[llvm-branch-commits] [llvm] [LV] Bundle (partial) reductions with a mul of a constant (PR #162503)

[Sander de Smalen via llvm-branch-commits]

================
@@ -3472,25 +3472,30 @@ tryToMatchAndCreateExtendedReduction(VPReductionRecipe *Red, VPCostContext &Ctx,
           auto *SrcVecTy = cast<VectorType>(toVectorTy(SrcTy, VF));
           TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
 
-          InstructionCost ExtRedCost;
           InstructionCost ExtCost =
               cast<VPWidenCastRecipe>(VecOp)->computeCost(VF, Ctx);
           InstructionCost RedCost = Red->computeCost(VF, Ctx);
+          InstructionCost BaseCost = ExtCost + RedCost;
 
           if (isa<VPPartialReductionRecipe>(Red)) {
             TargetTransformInfo::PartialReductionExtendKind ExtKind =
                 TargetTransformInfo::getPartialReductionExtendKind(ExtOpc);
             // FIXME: Move partial reduction creation, costing and clamping
             // here from LoopVectorize.cpp.
-            ExtRedCost = Ctx.TTI.getPartialReductionCost(
-                Opcode, SrcTy, nullptr, RedTy, VF, ExtKind,
-                llvm::TargetTransformInfo::PR_None, std::nullopt, Ctx.CostKind);
-          } else {
-            ExtRedCost = Ctx.TTI.getExtendedReductionCost(
-                Opcode, ExtOpc == Instruction::CastOps::ZExt, RedTy, SrcVecTy,
-                Red->getFastMathFlags(), CostKind);
+            InstructionCost PartialReductionCost =
+                Ctx.TTI.getPartialReductionCost(
+                    Opcode, SrcTy, nullptr, RedTy, VF, ExtKind,
+                    llvm::TargetTransformInfo::PR_None, std::nullopt,
+                    Ctx.CostKind);
+            assert(PartialReductionCost.isValid() &&
----------------
sdesmalen-arm wrote:

A little bit to my surprise, it seems none of the tests change by doing this.
My suggestion would be to keep the code exactly the same as it was, but just add the assert that the partial reduction cost is valid. And possibly to do that in a separate NFCI PR from this one, because they're functionally unrelated.

https://github.com/llvm/llvm-project/pull/162503