UndefElts2; break; + + // SSE4A instructions leave the upper 64-bits of the 128-bit result + // in an undefined state. + case Intrinsic::x86_sse4a_extrq: + case Intrinsic::x86_sse4a_extrqi: + case Intrinsic::x86_sse4a_insertq: + case Intrinsic::x86_sse4a_insertqi: + UndefElts |= APInt::getHighBitsSet(VWidth, VWidth / 2); + break; } break; } |