VPlan: thread plan to VPBuilder (NFC)

Construct VPBuilder with a VPlan object, and change VPRecipeBuilder to
own the VPBuilder, in preparation to use the VPlan object in VPBuilder
to implement constant-folding. The VPlan reference in VPBuilder is
unused for now.

Author: artagnon

llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h

@@ -25,7 +25,6 @@
 #define LLVM_TRANSFORMS_VECTORIZE_LOOPVECTORIZATIONPLANNER_H
 
 #include "VPlan.h"
-#include "llvm/ADT/SmallSet.h"
 #include "llvm/Support/InstructionCost.h"
 
 namespace llvm {
@@ -46,6 +45,7 @@ struct VFRange;
 class VPBuilder {
   VPBasicBlock *BB = nullptr;
   VPBasicBlock::iterator InsertPt = VPBasicBlock::iterator();
+  [[maybe_unused]] VPlan &Plan;
 
   /// Insert \p VPI in BB at InsertPt if BB is set.
   template <typename T> T *tryInsertInstruction(T *R) {
@@ -67,10 +67,15 @@ class VPBuilder {
   }
 
 public:
-  VPBuilder() = default;
-  VPBuilder(VPBasicBlock *InsertBB) { setInsertPoint(InsertBB); }
-  VPBuilder(VPRecipeBase *InsertPt) { setInsertPoint(InsertPt); }
-  VPBuilder(VPBasicBlock *TheBB, VPBasicBlock::iterator IP) {
+  VPBuilder(VPlan &Plan) : Plan(Plan) {}
+  VPBuilder(VPlan &Plan, VPBasicBlock *InsertBB) : Plan(Plan) {
+    setInsertPoint(InsertBB);
+  }
+  VPBuilder(VPlan &Plan, VPRecipeBase *InsertPt) : Plan(Plan) {
+    setInsertPoint(InsertPt);
+  }
+  VPBuilder(VPlan &Plan, VPBasicBlock *TheBB, VPBasicBlock::iterator IP)
+      : Plan(Plan) {
     setInsertPoint(TheBB, IP);
   }
 
@@ -84,13 +89,6 @@ class VPBuilder {
   VPBasicBlock *getInsertBlock() const { return BB; }
   VPBasicBlock::iterator getInsertPoint() const { return InsertPt; }
 
-  /// Create a VPBuilder to insert after \p R.
-  static VPBuilder getToInsertAfter(VPRecipeBase *R) {
-    VPBuilder B;
-    B.setInsertPoint(R->getParent(), std::next(R->getIterator()));
-    return B;
-  }
-
   /// InsertPoint - A saved insertion point.
   class VPInsertPoint {
     VPBasicBlock *Block = nullptr;
@@ -394,9 +392,6 @@ class LoopVectorizationPlanner {
   /// Profitable vector factors.
   SmallVector<VectorizationFactor, 8> ProfitableVFs;
 
-  /// A builder used to construct the current plan.
-  VPBuilder Builder;
-
   /// Computes the cost of \p Plan for vectorization factor \p VF.
   ///
   /// The current implementation requires access to the

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -8986,7 +8986,7 @@ static void addCanonicalIVRecipes(VPlan &Plan, Type *IdxTy, bool HasNUW,
   VPBasicBlock *Header = TopRegion->getEntryBasicBlock();
   Header->insert(CanonicalIVPHI, Header->begin());
 
-  VPBuilder Builder(TopRegion->getExitingBasicBlock());
+  VPBuilder Builder(Plan, TopRegion->getExitingBasicBlock());
   // Add a VPInstruction to increment the scalar canonical IV by VF * UF.
   auto *CanonicalIVIncrement = Builder.createOverflowingOp(
       Instruction::Add, {CanonicalIVPHI, &Plan.getVFxUF()}, {HasNUW, false}, DL,
@@ -9046,9 +9046,9 @@ static void addScalarResumePhis(VPRecipeBuilder &Builder, VPlan &Plan,
   auto *MiddleVPBB = cast<VPBasicBlock>(ScalarPH->getSinglePredecessor());
   VPRegionBlock *VectorRegion = Plan.getVectorLoopRegion();
   VPBuilder VectorPHBuilder(
-      cast<VPBasicBlock>(VectorRegion->getSinglePredecessor()));
-  VPBuilder MiddleBuilder(MiddleVPBB, MiddleVPBB->getFirstNonPhi());
-  VPBuilder ScalarPHBuilder(ScalarPH);
+      Plan, cast<VPBasicBlock>(VectorRegion->getSinglePredecessor()));
+  VPBuilder MiddleBuilder(Plan, MiddleVPBB, MiddleVPBB->getFirstNonPhi());
+  VPBuilder ScalarPHBuilder(Plan, ScalarPH);
   VPValue *OneVPV = Plan.getOrAddLiveIn(
       ConstantInt::get(Plan.getCanonicalIV()->getScalarType(), 1));
   for (VPRecipeBase &ScalarPhiR : *Plan.getScalarHeader()) {
@@ -9140,7 +9140,7 @@ addUsersInExitBlocks(VPlan &Plan,
     return;
 
   auto *MiddleVPBB = Plan.getMiddleBlock();
-  VPBuilder B(MiddleVPBB, MiddleVPBB->getFirstNonPhi());
+  VPBuilder B(Plan, MiddleVPBB, MiddleVPBB->getFirstNonPhi());
 
   // Introduce extract for exiting values and update the VPIRInstructions
   // modeling the corresponding LCSSA phis.
@@ -9162,8 +9162,8 @@ static void addExitUsersForFirstOrderRecurrences(
   VPRegionBlock *VectorRegion = Plan.getVectorLoopRegion();
   auto *ScalarPHVPBB = Plan.getScalarPreheader();
   auto *MiddleVPBB = Plan.getMiddleBlock();
-  VPBuilder ScalarPHBuilder(ScalarPHVPBB);
-  VPBuilder MiddleBuilder(MiddleVPBB, MiddleVPBB->getFirstNonPhi());
+  VPBuilder ScalarPHBuilder(Plan, ScalarPHVPBB);
+  VPBuilder MiddleBuilder(Plan, MiddleVPBB, MiddleVPBB->getFirstNonPhi());
   VPValue *TwoVPV = Plan.getOrAddLiveIn(
       ConstantInt::get(Plan.getCanonicalIV()->getScalarType(), 2));
 
@@ -9300,8 +9300,7 @@ LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(VFRange &Range) {
   bool HasNUW = !IVUpdateMayOverflow || Style == TailFoldingStyle::None;
   addCanonicalIVRecipes(*Plan, Legal->getWidestInductionType(), HasNUW, DL);
 
-  VPRecipeBuilder RecipeBuilder(*Plan, OrigLoop, TLI, &TTI, Legal, CM, PSE,
-                                Builder);
+  VPRecipeBuilder RecipeBuilder(*Plan, OrigLoop, TLI, &TTI, Legal, CM, PSE);
 
   // ---------------------------------------------------------------------------
   // Pre-construction: record ingredients whose recipes we'll need to further
@@ -9357,7 +9356,7 @@ LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(VFRange &Range) {
     // ingredients and fill a new VPBasicBlock.
     if (VPBB != HeaderVPBB)
       VPBB->setName(BB->getName());
-    Builder.setInsertPoint(VPBB);
+    RecipeBuilder.setInsertPoint(VPBB);
 
     if (VPBB == HeaderVPBB)
       RecipeBuilder.createHeaderMask();
@@ -9521,7 +9520,7 @@ LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(VFRange &Range) {
   // Sink users of fixed-order recurrence past the recipe defining the previous
   // value and introduce FirstOrderRecurrenceSplice VPInstructions.
   if (!VPlanTransforms::runPass(VPlanTransforms::adjustFixedOrderRecurrences,
-                                *Plan, Builder))
+                                *Plan, RecipeBuilder.getIRBuilder()))
     return nullptr;
 
   if (useActiveLaneMask(Style)) {
@@ -9571,8 +9570,7 @@ VPlanPtr LoopVectorizationPlanner::buildVPlan(VFRange &Range) {
 
   // Collect mapping of IR header phis to header phi recipes, to be used in
   // addScalarResumePhis.
-  VPRecipeBuilder RecipeBuilder(*Plan, OrigLoop, TLI, &TTI, Legal, CM, PSE,
-                                Builder);
+  VPRecipeBuilder RecipeBuilder(*Plan, OrigLoop, TLI, &TTI, Legal, CM, PSE);
   for (auto &R : Plan->getVectorLoopRegion()->getEntryBasicBlock()->phis()) {
     if (isa<VPCanonicalIVPHIRecipe>(&R))
       continue;
@@ -9737,6 +9735,7 @@ void LoopVectorizationPlanner::adjustRecipesForReductions(
     }
   }
   VPBasicBlock *LatchVPBB = VectorLoopRegion->getExitingBasicBlock();
+  VPBuilder Builder(*Plan);
   Builder.setInsertPoint(&*LatchVPBB->begin());
   VPBasicBlock::iterator IP = MiddleVPBB->getFirstNonPhi();
   for (VPRecipeBase &R :
@@ -10248,7 +10247,7 @@ static void preparePlanForMainVectorLoop(VPlan &MainPlan, VPlan &EpiPlan) {
                              m_Specific(VectorTC), m_SpecificInt(0)));
       }))
     return;
-  VPBuilder ScalarPHBuilder(MainScalarPH, MainScalarPH->begin());
+  VPBuilder ScalarPHBuilder(MainPlan, MainScalarPH, MainScalarPH->begin());
   ScalarPHBuilder.createNaryOp(
       VPInstruction::ResumePhi,
       {VectorTC, MainPlan.getCanonicalIV()->getStartValue()}, {},

llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h

@@ -12,7 +12,6 @@
 #include "LoopVectorizationPlanner.h"
 #include "VPlan.h"
 #include "llvm/ADT/DenseMap.h"
-#include "llvm/ADT/PointerUnion.h"
 #include "llvm/Analysis/ScalarEvolutionExpressions.h"
 #include "llvm/IR/IRBuilder.h"
 
@@ -66,7 +65,7 @@ class VPRecipeBuilder {
 
   PredicatedScalarEvolution &PSE;
 
-  VPBuilder &Builder;
+  VPBuilder Builder;
 
   /// When we if-convert we need to create edge masks. We have to cache values
   /// so that we don't end up with exponential recursion/IR. Note that
@@ -155,9 +154,13 @@ class VPRecipeBuilder {
                   const TargetTransformInfo *TTI,
                   LoopVectorizationLegality *Legal,
                   LoopVectorizationCostModel &CM,
-                  PredicatedScalarEvolution &PSE, VPBuilder &Builder)
+                  PredicatedScalarEvolution &PSE)
       : Plan(Plan), OrigLoop(OrigLoop), TLI(TLI), TTI(TTI), Legal(Legal),
-        CM(CM), PSE(PSE), Builder(Builder) {}
+        CM(CM), PSE(PSE), Builder(Plan) {}
+
+  void setInsertPoint(VPBasicBlock *VPBB) { Builder.setInsertPoint(VPBB); }
+
+  VPBuilder &getIRBuilder() { return Builder; }
 
   std::optional<unsigned> getScalingForReduction(const Instruction *ExitInst) {
     auto It = ScaledReductionMap.find(ExitInst);

llvm/lib/Transforms/Vectorize/VPlan.cpp

@@ -922,7 +922,7 @@ VPlanPtr VPlan::createInitialVPlan(Type *InductionTy,
   // of the corresponding compare because they may have ended up with
   // different line numbers and we want to avoid awkward line stepping while
   // debugging. Eg. if the compare has got a line number inside the loop.
-  VPBuilder Builder(MiddleVPBB);
+  VPBuilder Builder(*Plan, MiddleVPBB);
   VPValue *Cmp =
       TailFolded
           ? Plan->getOrAddLiveIn(ConstantInt::getTrue(

llvm/lib/Transforms/Vectorize/VPlanHCFGBuilder.cpp

@@ -72,7 +72,7 @@ class PlainCFGBuilder {
 
 public:
   PlainCFGBuilder(Loop *Lp, LoopInfo *LI, VPlan &P)
-      : TheLoop(Lp), LI(LI), Plan(P) {}
+      : TheLoop(Lp), LI(LI), Plan(P), VPIRBuilder(Plan) {}
 
   /// Build plain CFG for TheLoop  and connects it to Plan's entry.
   void buildPlainCFG();

llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

@@ -592,7 +592,7 @@ static void legalizeAndOptimizeInductions(VPlan &Plan) {
   using namespace llvm::VPlanPatternMatch;
   VPBasicBlock *HeaderVPBB = Plan.getVectorLoopRegion()->getEntryBasicBlock();
   bool HasOnlyVectorVFs = !Plan.hasScalarVFOnly();
-  VPBuilder Builder(HeaderVPBB, HeaderVPBB->getFirstNonPhi());
+  VPBuilder Builder(Plan, HeaderVPBB, HeaderVPBB->getFirstNonPhi());
   for (VPRecipeBase &Phi : HeaderVPBB->phis()) {
     auto *PhiR = dyn_cast<VPWidenInductionRecipe>(&Phi);
     if (!PhiR)
@@ -745,7 +745,7 @@ void VPlanTransforms::optimizeInductionExitUsers(
          "predecessor must be the middle block");
 
   VPTypeAnalysis TypeInfo(Plan.getCanonicalIV()->getScalarType());
-  VPBuilder B(Plan.getMiddleBlock()->getTerminator());
+  VPBuilder B(Plan, Plan.getMiddleBlock()->getTerminator());
   for (VPRecipeBase &R : *ExitVPBB) {
     auto *ExitIRI = cast<VPIRInstruction>(&R);
     if (!isa<PHINode>(ExitIRI->getInstruction()))
@@ -1503,7 +1503,7 @@ static VPActiveLaneMaskPHIRecipe *addVPLaneMaskPhiAndUpdateExitBranch(
   // we have to take unrolling into account. Each part needs to start at
   //   Part * VF
   auto *VecPreheader = Plan.getVectorPreheader();
-  VPBuilder Builder(VecPreheader);
+  VPBuilder Builder(Plan, VecPreheader);
 
   // Create the ActiveLaneMask instruction using the correct start values.
   VPValue *TC = Plan.getTripCount();
@@ -1622,7 +1622,8 @@ void VPlanTransforms::addActiveLaneMask(
     LaneMask = addVPLaneMaskPhiAndUpdateExitBranch(
         Plan, DataAndControlFlowWithoutRuntimeCheck);
   } else {
-    VPBuilder B = VPBuilder::getToInsertAfter(WideCanonicalIV);
+    VPBuilder B(Plan, WideCanonicalIV->getParent(),
+                std::next(WideCanonicalIV->getIterator()));
     LaneMask = B.createNaryOp(VPInstruction::ActiveLaneMask,
                               {WideCanonicalIV, Plan.getTripCount()}, nullptr,
                               "active.lane.mask");
@@ -1826,7 +1827,7 @@ bool VPlanTransforms::tryAddExplicitVectorLength(
   // Create the ExplicitVectorLengthPhi recipe in the main loop.
   auto *EVLPhi = new VPEVLBasedIVPHIRecipe(StartV, DebugLoc());
   EVLPhi->insertAfter(CanonicalIVPHI);
-  VPBuilder Builder(Header, Header->getFirstNonPhi());
+  VPBuilder Builder(Plan, Header, Header->getFirstNonPhi());
   // Compute original TC - IV as the AVL (application vector length).
   VPValue *AVL = Builder.createNaryOp(
       Instruction::Sub, {Plan.getTripCount(), EVLPhi}, DebugLoc(), "avl");
@@ -1907,7 +1908,7 @@ void VPlanTransforms::dropPoisonGeneratingRecipes(
         // where the operands are disjoint or poison otherwise.
         if (match(RecWithFlags, m_BinaryOr(m_VPValue(A), m_VPValue(B))) &&
             RecWithFlags->isDisjoint()) {
-          VPBuilder Builder(RecWithFlags);
+          VPBuilder Builder(Plan, RecWithFlags);
           VPInstruction *New = Builder.createOverflowingOp(
               Instruction::Add, {A, B}, {false, false},
               RecWithFlags->getDebugLoc());
@@ -2021,7 +2022,7 @@ void VPlanTransforms::createInterleaveGroups(
                    /*IsSigned=*/true);
       VPValue *OffsetVPV = Plan.getOrAddLiveIn(
           ConstantInt::get(IRInsertPos->getParent()->getContext(), -Offset));
-      VPBuilder B(InsertPos);
+      VPBuilder B(Plan, InsertPos);
       Addr = InBounds ? B.createInBoundsPtrAdd(InsertPos->getAddr(), OffsetVPV)
                       : B.createPtrAdd(InsertPos->getAddr(), OffsetVPV);
     }
@@ -2067,7 +2068,7 @@ void VPlanTransforms::handleUncountableEarlyExit(
     BasicBlock *UncountableExitingBlock, VPRecipeBuilder &RecipeBuilder) {
   VPRegionBlock *LoopRegion = Plan.getVectorLoopRegion();
   auto *LatchVPBB = cast<VPBasicBlock>(LoopRegion->getExiting());
-  VPBuilder Builder(LatchVPBB->getTerminator());
+  VPBuilder Builder(Plan, LatchVPBB->getTerminator());
   auto *MiddleVPBB = Plan.getMiddleBlock();
   VPValue *IsEarlyExitTaken = nullptr;
 
@@ -2108,8 +2109,8 @@ void VPlanTransforms::handleUncountableEarlyExit(
   VPBlockUtils::connectBlocks(VectorEarlyExitVPBB, VPEarlyExitBlock);
 
   // Update the exit phis in the early exit block.
-  VPBuilder MiddleBuilder(NewMiddle);
-  VPBuilder EarlyExitB(VectorEarlyExitVPBB);
+  VPBuilder MiddleBuilder(Plan, NewMiddle);
+  VPBuilder EarlyExitB(Plan, VectorEarlyExitVPBB);
   for (VPRecipeBase &R : *VPEarlyExitBlock) {
     auto *ExitIRI = cast<VPIRInstruction>(&R);
     auto *ExitPhi = dyn_cast<PHINode>(&ExitIRI->getInstruction());

llvm/lib/Transforms/Vectorize/VPlanUnroll.cpp

@@ -156,7 +156,7 @@ void UnrollState::unrollWidenInductionByUF(
     FMFs = ID.getInductionBinOp()->getFastMathFlags();
 
   VPValue *VectorStep = &Plan.getVF();
-  VPBuilder Builder(PH);
+  VPBuilder Builder(Plan, PH);
   if (TypeInfo.inferScalarType(VectorStep) != IVTy) {
     Instruction::CastOps CastOp =
         IVTy->isFloatingPointTy() ? Instruction::UIToFP : Instruction::Trunc;