[VPlan] Implement VPHeaderPHIRecipe::computeCost.

Fill out computeCost implementations for various header PHI recipes,
matching the legacy cost model for now.

Author: fhahn

llvm/lib/Transforms/Vectorize/VPlan.h

@@ -2050,6 +2050,10 @@ class VPHeaderPHIRecipe : public VPSingleDefRecipe {
   /// Generate the phi nodes.
   void execute(VPTransformState &State) override = 0;
 
+  /// Return the cost of this header phi recipe.
+  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,
@@ -2295,6 +2299,10 @@ struct VPFirstOrderRecurrencePHIRecipe : public VPHeaderPHIRecipe {
 
   void execute(VPTransformState &State) override;
 
+  /// Return the cost of this first-order recurrence phi recipe.
+  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,
@@ -3134,6 +3142,13 @@ class VPCanonicalIVPHIRecipe : public VPHeaderPHIRecipe {
   /// canonical, i.e.  has the same start and step (of 1) as the canonical IV.
   bool isCanonical(InductionDescriptor::InductionKind Kind, VPValue *Start,
                    VPValue *Step) const;
+
+  /// Return the cost of this VPCanonicalIVPHIRecipe.
+  InstructionCost computeCost(ElementCount VF,
+                              VPCostContext &Ctx) const override {
+    // For now, match the behavior of the legacy cost model.
+    return 0;
+  }
 };
 
 /// A recipe for generating the active lane mask for the vector loop that is
@@ -3196,6 +3211,13 @@ class VPEVLBasedIVPHIRecipe : public VPHeaderPHIRecipe {
   /// TODO: investigate if it can share the code with VPCanonicalIVPHIRecipe.
   void execute(VPTransformState &State) override;
 
+  /// Return the cost of this VPEVLBasedIVPHIRecipe.
+  InstructionCost computeCost(ElementCount VF,
+                              VPCostContext &Ctx) const override {
+    // For now, match the behavior of the legacy cost model.
+    return 0;
+  }
+
   /// Returns true if the recipe only uses the first lane of operand \p Op.
   bool onlyFirstLaneUsed(const VPValue *Op) const override {
     assert(is_contained(operands(), Op) &&

llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp

@@ -1589,6 +1589,11 @@ void VPWidenCastRecipe::print(raw_ostream &O, const Twine &Indent,
 }
 #endif
 
+InstructionCost VPHeaderPHIRecipe::computeCost(ElementCount VF,
+                                               VPCostContext &Ctx) const {
+  return Ctx.TTI.getCFInstrCost(Instruction::PHI, TTI::TCK_RecipThroughput);
+}
+
 /// This function adds
 /// (StartIdx * Step, (StartIdx + 1) * Step, (StartIdx + 2) * Step, ...)
 /// to each vector element of Val. The sequence starts at StartIndex.
@@ -3334,6 +3339,23 @@ void VPFirstOrderRecurrencePHIRecipe::execute(VPTransformState &State) {
   State.set(this, Phi);
 }
 
+InstructionCost
+VPFirstOrderRecurrencePHIRecipe::computeCost(ElementCount VF,
+                                             VPCostContext &Ctx) const {
+  if (VF.isScalable() && VF.getKnownMinValue() == 1)
+    return InstructionCost::getInvalid();
+
+  SmallVector<int> Mask(VF.getKnownMinValue());
+  std::iota(Mask.begin(), Mask.end(), VF.getKnownMinValue() - 1);
+  Type *VectorTy =
+      ToVectorTy(Ctx.Types.inferScalarType(this->getVPSingleValue()), VF);
+
+  TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
+  return Ctx.TTI.getShuffleCost(TargetTransformInfo::SK_Splice,
+                                cast<VectorType>(VectorTy), Mask, CostKind,
+                                VF.getKnownMinValue() - 1);
+}
+
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void VPFirstOrderRecurrencePHIRecipe::print(raw_ostream &O, const Twine &Indent,
                                             VPSlotTracker &SlotTracker) const {