diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp index 62c2c6ae6c4..2824fdb8860 100644 --- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp +++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp @@ -7181,19 +7181,23 @@ SDValue DAGCombiner::visitEXTRACT_SUBVECTOR(SDNode* N) { if (NVT != SmallVT || NVT.getSizeInBits()*2 != BigVT.getSizeInBits()) return SDValue(); - // Combine: - // (extract_subvec (insert_subvec V1, V2, InsIdx), ExtIdx) - // Into: - // indicies are equal => V1 - // otherwise => (extract_subvec V1, ExtIdx) - // - SDValue InsIdx = N->getOperand(1); - SDValue ExtIdx = V->getOperand(2); + // Only handle cases where both indexes are constants with the same type. + ConstantSDNode *InsIdx = dyn_cast(N->getOperand(1)); + ConstantSDNode *ExtIdx = dyn_cast(V->getOperand(2)); - if (InsIdx == ExtIdx) - return V->getOperand(1); - return DAG.getNode(ISD::EXTRACT_SUBVECTOR, N->getDebugLoc(), NVT, - V->getOperand(0), N->getOperand(1)); + if (InsIdx && ExtIdx && + InsIdx->getValueType(0).getSizeInBits() <= 64 && + ExtIdx->getValueType(0).getSizeInBits() <= 64) { + // Combine: + // (extract_subvec (insert_subvec V1, V2, InsIdx), ExtIdx) + // Into: + // indices are equal => V1 + // otherwise => (extract_subvec V1, ExtIdx) + if (InsIdx->getZExtValue() == ExtIdx->getZExtValue()) + return V->getOperand(1); + return DAG.getNode(ISD::EXTRACT_SUBVECTOR, N->getDebugLoc(), NVT, + V->getOperand(0), N->getOperand(1)); + } } return SDValue(); diff --git a/test/CodeGen/X86/2011-12-06-AVXVectorExtractCombine.ll b/test/CodeGen/X86/2011-12-06-AVXVectorExtractCombine.ll new file mode 100644 index 00000000000..fcaabddd2cc --- /dev/null +++ b/test/CodeGen/X86/2011-12-06-AVXVectorExtractCombine.ll @@ -0,0 +1,18 @@ +; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx -mattr=+avx | FileCheck %s +; PR11494 + +define void @test(<4 x i32>* nocapture %p) nounwind { + ; CHECK: test: + ; CHECK: vpxor %xmm0, %xmm0, %xmm0 + ; CHECK-NEXT: vpmaxsd {{.*}}, %xmm0, %xmm0 + ; CHECK-NEXT: vmovdqu %xmm0, (%rdi) + ; CHECK-NEXT: ret + %a = call <4 x i32> @llvm.x86.sse41.pmaxsd(<4 x i32> , <4 x i32> zeroinitializer) nounwind + %b = shufflevector <4 x i32> %a, <4 x i32> undef, <8 x i32> + %c = shufflevector <8 x i32> %b, <8 x i32> undef, <4 x i32> + store <4 x i32> %c, <4 x i32>* %p, align 1 + ret void +} + +declare <4 x i32> @llvm.x86.sse41.pminsd(<4 x i32>, <4 x i32>) nounwind readnone +declare <4 x i32> @llvm.x86.sse41.pmaxsd(<4 x i32>, <4 x i32>) nounwind readnone