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

[Simon Pilgrim via llvm-commits]

================
@@ -8743,6 +8745,43 @@ static SDValue lowerBuildVectorToBitOp(BuildVectorSDNode *Op, const SDLoc &DL,
   return LowerShift(Res, Subtarget, DAG);
 }
 
+/// 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);
+  auto const NumElems = VT.getVectorNumElements();
+
+  if (Subtarget.hasAVX() && VT == MVT::v4f64) {
----------------
RKSimon wrote:

to avoid the AVX1 'var' regression below we're going to need something along the lines of:
```
if (Subtarget.hasAVX2() || isShuffleFoldableLoad(Op0) || isShuffleFoldableLoad(Op1))
```
(we shouldn't need the AVX check as v4f64 is only legal on AVX+ targets)

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