[llvm] [X86][AVX] Match v4f64 blend from shuffle of scalar values. (PR #135753)

[Simon Pilgrim via llvm-commits]

================
@@ -8743,6 +8745,52 @@ static SDValue lowerBuildVectorToBitOp(BuildVectorSDNode *Op, const SDLoc &DL,
   return LowerShift(Res, Subtarget, DAG);
 }
 
+static bool isShuffleFoldableLoad(SDValue);
+
+/// Attempt to lower a BUILD_VECTOR of scalar values to a shuffle of splats
+/// representing a blend.
+static SDValue lowerBuildVectorAsBlend(BuildVectorSDNode *BVOp, SDLoc const &DL,
+                                       X86Subtarget const &Subtarget,
+                                       SelectionDAG &DAG) {
+  MVT VT = BVOp->getSimpleValueType(0u);
+
+  if (VT != MVT::v4f64)
+    return SDValue();
+
+  // Collect unique operands.
+  auto UniqueOps = SmallSet<SDValue, 16u>();
+  for (SDValue Op : BVOp->ops()) {
+    if (isIntOrFPConstant(Op) || Op.isUndef())
+      return SDValue();
+    UniqueOps.insert(Op);
+  }
+
+  // Candidate BUILD_VECTOR must have 2 unique operands.
+  if (UniqueOps.size() != 2u)
+    return SDValue();
+
+  SDValue Op0 = BVOp->getOperand(0u);
+  UniqueOps.erase(Op0);
+  SDValue Op1 = *UniqueOps.begin();
+
+  if (isShuffleFoldableLoad(Op0) || isShuffleFoldableLoad(Op1) ||
+      Subtarget.hasAVX2()) {
----------------
RKSimon wrote:

Check for hasAVX2 first - its a lot cheaper than the isShuffleFoldableLoad checks.

https://github.com/llvm/llvm-project/pull/135753