[VPlan] Change parent of VPReductionRecipe to VPRecipeWithIRFlags. NFC

This patch change the parent of the VPReductionRecipe from
VPSingleDefRecipe to VPRecipeWithIRFlags and also print/get/control
flags by the VPRecipeWithIRFlags. This will remove the dependency of the
underlying instruction.

This patch also add a new function `setFastMathFlags()` to the
VPRecipeWithIRFlags because the entire reduction chain may contains
multiple instructions. And the underlying instruction may not contains
the corresponding flags for this reduction.

Author: ElvisWang123

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -9797,9 +9797,9 @@ void LoopVectorizationPlanner::adjustRecipesForReductions(
       if (CM.blockNeedsPredicationForAnyReason(BB))
         CondOp = RecipeBuilder.getBlockInMask(BB);
 
-      auto *RedRecipe = new VPReductionRecipe(
-          RdxDesc, CurrentLinkI, PreviousLink, VecOp, CondOp,
-          CM.useOrderedReductions(RdxDesc), CurrentLinkI->getDebugLoc());
+      auto *RedRecipe =
+          new VPReductionRecipe(RdxDesc, CurrentLinkI, PreviousLink, VecOp,
+                                CondOp, CM.useOrderedReductions(RdxDesc));
       // Append the recipe to the end of the VPBasicBlock because we need to
       // ensure that it comes after all of it's inputs, including CondOp.
       // Delete CurrentLink as it will be invalid if its operand is replaced

llvm/lib/Transforms/Vectorize/VPlan.h

@@ -711,6 +711,8 @@ class VPRecipeWithIRFlags : public VPSingleDefRecipe {
            R->getVPDefID() == VPRecipeBase::VPWidenGEPSC ||
            R->getVPDefID() == VPRecipeBase::VPWidenCastSC ||
            R->getVPDefID() == VPRecipeBase::VPWidenIntrinsicSC ||
+           R->getVPDefID() == VPRecipeBase::VPReductionSC ||
+           R->getVPDefID() == VPRecipeBase::VPReductionEVLSC ||
            R->getVPDefID() == VPRecipeBase::VPReplicateSC ||
            R->getVPDefID() == VPRecipeBase::VPReverseVectorPointerSC ||
            R->getVPDefID() == VPRecipeBase::VPVectorPointerSC;
@@ -786,6 +788,12 @@ class VPRecipeWithIRFlags : public VPSingleDefRecipe {
     }
   }
 
+  /// Set fast-math flags for this recipe.
+  void setFastMathFlags(FastMathFlags FMFs) {
+    OpType = OperationType::FPMathOp;
+    this->FMFs = FMFs;
+  }
+
   CmpInst::Predicate getPredicate() const {
     assert(OpType == OperationType::Cmp &&
            "recipe doesn't have a compare predicate");
@@ -2236,7 +2244,7 @@ class VPInterleaveRecipe : public VPRecipeBase {
 /// A recipe to represent inloop reduction operations, performing a reduction on
 /// a vector operand into a scalar value, and adding the result to a chain.
 /// The Operands are {ChainOp, VecOp, [Condition]}.
-class VPReductionRecipe : public VPSingleDefRecipe {
+class VPReductionRecipe : public VPRecipeWithIRFlags {
   /// The recurrence decriptor for the reduction in question.
   const RecurrenceDescriptor &RdxDesc;
   bool IsOrdered;
@@ -2246,29 +2254,32 @@ class VPReductionRecipe : public VPSingleDefRecipe {
 protected:
   VPReductionRecipe(const unsigned char SC, const RecurrenceDescriptor &R,
                     Instruction *I, ArrayRef<VPValue *> Operands,
-                    VPValue *CondOp, bool IsOrdered, DebugLoc DL)
-      : VPSingleDefRecipe(SC, Operands, I, DL), RdxDesc(R),
+                    VPValue *CondOp, bool IsOrdered)
+      : VPRecipeWithIRFlags(SC, Operands, *I), RdxDesc(R),
         IsOrdered(IsOrdered) {
     if (CondOp) {
       IsConditional = true;
       addOperand(CondOp);
     }
+    // The inloop reduction may across multiple scalar instruction and the
+    // underlying instruction may not contains the corresponding flags. Set the
+    // flags explicit from the redurrence descriptor.
+    setFastMathFlags(R.getFastMathFlags());
   }
 
 public:
   VPReductionRecipe(const RecurrenceDescriptor &R, Instruction *I,
                     VPValue *ChainOp, VPValue *VecOp, VPValue *CondOp,
-                    bool IsOrdered, DebugLoc DL = {})
+                    bool IsOrdered)
       : VPReductionRecipe(VPDef::VPReductionSC, R, I,
                           ArrayRef<VPValue *>({ChainOp, VecOp}), CondOp,
-                          IsOrdered, DL) {}
+                          IsOrdered) {}
 
   ~VPReductionRecipe() override = default;
 
   VPReductionRecipe *clone() override {
     return new VPReductionRecipe(RdxDesc, getUnderlyingInstr(), getChainOp(),
-                                 getVecOp(), getCondOp(), IsOrdered,
-                                 getDebugLoc());
+                                 getVecOp(), getCondOp(), IsOrdered);
   }
 
   static inline bool classof(const VPRecipeBase *R) {
@@ -2323,7 +2334,7 @@ class VPReductionEVLRecipe : public VPReductionRecipe {
             VPDef::VPReductionEVLSC, R.getRecurrenceDescriptor(),
             cast_or_null<Instruction>(R.getUnderlyingValue()),
             ArrayRef<VPValue *>({R.getChainOp(), R.getVecOp(), &EVL}), CondOp,
-            R.isOrdered(), R.getDebugLoc()) {}
+            R.isOrdered()) {}
 
   ~VPReductionEVLRecipe() override = default;
 

llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp

@@ -2290,7 +2290,8 @@ void VPReductionRecipe::execute(VPTransformState &State) {
          "In-loop AnyOf reductions aren't currently supported");
   // Propagate the fast-math flags carried by the underlying instruction.
   IRBuilderBase::FastMathFlagGuard FMFGuard(State.Builder);
-  State.Builder.setFastMathFlags(RdxDesc.getFastMathFlags());
+  if (hasFastMathFlags())
+    State.Builder.setFastMathFlags(getFastMathFlags());
   State.setDebugLocFrom(getDebugLoc());
   Value *NewVecOp = State.get(getVecOp());
   if (VPValue *Cond = getCondOp()) {
@@ -2337,7 +2338,8 @@ void VPReductionEVLRecipe::execute(VPTransformState &State) {
   // Propagate the fast-math flags carried by the underlying instruction.
   IRBuilderBase::FastMathFlagGuard FMFGuard(Builder);
   const RecurrenceDescriptor &RdxDesc = getRecurrenceDescriptor();
-  Builder.setFastMathFlags(RdxDesc.getFastMathFlags());
+  if (hasFastMathFlags())
+    Builder.setFastMathFlags(getFastMathFlags());
 
   RecurKind Kind = RdxDesc.getRecurrenceKind();
   Value *Prev = State.get(getChainOp(), /*IsScalar*/ true);
@@ -2374,6 +2376,8 @@ InstructionCost VPReductionRecipe::computeCost(ElementCount VF,
   Type *ElementTy = Ctx.Types.inferScalarType(this);
   auto *VectorTy = cast<VectorType>(toVectorTy(ElementTy, VF));
   unsigned Opcode = RdxDesc.getOpcode();
+  FastMathFlags FMFs =
+      hasFastMathFlags() ? getFastMathFlags() : FastMathFlags();
 
   // TODO: Support any-of and in-loop reductions.
   assert(
@@ -2393,12 +2397,12 @@ InstructionCost VPReductionRecipe::computeCost(ElementCount VF,
       Ctx.TTI.getArithmeticInstrCost(Opcode, ElementTy, Ctx.CostKind);
   if (RecurrenceDescriptor::isMinMaxRecurrenceKind(RdxKind)) {
     Intrinsic::ID Id = getMinMaxReductionIntrinsicOp(RdxKind);
-    return Cost + Ctx.TTI.getMinMaxReductionCost(
-                      Id, VectorTy, RdxDesc.getFastMathFlags(), Ctx.CostKind);
+    return Cost +
+           Ctx.TTI.getMinMaxReductionCost(Id, VectorTy, FMFs, Ctx.CostKind);
   }
 
-  return Cost + Ctx.TTI.getArithmeticReductionCost(
-                    Opcode, VectorTy, RdxDesc.getFastMathFlags(), Ctx.CostKind);
+  return Cost + Ctx.TTI.getArithmeticReductionCost(Opcode, VectorTy, FMFs,
+                                                   Ctx.CostKind);
 }
 
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
@@ -2409,8 +2413,8 @@ void VPReductionRecipe::print(raw_ostream &O, const Twine &Indent,
   O << " = ";
   getChainOp()->printAsOperand(O, SlotTracker);
   O << " +";
-  if (isa<FPMathOperator>(getUnderlyingInstr()))
-    O << getUnderlyingInstr()->getFastMathFlags();
+  if (isa_and_nonnull<FPMathOperator>(getUnderlyingValue()))
+    printFlags(O);
   O << " reduce." << Instruction::getOpcodeName(RdxDesc.getOpcode()) << " (";
   getVecOp()->printAsOperand(O, SlotTracker);
   if (isConditional()) {
@@ -2431,8 +2435,8 @@ void VPReductionEVLRecipe::print(raw_ostream &O, const Twine &Indent,
   O << " = ";
   getChainOp()->printAsOperand(O, SlotTracker);
   O << " +";
-  if (isa<FPMathOperator>(getUnderlyingInstr()))
-    O << getUnderlyingInstr()->getFastMathFlags();
+  if (isa_and_nonnull<FPMathOperator>(getUnderlyingValue()))
+    printFlags(O);
   O << " vp.reduce." << Instruction::getOpcodeName(RdxDesc.getOpcode()) << " (";
   getVecOp()->printAsOperand(O, SlotTracker);
   O << ", ";