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

[Sander de Smalen via llvm-commits]

================
@@ -30233,6 +30142,15 @@ 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->noSVELD1RDUP() &&
+          Subtarget->isSVEorStreamingSVEAvailable() && Op->hasOneUse() &&
+          VT.getScalarType().isInteger() &&
+          VT.getScalarType() != LD->getMemoryVT().getScalarType()) {
----------------
sdesmalen-arm wrote:

Perhaps also add a check that we don't want to use SVE for this type? For example, for fixed-length vector types from using `-msve-vector-bits`, we still want to use LD1R. (Can you also add a test-case for this?)

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