[llvm] [AArch64] Fix regression from “Fold scalar-to-vector shuffles into DUP/FMOV" (PR #178227)

[David Green via llvm-commits]

================
@@ -16016,8 +16017,23 @@ SDValue AArch64TargetLowering::LowerBUILD_VECTOR(SDValue Op,
   SmallMapVector<SDValue, unsigned, 16> DifferentValueMap;
   unsigned ConsecutiveValCount = 0;
   SDValue PrevVal;
+  auto IsZero = [&](SDValue V) {
+    return isNullConstant(V) || isNullFPConstant(V);
+  };
+  bool MaybeLowHalfZeroHigh =
+      VT.isFixedLengthVector() && VT.getSizeInBits() == 128 && NumElts != 0;
----------------
davemgreen wrote:

When can NumElts == 0?

https://github.com/llvm/llvm-project/pull/178227