diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp index 2d76a9799e2..94452b1e31a 100644 --- a/lib/Target/X86/X86ISelLowering.cpp +++ b/lib/Target/X86/X86ISelLowering.cpp @@ -7556,10 +7556,16 @@ static SDValue lowerVectorShuffleAsZeroOrAnyExtend( static SDValue lowerIntegerElementInsertionVectorShuffle( MVT VT, SDLoc DL, SDValue V1, SDValue V2, ArrayRef Mask, const X86Subtarget *Subtarget, SelectionDAG &DAG) { + SmallBitVector Zeroable = computeZeroableShuffleElements(Mask, V1, V2); + int V2Index = std::find_if(Mask.begin(), Mask.end(), [&Mask](int M) { return M >= (int)Mask.size(); }) - Mask.begin(); + for (int i = 0, Size = Mask.size(); i < Size; ++i) + if (i != V2Index && !Zeroable[i]) + return SDValue(); // Not inserting into a zero vector. + // Check for a single input from a SCALAR_TO_VECTOR node. // FIXME: All of this should be canonicalized into INSERT_VECTOR_ELT and // all the smarts here sunk into that routine. However, the current @@ -7572,19 +7578,6 @@ static SDValue lowerIntegerElementInsertionVectorShuffle( SDValue V2S = V2.getOperand(Mask[V2Index] - Mask.size()); - if (V1.getOpcode() == ISD::BUILD_VECTOR) { - for (int M : Mask) { - if (M < 0 || M >= (int)Mask.size()) - continue; - SDValue Input = V1.getOperand(M); - if (Input.getOpcode() != ISD::UNDEF && !X86::isZeroNode(Input)) - // A non-zero input! - return SDValue(); - } - } else if (!ISD::isBuildVectorAllZeros(V1.getNode())) { - return SDValue(); - } - // First, we need to zext the scalar if it is smaller than an i32. MVT EltVT = VT.getVectorElementType(); assert(EltVT == V2S.getSimpleValueType() &&