[llvm] [AArch64] Use SVE for load-extend-broadcast (PR #221689)

[Benjamin Maxwell via llvm-commits]

================
@@ -30233,6 +30142,18 @@ static SDValue performDUPCombine(SDNode *N,
     //   v4i32 = SCALAR_TO_VECTOR (i32 (zextloadi8 addr)) ; Matches to ldr b0
     //   v4i32 = DUPLANE32 (v4i32), 0
     if (auto *LD = dyn_cast<LoadSDNode>(Op)) {
+      if (Subtarget->preferSVEVectors() &&
+          Subtarget->isSVEorStreamingSVEAvailable() && Op->hasOneUse() &&
+          VT.getScalarType().isInteger()) {
+        EVT ScalableVT = getContainerForFixedLengthVector(DCI.DAG, VT);
+        // Using SVE Vectors with the same scalar type is not profitable
+        if (!(ScalableVT.getScalarType() ==
+              LD->getMemoryVT().getScalarType())) {
----------------
MacDue wrote:

`getContainerForFixedLengthVector` does not change the element type of `VT`. So `VT.getScalarType() != LD->getMemoryVT().getScalarType())` can be added to the outer condition and the container type does not need to be created until we're going to do the combine.  

https://github.com/llvm/llvm-project/pull/221689