diff options
| -rw-r--r-- | lib/Target/X86/X86ISelLowering.cpp | 22 | ||||
| -rw-r--r-- | test/CodeGen/X86/pr18054.ll | 10 |
2 files changed, 25 insertions, 7 deletions
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp index 627aa86e75..12d40c488a 100644 --- a/lib/Target/X86/X86ISelLowering.cpp +++ b/lib/Target/X86/X86ISelLowering.cpp @@ -13120,19 +13120,27 @@ SDValue X86TargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op, // fall through case MVT::v4i32: case MVT::v8i16: { - // (sext (vzext x)) -> (vsext x) SDValue Op0 = Op.getOperand(0); SDValue Op00 = Op0.getOperand(0); SDValue Tmp1; // Hopefully, this VECTOR_SHUFFLE is just a VZEXT. if (Op0.getOpcode() == ISD::BITCAST && - Op00.getOpcode() == ISD::VECTOR_SHUFFLE) + Op00.getOpcode() == ISD::VECTOR_SHUFFLE) { + // (sext (vzext x)) -> (vsext x) Tmp1 = LowerVectorIntExtend(Op00, Subtarget, DAG); - if (Tmp1.getNode()) { - SDValue Tmp1Op0 = Tmp1.getOperand(0); - assert(Tmp1Op0.getOpcode() == X86ISD::VZEXT && - "This optimization is invalid without a VZEXT."); - return DAG.getNode(X86ISD::VSEXT, dl, VT, Tmp1Op0.getOperand(0)); + if (Tmp1.getNode()) { + EVT ExtraEltVT = ExtraVT.getVectorElementType(); + // This folding is only valid when the in-reg type is a vector of i8, + // i16, or i32. + if (ExtraEltVT == MVT::i8 || ExtraEltVT == MVT::i16 || + ExtraEltVT == MVT::i32) { + SDValue Tmp1Op0 = Tmp1.getOperand(0); + assert(Tmp1Op0.getOpcode() == X86ISD::VZEXT && + "This optimization is invalid without a VZEXT."); + return DAG.getNode(X86ISD::VSEXT, dl, VT, Tmp1Op0.getOperand(0)); + } + Op0 = Tmp1; + } } // If the above didn't work, then just use Shift-Left + Shift-Right. diff --git a/test/CodeGen/X86/pr18054.ll b/test/CodeGen/X86/pr18054.ll new file mode 100644 index 0000000000..b7af516180 --- /dev/null +++ b/test/CodeGen/X86/pr18054.ll @@ -0,0 +1,10 @@ +; RUN: llc < %s -mtriple=x86_64-pc-linux -mcpu=penryn | FileCheck %s + +define void @foo(<16 x i32>* %p, <16 x i1> %x) { + %ret = sext <16 x i1> %x to <16 x i32> + store <16 x i32> %ret, <16 x i32>* %p + ret void +; CHECK: foo +; CHECK-NOT: pmovsxbd +; CHECK: ret +} |