[llvm] [VectorCombine] foldShuffleChainsToReduce - add FADD/FMUL handling (PR #201302)

[Rajveer Singh Bharadwaj via llvm-commits]

================
@@ -4172,6 +4192,13 @@ bool VectorCombine::foldShuffleChainsToReduce(Instruction &I) {
   Intrinsic::ID ReducedOp =
       (CommonCallOp ? getMinMaxReductionIntrinsicID(*CommonCallOp)
                     : getReductionForBinop(*CommonBinOp));
+  // getReductionForBinop only covers integer ops; handle FP here.
+  if (ReducedOp == Intrinsic::not_intrinsic && CommonBinOp) {
+    if (*CommonBinOp == Instruction::FAdd)
+      ReducedOp = Intrinsic::vector_reduce_fadd;
+    else if (*CommonBinOp == Instruction::FMul)
+      ReducedOp = Intrinsic::vector_reduce_fmul;
+  }
----------------
Rajveer100 wrote:

I think we could add this in `getReductionForBinop` itself right? It would go well with `getArithmeticReductionInstruction` too since they are inverse of each other.

https://github.com/llvm/llvm-project/pull/201302