1 files changed, 39 insertions, 0 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 47389f2df32..34a85806f56 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -42430,6 +42430,41 @@ static SDValue combineVectorCompareAndMaskUnaryOp(SDNode *N,
   return SDValue();
 }
 
+/// If we are converting a value to floating-point, try to replace scalar
+/// truncate of an extracted vector element with a bitcast. This tries to keep
+/// the sequence on XMM registers rather than moving between vector and GPRs.
+static SDValue combineToFPTruncExtElt(SDNode *N, SelectionDAG &DAG) {
+  // TODO: This is currently only used by combineSIntToFP, but it is generalized
+  //       to allow being called by any similar cast opcode.
+  // TODO: Consider merging this into lowering: vectorizeExtractedCast().
+  SDValue Trunc = N->getOperand(0);
+  if (!Trunc.hasOneUse() || Trunc.getOpcode() != ISD::TRUNCATE)
+    return SDValue();
+
+  SDValue ExtElt = Trunc.getOperand(0);
+  if (!ExtElt.hasOneUse() || ExtElt.getOpcode() != ISD::EXTRACT_VECTOR_ELT ||
+      !isNullConstant(ExtElt.getOperand(1)))
+    return SDValue();
+
+  EVT TruncVT = Trunc.getValueType();
+  EVT SrcVT = ExtElt.getValueType();
+  unsigned DestWidth = TruncVT.getSizeInBits();
+  unsigned SrcWidth = SrcVT.getSizeInBits();
+  if (SrcWidth % DestWidth != 0)
+    return SDValue();
+
+  // inttofp (trunc (extelt X, 0)) --> inttofp (extelt (bitcast X), 0)
+  EVT SrcVecVT = ExtElt.getOperand(0).getValueType();
+  unsigned VecWidth = SrcVecVT.getSizeInBits();
+  unsigned NumElts = VecWidth / DestWidth;
+  EVT BitcastVT = EVT::getVectorVT(*DAG.getContext(), TruncVT, NumElts);
+  SDValue BitcastVec = DAG.getBitcast(BitcastVT, ExtElt.getOperand(0));
+  SDLoc DL(N);
+  SDValue NewExtElt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, TruncVT,
+                                  BitcastVec, ExtElt.getOperand(1));
+  return DAG.getNode(N->getOpcode(), DL, N->getValueType(0), NewExtElt);
+}
+
 static SDValue combineUIntToFP(SDNode *N, SelectionDAG &DAG,
                                const X86Subtarget &Subtarget) {
   SDValue Op0 = N->getOperand(0);
@@ -42523,6 +42558,10 @@ static SDValue combineSIntToFP(SDNode *N, SelectionDAG &DAG,
       return FILDChain;
     }
   }
+
+  if (SDValue V = combineToFPTruncExtElt(N, DAG))
+    return V;
+
   return SDValue();
 }