[llvm] [AArch64] Lower v8bf16 FMUL to BFMLAL top/bottom with +sve (PR #169655)

[Sam Tebbs via llvm-commits]

================
@@ -7697,37 +7701,59 @@ SDValue AArch64TargetLowering::LowerFMUL(SDValue Op, SelectionDAG &DAG) const {
     };
   };
 
-  auto ReinterpretCast = [&](SDValue Value, EVT VT) {
-    if (VT == Value.getValueType())
+  auto Reinterpret = [&](SDValue Value, EVT VT) {
+    EVT SrcVT = Value.getValueType();
+    if (VT == SrcVT)
       return Value;
+    if (SrcVT.isFixedLengthVector())
+      return convertToScalableVector(DAG, VT, Value);
----------------
SamTebbs33 wrote:

Should this be `SrcVT` instead of `VT`? If so then it looks like we need some test coverage for this line.

https://github.com/llvm/llvm-project/pull/169655