//===- VPlan.cpp - Vectorizer Plan ----------------------------------------===// // // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. // See https://llvm.org/LICENSE.txt for license information. // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception // //===----------------------------------------------------------------------===// /// /// \file /// This is the LLVM vectorization plan. It represents a candidate for /// vectorization, allowing to plan and optimize how to vectorize a given loop /// before generating LLVM-IR. /// The vectorizer uses vectorization plans to estimate the costs of potential /// candidates and if profitable to execute the desired plan, generating vector /// LLVM-IR code. /// //===----------------------------------------------------------------------===// #include "VPlan.h" #include "LoopVectorizationPlanner.h" #include "VPlanCFG.h" #include "VPlanDominatorTree.h" #include "VPlanHelpers.h" #include "VPlanPatternMatch.h" #include "VPlanTransforms.h" #include "VPlanUtils.h" #include "llvm/ADT/PostOrderIterator.h" #include "llvm/ADT/STLExtras.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/StringExtras.h" #include "llvm/ADT/Twine.h" #include "llvm/Analysis/DomTreeUpdater.h" #include "llvm/Analysis/LoopInfo.h" #include "llvm/IR/BasicBlock.h" #include "llvm/IR/CFG.h" #include "llvm/IR/IRBuilder.h" #include "llvm/IR/Instruction.h" #include "llvm/IR/Instructions.h" #include "llvm/IR/Type.h" #include "llvm/IR/Value.h" #include "llvm/Support/Casting.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/GraphWriter.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Transforms/Utils/BasicBlockUtils.h" #include "llvm/Transforms/Utils/LoopVersioning.h" #include #include using namespace llvm; using namespace llvm::VPlanPatternMatch; namespace llvm { extern cl::opt EnableVPlanNativePath; } extern cl::opt ForceTargetInstructionCost; static cl::opt PrintVPlansInDotFormat( "vplan-print-in-dot-format", cl::Hidden, cl::desc("Use dot format instead of plain text when dumping VPlans")); #define DEBUG_TYPE "loop-vectorize" #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) raw_ostream &llvm::operator<<(raw_ostream &OS, const VPRecipeBase &R) { const VPBasicBlock *Parent = R.getParent(); VPSlotTracker SlotTracker(Parent ? Parent->getPlan() : nullptr); R.print(OS, "", SlotTracker); return OS; } #endif Value *VPLane::getAsRuntimeExpr(IRBuilderBase &Builder, const ElementCount &VF) const { switch (LaneKind) { case VPLane::Kind::ScalableLast: // Lane = RuntimeVF - VF.getKnownMinValue() + Lane return Builder.CreateSub(getRuntimeVF(Builder, Builder.getInt32Ty(), VF), Builder.getInt32(VF.getKnownMinValue() - Lane)); case VPLane::Kind::First: return Builder.getInt32(Lane); } llvm_unreachable("Unknown lane kind"); } VPValue::VPValue(const unsigned char SC, Value *UV, VPDef *Def) : SubclassID(SC), UnderlyingVal(UV), Def(Def) { if (Def) Def->addDefinedValue(this); } VPValue::~VPValue() { assert(Users.empty() && "trying to delete a VPValue with remaining users"); if (Def) Def->removeDefinedValue(this); } #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) void VPValue::print(raw_ostream &OS, VPSlotTracker &SlotTracker) const { if (const VPRecipeBase *R = dyn_cast_or_null(Def)) R->print(OS, "", SlotTracker); else printAsOperand(OS, SlotTracker); } void VPValue::dump() const { const VPRecipeBase *Instr = dyn_cast_or_null(this->Def); VPSlotTracker SlotTracker( (Instr && Instr->getParent()) ? Instr->getParent()->getPlan() : nullptr); print(dbgs(), SlotTracker); dbgs() << "\n"; } void VPDef::dump() const { const VPRecipeBase *Instr = dyn_cast_or_null(this); VPSlotTracker SlotTracker( (Instr && Instr->getParent()) ? Instr->getParent()->getPlan() : nullptr); print(dbgs(), "", SlotTracker); dbgs() << "\n"; } #endif VPRecipeBase *VPValue::getDefiningRecipe() { return cast_or_null(Def); } const VPRecipeBase *VPValue::getDefiningRecipe() const { return cast_or_null(Def); } // Get the top-most entry block of \p Start. This is the entry block of the // containing VPlan. This function is templated to support both const and non-const blocks template static T *getPlanEntry(T *Start) { T *Next = Start; T *Current = Start; while ((Next = Next->getParent())) Current = Next; SmallSetVector WorkList; WorkList.insert(Current); for (unsigned i = 0; i < WorkList.size(); i++) { T *Current = WorkList[i]; if (Current->getNumPredecessors() == 0) return Current; auto &Predecessors = Current->getPredecessors(); WorkList.insert_range(Predecessors); } llvm_unreachable("VPlan without any entry node without predecessors"); } VPlan *VPBlockBase::getPlan() { return getPlanEntry(this)->Plan; } const VPlan *VPBlockBase::getPlan() const { return getPlanEntry(this)->Plan; } /// \return the VPBasicBlock that is the entry of Block, possibly indirectly. const VPBasicBlock *VPBlockBase::getEntryBasicBlock() const { const VPBlockBase *Block = this; while (const VPRegionBlock *Region = dyn_cast(Block)) Block = Region->getEntry(); return cast(Block); } VPBasicBlock *VPBlockBase::getEntryBasicBlock() { VPBlockBase *Block = this; while (VPRegionBlock *Region = dyn_cast(Block)) Block = Region->getEntry(); return cast(Block); } void VPBlockBase::setPlan(VPlan *ParentPlan) { assert(ParentPlan->getEntry() == this && "Can only set plan on its entry."); Plan = ParentPlan; } /// \return the VPBasicBlock that is the exit of Block, possibly indirectly. const VPBasicBlock *VPBlockBase::getExitingBasicBlock() const { const VPBlockBase *Block = this; while (const VPRegionBlock *Region = dyn_cast(Block)) Block = Region->getExiting(); return cast(Block); } VPBasicBlock *VPBlockBase::getExitingBasicBlock() { VPBlockBase *Block = this; while (VPRegionBlock *Region = dyn_cast(Block)) Block = Region->getExiting(); return cast(Block); } VPBlockBase *VPBlockBase::getEnclosingBlockWithSuccessors() { if (!Successors.empty() || !Parent) return this; assert(Parent->getExiting() == this && "Block w/o successors not the exiting block of its parent."); return Parent->getEnclosingBlockWithSuccessors(); } VPBlockBase *VPBlockBase::getEnclosingBlockWithPredecessors() { if (!Predecessors.empty() || !Parent) return this; assert(Parent->getEntry() == this && "Block w/o predecessors not the entry of its parent."); return Parent->getEnclosingBlockWithPredecessors(); } bool VPBlockUtils::isHeader(const VPBlockBase *VPB, const VPDominatorTree &VPDT) { auto *VPBB = dyn_cast(VPB); if (!VPBB) return false; // If VPBB is in a region R, VPBB is a loop header if R is a loop region with // VPBB as its entry, i.e., free of predecessors. if (auto *R = VPBB->getParent()) return !R->isReplicator() && VPBB->getNumPredecessors() == 0; // A header dominates its second predecessor (the latch), with the other // predecessor being the preheader return VPB->getPredecessors().size() == 2 && VPDT.dominates(VPB, VPB->getPredecessors()[1]); } bool VPBlockUtils::isLatch(const VPBlockBase *VPB, const VPDominatorTree &VPDT) { // A latch has a header as its second successor, with its other successor // leaving the loop. A preheader OTOH has a header as its first (and only) // successor. return VPB->getNumSuccessors() == 2 && VPBlockUtils::isHeader(VPB->getSuccessors()[1], VPDT); } VPBasicBlock::iterator VPBasicBlock::getFirstNonPhi() { iterator It = begin(); while (It != end() && It->isPhi()) It++; return It; } VPTransformState::VPTransformState(const TargetTransformInfo *TTI, ElementCount VF, LoopInfo *LI, DominatorTree *DT, AssumptionCache *AC, IRBuilderBase &Builder, VPlan *Plan, Loop *CurrentParentLoop, Type *CanonicalIVTy) : TTI(TTI), VF(VF), CFG(DT), LI(LI), AC(AC), Builder(Builder), Plan(Plan), CurrentParentLoop(CurrentParentLoop), TypeAnalysis(CanonicalIVTy), VPDT(*Plan) {} Value *VPTransformState::get(const VPValue *Def, const VPLane &Lane) { if (Def->isLiveIn()) return Def->getLiveInIRValue(); if (hasScalarValue(Def, Lane)) return Data.VPV2Scalars[Def][Lane.mapToCacheIndex(VF)]; if (!Lane.isFirstLane() && vputils::isSingleScalar(Def) && hasScalarValue(Def, VPLane::getFirstLane())) { return Data.VPV2Scalars[Def][0]; } // Look through BuildVector to avoid redundant extracts. // TODO: Remove once replicate regions are unrolled explicitly. if (Lane.getKind() == VPLane::Kind::First && match(Def, m_BuildVector())) { auto *BuildVector = cast(Def); return get(BuildVector->getOperand(Lane.getKnownLane()), true); } assert(hasVectorValue(Def)); auto *VecPart = Data.VPV2Vector[Def]; if (!VecPart->getType()->isVectorTy()) { assert(Lane.isFirstLane() && "cannot get lane > 0 for scalar"); return VecPart; } // TODO: Cache created scalar values. Value *LaneV = Lane.getAsRuntimeExpr(Builder, VF); auto *Extract = Builder.CreateExtractElement(VecPart, LaneV); // set(Def, Extract, Instance); return Extract; } Value *VPTransformState::get(const VPValue *Def, bool NeedsScalar) { if (NeedsScalar) { assert((VF.isScalar() || Def->isLiveIn() || hasVectorValue(Def) || !vputils::onlyFirstLaneUsed(Def) || (hasScalarValue(Def, VPLane(0)) && Data.VPV2Scalars[Def].size() == 1)) && "Trying to access a single scalar per part but has multiple scalars " "per part."); return get(Def, VPLane(0)); } // If Values have been set for this Def return the one relevant for \p Part. if (hasVectorValue(Def)) return Data.VPV2Vector[Def]; auto GetBroadcastInstrs = [this, Def](Value *V) { bool SafeToHoist = !Def->hasDefiningRecipe() || VPDT.properlyDominates(Def->getDefiningRecipe()->getParent(), Plan->getVectorPreheader()); if (VF.isScalar()) return V; // Place the code for broadcasting invariant variables in the new preheader. IRBuilder<>::InsertPointGuard Guard(Builder); if (SafeToHoist) { BasicBlock *LoopVectorPreHeader = CFG.VPBB2IRBB[Plan->getVectorPreheader()]; if (LoopVectorPreHeader) Builder.SetInsertPoint(LoopVectorPreHeader->getTerminator()); } // Place the code for broadcasting invariant variables in the new preheader. // Broadcast the scalar into all locations in the vector. Value *Shuf = Builder.CreateVectorSplat(VF, V, "broadcast"); return Shuf; }; if (!hasScalarValue(Def, {0})) { assert(Def->isLiveIn() && "expected a live-in"); Value *IRV = Def->getLiveInIRValue(); Value *B = GetBroadcastInstrs(IRV); set(Def, B); return B; } Value *ScalarValue = get(Def, VPLane(0)); // If we aren't vectorizing, we can just copy the scalar map values over // to the vector map. if (VF.isScalar()) { set(Def, ScalarValue); return ScalarValue; } bool IsSingleScalar = vputils::isSingleScalar(Def); VPLane LastLane(IsSingleScalar ? 0 : VF.getFixedValue() - 1); // Check if there is a scalar value for the selected lane. if (!hasScalarValue(Def, LastLane)) { // At the moment, VPWidenIntOrFpInductionRecipes, VPScalarIVStepsRecipes and // VPExpandSCEVRecipes can also be a single scalar. assert((isa(Def->getDefiningRecipe())) && "unexpected recipe found to be invariant"); IsSingleScalar = true; LastLane = 0; } auto *LastInst = cast(get(Def, LastLane)); // Set the insert point after the last scalarized instruction or after the // last PHI, if LastInst is a PHI. This ensures the insertelement sequence // will directly follow the scalar definitions. auto OldIP = Builder.saveIP(); auto NewIP = isa(LastInst) ? LastInst->getParent()->getFirstNonPHIIt() : std::next(BasicBlock::iterator(LastInst)); Builder.SetInsertPoint(&*NewIP); // However, if we are vectorizing, we need to construct the vector values. // If the value is known to be uniform after vectorization, we can just // broadcast the scalar value corresponding to lane zero. Otherwise, we // construct the vector values using insertelement instructions. Since the // resulting vectors are stored in State, we will only generate the // insertelements once. Value *VectorValue = nullptr; if (IsSingleScalar) { VectorValue = GetBroadcastInstrs(ScalarValue); set(Def, VectorValue); } else { assert(!VF.isScalable() && "VF is assumed to be non scalable."); // Initialize packing with insertelements to start from poison. VectorValue = PoisonValue::get(toVectorizedTy(LastInst->getType(), VF)); for (unsigned Lane = 0; Lane < VF.getFixedValue(); ++Lane) VectorValue = packScalarIntoVectorizedValue(Def, VectorValue, Lane); set(Def, VectorValue); } Builder.restoreIP(OldIP); return VectorValue; } void VPTransformState::setDebugLocFrom(DebugLoc DL) { const DILocation *DIL = DL; // When a FSDiscriminator is enabled, we don't need to add the multiply // factors to the discriminators. if (DIL && Builder.GetInsertBlock() ->getParent() ->shouldEmitDebugInfoForProfiling() && !EnableFSDiscriminator) { // FIXME: For scalable vectors, assume vscale=1. unsigned UF = Plan->getUF(); auto NewDIL = DIL->cloneByMultiplyingDuplicationFactor(UF * VF.getKnownMinValue()); if (NewDIL) Builder.SetCurrentDebugLocation(*NewDIL); else LLVM_DEBUG(dbgs() << "Failed to create new discriminator: " << DIL->getFilename() << " Line: " << DIL->getLine()); } else Builder.SetCurrentDebugLocation(DL); } Value *VPTransformState::packScalarIntoVectorizedValue(const VPValue *Def, Value *WideValue, const VPLane &Lane) { Value *ScalarInst = get(Def, Lane); Value *LaneExpr = Lane.getAsRuntimeExpr(Builder, VF); if (auto *StructTy = dyn_cast(WideValue->getType())) { // We must handle each element of a vectorized struct type. for (unsigned I = 0, E = StructTy->getNumElements(); I != E; I++) { Value *ScalarValue = Builder.CreateExtractValue(ScalarInst, I); Value *VectorValue = Builder.CreateExtractValue(WideValue, I); VectorValue = Builder.CreateInsertElement(VectorValue, ScalarValue, LaneExpr); WideValue = Builder.CreateInsertValue(WideValue, VectorValue, I); } } else { WideValue = Builder.CreateInsertElement(WideValue, ScalarInst, LaneExpr); } return WideValue; } BasicBlock *VPBasicBlock::createEmptyBasicBlock(VPTransformState &State) { auto &CFG = State.CFG; // BB stands for IR BasicBlocks. VPBB stands for VPlan VPBasicBlocks. // Pred stands for Predessor. Prev stands for Previous - last visited/created. BasicBlock *PrevBB = CFG.PrevBB; BasicBlock *NewBB = BasicBlock::Create(PrevBB->getContext(), getName(), PrevBB->getParent(), CFG.ExitBB); LLVM_DEBUG(dbgs() << "LV: created " << NewBB->getName() << '\n'); return NewBB; } void VPBasicBlock::connectToPredecessors(VPTransformState &State) { auto &CFG = State.CFG; BasicBlock *NewBB = CFG.VPBB2IRBB[this]; // Register NewBB in its loop. In innermost loops its the same for all // BB's. Loop *ParentLoop = State.CurrentParentLoop; // If this block has a sole successor that is an exit block or is an exit // block itself then it needs adding to the same parent loop as the exit // block. VPBlockBase *SuccOrExitVPB = getSingleSuccessor(); SuccOrExitVPB = SuccOrExitVPB ? SuccOrExitVPB : this; if (State.Plan->isExitBlock(SuccOrExitVPB)) { ParentLoop = State.LI->getLoopFor( cast(SuccOrExitVPB)->getIRBasicBlock()); } if (ParentLoop && !State.LI->getLoopFor(NewBB)) ParentLoop->addBasicBlockToLoop(NewBB, *State.LI); SmallVector Preds; if (VPBlockUtils::isHeader(this, State.VPDT)) { // There's no block for the latch yet, connect to the preheader only. Preds = {getPredecessors()[0]}; } else { Preds = to_vector(getPredecessors()); } // Hook up the new basic block to its predecessors. for (VPBlockBase *PredVPBlock : Preds) { VPBasicBlock *PredVPBB = PredVPBlock->getExitingBasicBlock(); auto &PredVPSuccessors = PredVPBB->getHierarchicalSuccessors(); assert(CFG.VPBB2IRBB.contains(PredVPBB) && "Predecessor basic-block not found building successor."); BasicBlock *PredBB = CFG.VPBB2IRBB[PredVPBB]; auto *PredBBTerminator = PredBB->getTerminator(); LLVM_DEBUG(dbgs() << "LV: draw edge from " << PredBB->getName() << '\n'); auto *TermBr = dyn_cast(PredBBTerminator); if (isa(PredBBTerminator)) { assert(PredVPSuccessors.size() == 1 && "Predecessor ending w/o branch must have single successor."); DebugLoc DL = PredBBTerminator->getDebugLoc(); PredBBTerminator->eraseFromParent(); auto *Br = BranchInst::Create(NewBB, PredBB); Br->setDebugLoc(DL); } else if (TermBr && !TermBr->isConditional()) { TermBr->setSuccessor(0, NewBB); } else { // Set each forward successor here when it is created, excluding // backedges. A backward successor is set when the branch is created. // Branches to VPIRBasicBlocks must have the same successors in VPlan as // in the original IR, except when the predecessor is the entry block. // This enables including SCEV and memory runtime check blocks in VPlan. // TODO: Remove exception by modeling the terminator of entry block using // BranchOnCond. unsigned idx = PredVPSuccessors.front() == this ? 0 : 1; assert((TermBr && (!TermBr->getSuccessor(idx) || (isa(this) && (TermBr->getSuccessor(idx) == NewBB || PredVPBlock == getPlan()->getEntry())))) && "Trying to reset an existing successor block."); TermBr->setSuccessor(idx, NewBB); } CFG.DTU.applyUpdates({{DominatorTree::Insert, PredBB, NewBB}}); } } void VPIRBasicBlock::execute(VPTransformState *State) { assert(getHierarchicalSuccessors().size() <= 2 && "VPIRBasicBlock can have at most two successors at the moment!"); State->Builder.SetInsertPoint(IRBB->getTerminator()); State->CFG.PrevBB = IRBB; State->CFG.VPBB2IRBB[this] = IRBB; executeRecipes(State, IRBB); // Create a branch instruction to terminate IRBB if one was not created yet // and is needed. if (getSingleSuccessor() && isa(IRBB->getTerminator())) { auto *Br = State->Builder.CreateBr(IRBB); Br->setOperand(0, nullptr); IRBB->getTerminator()->eraseFromParent(); } else { assert( (getNumSuccessors() == 0 || isa(IRBB->getTerminator())) && "other blocks must be terminated by a branch"); } connectToPredecessors(*State); } VPIRBasicBlock *VPIRBasicBlock::clone() { auto *NewBlock = getPlan()->createEmptyVPIRBasicBlock(IRBB); for (VPRecipeBase &R : Recipes) NewBlock->appendRecipe(R.clone()); return NewBlock; } void VPBasicBlock::execute(VPTransformState *State) { bool Replica = bool(State->Lane); BasicBlock *NewBB = State->CFG.PrevBB; // Reuse it if possible. if (VPBlockUtils::isHeader(this, State->VPDT)) { // Create and register the new vector loop. Loop *PrevParentLoop = State->CurrentParentLoop; State->CurrentParentLoop = State->LI->AllocateLoop(); // Insert the new loop into the loop nest and register the new basic blocks // before calling any utilities such as SCEV that require valid LoopInfo. if (PrevParentLoop) PrevParentLoop->addChildLoop(State->CurrentParentLoop); else State->LI->addTopLevelLoop(State->CurrentParentLoop); } auto IsReplicateRegion = [](VPBlockBase *BB) { auto *R = dyn_cast_or_null(BB); assert((!R || R->isReplicator()) && "only replicate region blocks should remain"); return R; }; // 1. Create an IR basic block. if ((Replica && this == getParent()->getEntry()) || IsReplicateRegion(getSingleHierarchicalPredecessor())) { // Reuse the previous basic block if the current VPBB is either // * the entry to a replicate region, or // * the exit of a replicate region. State->CFG.VPBB2IRBB[this] = NewBB; } else { NewBB = createEmptyBasicBlock(*State); State->Builder.SetInsertPoint(NewBB); // Temporarily terminate with unreachable until CFG is rewired. UnreachableInst *Terminator = State->Builder.CreateUnreachable(); State->Builder.SetInsertPoint(Terminator); State->CFG.PrevBB = NewBB; State->CFG.VPBB2IRBB[this] = NewBB; connectToPredecessors(*State); } // 2. Fill the IR basic block with IR instructions. executeRecipes(State, NewBB); // If this block is a latch, update CurrentParentLoop. if (VPBlockUtils::isLatch(this, State->VPDT)) State->CurrentParentLoop = State->CurrentParentLoop->getParentLoop(); } VPBasicBlock *VPBasicBlock::clone() { auto *NewBlock = getPlan()->createVPBasicBlock(getName()); for (VPRecipeBase &R : *this) NewBlock->appendRecipe(R.clone()); return NewBlock; } void VPBasicBlock::executeRecipes(VPTransformState *State, BasicBlock *BB) { LLVM_DEBUG(dbgs() << "LV: vectorizing VPBB: " << getName() << " in BB: " << BB->getName() << '\n'); State->CFG.PrevVPBB = this; for (VPRecipeBase &Recipe : Recipes) { State->setDebugLocFrom(Recipe.getDebugLoc()); Recipe.execute(*State); } LLVM_DEBUG(dbgs() << "LV: filled BB: " << *BB); } VPBasicBlock *VPBasicBlock::splitAt(iterator SplitAt) { assert((SplitAt == end() || SplitAt->getParent() == this) && "can only split at a position in the same block"); // Create new empty block after the block to split. auto *SplitBlock = getPlan()->createVPBasicBlock(getName() + ".split"); VPBlockUtils::insertBlockAfter(SplitBlock, this); // Finally, move the recipes starting at SplitAt to new block. for (VPRecipeBase &ToMove : make_early_inc_range(make_range(SplitAt, this->end()))) ToMove.moveBefore(*SplitBlock, SplitBlock->end()); return SplitBlock; } /// Return the enclosing loop region for region \p P. The templated version is /// used to support both const and non-const block arguments. template static T *getEnclosingLoopRegionForRegion(T *P) { if (P && P->isReplicator()) { P = P->getParent(); // Multiple loop regions can be nested, but replicate regions can only be // nested inside a loop region or must be outside any other region. assert((!P || !P->isReplicator()) && "unexpected nested replicate regions"); } return P; } VPRegionBlock *VPBasicBlock::getEnclosingLoopRegion() { return getEnclosingLoopRegionForRegion(getParent()); } const VPRegionBlock *VPBasicBlock::getEnclosingLoopRegion() const { return getEnclosingLoopRegionForRegion(getParent()); } static bool hasConditionalTerminator(const VPBasicBlock *VPBB) { if (VPBB->empty()) { assert( VPBB->getNumSuccessors() < 2 && "block with multiple successors doesn't have a recipe as terminator"); return false; } const VPRecipeBase *R = &VPBB->back(); bool IsSwitch = isa(R) && cast(R)->getOpcode() == Instruction::Switch; bool IsCondBranch = isa(R) || match(R, m_BranchOnCond(m_VPValue())) || match(R, m_BranchOnCount(m_VPValue(), m_VPValue())); (void)IsCondBranch; (void)IsSwitch; if (VPBB->getNumSuccessors() == 2 || (VPBB->isExiting() && !VPBB->getParent()->isReplicator())) { assert((IsCondBranch || IsSwitch) && "block with multiple successors not terminated by " "conditional branch nor switch recipe"); return true; } if (VPBB->getNumSuccessors() > 2) { assert(IsSwitch && "block with more than 2 successors not terminated by " "a switch recipe"); return true; } assert( !IsCondBranch && "block with 0 or 1 successors terminated by conditional branch recipe"); return false; } VPRecipeBase *VPBasicBlock::getTerminator() { if (hasConditionalTerminator(this)) return &back(); return nullptr; } const VPRecipeBase *VPBasicBlock::getTerminator() const { if (hasConditionalTerminator(this)) return &back(); return nullptr; } bool VPBasicBlock::isExiting() const { return getParent() && getParent()->getExitingBasicBlock() == this; } #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) void VPBlockBase::print(raw_ostream &O) const { VPSlotTracker SlotTracker(getPlan()); print(O, "", SlotTracker); } void VPBlockBase::printSuccessors(raw_ostream &O, const Twine &Indent) const { if (getSuccessors().empty()) { O << Indent << "No successors\n"; } else { O << Indent << "Successor(s): "; ListSeparator LS; for (auto *Succ : getSuccessors()) O << LS << Succ->getName(); O << '\n'; } } void VPBasicBlock::print(raw_ostream &O, const Twine &Indent, VPSlotTracker &SlotTracker) const { O << Indent << getName() << ":\n"; auto RecipeIndent = Indent + " "; for (const VPRecipeBase &Recipe : *this) { Recipe.print(O, RecipeIndent, SlotTracker); O << '\n'; } printSuccessors(O, Indent); } #endif static std::pair cloneFrom(VPBlockBase *Entry); // Clone the CFG for all nodes reachable from \p Entry, this includes cloning // the blocks and their recipes. Operands of cloned recipes will NOT be updated. // Remapping of operands must be done separately. Returns a pair with the new // entry and exiting blocks of the cloned region. If \p Entry isn't part of a // region, return nullptr for the exiting block. static std::pair cloneFrom(VPBlockBase *Entry) { DenseMap Old2NewVPBlocks; VPBlockBase *Exiting = nullptr; bool InRegion = Entry->getParent(); // First, clone blocks reachable from Entry. for (VPBlockBase *BB : vp_depth_first_shallow(Entry)) { VPBlockBase *NewBB = BB->clone(); Old2NewVPBlocks[BB] = NewBB; if (InRegion && BB->getNumSuccessors() == 0) { assert(!Exiting && "Multiple exiting blocks?"); Exiting = BB; } } assert((!InRegion || Exiting) && "regions must have a single exiting block"); // Second, update the predecessors & successors of the cloned blocks. for (VPBlockBase *BB : vp_depth_first_shallow(Entry)) { VPBlockBase *NewBB = Old2NewVPBlocks[BB]; SmallVector NewPreds; for (VPBlockBase *Pred : BB->getPredecessors()) { NewPreds.push_back(Old2NewVPBlocks[Pred]); } NewBB->setPredecessors(NewPreds); SmallVector NewSuccs; for (VPBlockBase *Succ : BB->successors()) { NewSuccs.push_back(Old2NewVPBlocks[Succ]); } NewBB->setSuccessors(NewSuccs); } #if !defined(NDEBUG) // Verify that the order of predecessors and successors matches in the cloned // version. for (const auto &[OldBB, NewBB] : zip(vp_depth_first_shallow(Entry), vp_depth_first_shallow(Old2NewVPBlocks[Entry]))) { for (const auto &[OldPred, NewPred] : zip(OldBB->getPredecessors(), NewBB->getPredecessors())) assert(NewPred == Old2NewVPBlocks[OldPred] && "Different predecessors"); for (const auto &[OldSucc, NewSucc] : zip(OldBB->successors(), NewBB->successors())) assert(NewSucc == Old2NewVPBlocks[OldSucc] && "Different successors"); } #endif return std::make_pair(Old2NewVPBlocks[Entry], Exiting ? Old2NewVPBlocks[Exiting] : nullptr); } VPRegionBlock *VPRegionBlock::clone() { const auto &[NewEntry, NewExiting] = cloneFrom(getEntry()); auto *NewRegion = getPlan()->createVPRegionBlock(NewEntry, NewExiting, getName(), isReplicator()); for (VPBlockBase *Block : vp_depth_first_shallow(NewEntry)) Block->setParent(NewRegion); return NewRegion; } void VPRegionBlock::execute(VPTransformState *State) { assert(isReplicator() && "Loop regions should have been lowered to plain CFG"); assert(!State->Lane && "Replicating a Region with non-null instance."); assert(!State->VF.isScalable() && "VF is assumed to be non scalable."); ReversePostOrderTraversal> RPOT( Entry); State->Lane = VPLane(0); for (unsigned Lane = 0, VF = State->VF.getFixedValue(); Lane < VF; ++Lane) { State->Lane = VPLane(Lane, VPLane::Kind::First); // Visit the VPBlocks connected to \p this, starting from it. for (VPBlockBase *Block : RPOT) { LLVM_DEBUG(dbgs() << "LV: VPBlock in RPO " << Block->getName() << '\n'); Block->execute(State); } } // Exit replicating mode. State->Lane.reset(); } InstructionCost VPBasicBlock::cost(ElementCount VF, VPCostContext &Ctx) { InstructionCost Cost = 0; for (VPRecipeBase &R : Recipes) Cost += R.cost(VF, Ctx); return Cost; } const VPBasicBlock *VPBasicBlock::getCFGPredecessor(unsigned Idx) const { const VPBlockBase *Pred = nullptr; if (getNumPredecessors() > 0) { Pred = getPredecessors()[Idx]; } else { auto *Region = getParent(); assert(Region && !Region->isReplicator() && Region->getEntry() == this && "must be in the entry block of a non-replicate region"); assert(Idx < 2 && Region->getNumPredecessors() == 1 && "loop region has a single predecessor (preheader), its entry block " "has 2 incoming blocks"); // Idx == 0 selects the predecessor of the region, Idx == 1 selects the // region itself whose exiting block feeds the phi across the backedge. Pred = Idx == 0 ? Region->getSinglePredecessor() : Region; } return Pred->getExitingBasicBlock(); } InstructionCost VPRegionBlock::cost(ElementCount VF, VPCostContext &Ctx) { if (!isReplicator()) { InstructionCost Cost = 0; for (VPBlockBase *Block : vp_depth_first_shallow(getEntry())) Cost += Block->cost(VF, Ctx); InstructionCost BackedgeCost = ForceTargetInstructionCost.getNumOccurrences() ? InstructionCost(ForceTargetInstructionCost.getNumOccurrences()) : Ctx.TTI.getCFInstrCost(Instruction::Br, Ctx.CostKind); LLVM_DEBUG(dbgs() << "Cost of " << BackedgeCost << " for VF " << VF << ": vector loop backedge\n"); Cost += BackedgeCost; return Cost; } // Compute the cost of a replicate region. Replicating isn't supported for // scalable vectors, return an invalid cost for them. // TODO: Discard scalable VPlans with replicate recipes earlier after // construction. if (VF.isScalable()) return InstructionCost::getInvalid(); // First compute the cost of the conditionally executed recipes, followed by // account for the branching cost, except if the mask is a header mask or // uniform condition. using namespace llvm::VPlanPatternMatch; VPBasicBlock *Then = cast(getEntry()->getSuccessors()[0]); InstructionCost ThenCost = Then->cost(VF, Ctx); // For the scalar case, we may not always execute the original predicated // block, Thus, scale the block's cost by the probability of executing it. if (VF.isScalar()) return ThenCost / getPredBlockCostDivisor(Ctx.CostKind); return ThenCost; } #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) void VPRegionBlock::print(raw_ostream &O, const Twine &Indent, VPSlotTracker &SlotTracker) const { O << Indent << (isReplicator() ? " " : " ") << getName() << ": {"; auto NewIndent = Indent + " "; for (auto *BlockBase : vp_depth_first_shallow(Entry)) { O << '\n'; BlockBase->print(O, NewIndent, SlotTracker); } O << Indent << "}\n"; printSuccessors(O, Indent); } #endif void VPRegionBlock::dissolveToCFGLoop() { auto *Header = cast(getEntry()); if (auto *CanIV = dyn_cast(&Header->front())) { assert(this == getPlan()->getVectorLoopRegion() && "Canonical IV must be in the entry of the top-level loop region"); auto *ScalarR = VPBuilder(CanIV).createScalarPhi( {CanIV->getStartValue(), CanIV->getBackedgeValue()}, CanIV->getDebugLoc(), "index"); CanIV->replaceAllUsesWith(ScalarR); CanIV->eraseFromParent(); } VPBlockBase *Preheader = getSinglePredecessor(); auto *ExitingLatch = cast(getExiting()); VPBlockBase *Middle = getSingleSuccessor(); VPBlockUtils::disconnectBlocks(Preheader, this); VPBlockUtils::disconnectBlocks(this, Middle); for (VPBlockBase *VPB : vp_depth_first_shallow(Entry)) VPB->setParent(getParent()); VPBlockUtils::connectBlocks(Preheader, Header); VPBlockUtils::connectBlocks(ExitingLatch, Middle); VPBlockUtils::connectBlocks(ExitingLatch, Header); } VPlan::VPlan(Loop *L) { setEntry(createVPIRBasicBlock(L->getLoopPreheader())); ScalarHeader = createVPIRBasicBlock(L->getHeader()); SmallVector IRExitBlocks; L->getUniqueExitBlocks(IRExitBlocks); for (BasicBlock *EB : IRExitBlocks) ExitBlocks.push_back(createVPIRBasicBlock(EB)); } VPlan::~VPlan() { VPValue DummyValue; for (auto *VPB : CreatedBlocks) { if (auto *VPBB = dyn_cast(VPB)) { // Replace all operands of recipes and all VPValues defined in VPBB with // DummyValue so the block can be deleted. for (VPRecipeBase &R : *VPBB) { for (auto *Def : R.definedValues()) Def->replaceAllUsesWith(&DummyValue); for (unsigned I = 0, E = R.getNumOperands(); I != E; I++) R.setOperand(I, &DummyValue); } } delete VPB; } for (VPValue *VPV : getLiveIns()) delete VPV; if (BackedgeTakenCount) delete BackedgeTakenCount; } void VPlan::prepareToExecute(Value *TripCountV, Value *VectorTripCountV, VPTransformState &State) { Type *TCTy = TripCountV->getType(); // Check if the backedge taken count is needed, and if so build it. if (BackedgeTakenCount && BackedgeTakenCount->getNumUsers()) { IRBuilder<> Builder(State.CFG.PrevBB->getTerminator()); auto *TCMO = Builder.CreateSub(TripCountV, ConstantInt::get(TCTy, 1), "trip.count.minus.1"); BackedgeTakenCount->setUnderlyingValue(TCMO); } VectorTripCount.setUnderlyingValue(VectorTripCountV); IRBuilder<> Builder(State.CFG.PrevBB->getTerminator()); // FIXME: Model VF * UF computation completely in VPlan. unsigned UF = getUF(); if (VF.getNumUsers()) { Value *RuntimeVF = getRuntimeVF(Builder, TCTy, State.VF); VF.setUnderlyingValue(RuntimeVF); VFxUF.setUnderlyingValue( UF > 1 ? Builder.CreateMul(RuntimeVF, ConstantInt::get(TCTy, UF)) : RuntimeVF); } else { VFxUF.setUnderlyingValue(createStepForVF(Builder, TCTy, State.VF, UF)); } } VPIRBasicBlock *VPlan::getExitBlock(BasicBlock *IRBB) const { auto Iter = find_if(getExitBlocks(), [IRBB](const VPIRBasicBlock *VPIRBB) { return VPIRBB->getIRBasicBlock() == IRBB; }); assert(Iter != getExitBlocks().end() && "no exit block found"); return *Iter; } bool VPlan::isExitBlock(VPBlockBase *VPBB) { return is_contained(ExitBlocks, VPBB); } /// Generate the code inside the preheader and body of the vectorized loop. /// Assumes a single pre-header basic-block was created for this. Introduce /// additional basic-blocks as needed, and fill them all. void VPlan::execute(VPTransformState *State) { // Initialize CFG state. State->CFG.PrevVPBB = nullptr; State->CFG.ExitBB = State->CFG.PrevBB->getSingleSuccessor(); // Update VPDominatorTree since VPBasicBlock may be removed after State was // constructed. State->VPDT.recalculate(*this); // Disconnect VectorPreHeader from ExitBB in both the CFG and DT. BasicBlock *VectorPreHeader = State->CFG.PrevBB; cast(VectorPreHeader->getTerminator())->setSuccessor(0, nullptr); State->CFG.DTU.applyUpdates( {{DominatorTree::Delete, VectorPreHeader, State->CFG.ExitBB}}); LLVM_DEBUG(dbgs() << "Executing best plan with VF=" << State->VF << ", UF=" << getUF() << '\n'); setName("Final VPlan"); LLVM_DEBUG(dump()); // Disconnect scalar preheader and scalar header, as the dominator tree edge // will be updated as part of VPlan execution. This allows keeping the DTU // logic generic during VPlan execution. BasicBlock *ScalarPh = State->CFG.ExitBB; State->CFG.DTU.applyUpdates( {{DominatorTree::Delete, ScalarPh, ScalarPh->getSingleSuccessor()}}); ReversePostOrderTraversal> RPOT( Entry); // Generate code for the VPlan, in parts of the vector skeleton, loop body and // successor blocks including the middle, exit and scalar preheader blocks. for (VPBlockBase *Block : RPOT) Block->execute(State); State->CFG.DTU.flush(); VPBasicBlock *Header = vputils::getFirstLoopHeader(*this, State->VPDT); if (!Header) return; auto *LatchVPBB = cast(Header->getPredecessors()[1]); BasicBlock *VectorLatchBB = State->CFG.VPBB2IRBB[LatchVPBB]; // Fix the latch value of canonical, reduction and first-order recurrences // phis in the vector loop. for (VPRecipeBase &R : Header->phis()) { // Skip phi-like recipes that generate their backedege values themselves. if (isa(&R)) continue; if (auto *WidenPhi = dyn_cast(&R)) { assert(!WidenPhi->onlyScalarsGenerated(State->VF.isScalable()) && "recipe generating only scalars should have been replaced"); auto *GEP = cast(State->get(WidenPhi)); PHINode *Phi = cast(GEP->getPointerOperand()); Phi->setIncomingBlock(1, VectorLatchBB); // Move the last step to the end of the latch block. This ensures // consistent placement of all induction updates. Instruction *Inc = cast(Phi->getIncomingValue(1)); Inc->moveBefore(std::prev(VectorLatchBB->getTerminator()->getIterator())); continue; } auto *PhiR = cast(&R); // VPInstructions currently model scalar Phis only. bool NeedsScalar = isa(PhiR) || (isa(PhiR) && cast(PhiR)->isInLoop()); Value *Phi = State->get(PhiR, NeedsScalar); // VPHeaderPHIRecipe supports getBackedgeValue() but VPInstruction does // not. Value *Val = State->get(PhiR->getOperand(1), NeedsScalar); cast(Phi)->addIncoming(Val, VectorLatchBB); } } InstructionCost VPlan::cost(ElementCount VF, VPCostContext &Ctx) { // For now only return the cost of the vector loop region, ignoring any other // blocks, like the preheader or middle blocks. InstructionCost Cost = getVectorLoopRegion()->cost(VF, Ctx); // If any instructions in the middle block are invalid return invalid. // TODO: Remove once no VPlans with VF == vscale x 1 and first-order recurrences are created. if (!getMiddleBlock()->cost(VF, Ctx).isValid()) return InstructionCost::getInvalid(); return Cost; } VPRegionBlock *VPlan::getVectorLoopRegion() { // TODO: Cache if possible. for (VPBlockBase *B : vp_depth_first_shallow(getEntry())) if (auto *R = dyn_cast(B)) return R->isReplicator() ? nullptr : R; return nullptr; } const VPRegionBlock *VPlan::getVectorLoopRegion() const { for (const VPBlockBase *B : vp_depth_first_shallow(getEntry())) if (auto *R = dyn_cast(B)) return R->isReplicator() ? nullptr : R; return nullptr; } #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) void VPlan::printLiveIns(raw_ostream &O) const { VPSlotTracker SlotTracker(this); if (VF.getNumUsers() > 0) { O << "\nLive-in "; VF.printAsOperand(O, SlotTracker); O << " = VF"; } if (VFxUF.getNumUsers() > 0) { O << "\nLive-in "; VFxUF.printAsOperand(O, SlotTracker); O << " = VF * UF"; } if (VectorTripCount.getNumUsers() > 0) { O << "\nLive-in "; VectorTripCount.printAsOperand(O, SlotTracker); O << " = vector-trip-count"; } if (BackedgeTakenCount && BackedgeTakenCount->getNumUsers()) { O << "\nLive-in "; BackedgeTakenCount->printAsOperand(O, SlotTracker); O << " = backedge-taken count"; } O << "\n"; if (TripCount) { if (TripCount->isLiveIn()) O << "Live-in "; TripCount->printAsOperand(O, SlotTracker); O << " = original trip-count"; O << "\n"; } } LLVM_DUMP_METHOD void VPlan::print(raw_ostream &O) const { VPSlotTracker SlotTracker(this); O << "VPlan '" << getName() << "' {"; printLiveIns(O); ReversePostOrderTraversal> RPOT(getEntry()); for (const VPBlockBase *Block : RPOT) { O << '\n'; Block->print(O, "", SlotTracker); } O << "}\n"; } std::string VPlan::getName() const { std::string Out; raw_string_ostream RSO(Out); RSO << Name << " for "; if (!VFs.empty()) { RSO << "VF={" << VFs[0]; for (ElementCount VF : drop_begin(VFs)) RSO << "," << VF; RSO << "},"; } if (UFs.empty()) { RSO << "UF>=1"; } else { RSO << "UF={" << UFs[0]; for (unsigned UF : drop_begin(UFs)) RSO << "," << UF; RSO << "}"; } return Out; } LLVM_DUMP_METHOD void VPlan::printDOT(raw_ostream &O) const { VPlanPrinter Printer(O, *this); Printer.dump(); } LLVM_DUMP_METHOD void VPlan::dump() const { print(dbgs()); } #endif static void remapOperands(VPBlockBase *Entry, VPBlockBase *NewEntry, DenseMap &Old2NewVPValues) { // Update the operands of all cloned recipes starting at NewEntry. This // traverses all reachable blocks. This is done in two steps, to handle cycles // in PHI recipes. ReversePostOrderTraversal> OldDeepRPOT(Entry); ReversePostOrderTraversal> NewDeepRPOT(NewEntry); // First, collect all mappings from old to new VPValues defined by cloned // recipes. for (const auto &[OldBB, NewBB] : zip(VPBlockUtils::blocksOnly(OldDeepRPOT), VPBlockUtils::blocksOnly(NewDeepRPOT))) { assert(OldBB->getRecipeList().size() == NewBB->getRecipeList().size() && "blocks must have the same number of recipes"); for (const auto &[OldR, NewR] : zip(*OldBB, *NewBB)) { assert(OldR.getNumOperands() == NewR.getNumOperands() && "recipes must have the same number of operands"); assert(OldR.getNumDefinedValues() == NewR.getNumDefinedValues() && "recipes must define the same number of operands"); for (const auto &[OldV, NewV] : zip(OldR.definedValues(), NewR.definedValues())) Old2NewVPValues[OldV] = NewV; } } // Update all operands to use cloned VPValues. for (VPBasicBlock *NewBB : VPBlockUtils::blocksOnly(NewDeepRPOT)) { for (VPRecipeBase &NewR : *NewBB) for (unsigned I = 0, E = NewR.getNumOperands(); I != E; ++I) { VPValue *NewOp = Old2NewVPValues.lookup(NewR.getOperand(I)); NewR.setOperand(I, NewOp); } } } VPlan *VPlan::duplicate() { unsigned NumBlocksBeforeCloning = CreatedBlocks.size(); // Clone blocks. const auto &[NewEntry, __] = cloneFrom(Entry); BasicBlock *ScalarHeaderIRBB = getScalarHeader()->getIRBasicBlock(); VPIRBasicBlock *NewScalarHeader = nullptr; if (getScalarHeader()->getNumPredecessors() == 0) { NewScalarHeader = createVPIRBasicBlock(ScalarHeaderIRBB); } else { NewScalarHeader = cast(*find_if( vp_depth_first_shallow(NewEntry), [ScalarHeaderIRBB](VPBlockBase *VPB) { auto *VPIRBB = dyn_cast(VPB); return VPIRBB && VPIRBB->getIRBasicBlock() == ScalarHeaderIRBB; })); } // Create VPlan, clone live-ins and remap operands in the cloned blocks. auto *NewPlan = new VPlan(cast(NewEntry), NewScalarHeader); DenseMap Old2NewVPValues; for (VPValue *OldLiveIn : getLiveIns()) { Old2NewVPValues[OldLiveIn] = NewPlan->getOrAddLiveIn(OldLiveIn->getLiveInIRValue()); } Old2NewVPValues[&VectorTripCount] = &NewPlan->VectorTripCount; Old2NewVPValues[&VF] = &NewPlan->VF; Old2NewVPValues[&VFxUF] = &NewPlan->VFxUF; if (BackedgeTakenCount) { NewPlan->BackedgeTakenCount = new VPValue(); Old2NewVPValues[BackedgeTakenCount] = NewPlan->BackedgeTakenCount; } if (TripCount && TripCount->isLiveIn()) Old2NewVPValues[TripCount] = NewPlan->getOrAddLiveIn(TripCount->getLiveInIRValue()); // else NewTripCount will be created and inserted into Old2NewVPValues when // TripCount is cloned. In any case NewPlan->TripCount is updated below. remapOperands(Entry, NewEntry, Old2NewVPValues); // Initialize remaining fields of cloned VPlan. NewPlan->VFs = VFs; NewPlan->UFs = UFs; // TODO: Adjust names. NewPlan->Name = Name; if (TripCount) { assert(Old2NewVPValues.contains(TripCount) && "TripCount must have been added to Old2NewVPValues"); NewPlan->TripCount = Old2NewVPValues[TripCount]; } // Transfer all cloned blocks (the second half of all current blocks) from // current to new VPlan. unsigned NumBlocksAfterCloning = CreatedBlocks.size(); for (unsigned I : seq(NumBlocksBeforeCloning, NumBlocksAfterCloning)) NewPlan->CreatedBlocks.push_back(this->CreatedBlocks[I]); CreatedBlocks.truncate(NumBlocksBeforeCloning); // Update ExitBlocks of the new plan. for (VPBlockBase *VPB : NewPlan->CreatedBlocks) { if (VPB->getNumSuccessors() == 0 && isa(VPB) && VPB != NewScalarHeader) NewPlan->ExitBlocks.push_back(cast(VPB)); } return NewPlan; } VPIRBasicBlock *VPlan::createEmptyVPIRBasicBlock(BasicBlock *IRBB) { auto *VPIRBB = new VPIRBasicBlock(IRBB); CreatedBlocks.push_back(VPIRBB); return VPIRBB; } VPIRBasicBlock *VPlan::createVPIRBasicBlock(BasicBlock *IRBB) { auto *VPIRBB = createEmptyVPIRBasicBlock(IRBB); for (Instruction &I : make_range(IRBB->begin(), IRBB->getTerminator()->getIterator())) VPIRBB->appendRecipe(VPIRInstruction::create(I)); return VPIRBB; } #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) Twine VPlanPrinter::getUID(const VPBlockBase *Block) { return (isa(Block) ? "cluster_N" : "N") + Twine(getOrCreateBID(Block)); } Twine VPlanPrinter::getOrCreateName(const VPBlockBase *Block) { const std::string &Name = Block->getName(); if (!Name.empty()) return Name; return "VPB" + Twine(getOrCreateBID(Block)); } void VPlanPrinter::dump() { Depth = 1; bumpIndent(0); OS << "digraph VPlan {\n"; OS << "graph [labelloc=t, fontsize=30; label=\"Vectorization Plan"; if (!Plan.getName().empty()) OS << "\\n" << DOT::EscapeString(Plan.getName()); { // Print live-ins. std::string Str; raw_string_ostream SS(Str); Plan.printLiveIns(SS); SmallVector Lines; StringRef(Str).rtrim('\n').split(Lines, "\n"); for (auto Line : Lines) OS << DOT::EscapeString(Line.str()) << "\\n"; } OS << "\"]\n"; OS << "node [shape=rect, fontname=Courier, fontsize=30]\n"; OS << "edge [fontname=Courier, fontsize=30]\n"; OS << "compound=true\n"; for (const VPBlockBase *Block : vp_depth_first_shallow(Plan.getEntry())) dumpBlock(Block); OS << "}\n"; } void VPlanPrinter::dumpBlock(const VPBlockBase *Block) { if (const VPBasicBlock *BasicBlock = dyn_cast(Block)) dumpBasicBlock(BasicBlock); else if (const VPRegionBlock *Region = dyn_cast(Block)) dumpRegion(Region); else llvm_unreachable("Unsupported kind of VPBlock."); } void VPlanPrinter::drawEdge(const VPBlockBase *From, const VPBlockBase *To, bool Hidden, const Twine &Label) { // Due to "dot" we print an edge between two regions as an edge between the // exiting basic block and the entry basic of the respective regions. const VPBlockBase *Tail = From->getExitingBasicBlock(); const VPBlockBase *Head = To->getEntryBasicBlock(); OS << Indent << getUID(Tail) << " -> " << getUID(Head); OS << " [ label=\"" << Label << '\"'; if (Tail != From) OS << " ltail=" << getUID(From); if (Head != To) OS << " lhead=" << getUID(To); if (Hidden) OS << "; splines=none"; OS << "]\n"; } void VPlanPrinter::dumpEdges(const VPBlockBase *Block) { auto &Successors = Block->getSuccessors(); if (Successors.size() == 1) drawEdge(Block, Successors.front(), false, ""); else if (Successors.size() == 2) { drawEdge(Block, Successors.front(), false, "T"); drawEdge(Block, Successors.back(), false, "F"); } else { unsigned SuccessorNumber = 0; for (auto *Successor : Successors) drawEdge(Block, Successor, false, Twine(SuccessorNumber++)); } } void VPlanPrinter::dumpBasicBlock(const VPBasicBlock *BasicBlock) { // Implement dot-formatted dump by performing plain-text dump into the // temporary storage followed by some post-processing. OS << Indent << getUID(BasicBlock) << " [label =\n"; bumpIndent(1); std::string Str; raw_string_ostream SS(Str); // Use no indentation as we need to wrap the lines into quotes ourselves. BasicBlock->print(SS, "", SlotTracker); // We need to process each line of the output separately, so split // single-string plain-text dump. SmallVector Lines; StringRef(Str).rtrim('\n').split(Lines, "\n"); auto EmitLine = [&](StringRef Line, StringRef Suffix) { OS << Indent << '"' << DOT::EscapeString(Line.str()) << "\\l\"" << Suffix; }; // Don't need the "+" after the last line. for (auto Line : make_range(Lines.begin(), Lines.end() - 1)) EmitLine(Line, " +\n"); EmitLine(Lines.back(), "\n"); bumpIndent(-1); OS << Indent << "]\n"; dumpEdges(BasicBlock); } void VPlanPrinter::dumpRegion(const VPRegionBlock *Region) { OS << Indent << "subgraph " << getUID(Region) << " {\n"; bumpIndent(1); OS << Indent << "fontname=Courier\n" << Indent << "label=\"" << DOT::EscapeString(Region->isReplicator() ? " " : " ") << DOT::EscapeString(Region->getName()) << "\"\n"; // Dump the blocks of the region. assert(Region->getEntry() && "Region contains no inner blocks."); for (const VPBlockBase *Block : vp_depth_first_shallow(Region->getEntry())) dumpBlock(Block); bumpIndent(-1); OS << Indent << "}\n"; dumpEdges(Region); } #endif /// Returns true if there is a vector loop region and \p VPV is defined in a /// loop region. static bool isDefinedInsideLoopRegions(const VPValue *VPV) { const VPRecipeBase *DefR = VPV->getDefiningRecipe(); return DefR && (!DefR->getParent()->getPlan()->getVectorLoopRegion() || DefR->getParent()->getEnclosingLoopRegion()); } bool VPValue::isDefinedOutsideLoopRegions() const { return !isDefinedInsideLoopRegions(this); } void VPValue::replaceAllUsesWith(VPValue *New) { replaceUsesWithIf(New, [](VPUser &, unsigned) { return true; }); } void VPValue::replaceUsesWithIf( VPValue *New, llvm::function_ref ShouldReplace) { // Note that this early exit is required for correctness; the implementation // below relies on the number of users for this VPValue to decrease, which // isn't the case if this == New. if (this == New) return; for (unsigned J = 0; J < getNumUsers();) { VPUser *User = Users[J]; bool RemovedUser = false; for (unsigned I = 0, E = User->getNumOperands(); I < E; ++I) { if (User->getOperand(I) != this || !ShouldReplace(*User, I)) continue; RemovedUser = true; User->setOperand(I, New); } // If a user got removed after updating the current user, the next user to // update will be moved to the current position, so we only need to // increment the index if the number of users did not change. if (!RemovedUser) J++; } } void VPUser::replaceUsesOfWith(VPValue *From, VPValue *To) { for (unsigned Idx = 0; Idx != getNumOperands(); ++Idx) { if (getOperand(Idx) == From) setOperand(Idx, To); } } #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) void VPValue::printAsOperand(raw_ostream &OS, VPSlotTracker &Tracker) const { OS << Tracker.getOrCreateName(this); } void VPUser::printOperands(raw_ostream &O, VPSlotTracker &SlotTracker) const { interleaveComma(operands(), O, [&O, &SlotTracker](VPValue *Op) { Op->printAsOperand(O, SlotTracker); }); } #endif void VPSlotTracker::assignName(const VPValue *V) { assert(!VPValue2Name.contains(V) && "VPValue already has a name!"); auto *UV = V->getUnderlyingValue(); auto *VPI = dyn_cast_or_null(V->getDefiningRecipe()); if (!UV && !(VPI && !VPI->getName().empty())) { VPValue2Name[V] = (Twine("vp<%") + Twine(NextSlot) + ">").str(); NextSlot++; return; } // Use the name of the underlying Value, wrapped in "ir<>", and versioned by // appending ".Number" to the name if there are multiple uses. std::string Name; if (UV) Name = getName(UV); else Name = VPI->getName(); assert(!Name.empty() && "Name cannot be empty."); StringRef Prefix = UV ? "ir<" : "vp<%"; std::string BaseName = (Twine(Prefix) + Name + Twine(">")).str(); // First assign the base name for V. const auto &[A, _] = VPValue2Name.insert({V, BaseName}); // Integer or FP constants with different types will result in he same string // due to stripping types. if (V->isLiveIn() && isa(UV)) return; // If it is already used by C > 0 other VPValues, increase the version counter // C and use it for V. const auto &[C, UseInserted] = BaseName2Version.insert({BaseName, 0}); if (!UseInserted) { C->second++; A->second = (BaseName + Twine(".") + Twine(C->second)).str(); } } void VPSlotTracker::assignNames(const VPlan &Plan) { if (Plan.VF.getNumUsers() > 0) assignName(&Plan.VF); if (Plan.VFxUF.getNumUsers() > 0) assignName(&Plan.VFxUF); assignName(&Plan.VectorTripCount); if (Plan.BackedgeTakenCount) assignName(Plan.BackedgeTakenCount); for (VPValue *LI : Plan.getLiveIns()) assignName(LI); ReversePostOrderTraversal> RPOT(VPBlockDeepTraversalWrapper(Plan.getEntry())); for (const VPBasicBlock *VPBB : VPBlockUtils::blocksOnly(RPOT)) assignNames(VPBB); } void VPSlotTracker::assignNames(const VPBasicBlock *VPBB) { for (const VPRecipeBase &Recipe : *VPBB) for (VPValue *Def : Recipe.definedValues()) assignName(Def); } std::string VPSlotTracker::getName(const Value *V) { std::string Name; raw_string_ostream S(Name); if (V->hasName() || !isa(V)) { V->printAsOperand(S, false); return Name; } if (!MST) { // Lazily create the ModuleSlotTracker when we first hit an unnamed // instruction. auto *I = cast(V); // This check is required to support unit tests with incomplete IR. if (I->getParent()) { MST = std::make_unique(I->getModule()); MST->incorporateFunction(*I->getFunction()); } else { MST = std::make_unique(nullptr); } } V->printAsOperand(S, false, *MST); return Name; } std::string VPSlotTracker::getOrCreateName(const VPValue *V) const { std::string Name = VPValue2Name.lookup(V); if (!Name.empty()) return Name; // If no name was assigned, no VPlan was provided when creating the slot // tracker or it is not reachable from the provided VPlan. This can happen, // e.g. when trying to print a recipe that has not been inserted into a VPlan // in a debugger. // TODO: Update VPSlotTracker constructor to assign names to recipes & // VPValues not associated with a VPlan, instead of constructing names ad-hoc // here. const VPRecipeBase *DefR = V->getDefiningRecipe(); (void)DefR; assert((!DefR || !DefR->getParent() || !DefR->getParent()->getPlan()) && "VPValue defined by a recipe in a VPlan?"); // Use the underlying value's name, if there is one. if (auto *UV = V->getUnderlyingValue()) { std::string Name; raw_string_ostream S(Name); UV->printAsOperand(S, false); return (Twine("ir<") + Name + ">").str(); } return ""; } bool LoopVectorizationPlanner::getDecisionAndClampRange( const std::function &Predicate, VFRange &Range) { assert(!Range.isEmpty() && "Trying to test an empty VF range."); bool PredicateAtRangeStart = Predicate(Range.Start); for (ElementCount TmpVF : VFRange(Range.Start * 2, Range.End)) if (Predicate(TmpVF) != PredicateAtRangeStart) { Range.End = TmpVF; break; } return PredicateAtRangeStart; } /// Build VPlans for the full range of feasible VF's = {\p MinVF, 2 * \p MinVF, /// 4 * \p MinVF, ..., \p MaxVF} by repeatedly building a VPlan for a sub-range /// of VF's starting at a given VF and extending it as much as possible. Each /// vectorization decision can potentially shorten this sub-range during /// buildVPlan(). void LoopVectorizationPlanner::buildVPlans(ElementCount MinVF, ElementCount MaxVF) { auto MaxVFTimes2 = MaxVF * 2; for (ElementCount VF = MinVF; ElementCount::isKnownLT(VF, MaxVFTimes2);) { VFRange SubRange = {VF, MaxVFTimes2}; if (auto Plan = tryToBuildVPlan(SubRange)) { VPlanTransforms::optimize(*Plan); // Update the name of the latch of the top-level vector loop region region // after optimizations which includes block folding. Plan->getVectorLoopRegion()->getExiting()->setName("vector.latch"); VPlans.push_back(std::move(Plan)); } VF = SubRange.End; } } VPlan &LoopVectorizationPlanner::getPlanFor(ElementCount VF) const { assert(count_if(VPlans, [VF](const VPlanPtr &Plan) { return Plan->hasVF(VF); }) == 1 && "Multiple VPlans for VF."); for (const VPlanPtr &Plan : VPlans) { if (Plan->hasVF(VF)) return *Plan.get(); } llvm_unreachable("No plan found!"); } #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) void LoopVectorizationPlanner::printPlans(raw_ostream &O) { if (VPlans.empty()) { O << "LV: No VPlans built.\n"; return; } for (const auto &Plan : VPlans) if (PrintVPlansInDotFormat) Plan->printDOT(O); else Plan->print(O); } #endif TargetTransformInfo::OperandValueInfo VPCostContext::getOperandInfo(VPValue *V) const { if (!V->isLiveIn()) return {}; return TTI::getOperandInfo(V->getLiveInIRValue()); }