Implement VPWidenStridedLoadRecipe::computeCost

Author: Mel-Chen

llvm/lib/Transforms/Vectorize/VPlan.h

@@ -3384,6 +3384,10 @@ struct VPWidenStridedLoadRecipe final : public VPWidenMemoryRecipe,
   /// Generate a strided load.
   void execute(VPTransformState &State) override;
 
+  /// Return the cost of this VPWidenStridedLoadRecipe.
+  InstructionCost computeCost(ElementCount VF,
+                              VPCostContext &Ctx) const override;
+
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
   /// Print the recipe.
   void print(raw_ostream &O, const Twine &Indent,

llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp

@@ -3508,11 +3508,9 @@ InstructionCost VPWidenMemoryRecipe::computeCost(ElementCount VF,
   const Align Alignment = getLoadStoreAlignment(&Ingredient);
   unsigned AS = cast<PointerType>(Ctx.Types.inferScalarType(getAddr()))
                     ->getAddressSpace();
-  unsigned Opcode =
-      isa<VPWidenLoadRecipe, VPWidenLoadEVLRecipe, VPWidenStridedLoadRecipe>(
-          this)
-          ? Instruction::Load
-          : Instruction::Store;
+  unsigned Opcode = isa<VPWidenLoadRecipe, VPWidenLoadEVLRecipe>(this)
+                        ? Instruction::Load
+                        : Instruction::Store;
 
   if (!Consecutive) {
     // TODO: Using the original IR may not be accurate.
@@ -3522,10 +3520,6 @@ InstructionCost VPWidenMemoryRecipe::computeCost(ElementCount VF,
            "Inconsecutive memory access should not have the order.");
 
     const Value *Ptr = getLoadStorePointerOperand(&Ingredient);
-    if (isa<VPWidenStridedLoadRecipe>(this))
-      return Ctx.TTI.getStridedMemoryOpCost(
-          Opcode, Ty, Ptr, IsMasked, Alignment, Ctx.CostKind, &Ingredient);
-
     Type *PtrTy = Ptr->getType();
     // If the address value is uniform across all lanes, then the address can be
     // calculated with scalar type and broadcast.
@@ -3708,6 +3702,16 @@ void VPWidenStridedLoadRecipe::execute(VPTransformState &State) {
   State.set(this, NewLI);
 }
 
+InstructionCost
+VPWidenStridedLoadRecipe::computeCost(ElementCount VF,
+                                      VPCostContext &Ctx) const {
+  Type *Ty = toVectorTy(getLoadStoreType(&Ingredient), VF);
+  const Value *Ptr = getLoadStorePointerOperand(&Ingredient);
+  const Align Alignment = getLoadStoreAlignment(&Ingredient);
+  return Ctx.TTI.getStridedMemoryOpCost(Instruction::Load, Ty, Ptr, IsMasked,
+                                        Alignment, Ctx.CostKind, &Ingredient);
+}
+
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void VPWidenStridedLoadRecipe::print(raw_ostream &O, const Twine &Indent,
                                      VPSlotTracker &SlotTracker) const {