[llvm-branch-commits] [llvm] [LoopVectorizer] Bundle partial reductions inside VPMulAccumulateReductionRecipe (PR #136173)

[Sander de Smalen via llvm-branch-commits]

================
@@ -4923,9 +4923,7 @@ InstructionCost AArch64TTIImpl::getPartialReductionCost(
         return Invalid;
       break;
     case 16:
-      if (AccumEVT == MVT::i64)
-        Cost *= 2;
-      else if (AccumEVT != MVT::i32)
+      if (AccumEVT != MVT::i32)
----------------
sdesmalen-arm wrote:

```llvm.partial.reduce(nxv2i64 %acc, mul(zext(nxv16i8 %a) -> nxv16i64, zext(nxv16i8 %b) -> nxv16i64))```
This case can be lowered with the correct support in selectiondag (see https://github.com/llvm/llvm-project/pull/130935#discussion_r2068162662)

I'd argue that this function will need a bit of a rewrite anyway once we get proper support in SelectionDAG for legalising these operations, so I wouldn't bother changing that in this PR.

https://github.com/llvm/llvm-project/pull/136173