[llvm] [AArch64][SVE] Fix v16i8 -> v2i64 partial_reduce for VL > 128 (PR #204938)

[Sander de Smalen via llvm-commits]

================
@@ -34635,6 +34635,7 @@ AArch64TargetLowering::LowerPARTIAL_REDUCE_MLA(SDValue Op,
   SDLoc DL(Op);
 
   SDValue Acc = Op.getOperand(0);
+  SDValue OrigAcc = Acc;
----------------
sdesmalen-arm wrote:

minor nit: my preference would be for us not to overwrite the original function parameter `Op` in the `ConvertToScalable` path, so that you can always write `Op.getOperand(0)` and we don't need to create `OrigAcc`.

https://github.com/llvm/llvm-project/pull/204938