diff options
Diffstat (limited to 'contrib/llvm/lib/Transforms/Vectorize')
| -rw-r--r-- | contrib/llvm/lib/Transforms/Vectorize/BBVectorize.cpp | 3251 | ||||
| -rw-r--r-- | contrib/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp | 5823 | ||||
| -rw-r--r-- | contrib/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp | 4222 | ||||
| -rw-r--r-- | contrib/llvm/lib/Transforms/Vectorize/Vectorize.cpp | 48 |
4 files changed, 13344 insertions, 0 deletions
diff --git a/contrib/llvm/lib/Transforms/Vectorize/BBVectorize.cpp b/contrib/llvm/lib/Transforms/Vectorize/BBVectorize.cpp new file mode 100644 index 0000000..8844d57 --- /dev/null +++ b/contrib/llvm/lib/Transforms/Vectorize/BBVectorize.cpp @@ -0,0 +1,3251 @@ +//===- BBVectorize.cpp - A Basic-Block Vectorizer -------------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file implements a basic-block vectorization pass. The algorithm was +// inspired by that used by the Vienna MAP Vectorizor by Franchetti and Kral, +// et al. It works by looking for chains of pairable operations and then +// pairing them. +// +//===----------------------------------------------------------------------===// + +#define BBV_NAME "bb-vectorize" +#include "llvm/Transforms/Vectorize.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/DenseSet.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/SmallSet.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/ADT/StringExtras.h" +#include "llvm/Analysis/AliasAnalysis.h" +#include "llvm/Analysis/AliasSetTracker.h" +#include "llvm/Analysis/GlobalsModRef.h" +#include "llvm/Analysis/ScalarEvolution.h" +#include "llvm/Analysis/ScalarEvolutionAliasAnalysis.h" +#include "llvm/Analysis/ScalarEvolutionExpressions.h" +#include "llvm/Analysis/TargetLibraryInfo.h" +#include "llvm/Analysis/TargetTransformInfo.h" +#include "llvm/Analysis/ValueTracking.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DataLayout.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Dominators.h" +#include "llvm/IR/Function.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/Intrinsics.h" +#include "llvm/IR/LLVMContext.h" +#include "llvm/IR/Metadata.h" +#include "llvm/IR/Module.h" +#include "llvm/IR/Type.h" +#include "llvm/IR/ValueHandle.h" +#include "llvm/Pass.h" +#include "llvm/Support/CommandLine.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/raw_ostream.h" +#include "llvm/Transforms/Utils/Local.h" +#include <algorithm> +using namespace llvm; + +#define DEBUG_TYPE BBV_NAME + +static cl::opt<bool> +IgnoreTargetInfo("bb-vectorize-ignore-target-info", cl::init(false), + cl::Hidden, cl::desc("Ignore target information")); + +static cl::opt<unsigned> +ReqChainDepth("bb-vectorize-req-chain-depth", cl::init(6), cl::Hidden, + cl::desc("The required chain depth for vectorization")); + +static cl::opt<bool> +UseChainDepthWithTI("bb-vectorize-use-chain-depth", cl::init(false), + cl::Hidden, cl::desc("Use the chain depth requirement with" + " target information")); + +static cl::opt<unsigned> +SearchLimit("bb-vectorize-search-limit", cl::init(400), cl::Hidden, + cl::desc("The maximum search distance for instruction pairs")); + +static cl::opt<bool> +SplatBreaksChain("bb-vectorize-splat-breaks-chain", cl::init(false), cl::Hidden, + cl::desc("Replicating one element to a pair breaks the chain")); + +static cl::opt<unsigned> +VectorBits("bb-vectorize-vector-bits", cl::init(128), cl::Hidden, + cl::desc("The size of the native vector registers")); + +static cl::opt<unsigned> +MaxIter("bb-vectorize-max-iter", cl::init(0), cl::Hidden, + cl::desc("The maximum number of pairing iterations")); + +static cl::opt<bool> +Pow2LenOnly("bb-vectorize-pow2-len-only", cl::init(false), cl::Hidden, + cl::desc("Don't try to form non-2^n-length vectors")); + +static cl::opt<unsigned> +MaxInsts("bb-vectorize-max-instr-per-group", cl::init(500), cl::Hidden, + cl::desc("The maximum number of pairable instructions per group")); + +static cl::opt<unsigned> +MaxPairs("bb-vectorize-max-pairs-per-group", cl::init(3000), cl::Hidden, + cl::desc("The maximum number of candidate instruction pairs per group")); + +static cl::opt<unsigned> +MaxCandPairsForCycleCheck("bb-vectorize-max-cycle-check-pairs", cl::init(200), + cl::Hidden, cl::desc("The maximum number of candidate pairs with which to use" + " a full cycle check")); + +static cl::opt<bool> +NoBools("bb-vectorize-no-bools", cl::init(false), cl::Hidden, + cl::desc("Don't try to vectorize boolean (i1) values")); + +static cl::opt<bool> +NoInts("bb-vectorize-no-ints", cl::init(false), cl::Hidden, + cl::desc("Don't try to vectorize integer values")); + +static cl::opt<bool> +NoFloats("bb-vectorize-no-floats", cl::init(false), cl::Hidden, + cl::desc("Don't try to vectorize floating-point values")); + +// FIXME: This should default to false once pointer vector support works. +static cl::opt<bool> +NoPointers("bb-vectorize-no-pointers", cl::init(/*false*/ true), cl::Hidden, + cl::desc("Don't try to vectorize pointer values")); + +static cl::opt<bool> +NoCasts("bb-vectorize-no-casts", cl::init(false), cl::Hidden, + cl::desc("Don't try to vectorize casting (conversion) operations")); + +static cl::opt<bool> +NoMath("bb-vectorize-no-math", cl::init(false), cl::Hidden, + cl::desc("Don't try to vectorize floating-point math intrinsics")); + +static cl::opt<bool> + NoBitManipulation("bb-vectorize-no-bitmanip", cl::init(false), cl::Hidden, + cl::desc("Don't try to vectorize BitManipulation intrinsics")); + +static cl::opt<bool> +NoFMA("bb-vectorize-no-fma", cl::init(false), cl::Hidden, + cl::desc("Don't try to vectorize the fused-multiply-add intrinsic")); + +static cl::opt<bool> +NoSelect("bb-vectorize-no-select", cl::init(false), cl::Hidden, + cl::desc("Don't try to vectorize select instructions")); + +static cl::opt<bool> +NoCmp("bb-vectorize-no-cmp", cl::init(false), cl::Hidden, + cl::desc("Don't try to vectorize comparison instructions")); + +static cl::opt<bool> +NoGEP("bb-vectorize-no-gep", cl::init(false), cl::Hidden, + cl::desc("Don't try to vectorize getelementptr instructions")); + +static cl::opt<bool> +NoMemOps("bb-vectorize-no-mem-ops", cl::init(false), cl::Hidden, + cl::desc("Don't try to vectorize loads and stores")); + +static cl::opt<bool> +AlignedOnly("bb-vectorize-aligned-only", cl::init(false), cl::Hidden, + cl::desc("Only generate aligned loads and stores")); + +static cl::opt<bool> +NoMemOpBoost("bb-vectorize-no-mem-op-boost", + cl::init(false), cl::Hidden, + cl::desc("Don't boost the chain-depth contribution of loads and stores")); + +static cl::opt<bool> +FastDep("bb-vectorize-fast-dep", cl::init(false), cl::Hidden, + cl::desc("Use a fast instruction dependency analysis")); + +#ifndef NDEBUG +static cl::opt<bool> +DebugInstructionExamination("bb-vectorize-debug-instruction-examination", + cl::init(false), cl::Hidden, + cl::desc("When debugging is enabled, output information on the" + " instruction-examination process")); +static cl::opt<bool> +DebugCandidateSelection("bb-vectorize-debug-candidate-selection", + cl::init(false), cl::Hidden, + cl::desc("When debugging is enabled, output information on the" + " candidate-selection process")); +static cl::opt<bool> +DebugPairSelection("bb-vectorize-debug-pair-selection", + cl::init(false), cl::Hidden, + cl::desc("When debugging is enabled, output information on the" + " pair-selection process")); +static cl::opt<bool> +DebugCycleCheck("bb-vectorize-debug-cycle-check", + cl::init(false), cl::Hidden, + cl::desc("When debugging is enabled, output information on the" + " cycle-checking process")); + +static cl::opt<bool> +PrintAfterEveryPair("bb-vectorize-debug-print-after-every-pair", + cl::init(false), cl::Hidden, + cl::desc("When debugging is enabled, dump the basic block after" + " every pair is fused")); +#endif + +STATISTIC(NumFusedOps, "Number of operations fused by bb-vectorize"); + +namespace { + struct BBVectorize : public BasicBlockPass { + static char ID; // Pass identification, replacement for typeid + + const VectorizeConfig Config; + + BBVectorize(const VectorizeConfig &C = VectorizeConfig()) + : BasicBlockPass(ID), Config(C) { + initializeBBVectorizePass(*PassRegistry::getPassRegistry()); + } + + BBVectorize(Pass *P, Function &F, const VectorizeConfig &C) + : BasicBlockPass(ID), Config(C) { + AA = &P->getAnalysis<AAResultsWrapperPass>().getAAResults(); + DT = &P->getAnalysis<DominatorTreeWrapperPass>().getDomTree(); + SE = &P->getAnalysis<ScalarEvolutionWrapperPass>().getSE(); + TLI = &P->getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(); + TTI = IgnoreTargetInfo + ? nullptr + : &P->getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F); + } + + typedef std::pair<Value *, Value *> ValuePair; + typedef std::pair<ValuePair, int> ValuePairWithCost; + typedef std::pair<ValuePair, size_t> ValuePairWithDepth; + typedef std::pair<ValuePair, ValuePair> VPPair; // A ValuePair pair + typedef std::pair<VPPair, unsigned> VPPairWithType; + + AliasAnalysis *AA; + DominatorTree *DT; + ScalarEvolution *SE; + const TargetLibraryInfo *TLI; + const TargetTransformInfo *TTI; + + // FIXME: const correct? + + bool vectorizePairs(BasicBlock &BB, bool NonPow2Len = false); + + bool getCandidatePairs(BasicBlock &BB, + BasicBlock::iterator &Start, + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &FixedOrderPairs, + DenseMap<ValuePair, int> &CandidatePairCostSavings, + std::vector<Value *> &PairableInsts, bool NonPow2Len); + + // FIXME: The current implementation does not account for pairs that + // are connected in multiple ways. For example: + // C1 = A1 / A2; C2 = A2 / A1 (which may be both direct and a swap) + enum PairConnectionType { + PairConnectionDirect, + PairConnectionSwap, + PairConnectionSplat + }; + + void computeConnectedPairs( + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &CandidatePairsSet, + std::vector<Value *> &PairableInsts, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseMap<VPPair, unsigned> &PairConnectionTypes); + + void buildDepMap(BasicBlock &BB, + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + std::vector<Value *> &PairableInsts, + DenseSet<ValuePair> &PairableInstUsers); + + void choosePairs(DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &CandidatePairsSet, + DenseMap<ValuePair, int> &CandidatePairCostSavings, + std::vector<Value *> &PairableInsts, + DenseSet<ValuePair> &FixedOrderPairs, + DenseMap<VPPair, unsigned> &PairConnectionTypes, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps, + DenseSet<ValuePair> &PairableInstUsers, + DenseMap<Value *, Value *>& ChosenPairs); + + void fuseChosenPairs(BasicBlock &BB, + std::vector<Value *> &PairableInsts, + DenseMap<Value *, Value *>& ChosenPairs, + DenseSet<ValuePair> &FixedOrderPairs, + DenseMap<VPPair, unsigned> &PairConnectionTypes, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps); + + + bool isInstVectorizable(Instruction *I, bool &IsSimpleLoadStore); + + bool areInstsCompatible(Instruction *I, Instruction *J, + bool IsSimpleLoadStore, bool NonPow2Len, + int &CostSavings, int &FixedOrder); + + bool trackUsesOfI(DenseSet<Value *> &Users, + AliasSetTracker &WriteSet, Instruction *I, + Instruction *J, bool UpdateUsers = true, + DenseSet<ValuePair> *LoadMoveSetPairs = nullptr); + + void computePairsConnectedTo( + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &CandidatePairsSet, + std::vector<Value *> &PairableInsts, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseMap<VPPair, unsigned> &PairConnectionTypes, + ValuePair P); + + bool pairsConflict(ValuePair P, ValuePair Q, + DenseSet<ValuePair> &PairableInstUsers, + DenseMap<ValuePair, std::vector<ValuePair> > + *PairableInstUserMap = nullptr, + DenseSet<VPPair> *PairableInstUserPairSet = nullptr); + + bool pairWillFormCycle(ValuePair P, + DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUsers, + DenseSet<ValuePair> &CurrentPairs); + + void pruneDAGFor( + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + std::vector<Value *> &PairableInsts, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseSet<ValuePair> &PairableInstUsers, + DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUserMap, + DenseSet<VPPair> &PairableInstUserPairSet, + DenseMap<Value *, Value *> &ChosenPairs, + DenseMap<ValuePair, size_t> &DAG, + DenseSet<ValuePair> &PrunedDAG, ValuePair J, + bool UseCycleCheck); + + void buildInitialDAGFor( + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &CandidatePairsSet, + std::vector<Value *> &PairableInsts, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseSet<ValuePair> &PairableInstUsers, + DenseMap<Value *, Value *> &ChosenPairs, + DenseMap<ValuePair, size_t> &DAG, ValuePair J); + + void findBestDAGFor( + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &CandidatePairsSet, + DenseMap<ValuePair, int> &CandidatePairCostSavings, + std::vector<Value *> &PairableInsts, + DenseSet<ValuePair> &FixedOrderPairs, + DenseMap<VPPair, unsigned> &PairConnectionTypes, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps, + DenseSet<ValuePair> &PairableInstUsers, + DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUserMap, + DenseSet<VPPair> &PairableInstUserPairSet, + DenseMap<Value *, Value *> &ChosenPairs, + DenseSet<ValuePair> &BestDAG, size_t &BestMaxDepth, + int &BestEffSize, Value *II, std::vector<Value *>&JJ, + bool UseCycleCheck); + + Value *getReplacementPointerInput(LLVMContext& Context, Instruction *I, + Instruction *J, unsigned o); + + void fillNewShuffleMask(LLVMContext& Context, Instruction *J, + unsigned MaskOffset, unsigned NumInElem, + unsigned NumInElem1, unsigned IdxOffset, + std::vector<Constant*> &Mask); + + Value *getReplacementShuffleMask(LLVMContext& Context, Instruction *I, + Instruction *J); + + bool expandIEChain(LLVMContext& Context, Instruction *I, Instruction *J, + unsigned o, Value *&LOp, unsigned numElemL, + Type *ArgTypeL, Type *ArgTypeR, bool IBeforeJ, + unsigned IdxOff = 0); + + Value *getReplacementInput(LLVMContext& Context, Instruction *I, + Instruction *J, unsigned o, bool IBeforeJ); + + void getReplacementInputsForPair(LLVMContext& Context, Instruction *I, + Instruction *J, SmallVectorImpl<Value *> &ReplacedOperands, + bool IBeforeJ); + + void replaceOutputsOfPair(LLVMContext& Context, Instruction *I, + Instruction *J, Instruction *K, + Instruction *&InsertionPt, Instruction *&K1, + Instruction *&K2); + + void collectPairLoadMoveSet(BasicBlock &BB, + DenseMap<Value *, Value *> &ChosenPairs, + DenseMap<Value *, std::vector<Value *> > &LoadMoveSet, + DenseSet<ValuePair> &LoadMoveSetPairs, + Instruction *I); + + void collectLoadMoveSet(BasicBlock &BB, + std::vector<Value *> &PairableInsts, + DenseMap<Value *, Value *> &ChosenPairs, + DenseMap<Value *, std::vector<Value *> > &LoadMoveSet, + DenseSet<ValuePair> &LoadMoveSetPairs); + + bool canMoveUsesOfIAfterJ(BasicBlock &BB, + DenseSet<ValuePair> &LoadMoveSetPairs, + Instruction *I, Instruction *J); + + void moveUsesOfIAfterJ(BasicBlock &BB, + DenseSet<ValuePair> &LoadMoveSetPairs, + Instruction *&InsertionPt, + Instruction *I, Instruction *J); + + bool vectorizeBB(BasicBlock &BB) { + if (skipOptnoneFunction(BB)) + return false; + if (!DT->isReachableFromEntry(&BB)) { + DEBUG(dbgs() << "BBV: skipping unreachable " << BB.getName() << + " in " << BB.getParent()->getName() << "\n"); + return false; + } + + DEBUG(if (TTI) dbgs() << "BBV: using target information\n"); + + bool changed = false; + // Iterate a sufficient number of times to merge types of size 1 bit, + // then 2 bits, then 4, etc. up to half of the target vector width of the + // target vector register. + unsigned n = 1; + for (unsigned v = 2; + (TTI || v <= Config.VectorBits) && + (!Config.MaxIter || n <= Config.MaxIter); + v *= 2, ++n) { + DEBUG(dbgs() << "BBV: fusing loop #" << n << + " for " << BB.getName() << " in " << + BB.getParent()->getName() << "...\n"); + if (vectorizePairs(BB)) + changed = true; + else + break; + } + + if (changed && !Pow2LenOnly) { + ++n; + for (; !Config.MaxIter || n <= Config.MaxIter; ++n) { + DEBUG(dbgs() << "BBV: fusing for non-2^n-length vectors loop #: " << + n << " for " << BB.getName() << " in " << + BB.getParent()->getName() << "...\n"); + if (!vectorizePairs(BB, true)) break; + } + } + + DEBUG(dbgs() << "BBV: done!\n"); + return changed; + } + + bool runOnBasicBlock(BasicBlock &BB) override { + // OptimizeNone check deferred to vectorizeBB(). + + AA = &getAnalysis<AAResultsWrapperPass>().getAAResults(); + DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree(); + SE = &getAnalysis<ScalarEvolutionWrapperPass>().getSE(); + TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(); + TTI = IgnoreTargetInfo + ? nullptr + : &getAnalysis<TargetTransformInfoWrapperPass>().getTTI( + *BB.getParent()); + + return vectorizeBB(BB); + } + + void getAnalysisUsage(AnalysisUsage &AU) const override { + BasicBlockPass::getAnalysisUsage(AU); + AU.addRequired<AAResultsWrapperPass>(); + AU.addRequired<DominatorTreeWrapperPass>(); + AU.addRequired<ScalarEvolutionWrapperPass>(); + AU.addRequired<TargetLibraryInfoWrapperPass>(); + AU.addRequired<TargetTransformInfoWrapperPass>(); + AU.addPreserved<DominatorTreeWrapperPass>(); + AU.addPreserved<GlobalsAAWrapperPass>(); + AU.addPreserved<ScalarEvolutionWrapperPass>(); + AU.addPreserved<SCEVAAWrapperPass>(); + AU.setPreservesCFG(); + } + + static inline VectorType *getVecTypeForPair(Type *ElemTy, Type *Elem2Ty) { + assert(ElemTy->getScalarType() == Elem2Ty->getScalarType() && + "Cannot form vector from incompatible scalar types"); + Type *STy = ElemTy->getScalarType(); + + unsigned numElem; + if (VectorType *VTy = dyn_cast<VectorType>(ElemTy)) { + numElem = VTy->getNumElements(); + } else { + numElem = 1; + } + + if (VectorType *VTy = dyn_cast<VectorType>(Elem2Ty)) { + numElem += VTy->getNumElements(); + } else { + numElem += 1; + } + + return VectorType::get(STy, numElem); + } + + static inline void getInstructionTypes(Instruction *I, + Type *&T1, Type *&T2) { + if (StoreInst *SI = dyn_cast<StoreInst>(I)) { + // For stores, it is the value type, not the pointer type that matters + // because the value is what will come from a vector register. + + Value *IVal = SI->getValueOperand(); + T1 = IVal->getType(); + } else { + T1 = I->getType(); + } + + if (CastInst *CI = dyn_cast<CastInst>(I)) + T2 = CI->getSrcTy(); + else + T2 = T1; + + if (SelectInst *SI = dyn_cast<SelectInst>(I)) { + T2 = SI->getCondition()->getType(); + } else if (ShuffleVectorInst *SI = dyn_cast<ShuffleVectorInst>(I)) { + T2 = SI->getOperand(0)->getType(); + } else if (CmpInst *CI = dyn_cast<CmpInst>(I)) { + T2 = CI->getOperand(0)->getType(); + } + } + + // Returns the weight associated with the provided value. A chain of + // candidate pairs has a length given by the sum of the weights of its + // members (one weight per pair; the weight of each member of the pair + // is assumed to be the same). This length is then compared to the + // chain-length threshold to determine if a given chain is significant + // enough to be vectorized. The length is also used in comparing + // candidate chains where longer chains are considered to be better. + // Note: when this function returns 0, the resulting instructions are + // not actually fused. + inline size_t getDepthFactor(Value *V) { + // InsertElement and ExtractElement have a depth factor of zero. This is + // for two reasons: First, they cannot be usefully fused. Second, because + // the pass generates a lot of these, they can confuse the simple metric + // used to compare the dags in the next iteration. Thus, giving them a + // weight of zero allows the pass to essentially ignore them in + // subsequent iterations when looking for vectorization opportunities + // while still tracking dependency chains that flow through those + // instructions. + if (isa<InsertElementInst>(V) || isa<ExtractElementInst>(V)) + return 0; + + // Give a load or store half of the required depth so that load/store + // pairs will vectorize. + if (!Config.NoMemOpBoost && (isa<LoadInst>(V) || isa<StoreInst>(V))) + return Config.ReqChainDepth/2; + + return 1; + } + + // Returns the cost of the provided instruction using TTI. + // This does not handle loads and stores. + unsigned getInstrCost(unsigned Opcode, Type *T1, Type *T2, + TargetTransformInfo::OperandValueKind Op1VK = + TargetTransformInfo::OK_AnyValue, + TargetTransformInfo::OperandValueKind Op2VK = + TargetTransformInfo::OK_AnyValue) { + switch (Opcode) { + default: break; + case Instruction::GetElementPtr: + // We mark this instruction as zero-cost because scalar GEPs are usually + // lowered to the instruction addressing mode. At the moment we don't + // generate vector GEPs. + return 0; + case Instruction::Br: + return TTI->getCFInstrCost(Opcode); + case Instruction::PHI: + return 0; + case Instruction::Add: + case Instruction::FAdd: + case Instruction::Sub: + case Instruction::FSub: + case Instruction::Mul: + case Instruction::FMul: + case Instruction::UDiv: + case Instruction::SDiv: + case Instruction::FDiv: + case Instruction::URem: + case Instruction::SRem: + case Instruction::FRem: + case Instruction::Shl: + case Instruction::LShr: + case Instruction::AShr: + case Instruction::And: + case Instruction::Or: + case Instruction::Xor: + return TTI->getArithmeticInstrCost(Opcode, T1, Op1VK, Op2VK); + case Instruction::Select: + case Instruction::ICmp: + case Instruction::FCmp: + return TTI->getCmpSelInstrCost(Opcode, T1, T2); + case Instruction::ZExt: + case Instruction::SExt: + case Instruction::FPToUI: + case Instruction::FPToSI: + case Instruction::FPExt: + case Instruction::PtrToInt: + case Instruction::IntToPtr: + case Instruction::SIToFP: + case Instruction::UIToFP: + case Instruction::Trunc: + case Instruction::FPTrunc: + case Instruction::BitCast: + case Instruction::ShuffleVector: + return TTI->getCastInstrCost(Opcode, T1, T2); + } + + return 1; + } + + // This determines the relative offset of two loads or stores, returning + // true if the offset could be determined to be some constant value. + // For example, if OffsetInElmts == 1, then J accesses the memory directly + // after I; if OffsetInElmts == -1 then I accesses the memory + // directly after J. + bool getPairPtrInfo(Instruction *I, Instruction *J, + Value *&IPtr, Value *&JPtr, unsigned &IAlignment, unsigned &JAlignment, + unsigned &IAddressSpace, unsigned &JAddressSpace, + int64_t &OffsetInElmts, bool ComputeOffset = true) { + OffsetInElmts = 0; + if (LoadInst *LI = dyn_cast<LoadInst>(I)) { + LoadInst *LJ = cast<LoadInst>(J); + IPtr = LI->getPointerOperand(); + JPtr = LJ->getPointerOperand(); + IAlignment = LI->getAlignment(); + JAlignment = LJ->getAlignment(); + IAddressSpace = LI->getPointerAddressSpace(); + JAddressSpace = LJ->getPointerAddressSpace(); + } else { + StoreInst *SI = cast<StoreInst>(I), *SJ = cast<StoreInst>(J); + IPtr = SI->getPointerOperand(); + JPtr = SJ->getPointerOperand(); + IAlignment = SI->getAlignment(); + JAlignment = SJ->getAlignment(); + IAddressSpace = SI->getPointerAddressSpace(); + JAddressSpace = SJ->getPointerAddressSpace(); + } + + if (!ComputeOffset) + return true; + + const SCEV *IPtrSCEV = SE->getSCEV(IPtr); + const SCEV *JPtrSCEV = SE->getSCEV(JPtr); + + // If this is a trivial offset, then we'll get something like + // 1*sizeof(type). With target data, which we need anyway, this will get + // constant folded into a number. + const SCEV *OffsetSCEV = SE->getMinusSCEV(JPtrSCEV, IPtrSCEV); + if (const SCEVConstant *ConstOffSCEV = + dyn_cast<SCEVConstant>(OffsetSCEV)) { + ConstantInt *IntOff = ConstOffSCEV->getValue(); + int64_t Offset = IntOff->getSExtValue(); + const DataLayout &DL = I->getModule()->getDataLayout(); + Type *VTy = IPtr->getType()->getPointerElementType(); + int64_t VTyTSS = (int64_t)DL.getTypeStoreSize(VTy); + + Type *VTy2 = JPtr->getType()->getPointerElementType(); + if (VTy != VTy2 && Offset < 0) { + int64_t VTy2TSS = (int64_t)DL.getTypeStoreSize(VTy2); + OffsetInElmts = Offset/VTy2TSS; + return (std::abs(Offset) % VTy2TSS) == 0; + } + + OffsetInElmts = Offset/VTyTSS; + return (std::abs(Offset) % VTyTSS) == 0; + } + + return false; + } + + // Returns true if the provided CallInst represents an intrinsic that can + // be vectorized. + bool isVectorizableIntrinsic(CallInst* I) { + Function *F = I->getCalledFunction(); + if (!F) return false; + + Intrinsic::ID IID = F->getIntrinsicID(); + if (!IID) return false; + + switch(IID) { + default: + return false; + case Intrinsic::sqrt: + case Intrinsic::powi: + case Intrinsic::sin: + case Intrinsic::cos: + case Intrinsic::log: + case Intrinsic::log2: + case Intrinsic::log10: + case Intrinsic::exp: + case Intrinsic::exp2: + case Intrinsic::pow: + case Intrinsic::round: + case Intrinsic::copysign: + case Intrinsic::ceil: + case Intrinsic::nearbyint: + case Intrinsic::rint: + case Intrinsic::trunc: + case Intrinsic::floor: + case Intrinsic::fabs: + case Intrinsic::minnum: + case Intrinsic::maxnum: + return Config.VectorizeMath; + case Intrinsic::bswap: + case Intrinsic::ctpop: + case Intrinsic::ctlz: + case Intrinsic::cttz: + return Config.VectorizeBitManipulations; + case Intrinsic::fma: + case Intrinsic::fmuladd: + return Config.VectorizeFMA; + } + } + + bool isPureIEChain(InsertElementInst *IE) { + InsertElementInst *IENext = IE; + do { + if (!isa<UndefValue>(IENext->getOperand(0)) && + !isa<InsertElementInst>(IENext->getOperand(0))) { + return false; + } + } while ((IENext = + dyn_cast<InsertElementInst>(IENext->getOperand(0)))); + + return true; + } + }; + + // This function implements one vectorization iteration on the provided + // basic block. It returns true if the block is changed. + bool BBVectorize::vectorizePairs(BasicBlock &BB, bool NonPow2Len) { + bool ShouldContinue; + BasicBlock::iterator Start = BB.getFirstInsertionPt(); + + std::vector<Value *> AllPairableInsts; + DenseMap<Value *, Value *> AllChosenPairs; + DenseSet<ValuePair> AllFixedOrderPairs; + DenseMap<VPPair, unsigned> AllPairConnectionTypes; + DenseMap<ValuePair, std::vector<ValuePair> > AllConnectedPairs, + AllConnectedPairDeps; + + do { + std::vector<Value *> PairableInsts; + DenseMap<Value *, std::vector<Value *> > CandidatePairs; + DenseSet<ValuePair> FixedOrderPairs; + DenseMap<ValuePair, int> CandidatePairCostSavings; + ShouldContinue = getCandidatePairs(BB, Start, CandidatePairs, + FixedOrderPairs, + CandidatePairCostSavings, + PairableInsts, NonPow2Len); + if (PairableInsts.empty()) continue; + + // Build the candidate pair set for faster lookups. + DenseSet<ValuePair> CandidatePairsSet; + for (DenseMap<Value *, std::vector<Value *> >::iterator I = + CandidatePairs.begin(), E = CandidatePairs.end(); I != E; ++I) + for (std::vector<Value *>::iterator J = I->second.begin(), + JE = I->second.end(); J != JE; ++J) + CandidatePairsSet.insert(ValuePair(I->first, *J)); + + // Now we have a map of all of the pairable instructions and we need to + // select the best possible pairing. A good pairing is one such that the + // users of the pair are also paired. This defines a (directed) forest + // over the pairs such that two pairs are connected iff the second pair + // uses the first. + + // Note that it only matters that both members of the second pair use some + // element of the first pair (to allow for splatting). + + DenseMap<ValuePair, std::vector<ValuePair> > ConnectedPairs, + ConnectedPairDeps; + DenseMap<VPPair, unsigned> PairConnectionTypes; + computeConnectedPairs(CandidatePairs, CandidatePairsSet, + PairableInsts, ConnectedPairs, PairConnectionTypes); + if (ConnectedPairs.empty()) continue; + + for (DenseMap<ValuePair, std::vector<ValuePair> >::iterator + I = ConnectedPairs.begin(), IE = ConnectedPairs.end(); + I != IE; ++I) + for (std::vector<ValuePair>::iterator J = I->second.begin(), + JE = I->second.end(); J != JE; ++J) + ConnectedPairDeps[*J].push_back(I->first); + + // Build the pairable-instruction dependency map + DenseSet<ValuePair> PairableInstUsers; + buildDepMap(BB, CandidatePairs, PairableInsts, PairableInstUsers); + + // There is now a graph of the connected pairs. For each variable, pick + // the pairing with the largest dag meeting the depth requirement on at + // least one branch. Then select all pairings that are part of that dag + // and remove them from the list of available pairings and pairable + // variables. + + DenseMap<Value *, Value *> ChosenPairs; + choosePairs(CandidatePairs, CandidatePairsSet, + CandidatePairCostSavings, + PairableInsts, FixedOrderPairs, PairConnectionTypes, + ConnectedPairs, ConnectedPairDeps, + PairableInstUsers, ChosenPairs); + + if (ChosenPairs.empty()) continue; + AllPairableInsts.insert(AllPairableInsts.end(), PairableInsts.begin(), + PairableInsts.end()); + AllChosenPairs.insert(ChosenPairs.begin(), ChosenPairs.end()); + + // Only for the chosen pairs, propagate information on fixed-order pairs, + // pair connections, and their types to the data structures used by the + // pair fusion procedures. + for (DenseMap<Value *, Value *>::iterator I = ChosenPairs.begin(), + IE = ChosenPairs.end(); I != IE; ++I) { + if (FixedOrderPairs.count(*I)) + AllFixedOrderPairs.insert(*I); + else if (FixedOrderPairs.count(ValuePair(I->second, I->first))) + AllFixedOrderPairs.insert(ValuePair(I->second, I->first)); + + for (DenseMap<Value *, Value *>::iterator J = ChosenPairs.begin(); + J != IE; ++J) { + DenseMap<VPPair, unsigned>::iterator K = + PairConnectionTypes.find(VPPair(*I, *J)); + if (K != PairConnectionTypes.end()) { + AllPairConnectionTypes.insert(*K); + } else { + K = PairConnectionTypes.find(VPPair(*J, *I)); + if (K != PairConnectionTypes.end()) + AllPairConnectionTypes.insert(*K); + } + } + } + + for (DenseMap<ValuePair, std::vector<ValuePair> >::iterator + I = ConnectedPairs.begin(), IE = ConnectedPairs.end(); + I != IE; ++I) + for (std::vector<ValuePair>::iterator J = I->second.begin(), + JE = I->second.end(); J != JE; ++J) + if (AllPairConnectionTypes.count(VPPair(I->first, *J))) { + AllConnectedPairs[I->first].push_back(*J); + AllConnectedPairDeps[*J].push_back(I->first); + } + } while (ShouldContinue); + + if (AllChosenPairs.empty()) return false; + NumFusedOps += AllChosenPairs.size(); + + // A set of pairs has now been selected. It is now necessary to replace the + // paired instructions with vector instructions. For this procedure each + // operand must be replaced with a vector operand. This vector is formed + // by using build_vector on the old operands. The replaced values are then + // replaced with a vector_extract on the result. Subsequent optimization + // passes should coalesce the build/extract combinations. + + fuseChosenPairs(BB, AllPairableInsts, AllChosenPairs, AllFixedOrderPairs, + AllPairConnectionTypes, + AllConnectedPairs, AllConnectedPairDeps); + + // It is important to cleanup here so that future iterations of this + // function have less work to do. + (void)SimplifyInstructionsInBlock(&BB, TLI); + return true; + } + + // This function returns true if the provided instruction is capable of being + // fused into a vector instruction. This determination is based only on the + // type and other attributes of the instruction. + bool BBVectorize::isInstVectorizable(Instruction *I, + bool &IsSimpleLoadStore) { + IsSimpleLoadStore = false; + + if (CallInst *C = dyn_cast<CallInst>(I)) { + if (!isVectorizableIntrinsic(C)) + return false; + } else if (LoadInst *L = dyn_cast<LoadInst>(I)) { + // Vectorize simple loads if possbile: + IsSimpleLoadStore = L->isSimple(); + if (!IsSimpleLoadStore || !Config.VectorizeMemOps) + return false; + } else if (StoreInst *S = dyn_cast<StoreInst>(I)) { + // Vectorize simple stores if possbile: + IsSimpleLoadStore = S->isSimple(); + if (!IsSimpleLoadStore || !Config.VectorizeMemOps) + return false; + } else if (CastInst *C = dyn_cast<CastInst>(I)) { + // We can vectorize casts, but not casts of pointer types, etc. + if (!Config.VectorizeCasts) + return false; + + Type *SrcTy = C->getSrcTy(); + if (!SrcTy->isSingleValueType()) + return false; + + Type *DestTy = C->getDestTy(); + if (!DestTy->isSingleValueType()) + return false; + } else if (isa<SelectInst>(I)) { + if (!Config.VectorizeSelect) + return false; + } else if (isa<CmpInst>(I)) { + if (!Config.VectorizeCmp) + return false; + } else if (GetElementPtrInst *G = dyn_cast<GetElementPtrInst>(I)) { + if (!Config.VectorizeGEP) + return false; + + // Currently, vector GEPs exist only with one index. + if (G->getNumIndices() != 1) + return false; + } else if (!(I->isBinaryOp() || isa<ShuffleVectorInst>(I) || + isa<ExtractElementInst>(I) || isa<InsertElementInst>(I))) { + return false; + } + + Type *T1, *T2; + getInstructionTypes(I, T1, T2); + + // Not every type can be vectorized... + if (!(VectorType::isValidElementType(T1) || T1->isVectorTy()) || + !(VectorType::isValidElementType(T2) || T2->isVectorTy())) + return false; + + if (T1->getScalarSizeInBits() == 1) { + if (!Config.VectorizeBools) + return false; + } else { + if (!Config.VectorizeInts && T1->isIntOrIntVectorTy()) + return false; + } + + if (T2->getScalarSizeInBits() == 1) { + if (!Config.VectorizeBools) + return false; + } else { + if (!Config.VectorizeInts && T2->isIntOrIntVectorTy()) + return false; + } + + if (!Config.VectorizeFloats + && (T1->isFPOrFPVectorTy() || T2->isFPOrFPVectorTy())) + return false; + + // Don't vectorize target-specific types. + if (T1->isX86_FP80Ty() || T1->isPPC_FP128Ty() || T1->isX86_MMXTy()) + return false; + if (T2->isX86_FP80Ty() || T2->isPPC_FP128Ty() || T2->isX86_MMXTy()) + return false; + + if (!Config.VectorizePointers && (T1->getScalarType()->isPointerTy() || + T2->getScalarType()->isPointerTy())) + return false; + + if (!TTI && (T1->getPrimitiveSizeInBits() >= Config.VectorBits || + T2->getPrimitiveSizeInBits() >= Config.VectorBits)) + return false; + + return true; + } + + // This function returns true if the two provided instructions are compatible + // (meaning that they can be fused into a vector instruction). This assumes + // that I has already been determined to be vectorizable and that J is not + // in the use dag of I. + bool BBVectorize::areInstsCompatible(Instruction *I, Instruction *J, + bool IsSimpleLoadStore, bool NonPow2Len, + int &CostSavings, int &FixedOrder) { + DEBUG(if (DebugInstructionExamination) dbgs() << "BBV: looking at " << *I << + " <-> " << *J << "\n"); + + CostSavings = 0; + FixedOrder = 0; + + // Loads and stores can be merged if they have different alignments, + // but are otherwise the same. + if (!J->isSameOperationAs(I, Instruction::CompareIgnoringAlignment | + (NonPow2Len ? Instruction::CompareUsingScalarTypes : 0))) + return false; + + Type *IT1, *IT2, *JT1, *JT2; + getInstructionTypes(I, IT1, IT2); + getInstructionTypes(J, JT1, JT2); + unsigned MaxTypeBits = std::max( + IT1->getPrimitiveSizeInBits() + JT1->getPrimitiveSizeInBits(), + IT2->getPrimitiveSizeInBits() + JT2->getPrimitiveSizeInBits()); + if (!TTI && MaxTypeBits > Config.VectorBits) + return false; + + // FIXME: handle addsub-type operations! + + if (IsSimpleLoadStore) { + Value *IPtr, *JPtr; + unsigned IAlignment, JAlignment, IAddressSpace, JAddressSpace; + int64_t OffsetInElmts = 0; + if (getPairPtrInfo(I, J, IPtr, JPtr, IAlignment, JAlignment, + IAddressSpace, JAddressSpace, OffsetInElmts) && + std::abs(OffsetInElmts) == 1) { + FixedOrder = (int) OffsetInElmts; + unsigned BottomAlignment = IAlignment; + if (OffsetInElmts < 0) BottomAlignment = JAlignment; + + Type *aTypeI = isa<StoreInst>(I) ? + cast<StoreInst>(I)->getValueOperand()->getType() : I->getType(); + Type *aTypeJ = isa<StoreInst>(J) ? + cast<StoreInst>(J)->getValueOperand()->getType() : J->getType(); + Type *VType = getVecTypeForPair(aTypeI, aTypeJ); + + if (Config.AlignedOnly) { + // An aligned load or store is possible only if the instruction + // with the lower offset has an alignment suitable for the + // vector type. + const DataLayout &DL = I->getModule()->getDataLayout(); + unsigned VecAlignment = DL.getPrefTypeAlignment(VType); + if (BottomAlignment < VecAlignment) + return false; + } + + if (TTI) { + unsigned ICost = TTI->getMemoryOpCost(I->getOpcode(), aTypeI, + IAlignment, IAddressSpace); + unsigned JCost = TTI->getMemoryOpCost(J->getOpcode(), aTypeJ, + JAlignment, JAddressSpace); + unsigned VCost = TTI->getMemoryOpCost(I->getOpcode(), VType, + BottomAlignment, + IAddressSpace); + + ICost += TTI->getAddressComputationCost(aTypeI); + JCost += TTI->getAddressComputationCost(aTypeJ); + VCost += TTI->getAddressComputationCost(VType); + + if (VCost > ICost + JCost) + return false; + + // We don't want to fuse to a type that will be split, even + // if the two input types will also be split and there is no other + // associated cost. + unsigned VParts = TTI->getNumberOfParts(VType); + if (VParts > 1) + return false; + else if (!VParts && VCost == ICost + JCost) + return false; + + CostSavings = ICost + JCost - VCost; + } + } else { + return false; + } + } else if (TTI) { + unsigned ICost = getInstrCost(I->getOpcode(), IT1, IT2); + unsigned JCost = getInstrCost(J->getOpcode(), JT1, JT2); + Type *VT1 = getVecTypeForPair(IT1, JT1), + *VT2 = getVecTypeForPair(IT2, JT2); + TargetTransformInfo::OperandValueKind Op1VK = + TargetTransformInfo::OK_AnyValue; + TargetTransformInfo::OperandValueKind Op2VK = + TargetTransformInfo::OK_AnyValue; + + // On some targets (example X86) the cost of a vector shift may vary + // depending on whether the second operand is a Uniform or + // NonUniform Constant. + switch (I->getOpcode()) { + default : break; + case Instruction::Shl: + case Instruction::LShr: + case Instruction::AShr: + + // If both I and J are scalar shifts by constant, then the + // merged vector shift count would be either a constant splat value + // or a non-uniform vector of constants. + if (ConstantInt *CII = dyn_cast<ConstantInt>(I->getOperand(1))) { + if (ConstantInt *CIJ = dyn_cast<ConstantInt>(J->getOperand(1))) + Op2VK = CII == CIJ ? TargetTransformInfo::OK_UniformConstantValue : + TargetTransformInfo::OK_NonUniformConstantValue; + } else { + // Check for a splat of a constant or for a non uniform vector + // of constants. + Value *IOp = I->getOperand(1); + Value *JOp = J->getOperand(1); + if ((isa<ConstantVector>(IOp) || isa<ConstantDataVector>(IOp)) && + (isa<ConstantVector>(JOp) || isa<ConstantDataVector>(JOp))) { + Op2VK = TargetTransformInfo::OK_NonUniformConstantValue; + Constant *SplatValue = cast<Constant>(IOp)->getSplatValue(); + if (SplatValue != nullptr && + SplatValue == cast<Constant>(JOp)->getSplatValue()) + Op2VK = TargetTransformInfo::OK_UniformConstantValue; + } + } + } + + // Note that this procedure is incorrect for insert and extract element + // instructions (because combining these often results in a shuffle), + // but this cost is ignored (because insert and extract element + // instructions are assigned a zero depth factor and are not really + // fused in general). + unsigned VCost = getInstrCost(I->getOpcode(), VT1, VT2, Op1VK, Op2VK); + + if (VCost > ICost + JCost) + return false; + + // We don't want to fuse to a type that will be split, even + // if the two input types will also be split and there is no other + // associated cost. + unsigned VParts1 = TTI->getNumberOfParts(VT1), + VParts2 = TTI->getNumberOfParts(VT2); + if (VParts1 > 1 || VParts2 > 1) + return false; + else if ((!VParts1 || !VParts2) && VCost == ICost + JCost) + return false; + + CostSavings = ICost + JCost - VCost; + } + + // The powi,ctlz,cttz intrinsics are special because only the first + // argument is vectorized, the second arguments must be equal. + CallInst *CI = dyn_cast<CallInst>(I); + Function *FI; + if (CI && (FI = CI->getCalledFunction())) { + Intrinsic::ID IID = FI->getIntrinsicID(); + if (IID == Intrinsic::powi || IID == Intrinsic::ctlz || + IID == Intrinsic::cttz) { + Value *A1I = CI->getArgOperand(1), + *A1J = cast<CallInst>(J)->getArgOperand(1); + const SCEV *A1ISCEV = SE->getSCEV(A1I), + *A1JSCEV = SE->getSCEV(A1J); + return (A1ISCEV == A1JSCEV); + } + + if (IID && TTI) { + SmallVector<Type*, 4> Tys; + for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) + Tys.push_back(CI->getArgOperand(i)->getType()); + unsigned ICost = TTI->getIntrinsicInstrCost(IID, IT1, Tys); + + Tys.clear(); + CallInst *CJ = cast<CallInst>(J); + for (unsigned i = 0, ie = CJ->getNumArgOperands(); i != ie; ++i) + Tys.push_back(CJ->getArgOperand(i)->getType()); + unsigned JCost = TTI->getIntrinsicInstrCost(IID, JT1, Tys); + + Tys.clear(); + assert(CI->getNumArgOperands() == CJ->getNumArgOperands() && + "Intrinsic argument counts differ"); + for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) { + if ((IID == Intrinsic::powi || IID == Intrinsic::ctlz || + IID == Intrinsic::cttz) && i == 1) + Tys.push_back(CI->getArgOperand(i)->getType()); + else + Tys.push_back(getVecTypeForPair(CI->getArgOperand(i)->getType(), + CJ->getArgOperand(i)->getType())); + } + + Type *RetTy = getVecTypeForPair(IT1, JT1); + unsigned VCost = TTI->getIntrinsicInstrCost(IID, RetTy, Tys); + + if (VCost > ICost + JCost) + return false; + + // We don't want to fuse to a type that will be split, even + // if the two input types will also be split and there is no other + // associated cost. + unsigned RetParts = TTI->getNumberOfParts(RetTy); + if (RetParts > 1) + return false; + else if (!RetParts && VCost == ICost + JCost) + return false; + + for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) { + if (!Tys[i]->isVectorTy()) + continue; + + unsigned NumParts = TTI->getNumberOfParts(Tys[i]); + if (NumParts > 1) + return false; + else if (!NumParts && VCost == ICost + JCost) + return false; + } + + CostSavings = ICost + JCost - VCost; + } + } + + return true; + } + + // Figure out whether or not J uses I and update the users and write-set + // structures associated with I. Specifically, Users represents the set of + // instructions that depend on I. WriteSet represents the set + // of memory locations that are dependent on I. If UpdateUsers is true, + // and J uses I, then Users is updated to contain J and WriteSet is updated + // to contain any memory locations to which J writes. The function returns + // true if J uses I. By default, alias analysis is used to determine + // whether J reads from memory that overlaps with a location in WriteSet. + // If LoadMoveSet is not null, then it is a previously-computed map + // where the key is the memory-based user instruction and the value is + // the instruction to be compared with I. So, if LoadMoveSet is provided, + // then the alias analysis is not used. This is necessary because this + // function is called during the process of moving instructions during + // vectorization and the results of the alias analysis are not stable during + // that process. + bool BBVectorize::trackUsesOfI(DenseSet<Value *> &Users, + AliasSetTracker &WriteSet, Instruction *I, + Instruction *J, bool UpdateUsers, + DenseSet<ValuePair> *LoadMoveSetPairs) { + bool UsesI = false; + + // This instruction may already be marked as a user due, for example, to + // being a member of a selected pair. + if (Users.count(J)) + UsesI = true; + + if (!UsesI) + for (User::op_iterator JU = J->op_begin(), JE = J->op_end(); + JU != JE; ++JU) { + Value *V = *JU; + if (I == V || Users.count(V)) { + UsesI = true; + break; + } + } + if (!UsesI && J->mayReadFromMemory()) { + if (LoadMoveSetPairs) { + UsesI = LoadMoveSetPairs->count(ValuePair(J, I)); + } else { + for (AliasSetTracker::iterator W = WriteSet.begin(), + WE = WriteSet.end(); W != WE; ++W) { + if (W->aliasesUnknownInst(J, *AA)) { + UsesI = true; + break; + } + } + } + } + + if (UsesI && UpdateUsers) { + if (J->mayWriteToMemory()) WriteSet.add(J); + Users.insert(J); + } + + return UsesI; + } + + // This function iterates over all instruction pairs in the provided + // basic block and collects all candidate pairs for vectorization. + bool BBVectorize::getCandidatePairs(BasicBlock &BB, + BasicBlock::iterator &Start, + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &FixedOrderPairs, + DenseMap<ValuePair, int> &CandidatePairCostSavings, + std::vector<Value *> &PairableInsts, bool NonPow2Len) { + size_t TotalPairs = 0; + BasicBlock::iterator E = BB.end(); + if (Start == E) return false; + + bool ShouldContinue = false, IAfterStart = false; + for (BasicBlock::iterator I = Start++; I != E; ++I) { + if (I == Start) IAfterStart = true; + + bool IsSimpleLoadStore; + if (!isInstVectorizable(&*I, IsSimpleLoadStore)) + continue; + + // Look for an instruction with which to pair instruction *I... + DenseSet<Value *> Users; + AliasSetTracker WriteSet(*AA); + if (I->mayWriteToMemory()) + WriteSet.add(&*I); + + bool JAfterStart = IAfterStart; + BasicBlock::iterator J = std::next(I); + for (unsigned ss = 0; J != E && ss <= Config.SearchLimit; ++J, ++ss) { + if (&*J == Start) + JAfterStart = true; + + // Determine if J uses I, if so, exit the loop. + bool UsesI = trackUsesOfI(Users, WriteSet, &*I, &*J, !Config.FastDep); + if (Config.FastDep) { + // Note: For this heuristic to be effective, independent operations + // must tend to be intermixed. This is likely to be true from some + // kinds of grouped loop unrolling (but not the generic LLVM pass), + // but otherwise may require some kind of reordering pass. + + // When using fast dependency analysis, + // stop searching after first use: + if (UsesI) break; + } else { + if (UsesI) continue; + } + + // J does not use I, and comes before the first use of I, so it can be + // merged with I if the instructions are compatible. + int CostSavings, FixedOrder; + if (!areInstsCompatible(&*I, &*J, IsSimpleLoadStore, NonPow2Len, + CostSavings, FixedOrder)) + continue; + + // J is a candidate for merging with I. + if (PairableInsts.empty() || + PairableInsts[PairableInsts.size() - 1] != &*I) { + PairableInsts.push_back(&*I); + } + + CandidatePairs[&*I].push_back(&*J); + ++TotalPairs; + if (TTI) + CandidatePairCostSavings.insert( + ValuePairWithCost(ValuePair(&*I, &*J), CostSavings)); + + if (FixedOrder == 1) + FixedOrderPairs.insert(ValuePair(&*I, &*J)); + else if (FixedOrder == -1) + FixedOrderPairs.insert(ValuePair(&*J, &*I)); + + // The next call to this function must start after the last instruction + // selected during this invocation. + if (JAfterStart) { + Start = std::next(J); + IAfterStart = JAfterStart = false; + } + + DEBUG(if (DebugCandidateSelection) dbgs() << "BBV: candidate pair " + << *I << " <-> " << *J << " (cost savings: " << + CostSavings << ")\n"); + + // If we have already found too many pairs, break here and this function + // will be called again starting after the last instruction selected + // during this invocation. + if (PairableInsts.size() >= Config.MaxInsts || + TotalPairs >= Config.MaxPairs) { + ShouldContinue = true; + break; + } + } + + if (ShouldContinue) + break; + } + + DEBUG(dbgs() << "BBV: found " << PairableInsts.size() + << " instructions with candidate pairs\n"); + + return ShouldContinue; + } + + // Finds candidate pairs connected to the pair P = <PI, PJ>. This means that + // it looks for pairs such that both members have an input which is an + // output of PI or PJ. + void BBVectorize::computePairsConnectedTo( + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &CandidatePairsSet, + std::vector<Value *> &PairableInsts, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseMap<VPPair, unsigned> &PairConnectionTypes, + ValuePair P) { + StoreInst *SI, *SJ; + + // For each possible pairing for this variable, look at the uses of + // the first value... + for (Value::user_iterator I = P.first->user_begin(), + E = P.first->user_end(); + I != E; ++I) { + User *UI = *I; + if (isa<LoadInst>(UI)) { + // A pair cannot be connected to a load because the load only takes one + // operand (the address) and it is a scalar even after vectorization. + continue; + } else if ((SI = dyn_cast<StoreInst>(UI)) && + P.first == SI->getPointerOperand()) { + // Similarly, a pair cannot be connected to a store through its + // pointer operand. + continue; + } + + // For each use of the first variable, look for uses of the second + // variable... + for (User *UJ : P.second->users()) { + if ((SJ = dyn_cast<StoreInst>(UJ)) && + P.second == SJ->getPointerOperand()) + continue; + + // Look for <I, J>: + if (CandidatePairsSet.count(ValuePair(UI, UJ))) { + VPPair VP(P, ValuePair(UI, UJ)); + ConnectedPairs[VP.first].push_back(VP.second); + PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionDirect)); + } + + // Look for <J, I>: + if (CandidatePairsSet.count(ValuePair(UJ, UI))) { + VPPair VP(P, ValuePair(UJ, UI)); + ConnectedPairs[VP.first].push_back(VP.second); + PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionSwap)); + } + } + + if (Config.SplatBreaksChain) continue; + // Look for cases where just the first value in the pair is used by + // both members of another pair (splatting). + for (Value::user_iterator J = P.first->user_begin(); J != E; ++J) { + User *UJ = *J; + if ((SJ = dyn_cast<StoreInst>(UJ)) && + P.first == SJ->getPointerOperand()) + continue; + + if (CandidatePairsSet.count(ValuePair(UI, UJ))) { + VPPair VP(P, ValuePair(UI, UJ)); + ConnectedPairs[VP.first].push_back(VP.second); + PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionSplat)); + } + } + } + + if (Config.SplatBreaksChain) return; + // Look for cases where just the second value in the pair is used by + // both members of another pair (splatting). + for (Value::user_iterator I = P.second->user_begin(), + E = P.second->user_end(); + I != E; ++I) { + User *UI = *I; + if (isa<LoadInst>(UI)) + continue; + else if ((SI = dyn_cast<StoreInst>(UI)) && + P.second == SI->getPointerOperand()) + continue; + + for (Value::user_iterator J = P.second->user_begin(); J != E; ++J) { + User *UJ = *J; + if ((SJ = dyn_cast<StoreInst>(UJ)) && + P.second == SJ->getPointerOperand()) + continue; + + if (CandidatePairsSet.count(ValuePair(UI, UJ))) { + VPPair VP(P, ValuePair(UI, UJ)); + ConnectedPairs[VP.first].push_back(VP.second); + PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionSplat)); + } + } + } + } + + // This function figures out which pairs are connected. Two pairs are + // connected if some output of the first pair forms an input to both members + // of the second pair. + void BBVectorize::computeConnectedPairs( + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &CandidatePairsSet, + std::vector<Value *> &PairableInsts, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseMap<VPPair, unsigned> &PairConnectionTypes) { + for (std::vector<Value *>::iterator PI = PairableInsts.begin(), + PE = PairableInsts.end(); PI != PE; ++PI) { + DenseMap<Value *, std::vector<Value *> >::iterator PP = + CandidatePairs.find(*PI); + if (PP == CandidatePairs.end()) + continue; + + for (std::vector<Value *>::iterator P = PP->second.begin(), + E = PP->second.end(); P != E; ++P) + computePairsConnectedTo(CandidatePairs, CandidatePairsSet, + PairableInsts, ConnectedPairs, + PairConnectionTypes, ValuePair(*PI, *P)); + } + + DEBUG(size_t TotalPairs = 0; + for (DenseMap<ValuePair, std::vector<ValuePair> >::iterator I = + ConnectedPairs.begin(), IE = ConnectedPairs.end(); I != IE; ++I) + TotalPairs += I->second.size(); + dbgs() << "BBV: found " << TotalPairs + << " pair connections.\n"); + } + + // This function builds a set of use tuples such that <A, B> is in the set + // if B is in the use dag of A. If B is in the use dag of A, then B + // depends on the output of A. + void BBVectorize::buildDepMap( + BasicBlock &BB, + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + std::vector<Value *> &PairableInsts, + DenseSet<ValuePair> &PairableInstUsers) { + DenseSet<Value *> IsInPair; + for (DenseMap<Value *, std::vector<Value *> >::iterator C = + CandidatePairs.begin(), E = CandidatePairs.end(); C != E; ++C) { + IsInPair.insert(C->first); + IsInPair.insert(C->second.begin(), C->second.end()); + } + + // Iterate through the basic block, recording all users of each + // pairable instruction. + + BasicBlock::iterator E = BB.end(), EL = + BasicBlock::iterator(cast<Instruction>(PairableInsts.back())); + for (BasicBlock::iterator I = BB.getFirstInsertionPt(); I != E; ++I) { + if (IsInPair.find(&*I) == IsInPair.end()) + continue; + + DenseSet<Value *> Users; + AliasSetTracker WriteSet(*AA); + if (I->mayWriteToMemory()) + WriteSet.add(&*I); + + for (BasicBlock::iterator J = std::next(I); J != E; ++J) { + (void)trackUsesOfI(Users, WriteSet, &*I, &*J); + + if (J == EL) + break; + } + + for (DenseSet<Value *>::iterator U = Users.begin(), E = Users.end(); + U != E; ++U) { + if (IsInPair.find(*U) == IsInPair.end()) continue; + PairableInstUsers.insert(ValuePair(&*I, *U)); + } + + if (I == EL) + break; + } + } + + // Returns true if an input to pair P is an output of pair Q and also an + // input of pair Q is an output of pair P. If this is the case, then these + // two pairs cannot be simultaneously fused. + bool BBVectorize::pairsConflict(ValuePair P, ValuePair Q, + DenseSet<ValuePair> &PairableInstUsers, + DenseMap<ValuePair, std::vector<ValuePair> > *PairableInstUserMap, + DenseSet<VPPair> *PairableInstUserPairSet) { + // Two pairs are in conflict if they are mutual Users of eachother. + bool QUsesP = PairableInstUsers.count(ValuePair(P.first, Q.first)) || + PairableInstUsers.count(ValuePair(P.first, Q.second)) || + PairableInstUsers.count(ValuePair(P.second, Q.first)) || + PairableInstUsers.count(ValuePair(P.second, Q.second)); + bool PUsesQ = PairableInstUsers.count(ValuePair(Q.first, P.first)) || + PairableInstUsers.count(ValuePair(Q.first, P.second)) || + PairableInstUsers.count(ValuePair(Q.second, P.first)) || + PairableInstUsers.count(ValuePair(Q.second, P.second)); + if (PairableInstUserMap) { + // FIXME: The expensive part of the cycle check is not so much the cycle + // check itself but this edge insertion procedure. This needs some + // profiling and probably a different data structure. + if (PUsesQ) { + if (PairableInstUserPairSet->insert(VPPair(Q, P)).second) + (*PairableInstUserMap)[Q].push_back(P); + } + if (QUsesP) { + if (PairableInstUserPairSet->insert(VPPair(P, Q)).second) + (*PairableInstUserMap)[P].push_back(Q); + } + } + + return (QUsesP && PUsesQ); + } + + // This function walks the use graph of current pairs to see if, starting + // from P, the walk returns to P. + bool BBVectorize::pairWillFormCycle(ValuePair P, + DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUserMap, + DenseSet<ValuePair> &CurrentPairs) { + DEBUG(if (DebugCycleCheck) + dbgs() << "BBV: starting cycle check for : " << *P.first << " <-> " + << *P.second << "\n"); + // A lookup table of visisted pairs is kept because the PairableInstUserMap + // contains non-direct associations. + DenseSet<ValuePair> Visited; + SmallVector<ValuePair, 32> Q; + // General depth-first post-order traversal: + Q.push_back(P); + do { + ValuePair QTop = Q.pop_back_val(); + Visited.insert(QTop); + + DEBUG(if (DebugCycleCheck) + dbgs() << "BBV: cycle check visiting: " << *QTop.first << " <-> " + << *QTop.second << "\n"); + DenseMap<ValuePair, std::vector<ValuePair> >::iterator QQ = + PairableInstUserMap.find(QTop); + if (QQ == PairableInstUserMap.end()) + continue; + + for (std::vector<ValuePair>::iterator C = QQ->second.begin(), + CE = QQ->second.end(); C != CE; ++C) { + if (*C == P) { + DEBUG(dbgs() + << "BBV: rejected to prevent non-trivial cycle formation: " + << QTop.first << " <-> " << C->second << "\n"); + return true; + } + + if (CurrentPairs.count(*C) && !Visited.count(*C)) + Q.push_back(*C); + } + } while (!Q.empty()); + + return false; + } + + // This function builds the initial dag of connected pairs with the + // pair J at the root. + void BBVectorize::buildInitialDAGFor( + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &CandidatePairsSet, + std::vector<Value *> &PairableInsts, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseSet<ValuePair> &PairableInstUsers, + DenseMap<Value *, Value *> &ChosenPairs, + DenseMap<ValuePair, size_t> &DAG, ValuePair J) { + // Each of these pairs is viewed as the root node of a DAG. The DAG + // is then walked (depth-first). As this happens, we keep track of + // the pairs that compose the DAG and the maximum depth of the DAG. + SmallVector<ValuePairWithDepth, 32> Q; + // General depth-first post-order traversal: + Q.push_back(ValuePairWithDepth(J, getDepthFactor(J.first))); + do { + ValuePairWithDepth QTop = Q.back(); + + // Push each child onto the queue: + bool MoreChildren = false; + size_t MaxChildDepth = QTop.second; + DenseMap<ValuePair, std::vector<ValuePair> >::iterator QQ = + ConnectedPairs.find(QTop.first); + if (QQ != ConnectedPairs.end()) + for (std::vector<ValuePair>::iterator k = QQ->second.begin(), + ke = QQ->second.end(); k != ke; ++k) { + // Make sure that this child pair is still a candidate: + if (CandidatePairsSet.count(*k)) { + DenseMap<ValuePair, size_t>::iterator C = DAG.find(*k); + if (C == DAG.end()) { + size_t d = getDepthFactor(k->first); + Q.push_back(ValuePairWithDepth(*k, QTop.second+d)); + MoreChildren = true; + } else { + MaxChildDepth = std::max(MaxChildDepth, C->second); + } + } + } + + if (!MoreChildren) { + // Record the current pair as part of the DAG: + DAG.insert(ValuePairWithDepth(QTop.first, MaxChildDepth)); + Q.pop_back(); + } + } while (!Q.empty()); + } + + // Given some initial dag, prune it by removing conflicting pairs (pairs + // that cannot be simultaneously chosen for vectorization). + void BBVectorize::pruneDAGFor( + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + std::vector<Value *> &PairableInsts, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseSet<ValuePair> &PairableInstUsers, + DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUserMap, + DenseSet<VPPair> &PairableInstUserPairSet, + DenseMap<Value *, Value *> &ChosenPairs, + DenseMap<ValuePair, size_t> &DAG, + DenseSet<ValuePair> &PrunedDAG, ValuePair J, + bool UseCycleCheck) { + SmallVector<ValuePairWithDepth, 32> Q; + // General depth-first post-order traversal: + Q.push_back(ValuePairWithDepth(J, getDepthFactor(J.first))); + do { + ValuePairWithDepth QTop = Q.pop_back_val(); + PrunedDAG.insert(QTop.first); + + // Visit each child, pruning as necessary... + SmallVector<ValuePairWithDepth, 8> BestChildren; + DenseMap<ValuePair, std::vector<ValuePair> >::iterator QQ = + ConnectedPairs.find(QTop.first); + if (QQ == ConnectedPairs.end()) + continue; + + for (std::vector<ValuePair>::iterator K = QQ->second.begin(), + KE = QQ->second.end(); K != KE; ++K) { + DenseMap<ValuePair, size_t>::iterator C = DAG.find(*K); + if (C == DAG.end()) continue; + + // This child is in the DAG, now we need to make sure it is the + // best of any conflicting children. There could be multiple + // conflicting children, so first, determine if we're keeping + // this child, then delete conflicting children as necessary. + + // It is also necessary to guard against pairing-induced + // dependencies. Consider instructions a .. x .. y .. b + // such that (a,b) are to be fused and (x,y) are to be fused + // but a is an input to x and b is an output from y. This + // means that y cannot be moved after b but x must be moved + // after b for (a,b) to be fused. In other words, after + // fusing (a,b) we have y .. a/b .. x where y is an input + // to a/b and x is an output to a/b: x and y can no longer + // be legally fused. To prevent this condition, we must + // make sure that a child pair added to the DAG is not + // both an input and output of an already-selected pair. + + // Pairing-induced dependencies can also form from more complicated + // cycles. The pair vs. pair conflicts are easy to check, and so + // that is done explicitly for "fast rejection", and because for + // child vs. child conflicts, we may prefer to keep the current + // pair in preference to the already-selected child. + DenseSet<ValuePair> CurrentPairs; + + bool CanAdd = true; + for (SmallVectorImpl<ValuePairWithDepth>::iterator C2 + = BestChildren.begin(), E2 = BestChildren.end(); + C2 != E2; ++C2) { + if (C2->first.first == C->first.first || + C2->first.first == C->first.second || + C2->first.second == C->first.first || + C2->first.second == C->first.second || + pairsConflict(C2->first, C->first, PairableInstUsers, + UseCycleCheck ? &PairableInstUserMap : nullptr, + UseCycleCheck ? &PairableInstUserPairSet + : nullptr)) { + if (C2->second >= C->second) { + CanAdd = false; + break; + } + + CurrentPairs.insert(C2->first); + } + } + if (!CanAdd) continue; + + // Even worse, this child could conflict with another node already + // selected for the DAG. If that is the case, ignore this child. + for (DenseSet<ValuePair>::iterator T = PrunedDAG.begin(), + E2 = PrunedDAG.end(); T != E2; ++T) { + if (T->first == C->first.first || + T->first == C->first.second || + T->second == C->first.first || + T->second == C->first.second || + pairsConflict(*T, C->first, PairableInstUsers, + UseCycleCheck ? &PairableInstUserMap : nullptr, + UseCycleCheck ? &PairableInstUserPairSet + : nullptr)) { + CanAdd = false; + break; + } + + CurrentPairs.insert(*T); + } + if (!CanAdd) continue; + + // And check the queue too... + for (SmallVectorImpl<ValuePairWithDepth>::iterator C2 = Q.begin(), + E2 = Q.end(); C2 != E2; ++C2) { + if (C2->first.first == C->first.first || + C2->first.first == C->first.second || + C2->first.second == C->first.first || + C2->first.second == C->first.second || + pairsConflict(C2->first, C->first, PairableInstUsers, + UseCycleCheck ? &PairableInstUserMap : nullptr, + UseCycleCheck ? &PairableInstUserPairSet + : nullptr)) { + CanAdd = false; + break; + } + + CurrentPairs.insert(C2->first); + } + if (!CanAdd) continue; + + // Last but not least, check for a conflict with any of the + // already-chosen pairs. + for (DenseMap<Value *, Value *>::iterator C2 = + ChosenPairs.begin(), E2 = ChosenPairs.end(); + C2 != E2; ++C2) { + if (pairsConflict(*C2, C->first, PairableInstUsers, + UseCycleCheck ? &PairableInstUserMap : nullptr, + UseCycleCheck ? &PairableInstUserPairSet + : nullptr)) { + CanAdd = false; + break; + } + + CurrentPairs.insert(*C2); + } + if (!CanAdd) continue; + + // To check for non-trivial cycles formed by the addition of the + // current pair we've formed a list of all relevant pairs, now use a + // graph walk to check for a cycle. We start from the current pair and + // walk the use dag to see if we again reach the current pair. If we + // do, then the current pair is rejected. + + // FIXME: It may be more efficient to use a topological-ordering + // algorithm to improve the cycle check. This should be investigated. + if (UseCycleCheck && + pairWillFormCycle(C->first, PairableInstUserMap, CurrentPairs)) + continue; + + // This child can be added, but we may have chosen it in preference + // to an already-selected child. Check for this here, and if a + // conflict is found, then remove the previously-selected child + // before adding this one in its place. + for (SmallVectorImpl<ValuePairWithDepth>::iterator C2 + = BestChildren.begin(); C2 != BestChildren.end();) { + if (C2->first.first == C->first.first || + C2->first.first == C->first.second || + C2->first.second == C->first.first || + C2->first.second == C->first.second || + pairsConflict(C2->first, C->first, PairableInstUsers)) + C2 = BestChildren.erase(C2); + else + ++C2; + } + + BestChildren.push_back(ValuePairWithDepth(C->first, C->second)); + } + + for (SmallVectorImpl<ValuePairWithDepth>::iterator C + = BestChildren.begin(), E2 = BestChildren.end(); + C != E2; ++C) { + size_t DepthF = getDepthFactor(C->first.first); + Q.push_back(ValuePairWithDepth(C->first, QTop.second+DepthF)); + } + } while (!Q.empty()); + } + + // This function finds the best dag of mututally-compatible connected + // pairs, given the choice of root pairs as an iterator range. + void BBVectorize::findBestDAGFor( + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &CandidatePairsSet, + DenseMap<ValuePair, int> &CandidatePairCostSavings, + std::vector<Value *> &PairableInsts, + DenseSet<ValuePair> &FixedOrderPairs, + DenseMap<VPPair, unsigned> &PairConnectionTypes, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps, + DenseSet<ValuePair> &PairableInstUsers, + DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUserMap, + DenseSet<VPPair> &PairableInstUserPairSet, + DenseMap<Value *, Value *> &ChosenPairs, + DenseSet<ValuePair> &BestDAG, size_t &BestMaxDepth, + int &BestEffSize, Value *II, std::vector<Value *>&JJ, + bool UseCycleCheck) { + for (std::vector<Value *>::iterator J = JJ.begin(), JE = JJ.end(); + J != JE; ++J) { + ValuePair IJ(II, *J); + if (!CandidatePairsSet.count(IJ)) + continue; + + // Before going any further, make sure that this pair does not + // conflict with any already-selected pairs (see comment below + // near the DAG pruning for more details). + DenseSet<ValuePair> ChosenPairSet; + bool DoesConflict = false; + for (DenseMap<Value *, Value *>::iterator C = ChosenPairs.begin(), + E = ChosenPairs.end(); C != E; ++C) { + if (pairsConflict(*C, IJ, PairableInstUsers, + UseCycleCheck ? &PairableInstUserMap : nullptr, + UseCycleCheck ? &PairableInstUserPairSet : nullptr)) { + DoesConflict = true; + break; + } + + ChosenPairSet.insert(*C); + } + if (DoesConflict) continue; + + if (UseCycleCheck && + pairWillFormCycle(IJ, PairableInstUserMap, ChosenPairSet)) + continue; + + DenseMap<ValuePair, size_t> DAG; + buildInitialDAGFor(CandidatePairs, CandidatePairsSet, + PairableInsts, ConnectedPairs, + PairableInstUsers, ChosenPairs, DAG, IJ); + + // Because we'll keep the child with the largest depth, the largest + // depth is still the same in the unpruned DAG. + size_t MaxDepth = DAG.lookup(IJ); + + DEBUG(if (DebugPairSelection) dbgs() << "BBV: found DAG for pair {" + << *IJ.first << " <-> " << *IJ.second << "} of depth " << + MaxDepth << " and size " << DAG.size() << "\n"); + + // At this point the DAG has been constructed, but, may contain + // contradictory children (meaning that different children of + // some dag node may be attempting to fuse the same instruction). + // So now we walk the dag again, in the case of a conflict, + // keep only the child with the largest depth. To break a tie, + // favor the first child. + + DenseSet<ValuePair> PrunedDAG; + pruneDAGFor(CandidatePairs, PairableInsts, ConnectedPairs, + PairableInstUsers, PairableInstUserMap, + PairableInstUserPairSet, + ChosenPairs, DAG, PrunedDAG, IJ, UseCycleCheck); + + int EffSize = 0; + if (TTI) { + DenseSet<Value *> PrunedDAGInstrs; + for (DenseSet<ValuePair>::iterator S = PrunedDAG.begin(), + E = PrunedDAG.end(); S != E; ++S) { + PrunedDAGInstrs.insert(S->first); + PrunedDAGInstrs.insert(S->second); + } + + // The set of pairs that have already contributed to the total cost. + DenseSet<ValuePair> IncomingPairs; + + // If the cost model were perfect, this might not be necessary; but we + // need to make sure that we don't get stuck vectorizing our own + // shuffle chains. + bool HasNontrivialInsts = false; + + // The node weights represent the cost savings associated with + // fusing the pair of instructions. + for (DenseSet<ValuePair>::iterator S = PrunedDAG.begin(), + E = PrunedDAG.end(); S != E; ++S) { + if (!isa<ShuffleVectorInst>(S->first) && + !isa<InsertElementInst>(S->first) && + !isa<ExtractElementInst>(S->first)) + HasNontrivialInsts = true; + + bool FlipOrder = false; + + if (getDepthFactor(S->first)) { + int ESContrib = CandidatePairCostSavings.find(*S)->second; + DEBUG(if (DebugPairSelection) dbgs() << "\tweight {" + << *S->first << " <-> " << *S->second << "} = " << + ESContrib << "\n"); + EffSize += ESContrib; + } + + // The edge weights contribute in a negative sense: they represent + // the cost of shuffles. + DenseMap<ValuePair, std::vector<ValuePair> >::iterator SS = + ConnectedPairDeps.find(*S); + if (SS != ConnectedPairDeps.end()) { + unsigned NumDepsDirect = 0, NumDepsSwap = 0; + for (std::vector<ValuePair>::iterator T = SS->second.begin(), + TE = SS->second.end(); T != TE; ++T) { + VPPair Q(*S, *T); + if (!PrunedDAG.count(Q.second)) + continue; + DenseMap<VPPair, unsigned>::iterator R = + PairConnectionTypes.find(VPPair(Q.second, Q.first)); + assert(R != PairConnectionTypes.end() && + "Cannot find pair connection type"); + if (R->second == PairConnectionDirect) + ++NumDepsDirect; + else if (R->second == PairConnectionSwap) + ++NumDepsSwap; + } + + // If there are more swaps than direct connections, then + // the pair order will be flipped during fusion. So the real + // number of swaps is the minimum number. + FlipOrder = !FixedOrderPairs.count(*S) && + ((NumDepsSwap > NumDepsDirect) || + FixedOrderPairs.count(ValuePair(S->second, S->first))); + + for (std::vector<ValuePair>::iterator T = SS->second.begin(), + TE = SS->second.end(); T != TE; ++T) { + VPPair Q(*S, *T); + if (!PrunedDAG.count(Q.second)) + continue; + DenseMap<VPPair, unsigned>::iterator R = + PairConnectionTypes.find(VPPair(Q.second, Q.first)); + assert(R != PairConnectionTypes.end() && + "Cannot find pair connection type"); + Type *Ty1 = Q.second.first->getType(), + *Ty2 = Q.second.second->getType(); + Type *VTy = getVecTypeForPair(Ty1, Ty2); + if ((R->second == PairConnectionDirect && FlipOrder) || + (R->second == PairConnectionSwap && !FlipOrder) || + R->second == PairConnectionSplat) { + int ESContrib = (int) getInstrCost(Instruction::ShuffleVector, + VTy, VTy); + + if (VTy->getVectorNumElements() == 2) { + if (R->second == PairConnectionSplat) + ESContrib = std::min(ESContrib, (int) TTI->getShuffleCost( + TargetTransformInfo::SK_Broadcast, VTy)); + else + ESContrib = std::min(ESContrib, (int) TTI->getShuffleCost( + TargetTransformInfo::SK_Reverse, VTy)); + } + + DEBUG(if (DebugPairSelection) dbgs() << "\tcost {" << + *Q.second.first << " <-> " << *Q.second.second << + "} -> {" << + *S->first << " <-> " << *S->second << "} = " << + ESContrib << "\n"); + EffSize -= ESContrib; + } + } + } + + // Compute the cost of outgoing edges. We assume that edges outgoing + // to shuffles, inserts or extracts can be merged, and so contribute + // no additional cost. + if (!S->first->getType()->isVoidTy()) { + Type *Ty1 = S->first->getType(), + *Ty2 = S->second->getType(); + Type *VTy = getVecTypeForPair(Ty1, Ty2); + + bool NeedsExtraction = false; + for (User *U : S->first->users()) { + if (ShuffleVectorInst *SI = dyn_cast<ShuffleVectorInst>(U)) { + // Shuffle can be folded if it has no other input + if (isa<UndefValue>(SI->getOperand(1))) + continue; + } + if (isa<ExtractElementInst>(U)) + continue; + if (PrunedDAGInstrs.count(U)) + continue; + NeedsExtraction = true; + break; + } + + if (NeedsExtraction) { + int ESContrib; + if (Ty1->isVectorTy()) { + ESContrib = (int) getInstrCost(Instruction::ShuffleVector, + Ty1, VTy); + ESContrib = std::min(ESContrib, (int) TTI->getShuffleCost( + TargetTransformInfo::SK_ExtractSubvector, VTy, 0, Ty1)); + } else + ESContrib = (int) TTI->getVectorInstrCost( + Instruction::ExtractElement, VTy, 0); + + DEBUG(if (DebugPairSelection) dbgs() << "\tcost {" << + *S->first << "} = " << ESContrib << "\n"); + EffSize -= ESContrib; + } + + NeedsExtraction = false; + for (User *U : S->second->users()) { + if (ShuffleVectorInst *SI = dyn_cast<ShuffleVectorInst>(U)) { + // Shuffle can be folded if it has no other input + if (isa<UndefValue>(SI->getOperand(1))) + continue; + } + if (isa<ExtractElementInst>(U)) + continue; + if (PrunedDAGInstrs.count(U)) + continue; + NeedsExtraction = true; + break; + } + + if (NeedsExtraction) { + int ESContrib; + if (Ty2->isVectorTy()) { + ESContrib = (int) getInstrCost(Instruction::ShuffleVector, + Ty2, VTy); + ESContrib = std::min(ESContrib, (int) TTI->getShuffleCost( + TargetTransformInfo::SK_ExtractSubvector, VTy, + Ty1->isVectorTy() ? Ty1->getVectorNumElements() : 1, Ty2)); + } else + ESContrib = (int) TTI->getVectorInstrCost( + Instruction::ExtractElement, VTy, 1); + DEBUG(if (DebugPairSelection) dbgs() << "\tcost {" << + *S->second << "} = " << ESContrib << "\n"); + EffSize -= ESContrib; + } + } + + // Compute the cost of incoming edges. + if (!isa<LoadInst>(S->first) && !isa<StoreInst>(S->first)) { + Instruction *S1 = cast<Instruction>(S->first), + *S2 = cast<Instruction>(S->second); + for (unsigned o = 0; o < S1->getNumOperands(); ++o) { + Value *O1 = S1->getOperand(o), *O2 = S2->getOperand(o); + + // Combining constants into vector constants (or small vector + // constants into larger ones are assumed free). + if (isa<Constant>(O1) && isa<Constant>(O2)) + continue; + + if (FlipOrder) + std::swap(O1, O2); + + ValuePair VP = ValuePair(O1, O2); + ValuePair VPR = ValuePair(O2, O1); + + // Internal edges are not handled here. + if (PrunedDAG.count(VP) || PrunedDAG.count(VPR)) + continue; + + Type *Ty1 = O1->getType(), + *Ty2 = O2->getType(); + Type *VTy = getVecTypeForPair(Ty1, Ty2); + + // Combining vector operations of the same type is also assumed + // folded with other operations. + if (Ty1 == Ty2) { + // If both are insert elements, then both can be widened. + InsertElementInst *IEO1 = dyn_cast<InsertElementInst>(O1), + *IEO2 = dyn_cast<InsertElementInst>(O2); + if (IEO1 && IEO2 && isPureIEChain(IEO1) && isPureIEChain(IEO2)) + continue; + // If both are extract elements, and both have the same input + // type, then they can be replaced with a shuffle + ExtractElementInst *EIO1 = dyn_cast<ExtractElementInst>(O1), + *EIO2 = dyn_cast<ExtractElementInst>(O2); + if (EIO1 && EIO2 && + EIO1->getOperand(0)->getType() == + EIO2->getOperand(0)->getType()) + continue; + // If both are a shuffle with equal operand types and only two + // unqiue operands, then they can be replaced with a single + // shuffle + ShuffleVectorInst *SIO1 = dyn_cast<ShuffleVectorInst>(O1), + *SIO2 = dyn_cast<ShuffleVectorInst>(O2); + if (SIO1 && SIO2 && + SIO1->getOperand(0)->getType() == + SIO2->getOperand(0)->getType()) { + SmallSet<Value *, 4> SIOps; + SIOps.insert(SIO1->getOperand(0)); + SIOps.insert(SIO1->getOperand(1)); + SIOps.insert(SIO2->getOperand(0)); + SIOps.insert(SIO2->getOperand(1)); + if (SIOps.size() <= 2) + continue; + } + } + + int ESContrib; + // This pair has already been formed. + if (IncomingPairs.count(VP)) { + continue; + } else if (IncomingPairs.count(VPR)) { + ESContrib = (int) getInstrCost(Instruction::ShuffleVector, + VTy, VTy); + + if (VTy->getVectorNumElements() == 2) + ESContrib = std::min(ESContrib, (int) TTI->getShuffleCost( + TargetTransformInfo::SK_Reverse, VTy)); + } else if (!Ty1->isVectorTy() && !Ty2->isVectorTy()) { + ESContrib = (int) TTI->getVectorInstrCost( + Instruction::InsertElement, VTy, 0); + ESContrib += (int) TTI->getVectorInstrCost( + Instruction::InsertElement, VTy, 1); + } else if (!Ty1->isVectorTy()) { + // O1 needs to be inserted into a vector of size O2, and then + // both need to be shuffled together. + ESContrib = (int) TTI->getVectorInstrCost( + Instruction::InsertElement, Ty2, 0); + ESContrib += (int) getInstrCost(Instruction::ShuffleVector, + VTy, Ty2); + } else if (!Ty2->isVectorTy()) { + // O2 needs to be inserted into a vector of size O1, and then + // both need to be shuffled together. + ESContrib = (int) TTI->getVectorInstrCost( + Instruction::InsertElement, Ty1, 0); + ESContrib += (int) getInstrCost(Instruction::ShuffleVector, + VTy, Ty1); + } else { + Type *TyBig = Ty1, *TySmall = Ty2; + if (Ty2->getVectorNumElements() > Ty1->getVectorNumElements()) + std::swap(TyBig, TySmall); + + ESContrib = (int) getInstrCost(Instruction::ShuffleVector, + VTy, TyBig); + if (TyBig != TySmall) + ESContrib += (int) getInstrCost(Instruction::ShuffleVector, + TyBig, TySmall); + } + + DEBUG(if (DebugPairSelection) dbgs() << "\tcost {" + << *O1 << " <-> " << *O2 << "} = " << + ESContrib << "\n"); + EffSize -= ESContrib; + IncomingPairs.insert(VP); + } + } + } + + if (!HasNontrivialInsts) { + DEBUG(if (DebugPairSelection) dbgs() << + "\tNo non-trivial instructions in DAG;" + " override to zero effective size\n"); + EffSize = 0; + } + } else { + for (DenseSet<ValuePair>::iterator S = PrunedDAG.begin(), + E = PrunedDAG.end(); S != E; ++S) + EffSize += (int) getDepthFactor(S->first); + } + + DEBUG(if (DebugPairSelection) + dbgs() << "BBV: found pruned DAG for pair {" + << *IJ.first << " <-> " << *IJ.second << "} of depth " << + MaxDepth << " and size " << PrunedDAG.size() << + " (effective size: " << EffSize << ")\n"); + if (((TTI && !UseChainDepthWithTI) || + MaxDepth >= Config.ReqChainDepth) && + EffSize > 0 && EffSize > BestEffSize) { + BestMaxDepth = MaxDepth; + BestEffSize = EffSize; + BestDAG = PrunedDAG; + } + } + } + + // Given the list of candidate pairs, this function selects those + // that will be fused into vector instructions. + void BBVectorize::choosePairs( + DenseMap<Value *, std::vector<Value *> > &CandidatePairs, + DenseSet<ValuePair> &CandidatePairsSet, + DenseMap<ValuePair, int> &CandidatePairCostSavings, + std::vector<Value *> &PairableInsts, + DenseSet<ValuePair> &FixedOrderPairs, + DenseMap<VPPair, unsigned> &PairConnectionTypes, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps, + DenseSet<ValuePair> &PairableInstUsers, + DenseMap<Value *, Value *>& ChosenPairs) { + bool UseCycleCheck = + CandidatePairsSet.size() <= Config.MaxCandPairsForCycleCheck; + + DenseMap<Value *, std::vector<Value *> > CandidatePairs2; + for (DenseSet<ValuePair>::iterator I = CandidatePairsSet.begin(), + E = CandidatePairsSet.end(); I != E; ++I) { + std::vector<Value *> &JJ = CandidatePairs2[I->second]; + if (JJ.empty()) JJ.reserve(32); + JJ.push_back(I->first); + } + + DenseMap<ValuePair, std::vector<ValuePair> > PairableInstUserMap; + DenseSet<VPPair> PairableInstUserPairSet; + for (std::vector<Value *>::iterator I = PairableInsts.begin(), + E = PairableInsts.end(); I != E; ++I) { + // The number of possible pairings for this variable: + size_t NumChoices = CandidatePairs.lookup(*I).size(); + if (!NumChoices) continue; + + std::vector<Value *> &JJ = CandidatePairs[*I]; + + // The best pair to choose and its dag: + size_t BestMaxDepth = 0; + int BestEffSize = 0; + DenseSet<ValuePair> BestDAG; + findBestDAGFor(CandidatePairs, CandidatePairsSet, + CandidatePairCostSavings, + PairableInsts, FixedOrderPairs, PairConnectionTypes, + ConnectedPairs, ConnectedPairDeps, + PairableInstUsers, PairableInstUserMap, + PairableInstUserPairSet, ChosenPairs, + BestDAG, BestMaxDepth, BestEffSize, *I, JJ, + UseCycleCheck); + + if (BestDAG.empty()) + continue; + + // A dag has been chosen (or not) at this point. If no dag was + // chosen, then this instruction, I, cannot be paired (and is no longer + // considered). + + DEBUG(dbgs() << "BBV: selected pairs in the best DAG for: " + << *cast<Instruction>(*I) << "\n"); + + for (DenseSet<ValuePair>::iterator S = BestDAG.begin(), + SE2 = BestDAG.end(); S != SE2; ++S) { + // Insert the members of this dag into the list of chosen pairs. + ChosenPairs.insert(ValuePair(S->first, S->second)); + DEBUG(dbgs() << "BBV: selected pair: " << *S->first << " <-> " << + *S->second << "\n"); + + // Remove all candidate pairs that have values in the chosen dag. + std::vector<Value *> &KK = CandidatePairs[S->first]; + for (std::vector<Value *>::iterator K = KK.begin(), KE = KK.end(); + K != KE; ++K) { + if (*K == S->second) + continue; + + CandidatePairsSet.erase(ValuePair(S->first, *K)); + } + + std::vector<Value *> &LL = CandidatePairs2[S->second]; + for (std::vector<Value *>::iterator L = LL.begin(), LE = LL.end(); + L != LE; ++L) { + if (*L == S->first) + continue; + + CandidatePairsSet.erase(ValuePair(*L, S->second)); + } + + std::vector<Value *> &MM = CandidatePairs[S->second]; + for (std::vector<Value *>::iterator M = MM.begin(), ME = MM.end(); + M != ME; ++M) { + assert(*M != S->first && "Flipped pair in candidate list?"); + CandidatePairsSet.erase(ValuePair(S->second, *M)); + } + + std::vector<Value *> &NN = CandidatePairs2[S->first]; + for (std::vector<Value *>::iterator N = NN.begin(), NE = NN.end(); + N != NE; ++N) { + assert(*N != S->second && "Flipped pair in candidate list?"); + CandidatePairsSet.erase(ValuePair(*N, S->first)); + } + } + } + + DEBUG(dbgs() << "BBV: selected " << ChosenPairs.size() << " pairs.\n"); + } + + std::string getReplacementName(Instruction *I, bool IsInput, unsigned o, + unsigned n = 0) { + if (!I->hasName()) + return ""; + + return (I->getName() + (IsInput ? ".v.i" : ".v.r") + utostr(o) + + (n > 0 ? "." + utostr(n) : "")).str(); + } + + // Returns the value that is to be used as the pointer input to the vector + // instruction that fuses I with J. + Value *BBVectorize::getReplacementPointerInput(LLVMContext& Context, + Instruction *I, Instruction *J, unsigned o) { + Value *IPtr, *JPtr; + unsigned IAlignment, JAlignment, IAddressSpace, JAddressSpace; + int64_t OffsetInElmts; + + // Note: the analysis might fail here, that is why the pair order has + // been precomputed (OffsetInElmts must be unused here). + (void) getPairPtrInfo(I, J, IPtr, JPtr, IAlignment, JAlignment, + IAddressSpace, JAddressSpace, + OffsetInElmts, false); + + // The pointer value is taken to be the one with the lowest offset. + Value *VPtr = IPtr; + + Type *ArgTypeI = IPtr->getType()->getPointerElementType(); + Type *ArgTypeJ = JPtr->getType()->getPointerElementType(); + Type *VArgType = getVecTypeForPair(ArgTypeI, ArgTypeJ); + Type *VArgPtrType + = PointerType::get(VArgType, + IPtr->getType()->getPointerAddressSpace()); + return new BitCastInst(VPtr, VArgPtrType, getReplacementName(I, true, o), + /* insert before */ I); + } + + void BBVectorize::fillNewShuffleMask(LLVMContext& Context, Instruction *J, + unsigned MaskOffset, unsigned NumInElem, + unsigned NumInElem1, unsigned IdxOffset, + std::vector<Constant*> &Mask) { + unsigned NumElem1 = J->getType()->getVectorNumElements(); + for (unsigned v = 0; v < NumElem1; ++v) { + int m = cast<ShuffleVectorInst>(J)->getMaskValue(v); + if (m < 0) { + Mask[v+MaskOffset] = UndefValue::get(Type::getInt32Ty(Context)); + } else { + unsigned mm = m + (int) IdxOffset; + if (m >= (int) NumInElem1) + mm += (int) NumInElem; + + Mask[v+MaskOffset] = + ConstantInt::get(Type::getInt32Ty(Context), mm); + } + } + } + + // Returns the value that is to be used as the vector-shuffle mask to the + // vector instruction that fuses I with J. + Value *BBVectorize::getReplacementShuffleMask(LLVMContext& Context, + Instruction *I, Instruction *J) { + // This is the shuffle mask. We need to append the second + // mask to the first, and the numbers need to be adjusted. + + Type *ArgTypeI = I->getType(); + Type *ArgTypeJ = J->getType(); + Type *VArgType = getVecTypeForPair(ArgTypeI, ArgTypeJ); + + unsigned NumElemI = ArgTypeI->getVectorNumElements(); + + // Get the total number of elements in the fused vector type. + // By definition, this must equal the number of elements in + // the final mask. + unsigned NumElem = VArgType->getVectorNumElements(); + std::vector<Constant*> Mask(NumElem); + + Type *OpTypeI = I->getOperand(0)->getType(); + unsigned NumInElemI = OpTypeI->getVectorNumElements(); + Type *OpTypeJ = J->getOperand(0)->getType(); + unsigned NumInElemJ = OpTypeJ->getVectorNumElements(); + + // The fused vector will be: + // ----------------------------------------------------- + // | NumInElemI | NumInElemJ | NumInElemI | NumInElemJ | + // ----------------------------------------------------- + // from which we'll extract NumElem total elements (where the first NumElemI + // of them come from the mask in I and the remainder come from the mask + // in J. + + // For the mask from the first pair... + fillNewShuffleMask(Context, I, 0, NumInElemJ, NumInElemI, + 0, Mask); + + // For the mask from the second pair... + fillNewShuffleMask(Context, J, NumElemI, NumInElemI, NumInElemJ, + NumInElemI, Mask); + + return ConstantVector::get(Mask); + } + + bool BBVectorize::expandIEChain(LLVMContext& Context, Instruction *I, + Instruction *J, unsigned o, Value *&LOp, + unsigned numElemL, + Type *ArgTypeL, Type *ArgTypeH, + bool IBeforeJ, unsigned IdxOff) { + bool ExpandedIEChain = false; + if (InsertElementInst *LIE = dyn_cast<InsertElementInst>(LOp)) { + // If we have a pure insertelement chain, then this can be rewritten + // into a chain that directly builds the larger type. + if (isPureIEChain(LIE)) { + SmallVector<Value *, 8> VectElemts(numElemL, + UndefValue::get(ArgTypeL->getScalarType())); + InsertElementInst *LIENext = LIE; + do { + unsigned Idx = + cast<ConstantInt>(LIENext->getOperand(2))->getSExtValue(); + VectElemts[Idx] = LIENext->getOperand(1); + } while ((LIENext = + dyn_cast<InsertElementInst>(LIENext->getOperand(0)))); + + LIENext = nullptr; + Value *LIEPrev = UndefValue::get(ArgTypeH); + for (unsigned i = 0; i < numElemL; ++i) { + if (isa<UndefValue>(VectElemts[i])) continue; + LIENext = InsertElementInst::Create(LIEPrev, VectElemts[i], + ConstantInt::get(Type::getInt32Ty(Context), + i + IdxOff), + getReplacementName(IBeforeJ ? I : J, + true, o, i+1)); + LIENext->insertBefore(IBeforeJ ? J : I); + LIEPrev = LIENext; + } + + LOp = LIENext ? (Value*) LIENext : UndefValue::get(ArgTypeH); + ExpandedIEChain = true; + } + } + + return ExpandedIEChain; + } + + static unsigned getNumScalarElements(Type *Ty) { + if (VectorType *VecTy = dyn_cast<VectorType>(Ty)) + return VecTy->getNumElements(); + return 1; + } + + // Returns the value to be used as the specified operand of the vector + // instruction that fuses I with J. + Value *BBVectorize::getReplacementInput(LLVMContext& Context, Instruction *I, + Instruction *J, unsigned o, bool IBeforeJ) { + Value *CV0 = ConstantInt::get(Type::getInt32Ty(Context), 0); + Value *CV1 = ConstantInt::get(Type::getInt32Ty(Context), 1); + + // Compute the fused vector type for this operand + Type *ArgTypeI = I->getOperand(o)->getType(); + Type *ArgTypeJ = J->getOperand(o)->getType(); + VectorType *VArgType = getVecTypeForPair(ArgTypeI, ArgTypeJ); + + Instruction *L = I, *H = J; + Type *ArgTypeL = ArgTypeI, *ArgTypeH = ArgTypeJ; + + unsigned numElemL = getNumScalarElements(ArgTypeL); + unsigned numElemH = getNumScalarElements(ArgTypeH); + + Value *LOp = L->getOperand(o); + Value *HOp = H->getOperand(o); + unsigned numElem = VArgType->getNumElements(); + + // First, we check if we can reuse the "original" vector outputs (if these + // exist). We might need a shuffle. + ExtractElementInst *LEE = dyn_cast<ExtractElementInst>(LOp); + ExtractElementInst *HEE = dyn_cast<ExtractElementInst>(HOp); + ShuffleVectorInst *LSV = dyn_cast<ShuffleVectorInst>(LOp); + ShuffleVectorInst *HSV = dyn_cast<ShuffleVectorInst>(HOp); + + // FIXME: If we're fusing shuffle instructions, then we can't apply this + // optimization. The input vectors to the shuffle might be a different + // length from the shuffle outputs. Unfortunately, the replacement + // shuffle mask has already been formed, and the mask entries are sensitive + // to the sizes of the inputs. + bool IsSizeChangeShuffle = + isa<ShuffleVectorInst>(L) && + (LOp->getType() != L->getType() || HOp->getType() != H->getType()); + + if ((LEE || LSV) && (HEE || HSV) && !IsSizeChangeShuffle) { + // We can have at most two unique vector inputs. + bool CanUseInputs = true; + Value *I1, *I2 = nullptr; + if (LEE) { + I1 = LEE->getOperand(0); + } else { + I1 = LSV->getOperand(0); + I2 = LSV->getOperand(1); + if (I2 == I1 || isa<UndefValue>(I2)) + I2 = nullptr; + } + + if (HEE) { + Value *I3 = HEE->getOperand(0); + if (!I2 && I3 != I1) + I2 = I3; + else if (I3 != I1 && I3 != I2) + CanUseInputs = false; + } else { + Value *I3 = HSV->getOperand(0); + if (!I2 && I3 != I1) + I2 = I3; + else if (I3 != I1 && I3 != I2) + CanUseInputs = false; + + if (CanUseInputs) { + Value *I4 = HSV->getOperand(1); + if (!isa<UndefValue>(I4)) { + if (!I2 && I4 != I1) + I2 = I4; + else if (I4 != I1 && I4 != I2) + CanUseInputs = false; + } + } + } + + if (CanUseInputs) { + unsigned LOpElem = + cast<Instruction>(LOp)->getOperand(0)->getType() + ->getVectorNumElements(); + + unsigned HOpElem = + cast<Instruction>(HOp)->getOperand(0)->getType() + ->getVectorNumElements(); + + // We have one or two input vectors. We need to map each index of the + // operands to the index of the original vector. + SmallVector<std::pair<int, int>, 8> II(numElem); + for (unsigned i = 0; i < numElemL; ++i) { + int Idx, INum; + if (LEE) { + Idx = + cast<ConstantInt>(LEE->getOperand(1))->getSExtValue(); + INum = LEE->getOperand(0) == I1 ? 0 : 1; + } else { + Idx = LSV->getMaskValue(i); + if (Idx < (int) LOpElem) { + INum = LSV->getOperand(0) == I1 ? 0 : 1; + } else { + Idx -= LOpElem; + INum = LSV->getOperand(1) == I1 ? 0 : 1; + } + } + + II[i] = std::pair<int, int>(Idx, INum); + } + for (unsigned i = 0; i < numElemH; ++i) { + int Idx, INum; + if (HEE) { + Idx = + cast<ConstantInt>(HEE->getOperand(1))->getSExtValue(); + INum = HEE->getOperand(0) == I1 ? 0 : 1; + } else { + Idx = HSV->getMaskValue(i); + if (Idx < (int) HOpElem) { + INum = HSV->getOperand(0) == I1 ? 0 : 1; + } else { + Idx -= HOpElem; + INum = HSV->getOperand(1) == I1 ? 0 : 1; + } + } + + II[i + numElemL] = std::pair<int, int>(Idx, INum); + } + + // We now have an array which tells us from which index of which + // input vector each element of the operand comes. + VectorType *I1T = cast<VectorType>(I1->getType()); + unsigned I1Elem = I1T->getNumElements(); + + if (!I2) { + // In this case there is only one underlying vector input. Check for + // the trivial case where we can use the input directly. + if (I1Elem == numElem) { + bool ElemInOrder = true; + for (unsigned i = 0; i < numElem; ++i) { + if (II[i].first != (int) i && II[i].first != -1) { + ElemInOrder = false; + break; + } + } + + if (ElemInOrder) + return I1; + } + + // A shuffle is needed. + std::vector<Constant *> Mask(numElem); + for (unsigned i = 0; i < numElem; ++i) { + int Idx = II[i].first; + if (Idx == -1) + Mask[i] = UndefValue::get(Type::getInt32Ty(Context)); + else + Mask[i] = ConstantInt::get(Type::getInt32Ty(Context), Idx); + } + + Instruction *S = + new ShuffleVectorInst(I1, UndefValue::get(I1T), + ConstantVector::get(Mask), + getReplacementName(IBeforeJ ? I : J, + true, o)); + S->insertBefore(IBeforeJ ? J : I); + return S; + } + + VectorType *I2T = cast<VectorType>(I2->getType()); + unsigned I2Elem = I2T->getNumElements(); + + // This input comes from two distinct vectors. The first step is to + // make sure that both vectors are the same length. If not, the + // smaller one will need to grow before they can be shuffled together. + if (I1Elem < I2Elem) { + std::vector<Constant *> Mask(I2Elem); + unsigned v = 0; + for (; v < I1Elem; ++v) + Mask[v] = ConstantInt::get(Type::getInt32Ty(Context), v); + for (; v < I2Elem; ++v) + Mask[v] = UndefValue::get(Type::getInt32Ty(Context)); + + Instruction *NewI1 = + new ShuffleVectorInst(I1, UndefValue::get(I1T), + ConstantVector::get(Mask), + getReplacementName(IBeforeJ ? I : J, + true, o, 1)); + NewI1->insertBefore(IBeforeJ ? J : I); + I1 = NewI1; + I1Elem = I2Elem; + } else if (I1Elem > I2Elem) { + std::vector<Constant *> Mask(I1Elem); + unsigned v = 0; + for (; v < I2Elem; ++v) + Mask[v] = ConstantInt::get(Type::getInt32Ty(Context), v); + for (; v < I1Elem; ++v) + Mask[v] = UndefValue::get(Type::getInt32Ty(Context)); + + Instruction *NewI2 = + new ShuffleVectorInst(I2, UndefValue::get(I2T), + ConstantVector::get(Mask), + getReplacementName(IBeforeJ ? I : J, + true, o, 1)); + NewI2->insertBefore(IBeforeJ ? J : I); + I2 = NewI2; + } + + // Now that both I1 and I2 are the same length we can shuffle them + // together (and use the result). + std::vector<Constant *> Mask(numElem); + for (unsigned v = 0; v < numElem; ++v) { + if (II[v].first == -1) { + Mask[v] = UndefValue::get(Type::getInt32Ty(Context)); + } else { + int Idx = II[v].first + II[v].second * I1Elem; + Mask[v] = ConstantInt::get(Type::getInt32Ty(Context), Idx); + } + } + + Instruction *NewOp = + new ShuffleVectorInst(I1, I2, ConstantVector::get(Mask), + getReplacementName(IBeforeJ ? I : J, true, o)); + NewOp->insertBefore(IBeforeJ ? J : I); + return NewOp; + } + } + + Type *ArgType = ArgTypeL; + if (numElemL < numElemH) { + if (numElemL == 1 && expandIEChain(Context, I, J, o, HOp, numElemH, + ArgTypeL, VArgType, IBeforeJ, 1)) { + // This is another short-circuit case: we're combining a scalar into + // a vector that is formed by an IE chain. We've just expanded the IE + // chain, now insert the scalar and we're done. + + Instruction *S = InsertElementInst::Create(HOp, LOp, CV0, + getReplacementName(IBeforeJ ? I : J, true, o)); + S->insertBefore(IBeforeJ ? J : I); + return S; + } else if (!expandIEChain(Context, I, J, o, LOp, numElemL, ArgTypeL, + ArgTypeH, IBeforeJ)) { + // The two vector inputs to the shuffle must be the same length, + // so extend the smaller vector to be the same length as the larger one. + Instruction *NLOp; + if (numElemL > 1) { + + std::vector<Constant *> Mask(numElemH); + unsigned v = 0; + for (; v < numElemL; ++v) + Mask[v] = ConstantInt::get(Type::getInt32Ty(Context), v); + for (; v < numElemH; ++v) + Mask[v] = UndefValue::get(Type::getInt32Ty(Context)); + + NLOp = new ShuffleVectorInst(LOp, UndefValue::get(ArgTypeL), + ConstantVector::get(Mask), + getReplacementName(IBeforeJ ? I : J, + true, o, 1)); + } else { + NLOp = InsertElementInst::Create(UndefValue::get(ArgTypeH), LOp, CV0, + getReplacementName(IBeforeJ ? I : J, + true, o, 1)); + } + + NLOp->insertBefore(IBeforeJ ? J : I); + LOp = NLOp; + } + + ArgType = ArgTypeH; + } else if (numElemL > numElemH) { + if (numElemH == 1 && expandIEChain(Context, I, J, o, LOp, numElemL, + ArgTypeH, VArgType, IBeforeJ)) { + Instruction *S = + InsertElementInst::Create(LOp, HOp, + ConstantInt::get(Type::getInt32Ty(Context), + numElemL), + getReplacementName(IBeforeJ ? I : J, + true, o)); + S->insertBefore(IBeforeJ ? J : I); + return S; + } else if (!expandIEChain(Context, I, J, o, HOp, numElemH, ArgTypeH, + ArgTypeL, IBeforeJ)) { + Instruction *NHOp; + if (numElemH > 1) { + std::vector<Constant *> Mask(numElemL); + unsigned v = 0; + for (; v < numElemH; ++v) + Mask[v] = ConstantInt::get(Type::getInt32Ty(Context), v); + for (; v < numElemL; ++v) + Mask[v] = UndefValue::get(Type::getInt32Ty(Context)); + + NHOp = new ShuffleVectorInst(HOp, UndefValue::get(ArgTypeH), + ConstantVector::get(Mask), + getReplacementName(IBeforeJ ? I : J, + true, o, 1)); + } else { + NHOp = InsertElementInst::Create(UndefValue::get(ArgTypeL), HOp, CV0, + getReplacementName(IBeforeJ ? I : J, + true, o, 1)); + } + + NHOp->insertBefore(IBeforeJ ? J : I); + HOp = NHOp; + } + } + + if (ArgType->isVectorTy()) { + unsigned numElem = VArgType->getVectorNumElements(); + std::vector<Constant*> Mask(numElem); + for (unsigned v = 0; v < numElem; ++v) { + unsigned Idx = v; + // If the low vector was expanded, we need to skip the extra + // undefined entries. + if (v >= numElemL && numElemH > numElemL) + Idx += (numElemH - numElemL); + Mask[v] = ConstantInt::get(Type::getInt32Ty(Context), Idx); + } + + Instruction *BV = new ShuffleVectorInst(LOp, HOp, + ConstantVector::get(Mask), + getReplacementName(IBeforeJ ? I : J, true, o)); + BV->insertBefore(IBeforeJ ? J : I); + return BV; + } + + Instruction *BV1 = InsertElementInst::Create( + UndefValue::get(VArgType), LOp, CV0, + getReplacementName(IBeforeJ ? I : J, + true, o, 1)); + BV1->insertBefore(IBeforeJ ? J : I); + Instruction *BV2 = InsertElementInst::Create(BV1, HOp, CV1, + getReplacementName(IBeforeJ ? I : J, + true, o, 2)); + BV2->insertBefore(IBeforeJ ? J : I); + return BV2; + } + + // This function creates an array of values that will be used as the inputs + // to the vector instruction that fuses I with J. + void BBVectorize::getReplacementInputsForPair(LLVMContext& Context, + Instruction *I, Instruction *J, + SmallVectorImpl<Value *> &ReplacedOperands, + bool IBeforeJ) { + unsigned NumOperands = I->getNumOperands(); + + for (unsigned p = 0, o = NumOperands-1; p < NumOperands; ++p, --o) { + // Iterate backward so that we look at the store pointer + // first and know whether or not we need to flip the inputs. + + if (isa<LoadInst>(I) || (o == 1 && isa<StoreInst>(I))) { + // This is the pointer for a load/store instruction. + ReplacedOperands[o] = getReplacementPointerInput(Context, I, J, o); + continue; + } else if (isa<CallInst>(I)) { + Function *F = cast<CallInst>(I)->getCalledFunction(); + Intrinsic::ID IID = F->getIntrinsicID(); + if (o == NumOperands-1) { + BasicBlock &BB = *I->getParent(); + + Module *M = BB.getParent()->getParent(); + Type *ArgTypeI = I->getType(); + Type *ArgTypeJ = J->getType(); + Type *VArgType = getVecTypeForPair(ArgTypeI, ArgTypeJ); + + ReplacedOperands[o] = Intrinsic::getDeclaration(M, IID, VArgType); + continue; + } else if ((IID == Intrinsic::powi || IID == Intrinsic::ctlz || + IID == Intrinsic::cttz) && o == 1) { + // The second argument of powi/ctlz/cttz is a single integer/constant + // and we've already checked that both arguments are equal. + // As a result, we just keep I's second argument. + ReplacedOperands[o] = I->getOperand(o); + continue; + } + } else if (isa<ShuffleVectorInst>(I) && o == NumOperands-1) { + ReplacedOperands[o] = getReplacementShuffleMask(Context, I, J); + continue; + } + + ReplacedOperands[o] = getReplacementInput(Context, I, J, o, IBeforeJ); + } + } + + // This function creates two values that represent the outputs of the + // original I and J instructions. These are generally vector shuffles + // or extracts. In many cases, these will end up being unused and, thus, + // eliminated by later passes. + void BBVectorize::replaceOutputsOfPair(LLVMContext& Context, Instruction *I, + Instruction *J, Instruction *K, + Instruction *&InsertionPt, + Instruction *&K1, Instruction *&K2) { + if (isa<StoreInst>(I)) + return; + + Type *IType = I->getType(); + Type *JType = J->getType(); + + VectorType *VType = getVecTypeForPair(IType, JType); + unsigned numElem = VType->getNumElements(); + + unsigned numElemI = getNumScalarElements(IType); + unsigned numElemJ = getNumScalarElements(JType); + + if (IType->isVectorTy()) { + std::vector<Constant *> Mask1(numElemI), Mask2(numElemI); + for (unsigned v = 0; v < numElemI; ++v) { + Mask1[v] = ConstantInt::get(Type::getInt32Ty(Context), v); + Mask2[v] = ConstantInt::get(Type::getInt32Ty(Context), numElemJ + v); + } + + K1 = new ShuffleVectorInst(K, UndefValue::get(VType), + ConstantVector::get(Mask1), + getReplacementName(K, false, 1)); + } else { + Value *CV0 = ConstantInt::get(Type::getInt32Ty(Context), 0); + K1 = ExtractElementInst::Create(K, CV0, getReplacementName(K, false, 1)); + } + + if (JType->isVectorTy()) { + std::vector<Constant *> Mask1(numElemJ), Mask2(numElemJ); + for (unsigned v = 0; v < numElemJ; ++v) { + Mask1[v] = ConstantInt::get(Type::getInt32Ty(Context), v); + Mask2[v] = ConstantInt::get(Type::getInt32Ty(Context), numElemI + v); + } + + K2 = new ShuffleVectorInst(K, UndefValue::get(VType), + ConstantVector::get(Mask2), + getReplacementName(K, false, 2)); + } else { + Value *CV1 = ConstantInt::get(Type::getInt32Ty(Context), numElem - 1); + K2 = ExtractElementInst::Create(K, CV1, getReplacementName(K, false, 2)); + } + + K1->insertAfter(K); + K2->insertAfter(K1); + InsertionPt = K2; + } + + // Move all uses of the function I (including pairing-induced uses) after J. + bool BBVectorize::canMoveUsesOfIAfterJ(BasicBlock &BB, + DenseSet<ValuePair> &LoadMoveSetPairs, + Instruction *I, Instruction *J) { + // Skip to the first instruction past I. + BasicBlock::iterator L = std::next(BasicBlock::iterator(I)); + + DenseSet<Value *> Users; + AliasSetTracker WriteSet(*AA); + if (I->mayWriteToMemory()) WriteSet.add(I); + + for (; cast<Instruction>(L) != J; ++L) + (void)trackUsesOfI(Users, WriteSet, I, &*L, true, &LoadMoveSetPairs); + + assert(cast<Instruction>(L) == J && + "Tracking has not proceeded far enough to check for dependencies"); + // If J is now in the use set of I, then trackUsesOfI will return true + // and we have a dependency cycle (and the fusing operation must abort). + return !trackUsesOfI(Users, WriteSet, I, J, true, &LoadMoveSetPairs); + } + + // Move all uses of the function I (including pairing-induced uses) after J. + void BBVectorize::moveUsesOfIAfterJ(BasicBlock &BB, + DenseSet<ValuePair> &LoadMoveSetPairs, + Instruction *&InsertionPt, + Instruction *I, Instruction *J) { + // Skip to the first instruction past I. + BasicBlock::iterator L = std::next(BasicBlock::iterator(I)); + + DenseSet<Value *> Users; + AliasSetTracker WriteSet(*AA); + if (I->mayWriteToMemory()) WriteSet.add(I); + + for (; cast<Instruction>(L) != J;) { + if (trackUsesOfI(Users, WriteSet, I, &*L, true, &LoadMoveSetPairs)) { + // Move this instruction + Instruction *InstToMove = &*L++; + + DEBUG(dbgs() << "BBV: moving: " << *InstToMove << + " to after " << *InsertionPt << "\n"); + InstToMove->removeFromParent(); + InstToMove->insertAfter(InsertionPt); + InsertionPt = InstToMove; + } else { + ++L; + } + } + } + + // Collect all load instruction that are in the move set of a given first + // pair member. These loads depend on the first instruction, I, and so need + // to be moved after J (the second instruction) when the pair is fused. + void BBVectorize::collectPairLoadMoveSet(BasicBlock &BB, + DenseMap<Value *, Value *> &ChosenPairs, + DenseMap<Value *, std::vector<Value *> > &LoadMoveSet, + DenseSet<ValuePair> &LoadMoveSetPairs, + Instruction *I) { + // Skip to the first instruction past I. + BasicBlock::iterator L = std::next(BasicBlock::iterator(I)); + + DenseSet<Value *> Users; + AliasSetTracker WriteSet(*AA); + if (I->mayWriteToMemory()) WriteSet.add(I); + + // Note: We cannot end the loop when we reach J because J could be moved + // farther down the use chain by another instruction pairing. Also, J + // could be before I if this is an inverted input. + for (BasicBlock::iterator E = BB.end(); L != E; ++L) { + if (trackUsesOfI(Users, WriteSet, I, &*L)) { + if (L->mayReadFromMemory()) { + LoadMoveSet[&*L].push_back(I); + LoadMoveSetPairs.insert(ValuePair(&*L, I)); + } + } + } + } + + // In cases where both load/stores and the computation of their pointers + // are chosen for vectorization, we can end up in a situation where the + // aliasing analysis starts returning different query results as the + // process of fusing instruction pairs continues. Because the algorithm + // relies on finding the same use dags here as were found earlier, we'll + // need to precompute the necessary aliasing information here and then + // manually update it during the fusion process. + void BBVectorize::collectLoadMoveSet(BasicBlock &BB, + std::vector<Value *> &PairableInsts, + DenseMap<Value *, Value *> &ChosenPairs, + DenseMap<Value *, std::vector<Value *> > &LoadMoveSet, + DenseSet<ValuePair> &LoadMoveSetPairs) { + for (std::vector<Value *>::iterator PI = PairableInsts.begin(), + PIE = PairableInsts.end(); PI != PIE; ++PI) { + DenseMap<Value *, Value *>::iterator P = ChosenPairs.find(*PI); + if (P == ChosenPairs.end()) continue; + + Instruction *I = cast<Instruction>(P->first); + collectPairLoadMoveSet(BB, ChosenPairs, LoadMoveSet, + LoadMoveSetPairs, I); + } + } + + // This function fuses the chosen instruction pairs into vector instructions, + // taking care preserve any needed scalar outputs and, then, it reorders the + // remaining instructions as needed (users of the first member of the pair + // need to be moved to after the location of the second member of the pair + // because the vector instruction is inserted in the location of the pair's + // second member). + void BBVectorize::fuseChosenPairs(BasicBlock &BB, + std::vector<Value *> &PairableInsts, + DenseMap<Value *, Value *> &ChosenPairs, + DenseSet<ValuePair> &FixedOrderPairs, + DenseMap<VPPair, unsigned> &PairConnectionTypes, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs, + DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps) { + LLVMContext& Context = BB.getContext(); + + // During the vectorization process, the order of the pairs to be fused + // could be flipped. So we'll add each pair, flipped, into the ChosenPairs + // list. After a pair is fused, the flipped pair is removed from the list. + DenseSet<ValuePair> FlippedPairs; + for (DenseMap<Value *, Value *>::iterator P = ChosenPairs.begin(), + E = ChosenPairs.end(); P != E; ++P) + FlippedPairs.insert(ValuePair(P->second, P->first)); + for (DenseSet<ValuePair>::iterator P = FlippedPairs.begin(), + E = FlippedPairs.end(); P != E; ++P) + ChosenPairs.insert(*P); + + DenseMap<Value *, std::vector<Value *> > LoadMoveSet; + DenseSet<ValuePair> LoadMoveSetPairs; + collectLoadMoveSet(BB, PairableInsts, ChosenPairs, + LoadMoveSet, LoadMoveSetPairs); + + DEBUG(dbgs() << "BBV: initial: \n" << BB << "\n"); + + for (BasicBlock::iterator PI = BB.getFirstInsertionPt(); PI != BB.end();) { + DenseMap<Value *, Value *>::iterator P = ChosenPairs.find(&*PI); + if (P == ChosenPairs.end()) { + ++PI; + continue; + } + + if (getDepthFactor(P->first) == 0) { + // These instructions are not really fused, but are tracked as though + // they are. Any case in which it would be interesting to fuse them + // will be taken care of by InstCombine. + --NumFusedOps; + ++PI; + continue; + } + + Instruction *I = cast<Instruction>(P->first), + *J = cast<Instruction>(P->second); + + DEBUG(dbgs() << "BBV: fusing: " << *I << + " <-> " << *J << "\n"); + + // Remove the pair and flipped pair from the list. + DenseMap<Value *, Value *>::iterator FP = ChosenPairs.find(P->second); + assert(FP != ChosenPairs.end() && "Flipped pair not found in list"); + ChosenPairs.erase(FP); + ChosenPairs.erase(P); + + if (!canMoveUsesOfIAfterJ(BB, LoadMoveSetPairs, I, J)) { + DEBUG(dbgs() << "BBV: fusion of: " << *I << + " <-> " << *J << + " aborted because of non-trivial dependency cycle\n"); + --NumFusedOps; + ++PI; + continue; + } + + // If the pair must have the other order, then flip it. + bool FlipPairOrder = FixedOrderPairs.count(ValuePair(J, I)); + if (!FlipPairOrder && !FixedOrderPairs.count(ValuePair(I, J))) { + // This pair does not have a fixed order, and so we might want to + // flip it if that will yield fewer shuffles. We count the number + // of dependencies connected via swaps, and those directly connected, + // and flip the order if the number of swaps is greater. + bool OrigOrder = true; + DenseMap<ValuePair, std::vector<ValuePair> >::iterator IJ = + ConnectedPairDeps.find(ValuePair(I, J)); + if (IJ == ConnectedPairDeps.end()) { + IJ = ConnectedPairDeps.find(ValuePair(J, I)); + OrigOrder = false; + } + + if (IJ != ConnectedPairDeps.end()) { + unsigned NumDepsDirect = 0, NumDepsSwap = 0; + for (std::vector<ValuePair>::iterator T = IJ->second.begin(), + TE = IJ->second.end(); T != TE; ++T) { + VPPair Q(IJ->first, *T); + DenseMap<VPPair, unsigned>::iterator R = + PairConnectionTypes.find(VPPair(Q.second, Q.first)); + assert(R != PairConnectionTypes.end() && + "Cannot find pair connection type"); + if (R->second == PairConnectionDirect) + ++NumDepsDirect; + else if (R->second == PairConnectionSwap) + ++NumDepsSwap; + } + + if (!OrigOrder) + std::swap(NumDepsDirect, NumDepsSwap); + + if (NumDepsSwap > NumDepsDirect) { + FlipPairOrder = true; + DEBUG(dbgs() << "BBV: reordering pair: " << *I << + " <-> " << *J << "\n"); + } + } + } + + Instruction *L = I, *H = J; + if (FlipPairOrder) + std::swap(H, L); + + // If the pair being fused uses the opposite order from that in the pair + // connection map, then we need to flip the types. + DenseMap<ValuePair, std::vector<ValuePair> >::iterator HL = + ConnectedPairs.find(ValuePair(H, L)); + if (HL != ConnectedPairs.end()) + for (std::vector<ValuePair>::iterator T = HL->second.begin(), + TE = HL->second.end(); T != TE; ++T) { + VPPair Q(HL->first, *T); + DenseMap<VPPair, unsigned>::iterator R = PairConnectionTypes.find(Q); + assert(R != PairConnectionTypes.end() && + "Cannot find pair connection type"); + if (R->second == PairConnectionDirect) + R->second = PairConnectionSwap; + else if (R->second == PairConnectionSwap) + R->second = PairConnectionDirect; + } + + bool LBeforeH = !FlipPairOrder; + unsigned NumOperands = I->getNumOperands(); + SmallVector<Value *, 3> ReplacedOperands(NumOperands); + getReplacementInputsForPair(Context, L, H, ReplacedOperands, + LBeforeH); + + // Make a copy of the original operation, change its type to the vector + // type and replace its operands with the vector operands. + Instruction *K = L->clone(); + if (L->hasName()) + K->takeName(L); + else if (H->hasName()) + K->takeName(H); + + if (auto CS = CallSite(K)) { + SmallVector<Type *, 3> Tys; + FunctionType *Old = CS.getFunctionType(); + unsigned NumOld = Old->getNumParams(); + assert(NumOld <= ReplacedOperands.size()); + for (unsigned i = 0; i != NumOld; ++i) + Tys.push_back(ReplacedOperands[i]->getType()); + CS.mutateFunctionType( + FunctionType::get(getVecTypeForPair(L->getType(), H->getType()), + Tys, Old->isVarArg())); + } else if (!isa<StoreInst>(K)) + K->mutateType(getVecTypeForPair(L->getType(), H->getType())); + + unsigned KnownIDs[] = {LLVMContext::MD_tbaa, LLVMContext::MD_alias_scope, + LLVMContext::MD_noalias, LLVMContext::MD_fpmath, + LLVMContext::MD_invariant_group}; + combineMetadata(K, H, KnownIDs); + K->intersectOptionalDataWith(H); + + for (unsigned o = 0; o < NumOperands; ++o) + K->setOperand(o, ReplacedOperands[o]); + + K->insertAfter(J); + + // Instruction insertion point: + Instruction *InsertionPt = K; + Instruction *K1 = nullptr, *K2 = nullptr; + replaceOutputsOfPair(Context, L, H, K, InsertionPt, K1, K2); + + // The use dag of the first original instruction must be moved to after + // the location of the second instruction. The entire use dag of the + // first instruction is disjoint from the input dag of the second + // (by definition), and so commutes with it. + + moveUsesOfIAfterJ(BB, LoadMoveSetPairs, InsertionPt, I, J); + + if (!isa<StoreInst>(I)) { + L->replaceAllUsesWith(K1); + H->replaceAllUsesWith(K2); + } + + // Instructions that may read from memory may be in the load move set. + // Once an instruction is fused, we no longer need its move set, and so + // the values of the map never need to be updated. However, when a load + // is fused, we need to merge the entries from both instructions in the + // pair in case those instructions were in the move set of some other + // yet-to-be-fused pair. The loads in question are the keys of the map. + if (I->mayReadFromMemory()) { + std::vector<ValuePair> NewSetMembers; + DenseMap<Value *, std::vector<Value *> >::iterator II = + LoadMoveSet.find(I); + if (II != LoadMoveSet.end()) + for (std::vector<Value *>::iterator N = II->second.begin(), + NE = II->second.end(); N != NE; ++N) + NewSetMembers.push_back(ValuePair(K, *N)); + DenseMap<Value *, std::vector<Value *> >::iterator JJ = + LoadMoveSet.find(J); + if (JJ != LoadMoveSet.end()) + for (std::vector<Value *>::iterator N = JJ->second.begin(), + NE = JJ->second.end(); N != NE; ++N) + NewSetMembers.push_back(ValuePair(K, *N)); + for (std::vector<ValuePair>::iterator A = NewSetMembers.begin(), + AE = NewSetMembers.end(); A != AE; ++A) { + LoadMoveSet[A->first].push_back(A->second); + LoadMoveSetPairs.insert(*A); + } + } + + // Before removing I, set the iterator to the next instruction. + PI = std::next(BasicBlock::iterator(I)); + if (cast<Instruction>(PI) == J) + ++PI; + + SE->forgetValue(I); + SE->forgetValue(J); + I->eraseFromParent(); + J->eraseFromParent(); + + DEBUG(if (PrintAfterEveryPair) dbgs() << "BBV: block is now: \n" << + BB << "\n"); + } + + DEBUG(dbgs() << "BBV: final: \n" << BB << "\n"); + } +} + +char BBVectorize::ID = 0; +static const char bb_vectorize_name[] = "Basic-Block Vectorization"; +INITIALIZE_PASS_BEGIN(BBVectorize, BBV_NAME, bb_vectorize_name, false, false) +INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass) +INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass) +INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass) +INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass) +INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass) +INITIALIZE_PASS_DEPENDENCY(ScalarEvolutionWrapperPass) +INITIALIZE_PASS_DEPENDENCY(GlobalsAAWrapperPass) +INITIALIZE_PASS_DEPENDENCY(SCEVAAWrapperPass) +INITIALIZE_PASS_END(BBVectorize, BBV_NAME, bb_vectorize_name, false, false) + +BasicBlockPass *llvm::createBBVectorizePass(const VectorizeConfig &C) { + return new BBVectorize(C); +} + +bool +llvm::vectorizeBasicBlock(Pass *P, BasicBlock &BB, const VectorizeConfig &C) { + BBVectorize BBVectorizer(P, *BB.getParent(), C); + return BBVectorizer.vectorizeBB(BB); +} + +//===----------------------------------------------------------------------===// +VectorizeConfig::VectorizeConfig() { + VectorBits = ::VectorBits; + VectorizeBools = !::NoBools; + VectorizeInts = !::NoInts; + VectorizeFloats = !::NoFloats; + VectorizePointers = !::NoPointers; + VectorizeCasts = !::NoCasts; + VectorizeMath = !::NoMath; + VectorizeBitManipulations = !::NoBitManipulation; + VectorizeFMA = !::NoFMA; + VectorizeSelect = !::NoSelect; + VectorizeCmp = !::NoCmp; + VectorizeGEP = !::NoGEP; + VectorizeMemOps = !::NoMemOps; + AlignedOnly = ::AlignedOnly; + ReqChainDepth= ::ReqChainDepth; + SearchLimit = ::SearchLimit; + MaxCandPairsForCycleCheck = ::MaxCandPairsForCycleCheck; + SplatBreaksChain = ::SplatBreaksChain; + MaxInsts = ::MaxInsts; + MaxPairs = ::MaxPairs; + MaxIter = ::MaxIter; + Pow2LenOnly = ::Pow2LenOnly; + NoMemOpBoost = ::NoMemOpBoost; + FastDep = ::FastDep; +} diff --git a/contrib/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/contrib/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp new file mode 100644 index 0000000..2c0d317 --- /dev/null +++ b/contrib/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp @@ -0,0 +1,5823 @@ +//===- LoopVectorize.cpp - A Loop Vectorizer ------------------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This is the LLVM loop vectorizer. This pass modifies 'vectorizable' loops +// and generates target-independent LLVM-IR. +// The vectorizer uses the TargetTransformInfo analysis to estimate the costs +// of instructions in order to estimate the profitability of vectorization. +// +// The loop vectorizer combines consecutive loop iterations into a single +// 'wide' iteration. After this transformation the index is incremented +// by the SIMD vector width, and not by one. +// +// This pass has three parts: +// 1. The main loop pass that drives the different parts. +// 2. LoopVectorizationLegality - A unit that checks for the legality +// of the vectorization. +// 3. InnerLoopVectorizer - A unit that performs the actual +// widening of instructions. +// 4. LoopVectorizationCostModel - A unit that checks for the profitability +// of vectorization. It decides on the optimal vector width, which +// can be one, if vectorization is not profitable. +// +//===----------------------------------------------------------------------===// +// +// The reduction-variable vectorization is based on the paper: +// D. Nuzman and R. Henderson. Multi-platform Auto-vectorization. +// +// Variable uniformity checks are inspired by: +// Karrenberg, R. and Hack, S. Whole Function Vectorization. +// +// The interleaved access vectorization is based on the paper: +// Dorit Nuzman, Ira Rosen and Ayal Zaks. Auto-Vectorization of Interleaved +// Data for SIMD +// +// Other ideas/concepts are from: +// A. Zaks and D. Nuzman. Autovectorization in GCC-two years later. +// +// S. Maleki, Y. Gao, M. Garzaran, T. Wong and D. Padua. An Evaluation of +// Vectorizing Compilers. +// +//===----------------------------------------------------------------------===// + +#include "llvm/Transforms/Vectorize.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/Hashing.h" +#include "llvm/ADT/MapVector.h" +#include "llvm/ADT/SetVector.h" +#include "llvm/ADT/SmallPtrSet.h" +#include "llvm/ADT/SmallSet.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/ADT/StringExtras.h" +#include "llvm/Analysis/AliasAnalysis.h" +#include "llvm/Analysis/BasicAliasAnalysis.h" +#include "llvm/Analysis/AliasSetTracker.h" +#include "llvm/Analysis/AssumptionCache.h" +#include "llvm/Analysis/BlockFrequencyInfo.h" +#include "llvm/Analysis/CodeMetrics.h" +#include "llvm/Analysis/DemandedBits.h" +#include "llvm/Analysis/GlobalsModRef.h" +#include "llvm/Analysis/LoopAccessAnalysis.h" +#include "llvm/Analysis/LoopInfo.h" +#include "llvm/Analysis/LoopIterator.h" +#include "llvm/Analysis/LoopPass.h" +#include "llvm/Analysis/ScalarEvolution.h" +#include "llvm/Analysis/ScalarEvolutionExpander.h" +#include "llvm/Analysis/ScalarEvolutionExpressions.h" +#include "llvm/Analysis/TargetTransformInfo.h" +#include "llvm/Analysis/ValueTracking.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DataLayout.h" +#include "llvm/IR/DebugInfo.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/DiagnosticInfo.h" +#include "llvm/IR/Dominators.h" +#include "llvm/IR/Function.h" +#include "llvm/IR/IRBuilder.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/LLVMContext.h" +#include "llvm/IR/Module.h" +#include "llvm/IR/PatternMatch.h" +#include "llvm/IR/Type.h" +#include "llvm/IR/Value.h" +#include "llvm/IR/ValueHandle.h" +#include "llvm/IR/Verifier.h" +#include "llvm/Pass.h" +#include "llvm/Support/BranchProbability.h" +#include "llvm/Support/CommandLine.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/raw_ostream.h" +#include "llvm/Transforms/Scalar.h" +#include "llvm/Transforms/Utils/BasicBlockUtils.h" +#include "llvm/Transforms/Utils/Local.h" +#include "llvm/Analysis/VectorUtils.h" +#include "llvm/Transforms/Utils/LoopUtils.h" +#include <algorithm> +#include <functional> +#include <map> +#include <tuple> + +using namespace llvm; +using namespace llvm::PatternMatch; + +#define LV_NAME "loop-vectorize" +#define DEBUG_TYPE LV_NAME + +STATISTIC(LoopsVectorized, "Number of loops vectorized"); +STATISTIC(LoopsAnalyzed, "Number of loops analyzed for vectorization"); + +static cl::opt<bool> +EnableIfConversion("enable-if-conversion", cl::init(true), cl::Hidden, + cl::desc("Enable if-conversion during vectorization.")); + +/// We don't vectorize loops with a known constant trip count below this number. +static cl::opt<unsigned> +TinyTripCountVectorThreshold("vectorizer-min-trip-count", cl::init(16), + cl::Hidden, + cl::desc("Don't vectorize loops with a constant " + "trip count that is smaller than this " + "value.")); + +static cl::opt<bool> MaximizeBandwidth( + "vectorizer-maximize-bandwidth", cl::init(false), cl::Hidden, + cl::desc("Maximize bandwidth when selecting vectorization factor which " + "will be determined by the smallest type in loop.")); + +/// This enables versioning on the strides of symbolically striding memory +/// accesses in code like the following. +/// for (i = 0; i < N; ++i) +/// A[i * Stride1] += B[i * Stride2] ... +/// +/// Will be roughly translated to +/// if (Stride1 == 1 && Stride2 == 1) { +/// for (i = 0; i < N; i+=4) +/// A[i:i+3] += ... +/// } else +/// ... +static cl::opt<bool> EnableMemAccessVersioning( + "enable-mem-access-versioning", cl::init(true), cl::Hidden, + cl::desc("Enable symbolic stride memory access versioning")); + +static cl::opt<bool> EnableInterleavedMemAccesses( + "enable-interleaved-mem-accesses", cl::init(false), cl::Hidden, + cl::desc("Enable vectorization on interleaved memory accesses in a loop")); + +/// Maximum factor for an interleaved memory access. +static cl::opt<unsigned> MaxInterleaveGroupFactor( + "max-interleave-group-factor", cl::Hidden, + cl::desc("Maximum factor for an interleaved access group (default = 8)"), + cl::init(8)); + +/// We don't interleave loops with a known constant trip count below this +/// number. +static const unsigned TinyTripCountInterleaveThreshold = 128; + +static cl::opt<unsigned> ForceTargetNumScalarRegs( + "force-target-num-scalar-regs", cl::init(0), cl::Hidden, + cl::desc("A flag that overrides the target's number of scalar registers.")); + +static cl::opt<unsigned> ForceTargetNumVectorRegs( + "force-target-num-vector-regs", cl::init(0), cl::Hidden, + cl::desc("A flag that overrides the target's number of vector registers.")); + +/// Maximum vectorization interleave count. +static const unsigned MaxInterleaveFactor = 16; + +static cl::opt<unsigned> ForceTargetMaxScalarInterleaveFactor( + "force-target-max-scalar-interleave", cl::init(0), cl::Hidden, + cl::desc("A flag that overrides the target's max interleave factor for " + "scalar loops.")); + +static cl::opt<unsigned> ForceTargetMaxVectorInterleaveFactor( + "force-target-max-vector-interleave", cl::init(0), cl::Hidden, + cl::desc("A flag that overrides the target's max interleave factor for " + "vectorized loops.")); + +static cl::opt<unsigned> ForceTargetInstructionCost( + "force-target-instruction-cost", cl::init(0), cl::Hidden, + cl::desc("A flag that overrides the target's expected cost for " + "an instruction to a single constant value. Mostly " + "useful for getting consistent testing.")); + +static cl::opt<unsigned> SmallLoopCost( + "small-loop-cost", cl::init(20), cl::Hidden, + cl::desc( + "The cost of a loop that is considered 'small' by the interleaver.")); + +static cl::opt<bool> LoopVectorizeWithBlockFrequency( + "loop-vectorize-with-block-frequency", cl::init(false), cl::Hidden, + cl::desc("Enable the use of the block frequency analysis to access PGO " + "heuristics minimizing code growth in cold regions and being more " + "aggressive in hot regions.")); + +// Runtime interleave loops for load/store throughput. +static cl::opt<bool> EnableLoadStoreRuntimeInterleave( + "enable-loadstore-runtime-interleave", cl::init(true), cl::Hidden, + cl::desc( + "Enable runtime interleaving until load/store ports are saturated")); + +/// The number of stores in a loop that are allowed to need predication. +static cl::opt<unsigned> NumberOfStoresToPredicate( + "vectorize-num-stores-pred", cl::init(1), cl::Hidden, + cl::desc("Max number of stores to be predicated behind an if.")); + +static cl::opt<bool> EnableIndVarRegisterHeur( + "enable-ind-var-reg-heur", cl::init(true), cl::Hidden, + cl::desc("Count the induction variable only once when interleaving")); + +static cl::opt<bool> EnableCondStoresVectorization( + "enable-cond-stores-vec", cl::init(false), cl::Hidden, + cl::desc("Enable if predication of stores during vectorization.")); + +static cl::opt<unsigned> MaxNestedScalarReductionIC( + "max-nested-scalar-reduction-interleave", cl::init(2), cl::Hidden, + cl::desc("The maximum interleave count to use when interleaving a scalar " + "reduction in a nested loop.")); + +static cl::opt<unsigned> PragmaVectorizeMemoryCheckThreshold( + "pragma-vectorize-memory-check-threshold", cl::init(128), cl::Hidden, + cl::desc("The maximum allowed number of runtime memory checks with a " + "vectorize(enable) pragma.")); + +static cl::opt<unsigned> VectorizeSCEVCheckThreshold( + "vectorize-scev-check-threshold", cl::init(16), cl::Hidden, + cl::desc("The maximum number of SCEV checks allowed.")); + +static cl::opt<unsigned> PragmaVectorizeSCEVCheckThreshold( + "pragma-vectorize-scev-check-threshold", cl::init(128), cl::Hidden, + cl::desc("The maximum number of SCEV checks allowed with a " + "vectorize(enable) pragma")); + +namespace { + +// Forward declarations. +class LoopVectorizeHints; +class LoopVectorizationLegality; +class LoopVectorizationCostModel; +class LoopVectorizationRequirements; + +/// \brief This modifies LoopAccessReport to initialize message with +/// loop-vectorizer-specific part. +class VectorizationReport : public LoopAccessReport { +public: + VectorizationReport(Instruction *I = nullptr) + : LoopAccessReport("loop not vectorized: ", I) {} + + /// \brief This allows promotion of the loop-access analysis report into the + /// loop-vectorizer report. It modifies the message to add the + /// loop-vectorizer-specific part of the message. + explicit VectorizationReport(const LoopAccessReport &R) + : LoopAccessReport(Twine("loop not vectorized: ") + R.str(), + R.getInstr()) {} +}; + +/// A helper function for converting Scalar types to vector types. +/// If the incoming type is void, we return void. If the VF is 1, we return +/// the scalar type. +static Type* ToVectorTy(Type *Scalar, unsigned VF) { + if (Scalar->isVoidTy() || VF == 1) + return Scalar; + return VectorType::get(Scalar, VF); +} + +/// A helper function that returns GEP instruction and knows to skip a +/// 'bitcast'. The 'bitcast' may be skipped if the source and the destination +/// pointee types of the 'bitcast' have the same size. +/// For example: +/// bitcast double** %var to i64* - can be skipped +/// bitcast double** %var to i8* - can not +static GetElementPtrInst *getGEPInstruction(Value *Ptr) { + + if (isa<GetElementPtrInst>(Ptr)) + return cast<GetElementPtrInst>(Ptr); + + if (isa<BitCastInst>(Ptr) && + isa<GetElementPtrInst>(cast<BitCastInst>(Ptr)->getOperand(0))) { + Type *BitcastTy = Ptr->getType(); + Type *GEPTy = cast<BitCastInst>(Ptr)->getSrcTy(); + if (!isa<PointerType>(BitcastTy) || !isa<PointerType>(GEPTy)) + return nullptr; + Type *Pointee1Ty = cast<PointerType>(BitcastTy)->getPointerElementType(); + Type *Pointee2Ty = cast<PointerType>(GEPTy)->getPointerElementType(); + const DataLayout &DL = cast<BitCastInst>(Ptr)->getModule()->getDataLayout(); + if (DL.getTypeSizeInBits(Pointee1Ty) == DL.getTypeSizeInBits(Pointee2Ty)) + return cast<GetElementPtrInst>(cast<BitCastInst>(Ptr)->getOperand(0)); + } + return nullptr; +} + +/// InnerLoopVectorizer vectorizes loops which contain only one basic +/// block to a specified vectorization factor (VF). +/// This class performs the widening of scalars into vectors, or multiple +/// scalars. This class also implements the following features: +/// * It inserts an epilogue loop for handling loops that don't have iteration +/// counts that are known to be a multiple of the vectorization factor. +/// * It handles the code generation for reduction variables. +/// * Scalarization (implementation using scalars) of un-vectorizable +/// instructions. +/// InnerLoopVectorizer does not perform any vectorization-legality +/// checks, and relies on the caller to check for the different legality +/// aspects. The InnerLoopVectorizer relies on the +/// LoopVectorizationLegality class to provide information about the induction +/// and reduction variables that were found to a given vectorization factor. +class InnerLoopVectorizer { +public: + InnerLoopVectorizer(Loop *OrigLoop, PredicatedScalarEvolution &PSE, + LoopInfo *LI, DominatorTree *DT, + const TargetLibraryInfo *TLI, + const TargetTransformInfo *TTI, unsigned VecWidth, + unsigned UnrollFactor) + : OrigLoop(OrigLoop), PSE(PSE), LI(LI), DT(DT), TLI(TLI), TTI(TTI), + VF(VecWidth), UF(UnrollFactor), Builder(PSE.getSE()->getContext()), + Induction(nullptr), OldInduction(nullptr), WidenMap(UnrollFactor), + TripCount(nullptr), VectorTripCount(nullptr), Legal(nullptr), + AddedSafetyChecks(false) {} + + // Perform the actual loop widening (vectorization). + // MinimumBitWidths maps scalar integer values to the smallest bitwidth they + // can be validly truncated to. The cost model has assumed this truncation + // will happen when vectorizing. + void vectorize(LoopVectorizationLegality *L, + MapVector<Instruction*,uint64_t> MinimumBitWidths) { + MinBWs = MinimumBitWidths; + Legal = L; + // Create a new empty loop. Unlink the old loop and connect the new one. + createEmptyLoop(); + // Widen each instruction in the old loop to a new one in the new loop. + // Use the Legality module to find the induction and reduction variables. + vectorizeLoop(); + } + + // Return true if any runtime check is added. + bool IsSafetyChecksAdded() { + return AddedSafetyChecks; + } + + virtual ~InnerLoopVectorizer() {} + +protected: + /// A small list of PHINodes. + typedef SmallVector<PHINode*, 4> PhiVector; + /// When we unroll loops we have multiple vector values for each scalar. + /// This data structure holds the unrolled and vectorized values that + /// originated from one scalar instruction. + typedef SmallVector<Value*, 2> VectorParts; + + // 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. + typedef DenseMap<std::pair<BasicBlock*, BasicBlock*>, + VectorParts> EdgeMaskCache; + + /// Create an empty loop, based on the loop ranges of the old loop. + void createEmptyLoop(); + /// Create a new induction variable inside L. + PHINode *createInductionVariable(Loop *L, Value *Start, Value *End, + Value *Step, Instruction *DL); + /// Copy and widen the instructions from the old loop. + virtual void vectorizeLoop(); + + /// \brief The Loop exit block may have single value PHI nodes where the + /// incoming value is 'Undef'. While vectorizing we only handled real values + /// that were defined inside the loop. Here we fix the 'undef case'. + /// See PR14725. + void fixLCSSAPHIs(); + + /// Shrinks vector element sizes based on information in "MinBWs". + void truncateToMinimalBitwidths(); + + /// A helper function that computes the predicate of the block BB, assuming + /// that the header block of the loop is set to True. It returns the *entry* + /// mask for the block BB. + VectorParts createBlockInMask(BasicBlock *BB); + /// A helper function that computes the predicate of the edge between SRC + /// and DST. + VectorParts createEdgeMask(BasicBlock *Src, BasicBlock *Dst); + + /// A helper function to vectorize a single BB within the innermost loop. + void vectorizeBlockInLoop(BasicBlock *BB, PhiVector *PV); + + /// Vectorize a single PHINode in a block. This method handles the induction + /// variable canonicalization. It supports both VF = 1 for unrolled loops and + /// arbitrary length vectors. + void widenPHIInstruction(Instruction *PN, VectorParts &Entry, + unsigned UF, unsigned VF, PhiVector *PV); + + /// Insert the new loop to the loop hierarchy and pass manager + /// and update the analysis passes. + void updateAnalysis(); + + /// This instruction is un-vectorizable. Implement it as a sequence + /// of scalars. If \p IfPredicateStore is true we need to 'hide' each + /// scalarized instruction behind an if block predicated on the control + /// dependence of the instruction. + virtual void scalarizeInstruction(Instruction *Instr, + bool IfPredicateStore=false); + + /// Vectorize Load and Store instructions, + virtual void vectorizeMemoryInstruction(Instruction *Instr); + + /// Create a broadcast instruction. This method generates a broadcast + /// instruction (shuffle) for loop invariant values and for the induction + /// value. If this is the induction variable then we extend it to N, N+1, ... + /// this is needed because each iteration in the loop corresponds to a SIMD + /// element. + virtual Value *getBroadcastInstrs(Value *V); + + /// This function adds (StartIdx, StartIdx + Step, StartIdx + 2*Step, ...) + /// to each vector element of Val. The sequence starts at StartIndex. + virtual Value *getStepVector(Value *Val, int StartIdx, Value *Step); + + /// When we go over instructions in the basic block we rely on previous + /// values within the current basic block or on loop invariant values. + /// When we widen (vectorize) values we place them in the map. If the values + /// are not within the map, they have to be loop invariant, so we simply + /// broadcast them into a vector. + VectorParts &getVectorValue(Value *V); + + /// Try to vectorize the interleaved access group that \p Instr belongs to. + void vectorizeInterleaveGroup(Instruction *Instr); + + /// Generate a shuffle sequence that will reverse the vector Vec. + virtual Value *reverseVector(Value *Vec); + + /// Returns (and creates if needed) the original loop trip count. + Value *getOrCreateTripCount(Loop *NewLoop); + + /// Returns (and creates if needed) the trip count of the widened loop. + Value *getOrCreateVectorTripCount(Loop *NewLoop); + + /// Emit a bypass check to see if the trip count would overflow, or we + /// wouldn't have enough iterations to execute one vector loop. + void emitMinimumIterationCountCheck(Loop *L, BasicBlock *Bypass); + /// Emit a bypass check to see if the vector trip count is nonzero. + void emitVectorLoopEnteredCheck(Loop *L, BasicBlock *Bypass); + /// Emit a bypass check to see if all of the SCEV assumptions we've + /// had to make are correct. + void emitSCEVChecks(Loop *L, BasicBlock *Bypass); + /// Emit bypass checks to check any memory assumptions we may have made. + void emitMemRuntimeChecks(Loop *L, BasicBlock *Bypass); + + /// This is a helper class that holds the vectorizer state. It maps scalar + /// instructions to vector instructions. When the code is 'unrolled' then + /// then a single scalar value is mapped to multiple vector parts. The parts + /// are stored in the VectorPart type. + struct ValueMap { + /// C'tor. UnrollFactor controls the number of vectors ('parts') that + /// are mapped. + ValueMap(unsigned UnrollFactor) : UF(UnrollFactor) {} + + /// \return True if 'Key' is saved in the Value Map. + bool has(Value *Key) const { return MapStorage.count(Key); } + + /// Initializes a new entry in the map. Sets all of the vector parts to the + /// save value in 'Val'. + /// \return A reference to a vector with splat values. + VectorParts &splat(Value *Key, Value *Val) { + VectorParts &Entry = MapStorage[Key]; + Entry.assign(UF, Val); + return Entry; + } + + ///\return A reference to the value that is stored at 'Key'. + VectorParts &get(Value *Key) { + VectorParts &Entry = MapStorage[Key]; + if (Entry.empty()) + Entry.resize(UF); + assert(Entry.size() == UF); + return Entry; + } + + private: + /// The unroll factor. Each entry in the map stores this number of vector + /// elements. + unsigned UF; + + /// Map storage. We use std::map and not DenseMap because insertions to a + /// dense map invalidates its iterators. + std::map<Value *, VectorParts> MapStorage; + }; + + /// The original loop. + Loop *OrigLoop; + /// A wrapper around ScalarEvolution used to add runtime SCEV checks. Applies + /// dynamic knowledge to simplify SCEV expressions and converts them to a + /// more usable form. + PredicatedScalarEvolution &PSE; + /// Loop Info. + LoopInfo *LI; + /// Dominator Tree. + DominatorTree *DT; + /// Alias Analysis. + AliasAnalysis *AA; + /// Target Library Info. + const TargetLibraryInfo *TLI; + /// Target Transform Info. + const TargetTransformInfo *TTI; + + /// The vectorization SIMD factor to use. Each vector will have this many + /// vector elements. + unsigned VF; + +protected: + /// The vectorization unroll factor to use. Each scalar is vectorized to this + /// many different vector instructions. + unsigned UF; + + /// The builder that we use + IRBuilder<> Builder; + + // --- Vectorization state --- + + /// The vector-loop preheader. + BasicBlock *LoopVectorPreHeader; + /// The scalar-loop preheader. + BasicBlock *LoopScalarPreHeader; + /// Middle Block between the vector and the scalar. + BasicBlock *LoopMiddleBlock; + ///The ExitBlock of the scalar loop. + BasicBlock *LoopExitBlock; + ///The vector loop body. + SmallVector<BasicBlock *, 4> LoopVectorBody; + ///The scalar loop body. + BasicBlock *LoopScalarBody; + /// A list of all bypass blocks. The first block is the entry of the loop. + SmallVector<BasicBlock *, 4> LoopBypassBlocks; + + /// The new Induction variable which was added to the new block. + PHINode *Induction; + /// The induction variable of the old basic block. + PHINode *OldInduction; + /// Maps scalars to widened vectors. + ValueMap WidenMap; + /// Store instructions that should be predicated, as a pair + /// <StoreInst, Predicate> + SmallVector<std::pair<StoreInst*,Value*>, 4> PredicatedStores; + EdgeMaskCache MaskCache; + /// Trip count of the original loop. + Value *TripCount; + /// Trip count of the widened loop (TripCount - TripCount % (VF*UF)) + Value *VectorTripCount; + + /// Map of scalar integer values to the smallest bitwidth they can be legally + /// represented as. The vector equivalents of these values should be truncated + /// to this type. + MapVector<Instruction*,uint64_t> MinBWs; + LoopVectorizationLegality *Legal; + + // Record whether runtime check is added. + bool AddedSafetyChecks; +}; + +class InnerLoopUnroller : public InnerLoopVectorizer { +public: + InnerLoopUnroller(Loop *OrigLoop, PredicatedScalarEvolution &PSE, + LoopInfo *LI, DominatorTree *DT, + const TargetLibraryInfo *TLI, + const TargetTransformInfo *TTI, unsigned UnrollFactor) + : InnerLoopVectorizer(OrigLoop, PSE, LI, DT, TLI, TTI, 1, UnrollFactor) {} + +private: + void scalarizeInstruction(Instruction *Instr, + bool IfPredicateStore = false) override; + void vectorizeMemoryInstruction(Instruction *Instr) override; + Value *getBroadcastInstrs(Value *V) override; + Value *getStepVector(Value *Val, int StartIdx, Value *Step) override; + Value *reverseVector(Value *Vec) override; +}; + +/// \brief Look for a meaningful debug location on the instruction or it's +/// operands. +static Instruction *getDebugLocFromInstOrOperands(Instruction *I) { + if (!I) + return I; + + DebugLoc Empty; + if (I->getDebugLoc() != Empty) + return I; + + for (User::op_iterator OI = I->op_begin(), OE = I->op_end(); OI != OE; ++OI) { + if (Instruction *OpInst = dyn_cast<Instruction>(*OI)) + if (OpInst->getDebugLoc() != Empty) + return OpInst; + } + + return I; +} + +/// \brief Set the debug location in the builder using the debug location in the +/// instruction. +static void setDebugLocFromInst(IRBuilder<> &B, const Value *Ptr) { + if (const Instruction *Inst = dyn_cast_or_null<Instruction>(Ptr)) + B.SetCurrentDebugLocation(Inst->getDebugLoc()); + else + B.SetCurrentDebugLocation(DebugLoc()); +} + +#ifndef NDEBUG +/// \return string containing a file name and a line # for the given loop. +static std::string getDebugLocString(const Loop *L) { + std::string Result; + if (L) { + raw_string_ostream OS(Result); + if (const DebugLoc LoopDbgLoc = L->getStartLoc()) + LoopDbgLoc.print(OS); + else + // Just print the module name. + OS << L->getHeader()->getParent()->getParent()->getModuleIdentifier(); + OS.flush(); + } + return Result; +} +#endif + +/// \brief Propagate known metadata from one instruction to another. +static void propagateMetadata(Instruction *To, const Instruction *From) { + SmallVector<std::pair<unsigned, MDNode *>, 4> Metadata; + From->getAllMetadataOtherThanDebugLoc(Metadata); + + for (auto M : Metadata) { + unsigned Kind = M.first; + + // These are safe to transfer (this is safe for TBAA, even when we + // if-convert, because should that metadata have had a control dependency + // on the condition, and thus actually aliased with some other + // non-speculated memory access when the condition was false, this would be + // caught by the runtime overlap checks). + if (Kind != LLVMContext::MD_tbaa && + Kind != LLVMContext::MD_alias_scope && + Kind != LLVMContext::MD_noalias && + Kind != LLVMContext::MD_fpmath && + Kind != LLVMContext::MD_nontemporal) + continue; + + To->setMetadata(Kind, M.second); + } +} + +/// \brief Propagate known metadata from one instruction to a vector of others. +static void propagateMetadata(SmallVectorImpl<Value *> &To, + const Instruction *From) { + for (Value *V : To) + if (Instruction *I = dyn_cast<Instruction>(V)) + propagateMetadata(I, From); +} + +/// \brief The group of interleaved loads/stores sharing the same stride and +/// close to each other. +/// +/// Each member in this group has an index starting from 0, and the largest +/// index should be less than interleaved factor, which is equal to the absolute +/// value of the access's stride. +/// +/// E.g. An interleaved load group of factor 4: +/// for (unsigned i = 0; i < 1024; i+=4) { +/// a = A[i]; // Member of index 0 +/// b = A[i+1]; // Member of index 1 +/// d = A[i+3]; // Member of index 3 +/// ... +/// } +/// +/// An interleaved store group of factor 4: +/// for (unsigned i = 0; i < 1024; i+=4) { +/// ... +/// A[i] = a; // Member of index 0 +/// A[i+1] = b; // Member of index 1 +/// A[i+2] = c; // Member of index 2 +/// A[i+3] = d; // Member of index 3 +/// } +/// +/// Note: the interleaved load group could have gaps (missing members), but +/// the interleaved store group doesn't allow gaps. +class InterleaveGroup { +public: + InterleaveGroup(Instruction *Instr, int Stride, unsigned Align) + : Align(Align), SmallestKey(0), LargestKey(0), InsertPos(Instr) { + assert(Align && "The alignment should be non-zero"); + + Factor = std::abs(Stride); + assert(Factor > 1 && "Invalid interleave factor"); + + Reverse = Stride < 0; + Members[0] = Instr; + } + + bool isReverse() const { return Reverse; } + unsigned getFactor() const { return Factor; } + unsigned getAlignment() const { return Align; } + unsigned getNumMembers() const { return Members.size(); } + + /// \brief Try to insert a new member \p Instr with index \p Index and + /// alignment \p NewAlign. The index is related to the leader and it could be + /// negative if it is the new leader. + /// + /// \returns false if the instruction doesn't belong to the group. + bool insertMember(Instruction *Instr, int Index, unsigned NewAlign) { + assert(NewAlign && "The new member's alignment should be non-zero"); + + int Key = Index + SmallestKey; + + // Skip if there is already a member with the same index. + if (Members.count(Key)) + return false; + + if (Key > LargestKey) { + // The largest index is always less than the interleave factor. + if (Index >= static_cast<int>(Factor)) + return false; + + LargestKey = Key; + } else if (Key < SmallestKey) { + // The largest index is always less than the interleave factor. + if (LargestKey - Key >= static_cast<int>(Factor)) + return false; + + SmallestKey = Key; + } + + // It's always safe to select the minimum alignment. + Align = std::min(Align, NewAlign); + Members[Key] = Instr; + return true; + } + + /// \brief Get the member with the given index \p Index + /// + /// \returns nullptr if contains no such member. + Instruction *getMember(unsigned Index) const { + int Key = SmallestKey + Index; + if (!Members.count(Key)) + return nullptr; + + return Members.find(Key)->second; + } + + /// \brief Get the index for the given member. Unlike the key in the member + /// map, the index starts from 0. + unsigned getIndex(Instruction *Instr) const { + for (auto I : Members) + if (I.second == Instr) + return I.first - SmallestKey; + + llvm_unreachable("InterleaveGroup contains no such member"); + } + + Instruction *getInsertPos() const { return InsertPos; } + void setInsertPos(Instruction *Inst) { InsertPos = Inst; } + +private: + unsigned Factor; // Interleave Factor. + bool Reverse; + unsigned Align; + DenseMap<int, Instruction *> Members; + int SmallestKey; + int LargestKey; + + // To avoid breaking dependences, vectorized instructions of an interleave + // group should be inserted at either the first load or the last store in + // program order. + // + // E.g. %even = load i32 // Insert Position + // %add = add i32 %even // Use of %even + // %odd = load i32 + // + // store i32 %even + // %odd = add i32 // Def of %odd + // store i32 %odd // Insert Position + Instruction *InsertPos; +}; + +/// \brief Drive the analysis of interleaved memory accesses in the loop. +/// +/// Use this class to analyze interleaved accesses only when we can vectorize +/// a loop. Otherwise it's meaningless to do analysis as the vectorization +/// on interleaved accesses is unsafe. +/// +/// The analysis collects interleave groups and records the relationships +/// between the member and the group in a map. +class InterleavedAccessInfo { +public: + InterleavedAccessInfo(PredicatedScalarEvolution &PSE, Loop *L, + DominatorTree *DT) + : PSE(PSE), TheLoop(L), DT(DT) {} + + ~InterleavedAccessInfo() { + SmallSet<InterleaveGroup *, 4> DelSet; + // Avoid releasing a pointer twice. + for (auto &I : InterleaveGroupMap) + DelSet.insert(I.second); + for (auto *Ptr : DelSet) + delete Ptr; + } + + /// \brief Analyze the interleaved accesses and collect them in interleave + /// groups. Substitute symbolic strides using \p Strides. + void analyzeInterleaving(const ValueToValueMap &Strides); + + /// \brief Check if \p Instr belongs to any interleave group. + bool isInterleaved(Instruction *Instr) const { + return InterleaveGroupMap.count(Instr); + } + + /// \brief Get the interleave group that \p Instr belongs to. + /// + /// \returns nullptr if doesn't have such group. + InterleaveGroup *getInterleaveGroup(Instruction *Instr) const { + if (InterleaveGroupMap.count(Instr)) + return InterleaveGroupMap.find(Instr)->second; + return nullptr; + } + +private: + /// A wrapper around ScalarEvolution, used to add runtime SCEV checks. + /// Simplifies SCEV expressions in the context of existing SCEV assumptions. + /// The interleaved access analysis can also add new predicates (for example + /// by versioning strides of pointers). + PredicatedScalarEvolution &PSE; + Loop *TheLoop; + DominatorTree *DT; + + /// Holds the relationships between the members and the interleave group. + DenseMap<Instruction *, InterleaveGroup *> InterleaveGroupMap; + + /// \brief The descriptor for a strided memory access. + struct StrideDescriptor { + StrideDescriptor(int Stride, const SCEV *Scev, unsigned Size, + unsigned Align) + : Stride(Stride), Scev(Scev), Size(Size), Align(Align) {} + + StrideDescriptor() : Stride(0), Scev(nullptr), Size(0), Align(0) {} + + int Stride; // The access's stride. It is negative for a reverse access. + const SCEV *Scev; // The scalar expression of this access + unsigned Size; // The size of the memory object. + unsigned Align; // The alignment of this access. + }; + + /// \brief Create a new interleave group with the given instruction \p Instr, + /// stride \p Stride and alignment \p Align. + /// + /// \returns the newly created interleave group. + InterleaveGroup *createInterleaveGroup(Instruction *Instr, int Stride, + unsigned Align) { + assert(!InterleaveGroupMap.count(Instr) && + "Already in an interleaved access group"); + InterleaveGroupMap[Instr] = new InterleaveGroup(Instr, Stride, Align); + return InterleaveGroupMap[Instr]; + } + + /// \brief Release the group and remove all the relationships. + void releaseGroup(InterleaveGroup *Group) { + for (unsigned i = 0; i < Group->getFactor(); i++) + if (Instruction *Member = Group->getMember(i)) + InterleaveGroupMap.erase(Member); + + delete Group; + } + + /// \brief Collect all the accesses with a constant stride in program order. + void collectConstStridedAccesses( + MapVector<Instruction *, StrideDescriptor> &StrideAccesses, + const ValueToValueMap &Strides); +}; + +/// Utility class for getting and setting loop vectorizer hints in the form +/// of loop metadata. +/// This class keeps a number of loop annotations locally (as member variables) +/// and can, upon request, write them back as metadata on the loop. It will +/// initially scan the loop for existing metadata, and will update the local +/// values based on information in the loop. +/// We cannot write all values to metadata, as the mere presence of some info, +/// for example 'force', means a decision has been made. So, we need to be +/// careful NOT to add them if the user hasn't specifically asked so. +class LoopVectorizeHints { + enum HintKind { + HK_WIDTH, + HK_UNROLL, + HK_FORCE + }; + + /// Hint - associates name and validation with the hint value. + struct Hint { + const char * Name; + unsigned Value; // This may have to change for non-numeric values. + HintKind Kind; + + Hint(const char * Name, unsigned Value, HintKind Kind) + : Name(Name), Value(Value), Kind(Kind) { } + + bool validate(unsigned Val) { + switch (Kind) { + case HK_WIDTH: + return isPowerOf2_32(Val) && Val <= VectorizerParams::MaxVectorWidth; + case HK_UNROLL: + return isPowerOf2_32(Val) && Val <= MaxInterleaveFactor; + case HK_FORCE: + return (Val <= 1); + } + return false; + } + }; + + /// Vectorization width. + Hint Width; + /// Vectorization interleave factor. + Hint Interleave; + /// Vectorization forced + Hint Force; + + /// Return the loop metadata prefix. + static StringRef Prefix() { return "llvm.loop."; } + +public: + enum ForceKind { + FK_Undefined = -1, ///< Not selected. + FK_Disabled = 0, ///< Forcing disabled. + FK_Enabled = 1, ///< Forcing enabled. + }; + + LoopVectorizeHints(const Loop *L, bool DisableInterleaving) + : Width("vectorize.width", VectorizerParams::VectorizationFactor, + HK_WIDTH), + Interleave("interleave.count", DisableInterleaving, HK_UNROLL), + Force("vectorize.enable", FK_Undefined, HK_FORCE), + TheLoop(L) { + // Populate values with existing loop metadata. + getHintsFromMetadata(); + + // force-vector-interleave overrides DisableInterleaving. + if (VectorizerParams::isInterleaveForced()) + Interleave.Value = VectorizerParams::VectorizationInterleave; + + DEBUG(if (DisableInterleaving && Interleave.Value == 1) dbgs() + << "LV: Interleaving disabled by the pass manager\n"); + } + + /// Mark the loop L as already vectorized by setting the width to 1. + void setAlreadyVectorized() { + Width.Value = Interleave.Value = 1; + Hint Hints[] = {Width, Interleave}; + writeHintsToMetadata(Hints); + } + + bool allowVectorization(Function *F, Loop *L, bool AlwaysVectorize) const { + if (getForce() == LoopVectorizeHints::FK_Disabled) { + DEBUG(dbgs() << "LV: Not vectorizing: #pragma vectorize disable.\n"); + emitOptimizationRemarkAnalysis(F->getContext(), + vectorizeAnalysisPassName(), *F, + L->getStartLoc(), emitRemark()); + return false; + } + + if (!AlwaysVectorize && getForce() != LoopVectorizeHints::FK_Enabled) { + DEBUG(dbgs() << "LV: Not vectorizing: No #pragma vectorize enable.\n"); + emitOptimizationRemarkAnalysis(F->getContext(), + vectorizeAnalysisPassName(), *F, + L->getStartLoc(), emitRemark()); + return false; + } + + if (getWidth() == 1 && getInterleave() == 1) { + // FIXME: Add a separate metadata to indicate when the loop has already + // been vectorized instead of setting width and count to 1. + DEBUG(dbgs() << "LV: Not vectorizing: Disabled/already vectorized.\n"); + // FIXME: Add interleave.disable metadata. This will allow + // vectorize.disable to be used without disabling the pass and errors + // to differentiate between disabled vectorization and a width of 1. + emitOptimizationRemarkAnalysis( + F->getContext(), vectorizeAnalysisPassName(), *F, L->getStartLoc(), + "loop not vectorized: vectorization and interleaving are explicitly " + "disabled, or vectorize width and interleave count are both set to " + "1"); + return false; + } + + return true; + } + + /// Dumps all the hint information. + std::string emitRemark() const { + VectorizationReport R; + if (Force.Value == LoopVectorizeHints::FK_Disabled) + R << "vectorization is explicitly disabled"; + else { + R << "use -Rpass-analysis=loop-vectorize for more info"; + if (Force.Value == LoopVectorizeHints::FK_Enabled) { + R << " (Force=true"; + if (Width.Value != 0) + R << ", Vector Width=" << Width.Value; + if (Interleave.Value != 0) + R << ", Interleave Count=" << Interleave.Value; + R << ")"; + } + } + + return R.str(); + } + + unsigned getWidth() const { return Width.Value; } + unsigned getInterleave() const { return Interleave.Value; } + enum ForceKind getForce() const { return (ForceKind)Force.Value; } + const char *vectorizeAnalysisPassName() const { + // If hints are provided that don't disable vectorization use the + // AlwaysPrint pass name to force the frontend to print the diagnostic. + if (getWidth() == 1) + return LV_NAME; + if (getForce() == LoopVectorizeHints::FK_Disabled) + return LV_NAME; + if (getForce() == LoopVectorizeHints::FK_Undefined && getWidth() == 0) + return LV_NAME; + return DiagnosticInfo::AlwaysPrint; + } + + bool allowReordering() const { + // When enabling loop hints are provided we allow the vectorizer to change + // the order of operations that is given by the scalar loop. This is not + // enabled by default because can be unsafe or inefficient. For example, + // reordering floating-point operations will change the way round-off + // error accumulates in the loop. + return getForce() == LoopVectorizeHints::FK_Enabled || getWidth() > 1; + } + +private: + /// Find hints specified in the loop metadata and update local values. + void getHintsFromMetadata() { + MDNode *LoopID = TheLoop->getLoopID(); + if (!LoopID) + return; + + // First operand should refer to the loop id itself. + assert(LoopID->getNumOperands() > 0 && "requires at least one operand"); + assert(LoopID->getOperand(0) == LoopID && "invalid loop id"); + + for (unsigned i = 1, ie = LoopID->getNumOperands(); i < ie; ++i) { + const MDString *S = nullptr; + SmallVector<Metadata *, 4> Args; + + // The expected hint is either a MDString or a MDNode with the first + // operand a MDString. + if (const MDNode *MD = dyn_cast<MDNode>(LoopID->getOperand(i))) { + if (!MD || MD->getNumOperands() == 0) + continue; + S = dyn_cast<MDString>(MD->getOperand(0)); + for (unsigned i = 1, ie = MD->getNumOperands(); i < ie; ++i) + Args.push_back(MD->getOperand(i)); + } else { + S = dyn_cast<MDString>(LoopID->getOperand(i)); + assert(Args.size() == 0 && "too many arguments for MDString"); + } + + if (!S) + continue; + + // Check if the hint starts with the loop metadata prefix. + StringRef Name = S->getString(); + if (Args.size() == 1) + setHint(Name, Args[0]); + } + } + + /// Checks string hint with one operand and set value if valid. + void setHint(StringRef Name, Metadata *Arg) { + if (!Name.startswith(Prefix())) + return; + Name = Name.substr(Prefix().size(), StringRef::npos); + + const ConstantInt *C = mdconst::dyn_extract<ConstantInt>(Arg); + if (!C) return; + unsigned Val = C->getZExtValue(); + + Hint *Hints[] = {&Width, &Interleave, &Force}; + for (auto H : Hints) { + if (Name == H->Name) { + if (H->validate(Val)) + H->Value = Val; + else + DEBUG(dbgs() << "LV: ignoring invalid hint '" << Name << "'\n"); + break; + } + } + } + + /// Create a new hint from name / value pair. + MDNode *createHintMetadata(StringRef Name, unsigned V) const { + LLVMContext &Context = TheLoop->getHeader()->getContext(); + Metadata *MDs[] = {MDString::get(Context, Name), + ConstantAsMetadata::get( + ConstantInt::get(Type::getInt32Ty(Context), V))}; + return MDNode::get(Context, MDs); + } + + /// Matches metadata with hint name. + bool matchesHintMetadataName(MDNode *Node, ArrayRef<Hint> HintTypes) { + MDString* Name = dyn_cast<MDString>(Node->getOperand(0)); + if (!Name) + return false; + + for (auto H : HintTypes) + if (Name->getString().endswith(H.Name)) + return true; + return false; + } + + /// Sets current hints into loop metadata, keeping other values intact. + void writeHintsToMetadata(ArrayRef<Hint> HintTypes) { + if (HintTypes.size() == 0) + return; + + // Reserve the first element to LoopID (see below). + SmallVector<Metadata *, 4> MDs(1); + // If the loop already has metadata, then ignore the existing operands. + MDNode *LoopID = TheLoop->getLoopID(); + if (LoopID) { + for (unsigned i = 1, ie = LoopID->getNumOperands(); i < ie; ++i) { + MDNode *Node = cast<MDNode>(LoopID->getOperand(i)); + // If node in update list, ignore old value. + if (!matchesHintMetadataName(Node, HintTypes)) + MDs.push_back(Node); + } + } + + // Now, add the missing hints. + for (auto H : HintTypes) + MDs.push_back(createHintMetadata(Twine(Prefix(), H.Name).str(), H.Value)); + + // Replace current metadata node with new one. + LLVMContext &Context = TheLoop->getHeader()->getContext(); + MDNode *NewLoopID = MDNode::get(Context, MDs); + // Set operand 0 to refer to the loop id itself. + NewLoopID->replaceOperandWith(0, NewLoopID); + + TheLoop->setLoopID(NewLoopID); + } + + /// The loop these hints belong to. + const Loop *TheLoop; +}; + +static void emitAnalysisDiag(const Function *TheFunction, const Loop *TheLoop, + const LoopVectorizeHints &Hints, + const LoopAccessReport &Message) { + const char *Name = Hints.vectorizeAnalysisPassName(); + LoopAccessReport::emitAnalysis(Message, TheFunction, TheLoop, Name); +} + +static void emitMissedWarning(Function *F, Loop *L, + const LoopVectorizeHints &LH) { + emitOptimizationRemarkMissed(F->getContext(), LV_NAME, *F, L->getStartLoc(), + LH.emitRemark()); + + if (LH.getForce() == LoopVectorizeHints::FK_Enabled) { + if (LH.getWidth() != 1) + emitLoopVectorizeWarning( + F->getContext(), *F, L->getStartLoc(), + "failed explicitly specified loop vectorization"); + else if (LH.getInterleave() != 1) + emitLoopInterleaveWarning( + F->getContext(), *F, L->getStartLoc(), + "failed explicitly specified loop interleaving"); + } +} + +/// LoopVectorizationLegality checks if it is legal to vectorize a loop, and +/// to what vectorization factor. +/// This class does not look at the profitability of vectorization, only the +/// legality. This class has two main kinds of checks: +/// * Memory checks - The code in canVectorizeMemory checks if vectorization +/// will change the order of memory accesses in a way that will change the +/// correctness of the program. +/// * Scalars checks - The code in canVectorizeInstrs and canVectorizeMemory +/// checks for a number of different conditions, such as the availability of a +/// single induction variable, that all types are supported and vectorize-able, +/// etc. This code reflects the capabilities of InnerLoopVectorizer. +/// This class is also used by InnerLoopVectorizer for identifying +/// induction variable and the different reduction variables. +class LoopVectorizationLegality { +public: + LoopVectorizationLegality(Loop *L, PredicatedScalarEvolution &PSE, + DominatorTree *DT, TargetLibraryInfo *TLI, + AliasAnalysis *AA, Function *F, + const TargetTransformInfo *TTI, + LoopAccessAnalysis *LAA, + LoopVectorizationRequirements *R, + const LoopVectorizeHints *H) + : NumPredStores(0), TheLoop(L), PSE(PSE), TLI(TLI), TheFunction(F), + TTI(TTI), DT(DT), LAA(LAA), LAI(nullptr), InterleaveInfo(PSE, L, DT), + Induction(nullptr), WidestIndTy(nullptr), HasFunNoNaNAttr(false), + Requirements(R), Hints(H) {} + + /// ReductionList contains the reduction descriptors for all + /// of the reductions that were found in the loop. + typedef DenseMap<PHINode *, RecurrenceDescriptor> ReductionList; + + /// InductionList saves induction variables and maps them to the + /// induction descriptor. + typedef MapVector<PHINode*, InductionDescriptor> InductionList; + + /// Returns true if it is legal to vectorize this loop. + /// This does not mean that it is profitable to vectorize this + /// loop, only that it is legal to do so. + bool canVectorize(); + + /// Returns the Induction variable. + PHINode *getInduction() { return Induction; } + + /// Returns the reduction variables found in the loop. + ReductionList *getReductionVars() { return &Reductions; } + + /// Returns the induction variables found in the loop. + InductionList *getInductionVars() { return &Inductions; } + + /// Returns the widest induction type. + Type *getWidestInductionType() { return WidestIndTy; } + + /// Returns True if V is an induction variable in this loop. + bool isInductionVariable(const Value *V); + + /// Returns True if PN is a reduction variable in this loop. + bool isReductionVariable(PHINode *PN) { return Reductions.count(PN); } + + /// Return true if the block BB needs to be predicated in order for the loop + /// to be vectorized. + bool blockNeedsPredication(BasicBlock *BB); + + /// Check if this pointer is consecutive when vectorizing. This happens + /// when the last index of the GEP is the induction variable, or that the + /// pointer itself is an induction variable. + /// This check allows us to vectorize A[idx] into a wide load/store. + /// Returns: + /// 0 - Stride is unknown or non-consecutive. + /// 1 - Address is consecutive. + /// -1 - Address is consecutive, and decreasing. + int isConsecutivePtr(Value *Ptr); + + /// Returns true if the value V is uniform within the loop. + bool isUniform(Value *V); + + /// Returns true if this instruction will remain scalar after vectorization. + bool isUniformAfterVectorization(Instruction* I) { return Uniforms.count(I); } + + /// Returns the information that we collected about runtime memory check. + const RuntimePointerChecking *getRuntimePointerChecking() const { + return LAI->getRuntimePointerChecking(); + } + + const LoopAccessInfo *getLAI() const { + return LAI; + } + + /// \brief Check if \p Instr belongs to any interleaved access group. + bool isAccessInterleaved(Instruction *Instr) { + return InterleaveInfo.isInterleaved(Instr); + } + + /// \brief Get the interleaved access group that \p Instr belongs to. + const InterleaveGroup *getInterleavedAccessGroup(Instruction *Instr) { + return InterleaveInfo.getInterleaveGroup(Instr); + } + + unsigned getMaxSafeDepDistBytes() { return LAI->getMaxSafeDepDistBytes(); } + + bool hasStride(Value *V) { return StrideSet.count(V); } + bool mustCheckStrides() { return !StrideSet.empty(); } + SmallPtrSet<Value *, 8>::iterator strides_begin() { + return StrideSet.begin(); + } + SmallPtrSet<Value *, 8>::iterator strides_end() { return StrideSet.end(); } + + /// Returns true if the target machine supports masked store operation + /// for the given \p DataType and kind of access to \p Ptr. + bool isLegalMaskedStore(Type *DataType, Value *Ptr) { + return isConsecutivePtr(Ptr) && TTI->isLegalMaskedStore(DataType); + } + /// Returns true if the target machine supports masked load operation + /// for the given \p DataType and kind of access to \p Ptr. + bool isLegalMaskedLoad(Type *DataType, Value *Ptr) { + return isConsecutivePtr(Ptr) && TTI->isLegalMaskedLoad(DataType); + } + /// Returns true if vector representation of the instruction \p I + /// requires mask. + bool isMaskRequired(const Instruction* I) { + return (MaskedOp.count(I) != 0); + } + unsigned getNumStores() const { + return LAI->getNumStores(); + } + unsigned getNumLoads() const { + return LAI->getNumLoads(); + } + unsigned getNumPredStores() const { + return NumPredStores; + } +private: + /// Check if a single basic block loop is vectorizable. + /// At this point we know that this is a loop with a constant trip count + /// and we only need to check individual instructions. + bool canVectorizeInstrs(); + + /// When we vectorize loops we may change the order in which + /// we read and write from memory. This method checks if it is + /// legal to vectorize the code, considering only memory constrains. + /// Returns true if the loop is vectorizable + bool canVectorizeMemory(); + + /// Return true if we can vectorize this loop using the IF-conversion + /// transformation. + bool canVectorizeWithIfConvert(); + + /// Collect the variables that need to stay uniform after vectorization. + void collectLoopUniforms(); + + /// Return true if all of the instructions in the block can be speculatively + /// executed. \p SafePtrs is a list of addresses that are known to be legal + /// and we know that we can read from them without segfault. + bool blockCanBePredicated(BasicBlock *BB, SmallPtrSetImpl<Value *> &SafePtrs); + + /// \brief Collect memory access with loop invariant strides. + /// + /// Looks for accesses like "a[i * StrideA]" where "StrideA" is loop + /// invariant. + void collectStridedAccess(Value *LoadOrStoreInst); + + /// Report an analysis message to assist the user in diagnosing loops that are + /// not vectorized. These are handled as LoopAccessReport rather than + /// VectorizationReport because the << operator of VectorizationReport returns + /// LoopAccessReport. + void emitAnalysis(const LoopAccessReport &Message) const { + emitAnalysisDiag(TheFunction, TheLoop, *Hints, Message); + } + + unsigned NumPredStores; + + /// The loop that we evaluate. + Loop *TheLoop; + /// A wrapper around ScalarEvolution used to add runtime SCEV checks. + /// Applies dynamic knowledge to simplify SCEV expressions in the context + /// of existing SCEV assumptions. The analysis will also add a minimal set + /// of new predicates if this is required to enable vectorization and + /// unrolling. + PredicatedScalarEvolution &PSE; + /// Target Library Info. + TargetLibraryInfo *TLI; + /// Parent function + Function *TheFunction; + /// Target Transform Info + const TargetTransformInfo *TTI; + /// Dominator Tree. + DominatorTree *DT; + // LoopAccess analysis. + LoopAccessAnalysis *LAA; + // And the loop-accesses info corresponding to this loop. This pointer is + // null until canVectorizeMemory sets it up. + const LoopAccessInfo *LAI; + + /// The interleave access information contains groups of interleaved accesses + /// with the same stride and close to each other. + InterleavedAccessInfo InterleaveInfo; + + // --- vectorization state --- // + + /// Holds the integer induction variable. This is the counter of the + /// loop. + PHINode *Induction; + /// Holds the reduction variables. + ReductionList Reductions; + /// Holds all of the induction variables that we found in the loop. + /// Notice that inductions don't need to start at zero and that induction + /// variables can be pointers. + InductionList Inductions; + /// Holds the widest induction type encountered. + Type *WidestIndTy; + + /// Allowed outside users. This holds the reduction + /// vars which can be accessed from outside the loop. + SmallPtrSet<Value*, 4> AllowedExit; + /// This set holds the variables which are known to be uniform after + /// vectorization. + SmallPtrSet<Instruction*, 4> Uniforms; + + /// Can we assume the absence of NaNs. + bool HasFunNoNaNAttr; + + /// Vectorization requirements that will go through late-evaluation. + LoopVectorizationRequirements *Requirements; + + /// Used to emit an analysis of any legality issues. + const LoopVectorizeHints *Hints; + + ValueToValueMap Strides; + SmallPtrSet<Value *, 8> StrideSet; + + /// While vectorizing these instructions we have to generate a + /// call to the appropriate masked intrinsic + SmallPtrSet<const Instruction *, 8> MaskedOp; +}; + +/// LoopVectorizationCostModel - estimates the expected speedups due to +/// vectorization. +/// In many cases vectorization is not profitable. This can happen because of +/// a number of reasons. In this class we mainly attempt to predict the +/// expected speedup/slowdowns due to the supported instruction set. We use the +/// TargetTransformInfo to query the different backends for the cost of +/// different operations. +class LoopVectorizationCostModel { +public: + LoopVectorizationCostModel(Loop *L, PredicatedScalarEvolution &PSE, + LoopInfo *LI, LoopVectorizationLegality *Legal, + const TargetTransformInfo &TTI, + const TargetLibraryInfo *TLI, DemandedBits *DB, + AssumptionCache *AC, const Function *F, + const LoopVectorizeHints *Hints) + : TheLoop(L), PSE(PSE), LI(LI), Legal(Legal), TTI(TTI), TLI(TLI), DB(DB), + AC(AC), TheFunction(F), Hints(Hints) {} + + /// Information about vectorization costs + struct VectorizationFactor { + unsigned Width; // Vector width with best cost + unsigned Cost; // Cost of the loop with that width + }; + /// \return The most profitable vectorization factor and the cost of that VF. + /// This method checks every power of two up to VF. If UserVF is not ZERO + /// then this vectorization factor will be selected if vectorization is + /// possible. + VectorizationFactor selectVectorizationFactor(bool OptForSize); + + /// \return The size (in bits) of the smallest and widest types in the code + /// that needs to be vectorized. We ignore values that remain scalar such as + /// 64 bit loop indices. + std::pair<unsigned, unsigned> getSmallestAndWidestTypes(); + + /// \return The desired interleave count. + /// If interleave count has been specified by metadata it will be returned. + /// Otherwise, the interleave count is computed and returned. VF and LoopCost + /// are the selected vectorization factor and the cost of the selected VF. + unsigned selectInterleaveCount(bool OptForSize, unsigned VF, + unsigned LoopCost); + + /// \return The most profitable unroll factor. + /// This method finds the best unroll-factor based on register pressure and + /// other parameters. VF and LoopCost are the selected vectorization factor + /// and the cost of the selected VF. + unsigned computeInterleaveCount(bool OptForSize, unsigned VF, + unsigned LoopCost); + + /// \brief A struct that represents some properties of the register usage + /// of a loop. + struct RegisterUsage { + /// Holds the number of loop invariant values that are used in the loop. + unsigned LoopInvariantRegs; + /// Holds the maximum number of concurrent live intervals in the loop. + unsigned MaxLocalUsers; + /// Holds the number of instructions in the loop. + unsigned NumInstructions; + }; + + /// \return Returns information about the register usages of the loop for the + /// given vectorization factors. + SmallVector<RegisterUsage, 8> + calculateRegisterUsage(const SmallVector<unsigned, 8> &VFs); + + /// Collect values we want to ignore in the cost model. + void collectValuesToIgnore(); + +private: + /// Returns the expected execution cost. The unit of the cost does + /// not matter because we use the 'cost' units to compare different + /// vector widths. The cost that is returned is *not* normalized by + /// the factor width. + unsigned expectedCost(unsigned VF); + + /// Returns the execution time cost of an instruction for a given vector + /// width. Vector width of one means scalar. + unsigned getInstructionCost(Instruction *I, unsigned VF); + + /// Returns whether the instruction is a load or store and will be a emitted + /// as a vector operation. + bool isConsecutiveLoadOrStore(Instruction *I); + + /// Report an analysis message to assist the user in diagnosing loops that are + /// not vectorized. These are handled as LoopAccessReport rather than + /// VectorizationReport because the << operator of VectorizationReport returns + /// LoopAccessReport. + void emitAnalysis(const LoopAccessReport &Message) const { + emitAnalysisDiag(TheFunction, TheLoop, *Hints, Message); + } + +public: + /// Map of scalar integer values to the smallest bitwidth they can be legally + /// represented as. The vector equivalents of these values should be truncated + /// to this type. + MapVector<Instruction*,uint64_t> MinBWs; + + /// The loop that we evaluate. + Loop *TheLoop; + /// Predicated scalar evolution analysis. + PredicatedScalarEvolution &PSE; + /// Loop Info analysis. + LoopInfo *LI; + /// Vectorization legality. + LoopVectorizationLegality *Legal; + /// Vector target information. + const TargetTransformInfo &TTI; + /// Target Library Info. + const TargetLibraryInfo *TLI; + /// Demanded bits analysis. + DemandedBits *DB; + /// Assumption cache. + AssumptionCache *AC; + const Function *TheFunction; + /// Loop Vectorize Hint. + const LoopVectorizeHints *Hints; + /// Values to ignore in the cost model. + SmallPtrSet<const Value *, 16> ValuesToIgnore; + /// Values to ignore in the cost model when VF > 1. + SmallPtrSet<const Value *, 16> VecValuesToIgnore; +}; + +/// \brief This holds vectorization requirements that must be verified late in +/// the process. The requirements are set by legalize and costmodel. Once +/// vectorization has been determined to be possible and profitable the +/// requirements can be verified by looking for metadata or compiler options. +/// For example, some loops require FP commutativity which is only allowed if +/// vectorization is explicitly specified or if the fast-math compiler option +/// has been provided. +/// Late evaluation of these requirements allows helpful diagnostics to be +/// composed that tells the user what need to be done to vectorize the loop. For +/// example, by specifying #pragma clang loop vectorize or -ffast-math. Late +/// evaluation should be used only when diagnostics can generated that can be +/// followed by a non-expert user. +class LoopVectorizationRequirements { +public: + LoopVectorizationRequirements() + : NumRuntimePointerChecks(0), UnsafeAlgebraInst(nullptr) {} + + void addUnsafeAlgebraInst(Instruction *I) { + // First unsafe algebra instruction. + if (!UnsafeAlgebraInst) + UnsafeAlgebraInst = I; + } + + void addRuntimePointerChecks(unsigned Num) { NumRuntimePointerChecks = Num; } + + bool doesNotMeet(Function *F, Loop *L, const LoopVectorizeHints &Hints) { + const char *Name = Hints.vectorizeAnalysisPassName(); + bool Failed = false; + if (UnsafeAlgebraInst && !Hints.allowReordering()) { + emitOptimizationRemarkAnalysisFPCommute( + F->getContext(), Name, *F, UnsafeAlgebraInst->getDebugLoc(), + VectorizationReport() << "cannot prove it is safe to reorder " + "floating-point operations"); + Failed = true; + } + + // Test if runtime memcheck thresholds are exceeded. + bool PragmaThresholdReached = + NumRuntimePointerChecks > PragmaVectorizeMemoryCheckThreshold; + bool ThresholdReached = + NumRuntimePointerChecks > VectorizerParams::RuntimeMemoryCheckThreshold; + if ((ThresholdReached && !Hints.allowReordering()) || + PragmaThresholdReached) { + emitOptimizationRemarkAnalysisAliasing( + F->getContext(), Name, *F, L->getStartLoc(), + VectorizationReport() + << "cannot prove it is safe to reorder memory operations"); + DEBUG(dbgs() << "LV: Too many memory checks needed.\n"); + Failed = true; + } + + return Failed; + } + +private: + unsigned NumRuntimePointerChecks; + Instruction *UnsafeAlgebraInst; +}; + +static void addInnerLoop(Loop &L, SmallVectorImpl<Loop *> &V) { + if (L.empty()) + return V.push_back(&L); + + for (Loop *InnerL : L) + addInnerLoop(*InnerL, V); +} + +/// The LoopVectorize Pass. +struct LoopVectorize : public FunctionPass { + /// Pass identification, replacement for typeid + static char ID; + + explicit LoopVectorize(bool NoUnrolling = false, bool AlwaysVectorize = true) + : FunctionPass(ID), + DisableUnrolling(NoUnrolling), + AlwaysVectorize(AlwaysVectorize) { + initializeLoopVectorizePass(*PassRegistry::getPassRegistry()); + } + + ScalarEvolution *SE; + LoopInfo *LI; + TargetTransformInfo *TTI; + DominatorTree *DT; + BlockFrequencyInfo *BFI; + TargetLibraryInfo *TLI; + DemandedBits *DB; + AliasAnalysis *AA; + AssumptionCache *AC; + LoopAccessAnalysis *LAA; + bool DisableUnrolling; + bool AlwaysVectorize; + + BlockFrequency ColdEntryFreq; + + bool runOnFunction(Function &F) override { + SE = &getAnalysis<ScalarEvolutionWrapperPass>().getSE(); + LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo(); + TTI = &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F); + DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree(); + BFI = &getAnalysis<BlockFrequencyInfoWrapperPass>().getBFI(); + auto *TLIP = getAnalysisIfAvailable<TargetLibraryInfoWrapperPass>(); + TLI = TLIP ? &TLIP->getTLI() : nullptr; + AA = &getAnalysis<AAResultsWrapperPass>().getAAResults(); + AC = &getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F); + LAA = &getAnalysis<LoopAccessAnalysis>(); + DB = &getAnalysis<DemandedBits>(); + + // Compute some weights outside of the loop over the loops. Compute this + // using a BranchProbability to re-use its scaling math. + const BranchProbability ColdProb(1, 5); // 20% + ColdEntryFreq = BlockFrequency(BFI->getEntryFreq()) * ColdProb; + + // Don't attempt if + // 1. the target claims to have no vector registers, and + // 2. interleaving won't help ILP. + // + // The second condition is necessary because, even if the target has no + // vector registers, loop vectorization may still enable scalar + // interleaving. + if (!TTI->getNumberOfRegisters(true) && TTI->getMaxInterleaveFactor(1) < 2) + return false; + + // Build up a worklist of inner-loops to vectorize. This is necessary as + // the act of vectorizing or partially unrolling a loop creates new loops + // and can invalidate iterators across the loops. + SmallVector<Loop *, 8> Worklist; + + for (Loop *L : *LI) + addInnerLoop(*L, Worklist); + + LoopsAnalyzed += Worklist.size(); + + // Now walk the identified inner loops. + bool Changed = false; + while (!Worklist.empty()) + Changed |= processLoop(Worklist.pop_back_val()); + + // Process each loop nest in the function. + return Changed; + } + + static void AddRuntimeUnrollDisableMetaData(Loop *L) { + SmallVector<Metadata *, 4> MDs; + // Reserve first location for self reference to the LoopID metadata node. + MDs.push_back(nullptr); + bool IsUnrollMetadata = false; + MDNode *LoopID = L->getLoopID(); + if (LoopID) { + // First find existing loop unrolling disable metadata. + for (unsigned i = 1, ie = LoopID->getNumOperands(); i < ie; ++i) { + MDNode *MD = dyn_cast<MDNode>(LoopID->getOperand(i)); + if (MD) { + const MDString *S = dyn_cast<MDString>(MD->getOperand(0)); + IsUnrollMetadata = + S && S->getString().startswith("llvm.loop.unroll.disable"); + } + MDs.push_back(LoopID->getOperand(i)); + } + } + + if (!IsUnrollMetadata) { + // Add runtime unroll disable metadata. + LLVMContext &Context = L->getHeader()->getContext(); + SmallVector<Metadata *, 1> DisableOperands; + DisableOperands.push_back( + MDString::get(Context, "llvm.loop.unroll.runtime.disable")); + MDNode *DisableNode = MDNode::get(Context, DisableOperands); + MDs.push_back(DisableNode); + MDNode *NewLoopID = MDNode::get(Context, MDs); + // Set operand 0 to refer to the loop id itself. + NewLoopID->replaceOperandWith(0, NewLoopID); + L->setLoopID(NewLoopID); + } + } + + bool processLoop(Loop *L) { + assert(L->empty() && "Only process inner loops."); + +#ifndef NDEBUG + const std::string DebugLocStr = getDebugLocString(L); +#endif /* NDEBUG */ + + DEBUG(dbgs() << "\nLV: Checking a loop in \"" + << L->getHeader()->getParent()->getName() << "\" from " + << DebugLocStr << "\n"); + + LoopVectorizeHints Hints(L, DisableUnrolling); + + DEBUG(dbgs() << "LV: Loop hints:" + << " force=" + << (Hints.getForce() == LoopVectorizeHints::FK_Disabled + ? "disabled" + : (Hints.getForce() == LoopVectorizeHints::FK_Enabled + ? "enabled" + : "?")) << " width=" << Hints.getWidth() + << " unroll=" << Hints.getInterleave() << "\n"); + + // Function containing loop + Function *F = L->getHeader()->getParent(); + + // Looking at the diagnostic output is the only way to determine if a loop + // was vectorized (other than looking at the IR or machine code), so it + // is important to generate an optimization remark for each loop. Most of + // these messages are generated by emitOptimizationRemarkAnalysis. Remarks + // generated by emitOptimizationRemark and emitOptimizationRemarkMissed are + // less verbose reporting vectorized loops and unvectorized loops that may + // benefit from vectorization, respectively. + + if (!Hints.allowVectorization(F, L, AlwaysVectorize)) { + DEBUG(dbgs() << "LV: Loop hints prevent vectorization.\n"); + return false; + } + + // Check the loop for a trip count threshold: + // do not vectorize loops with a tiny trip count. + const unsigned TC = SE->getSmallConstantTripCount(L); + if (TC > 0u && TC < TinyTripCountVectorThreshold) { + DEBUG(dbgs() << "LV: Found a loop with a very small trip count. " + << "This loop is not worth vectorizing."); + if (Hints.getForce() == LoopVectorizeHints::FK_Enabled) + DEBUG(dbgs() << " But vectorizing was explicitly forced.\n"); + else { + DEBUG(dbgs() << "\n"); + emitAnalysisDiag(F, L, Hints, VectorizationReport() + << "vectorization is not beneficial " + "and is not explicitly forced"); + return false; + } + } + + PredicatedScalarEvolution PSE(*SE); + + // Check if it is legal to vectorize the loop. + LoopVectorizationRequirements Requirements; + LoopVectorizationLegality LVL(L, PSE, DT, TLI, AA, F, TTI, LAA, + &Requirements, &Hints); + if (!LVL.canVectorize()) { + DEBUG(dbgs() << "LV: Not vectorizing: Cannot prove legality.\n"); + emitMissedWarning(F, L, Hints); + return false; + } + + // Use the cost model. + LoopVectorizationCostModel CM(L, PSE, LI, &LVL, *TTI, TLI, DB, AC, F, + &Hints); + CM.collectValuesToIgnore(); + + // Check the function attributes to find out if this function should be + // optimized for size. + bool OptForSize = Hints.getForce() != LoopVectorizeHints::FK_Enabled && + F->optForSize(); + + // Compute the weighted frequency of this loop being executed and see if it + // is less than 20% of the function entry baseline frequency. Note that we + // always have a canonical loop here because we think we *can* vectorize. + // FIXME: This is hidden behind a flag due to pervasive problems with + // exactly what block frequency models. + if (LoopVectorizeWithBlockFrequency) { + BlockFrequency LoopEntryFreq = BFI->getBlockFreq(L->getLoopPreheader()); + if (Hints.getForce() != LoopVectorizeHints::FK_Enabled && + LoopEntryFreq < ColdEntryFreq) + OptForSize = true; + } + + // Check the function attributes to see if implicit floats are allowed. + // FIXME: This check doesn't seem possibly correct -- what if the loop is + // an integer loop and the vector instructions selected are purely integer + // vector instructions? + if (F->hasFnAttribute(Attribute::NoImplicitFloat)) { + DEBUG(dbgs() << "LV: Can't vectorize when the NoImplicitFloat" + "attribute is used.\n"); + emitAnalysisDiag( + F, L, Hints, + VectorizationReport() + << "loop not vectorized due to NoImplicitFloat attribute"); + emitMissedWarning(F, L, Hints); + return false; + } + + // Select the optimal vectorization factor. + const LoopVectorizationCostModel::VectorizationFactor VF = + CM.selectVectorizationFactor(OptForSize); + + // Select the interleave count. + unsigned IC = CM.selectInterleaveCount(OptForSize, VF.Width, VF.Cost); + + // Get user interleave count. + unsigned UserIC = Hints.getInterleave(); + + // Identify the diagnostic messages that should be produced. + std::string VecDiagMsg, IntDiagMsg; + bool VectorizeLoop = true, InterleaveLoop = true; + + if (Requirements.doesNotMeet(F, L, Hints)) { + DEBUG(dbgs() << "LV: Not vectorizing: loop did not meet vectorization " + "requirements.\n"); + emitMissedWarning(F, L, Hints); + return false; + } + + if (VF.Width == 1) { + DEBUG(dbgs() << "LV: Vectorization is possible but not beneficial.\n"); + VecDiagMsg = + "the cost-model indicates that vectorization is not beneficial"; + VectorizeLoop = false; + } + + if (IC == 1 && UserIC <= 1) { + // Tell the user interleaving is not beneficial. + DEBUG(dbgs() << "LV: Interleaving is not beneficial.\n"); + IntDiagMsg = + "the cost-model indicates that interleaving is not beneficial"; + InterleaveLoop = false; + if (UserIC == 1) + IntDiagMsg += + " and is explicitly disabled or interleave count is set to 1"; + } else if (IC > 1 && UserIC == 1) { + // Tell the user interleaving is beneficial, but it explicitly disabled. + DEBUG(dbgs() + << "LV: Interleaving is beneficial but is explicitly disabled."); + IntDiagMsg = "the cost-model indicates that interleaving is beneficial " + "but is explicitly disabled or interleave count is set to 1"; + InterleaveLoop = false; + } + + // Override IC if user provided an interleave count. + IC = UserIC > 0 ? UserIC : IC; + + // Emit diagnostic messages, if any. + const char *VAPassName = Hints.vectorizeAnalysisPassName(); + if (!VectorizeLoop && !InterleaveLoop) { + // Do not vectorize or interleaving the loop. + emitOptimizationRemarkAnalysis(F->getContext(), VAPassName, *F, + L->getStartLoc(), VecDiagMsg); + emitOptimizationRemarkAnalysis(F->getContext(), LV_NAME, *F, + L->getStartLoc(), IntDiagMsg); + return false; + } else if (!VectorizeLoop && InterleaveLoop) { + DEBUG(dbgs() << "LV: Interleave Count is " << IC << '\n'); + emitOptimizationRemarkAnalysis(F->getContext(), VAPassName, *F, + L->getStartLoc(), VecDiagMsg); + } else if (VectorizeLoop && !InterleaveLoop) { + DEBUG(dbgs() << "LV: Found a vectorizable loop (" << VF.Width << ") in " + << DebugLocStr << '\n'); + emitOptimizationRemarkAnalysis(F->getContext(), LV_NAME, *F, + L->getStartLoc(), IntDiagMsg); + } else if (VectorizeLoop && InterleaveLoop) { + DEBUG(dbgs() << "LV: Found a vectorizable loop (" << VF.Width << ") in " + << DebugLocStr << '\n'); + DEBUG(dbgs() << "LV: Interleave Count is " << IC << '\n'); + } + + if (!VectorizeLoop) { + assert(IC > 1 && "interleave count should not be 1 or 0"); + // If we decided that it is not legal to vectorize the loop then + // interleave it. + InnerLoopUnroller Unroller(L, PSE, LI, DT, TLI, TTI, IC); + Unroller.vectorize(&LVL, CM.MinBWs); + + emitOptimizationRemark(F->getContext(), LV_NAME, *F, L->getStartLoc(), + Twine("interleaved loop (interleaved count: ") + + Twine(IC) + ")"); + } else { + // If we decided that it is *legal* to vectorize the loop then do it. + InnerLoopVectorizer LB(L, PSE, LI, DT, TLI, TTI, VF.Width, IC); + LB.vectorize(&LVL, CM.MinBWs); + ++LoopsVectorized; + + // Add metadata to disable runtime unrolling scalar loop when there's no + // runtime check about strides and memory. Because at this situation, + // scalar loop is rarely used not worthy to be unrolled. + if (!LB.IsSafetyChecksAdded()) + AddRuntimeUnrollDisableMetaData(L); + + // Report the vectorization decision. + emitOptimizationRemark(F->getContext(), LV_NAME, *F, L->getStartLoc(), + Twine("vectorized loop (vectorization width: ") + + Twine(VF.Width) + ", interleaved count: " + + Twine(IC) + ")"); + } + + // Mark the loop as already vectorized to avoid vectorizing again. + Hints.setAlreadyVectorized(); + + DEBUG(verifyFunction(*L->getHeader()->getParent())); + return true; + } + + void getAnalysisUsage(AnalysisUsage &AU) const override { + AU.addRequired<AssumptionCacheTracker>(); + AU.addRequiredID(LoopSimplifyID); + AU.addRequiredID(LCSSAID); + AU.addRequired<BlockFrequencyInfoWrapperPass>(); + AU.addRequired<DominatorTreeWrapperPass>(); + AU.addRequired<LoopInfoWrapperPass>(); + AU.addRequired<ScalarEvolutionWrapperPass>(); + AU.addRequired<TargetTransformInfoWrapperPass>(); + AU.addRequired<AAResultsWrapperPass>(); + AU.addRequired<LoopAccessAnalysis>(); + AU.addRequired<DemandedBits>(); + AU.addPreserved<LoopInfoWrapperPass>(); + AU.addPreserved<DominatorTreeWrapperPass>(); + AU.addPreserved<BasicAAWrapperPass>(); + AU.addPreserved<AAResultsWrapperPass>(); + AU.addPreserved<GlobalsAAWrapperPass>(); + } + +}; + +} // end anonymous namespace + +//===----------------------------------------------------------------------===// +// Implementation of LoopVectorizationLegality, InnerLoopVectorizer and +// LoopVectorizationCostModel. +//===----------------------------------------------------------------------===// + +Value *InnerLoopVectorizer::getBroadcastInstrs(Value *V) { + // We need to place the broadcast of invariant variables outside the loop. + Instruction *Instr = dyn_cast<Instruction>(V); + bool NewInstr = + (Instr && std::find(LoopVectorBody.begin(), LoopVectorBody.end(), + Instr->getParent()) != LoopVectorBody.end()); + bool Invariant = OrigLoop->isLoopInvariant(V) && !NewInstr; + + // Place the code for broadcasting invariant variables in the new preheader. + IRBuilder<>::InsertPointGuard Guard(Builder); + if (Invariant) + Builder.SetInsertPoint(LoopVectorPreHeader->getTerminator()); + + // Broadcast the scalar into all locations in the vector. + Value *Shuf = Builder.CreateVectorSplat(VF, V, "broadcast"); + + return Shuf; +} + +Value *InnerLoopVectorizer::getStepVector(Value *Val, int StartIdx, + Value *Step) { + assert(Val->getType()->isVectorTy() && "Must be a vector"); + assert(Val->getType()->getScalarType()->isIntegerTy() && + "Elem must be an integer"); + assert(Step->getType() == Val->getType()->getScalarType() && + "Step has wrong type"); + // Create the types. + Type *ITy = Val->getType()->getScalarType(); + VectorType *Ty = cast<VectorType>(Val->getType()); + int VLen = Ty->getNumElements(); + SmallVector<Constant*, 8> Indices; + + // Create a vector of consecutive numbers from zero to VF. + for (int i = 0; i < VLen; ++i) + Indices.push_back(ConstantInt::get(ITy, StartIdx + i)); + + // Add the consecutive indices to the vector value. + Constant *Cv = ConstantVector::get(Indices); + assert(Cv->getType() == Val->getType() && "Invalid consecutive vec"); + Step = Builder.CreateVectorSplat(VLen, Step); + assert(Step->getType() == Val->getType() && "Invalid step vec"); + // FIXME: The newly created binary instructions should contain nsw/nuw flags, + // which can be found from the original scalar operations. + Step = Builder.CreateMul(Cv, Step); + return Builder.CreateAdd(Val, Step, "induction"); +} + +int LoopVectorizationLegality::isConsecutivePtr(Value *Ptr) { + assert(Ptr->getType()->isPointerTy() && "Unexpected non-ptr"); + auto *SE = PSE.getSE(); + // Make sure that the pointer does not point to structs. + if (Ptr->getType()->getPointerElementType()->isAggregateType()) + return 0; + + // If this value is a pointer induction variable we know it is consecutive. + PHINode *Phi = dyn_cast_or_null<PHINode>(Ptr); + if (Phi && Inductions.count(Phi)) { + InductionDescriptor II = Inductions[Phi]; + return II.getConsecutiveDirection(); + } + + GetElementPtrInst *Gep = getGEPInstruction(Ptr); + if (!Gep) + return 0; + + unsigned NumOperands = Gep->getNumOperands(); + Value *GpPtr = Gep->getPointerOperand(); + // If this GEP value is a consecutive pointer induction variable and all of + // the indices are constant then we know it is consecutive. We can + Phi = dyn_cast<PHINode>(GpPtr); + if (Phi && Inductions.count(Phi)) { + + // Make sure that the pointer does not point to structs. + PointerType *GepPtrType = cast<PointerType>(GpPtr->getType()); + if (GepPtrType->getElementType()->isAggregateType()) + return 0; + + // Make sure that all of the index operands are loop invariant. + for (unsigned i = 1; i < NumOperands; ++i) + if (!SE->isLoopInvariant(PSE.getSCEV(Gep->getOperand(i)), TheLoop)) + return 0; + + InductionDescriptor II = Inductions[Phi]; + return II.getConsecutiveDirection(); + } + + unsigned InductionOperand = getGEPInductionOperand(Gep); + + // Check that all of the gep indices are uniform except for our induction + // operand. + for (unsigned i = 0; i != NumOperands; ++i) + if (i != InductionOperand && + !SE->isLoopInvariant(PSE.getSCEV(Gep->getOperand(i)), TheLoop)) + return 0; + + // We can emit wide load/stores only if the last non-zero index is the + // induction variable. + const SCEV *Last = nullptr; + if (!Strides.count(Gep)) + Last = PSE.getSCEV(Gep->getOperand(InductionOperand)); + else { + // Because of the multiplication by a stride we can have a s/zext cast. + // We are going to replace this stride by 1 so the cast is safe to ignore. + // + // %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] + // %0 = trunc i64 %indvars.iv to i32 + // %mul = mul i32 %0, %Stride1 + // %idxprom = zext i32 %mul to i64 << Safe cast. + // %arrayidx = getelementptr inbounds i32* %B, i64 %idxprom + // + Last = replaceSymbolicStrideSCEV(PSE, Strides, + Gep->getOperand(InductionOperand), Gep); + if (const SCEVCastExpr *C = dyn_cast<SCEVCastExpr>(Last)) + Last = + (C->getSCEVType() == scSignExtend || C->getSCEVType() == scZeroExtend) + ? C->getOperand() + : Last; + } + if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Last)) { + const SCEV *Step = AR->getStepRecurrence(*SE); + + // The memory is consecutive because the last index is consecutive + // and all other indices are loop invariant. + if (Step->isOne()) + return 1; + if (Step->isAllOnesValue()) + return -1; + } + + return 0; +} + +bool LoopVectorizationLegality::isUniform(Value *V) { + return LAI->isUniform(V); +} + +InnerLoopVectorizer::VectorParts& +InnerLoopVectorizer::getVectorValue(Value *V) { + assert(V != Induction && "The new induction variable should not be used."); + assert(!V->getType()->isVectorTy() && "Can't widen a vector"); + + // If we have a stride that is replaced by one, do it here. + if (Legal->hasStride(V)) + V = ConstantInt::get(V->getType(), 1); + + // If we have this scalar in the map, return it. + if (WidenMap.has(V)) + return WidenMap.get(V); + + // If this scalar is unknown, assume that it is a constant or that it is + // loop invariant. Broadcast V and save the value for future uses. + Value *B = getBroadcastInstrs(V); + return WidenMap.splat(V, B); +} + +Value *InnerLoopVectorizer::reverseVector(Value *Vec) { + assert(Vec->getType()->isVectorTy() && "Invalid type"); + SmallVector<Constant*, 8> ShuffleMask; + for (unsigned i = 0; i < VF; ++i) + ShuffleMask.push_back(Builder.getInt32(VF - i - 1)); + + return Builder.CreateShuffleVector(Vec, UndefValue::get(Vec->getType()), + ConstantVector::get(ShuffleMask), + "reverse"); +} + +// Get a mask to interleave \p NumVec vectors into a wide vector. +// I.e. <0, VF, VF*2, ..., VF*(NumVec-1), 1, VF+1, VF*2+1, ...> +// E.g. For 2 interleaved vectors, if VF is 4, the mask is: +// <0, 4, 1, 5, 2, 6, 3, 7> +static Constant *getInterleavedMask(IRBuilder<> &Builder, unsigned VF, + unsigned NumVec) { + SmallVector<Constant *, 16> Mask; + for (unsigned i = 0; i < VF; i++) + for (unsigned j = 0; j < NumVec; j++) + Mask.push_back(Builder.getInt32(j * VF + i)); + + return ConstantVector::get(Mask); +} + +// Get the strided mask starting from index \p Start. +// I.e. <Start, Start + Stride, ..., Start + Stride*(VF-1)> +static Constant *getStridedMask(IRBuilder<> &Builder, unsigned Start, + unsigned Stride, unsigned VF) { + SmallVector<Constant *, 16> Mask; + for (unsigned i = 0; i < VF; i++) + Mask.push_back(Builder.getInt32(Start + i * Stride)); + + return ConstantVector::get(Mask); +} + +// Get a mask of two parts: The first part consists of sequential integers +// starting from 0, The second part consists of UNDEFs. +// I.e. <0, 1, 2, ..., NumInt - 1, undef, ..., undef> +static Constant *getSequentialMask(IRBuilder<> &Builder, unsigned NumInt, + unsigned NumUndef) { + SmallVector<Constant *, 16> Mask; + for (unsigned i = 0; i < NumInt; i++) + Mask.push_back(Builder.getInt32(i)); + + Constant *Undef = UndefValue::get(Builder.getInt32Ty()); + for (unsigned i = 0; i < NumUndef; i++) + Mask.push_back(Undef); + + return ConstantVector::get(Mask); +} + +// Concatenate two vectors with the same element type. The 2nd vector should +// not have more elements than the 1st vector. If the 2nd vector has less +// elements, extend it with UNDEFs. +static Value *ConcatenateTwoVectors(IRBuilder<> &Builder, Value *V1, + Value *V2) { + VectorType *VecTy1 = dyn_cast<VectorType>(V1->getType()); + VectorType *VecTy2 = dyn_cast<VectorType>(V2->getType()); + assert(VecTy1 && VecTy2 && + VecTy1->getScalarType() == VecTy2->getScalarType() && + "Expect two vectors with the same element type"); + + unsigned NumElts1 = VecTy1->getNumElements(); + unsigned NumElts2 = VecTy2->getNumElements(); + assert(NumElts1 >= NumElts2 && "Unexpect the first vector has less elements"); + + if (NumElts1 > NumElts2) { + // Extend with UNDEFs. + Constant *ExtMask = + getSequentialMask(Builder, NumElts2, NumElts1 - NumElts2); + V2 = Builder.CreateShuffleVector(V2, UndefValue::get(VecTy2), ExtMask); + } + + Constant *Mask = getSequentialMask(Builder, NumElts1 + NumElts2, 0); + return Builder.CreateShuffleVector(V1, V2, Mask); +} + +// Concatenate vectors in the given list. All vectors have the same type. +static Value *ConcatenateVectors(IRBuilder<> &Builder, + ArrayRef<Value *> InputList) { + unsigned NumVec = InputList.size(); + assert(NumVec > 1 && "Should be at least two vectors"); + + SmallVector<Value *, 8> ResList; + ResList.append(InputList.begin(), InputList.end()); + do { + SmallVector<Value *, 8> TmpList; + for (unsigned i = 0; i < NumVec - 1; i += 2) { + Value *V0 = ResList[i], *V1 = ResList[i + 1]; + assert((V0->getType() == V1->getType() || i == NumVec - 2) && + "Only the last vector may have a different type"); + + TmpList.push_back(ConcatenateTwoVectors(Builder, V0, V1)); + } + + // Push the last vector if the total number of vectors is odd. + if (NumVec % 2 != 0) + TmpList.push_back(ResList[NumVec - 1]); + + ResList = TmpList; + NumVec = ResList.size(); + } while (NumVec > 1); + + return ResList[0]; +} + +// Try to vectorize the interleave group that \p Instr belongs to. +// +// E.g. Translate following interleaved load group (factor = 3): +// for (i = 0; i < N; i+=3) { +// R = Pic[i]; // Member of index 0 +// G = Pic[i+1]; // Member of index 1 +// B = Pic[i+2]; // Member of index 2 +// ... // do something to R, G, B +// } +// To: +// %wide.vec = load <12 x i32> ; Read 4 tuples of R,G,B +// %R.vec = shuffle %wide.vec, undef, <0, 3, 6, 9> ; R elements +// %G.vec = shuffle %wide.vec, undef, <1, 4, 7, 10> ; G elements +// %B.vec = shuffle %wide.vec, undef, <2, 5, 8, 11> ; B elements +// +// Or translate following interleaved store group (factor = 3): +// for (i = 0; i < N; i+=3) { +// ... do something to R, G, B +// Pic[i] = R; // Member of index 0 +// Pic[i+1] = G; // Member of index 1 +// Pic[i+2] = B; // Member of index 2 +// } +// To: +// %R_G.vec = shuffle %R.vec, %G.vec, <0, 1, 2, ..., 7> +// %B_U.vec = shuffle %B.vec, undef, <0, 1, 2, 3, u, u, u, u> +// %interleaved.vec = shuffle %R_G.vec, %B_U.vec, +// <0, 4, 8, 1, 5, 9, 2, 6, 10, 3, 7, 11> ; Interleave R,G,B elements +// store <12 x i32> %interleaved.vec ; Write 4 tuples of R,G,B +void InnerLoopVectorizer::vectorizeInterleaveGroup(Instruction *Instr) { + const InterleaveGroup *Group = Legal->getInterleavedAccessGroup(Instr); + assert(Group && "Fail to get an interleaved access group."); + + // Skip if current instruction is not the insert position. + if (Instr != Group->getInsertPos()) + return; + + LoadInst *LI = dyn_cast<LoadInst>(Instr); + StoreInst *SI = dyn_cast<StoreInst>(Instr); + Value *Ptr = LI ? LI->getPointerOperand() : SI->getPointerOperand(); + + // Prepare for the vector type of the interleaved load/store. + Type *ScalarTy = LI ? LI->getType() : SI->getValueOperand()->getType(); + unsigned InterleaveFactor = Group->getFactor(); + Type *VecTy = VectorType::get(ScalarTy, InterleaveFactor * VF); + Type *PtrTy = VecTy->getPointerTo(Ptr->getType()->getPointerAddressSpace()); + + // Prepare for the new pointers. + setDebugLocFromInst(Builder, Ptr); + VectorParts &PtrParts = getVectorValue(Ptr); + SmallVector<Value *, 2> NewPtrs; + unsigned Index = Group->getIndex(Instr); + for (unsigned Part = 0; Part < UF; Part++) { + // Extract the pointer for current instruction from the pointer vector. A + // reverse access uses the pointer in the last lane. + Value *NewPtr = Builder.CreateExtractElement( + PtrParts[Part], + Group->isReverse() ? Builder.getInt32(VF - 1) : Builder.getInt32(0)); + + // Notice current instruction could be any index. Need to adjust the address + // to the member of index 0. + // + // E.g. a = A[i+1]; // Member of index 1 (Current instruction) + // b = A[i]; // Member of index 0 + // Current pointer is pointed to A[i+1], adjust it to A[i]. + // + // E.g. A[i+1] = a; // Member of index 1 + // A[i] = b; // Member of index 0 + // A[i+2] = c; // Member of index 2 (Current instruction) + // Current pointer is pointed to A[i+2], adjust it to A[i]. + NewPtr = Builder.CreateGEP(NewPtr, Builder.getInt32(-Index)); + + // Cast to the vector pointer type. + NewPtrs.push_back(Builder.CreateBitCast(NewPtr, PtrTy)); + } + + setDebugLocFromInst(Builder, Instr); + Value *UndefVec = UndefValue::get(VecTy); + + // Vectorize the interleaved load group. + if (LI) { + for (unsigned Part = 0; Part < UF; Part++) { + Instruction *NewLoadInstr = Builder.CreateAlignedLoad( + NewPtrs[Part], Group->getAlignment(), "wide.vec"); + + for (unsigned i = 0; i < InterleaveFactor; i++) { + Instruction *Member = Group->getMember(i); + + // Skip the gaps in the group. + if (!Member) + continue; + + Constant *StrideMask = getStridedMask(Builder, i, InterleaveFactor, VF); + Value *StridedVec = Builder.CreateShuffleVector( + NewLoadInstr, UndefVec, StrideMask, "strided.vec"); + + // If this member has different type, cast the result type. + if (Member->getType() != ScalarTy) { + VectorType *OtherVTy = VectorType::get(Member->getType(), VF); + StridedVec = Builder.CreateBitOrPointerCast(StridedVec, OtherVTy); + } + + VectorParts &Entry = WidenMap.get(Member); + Entry[Part] = + Group->isReverse() ? reverseVector(StridedVec) : StridedVec; + } + + propagateMetadata(NewLoadInstr, Instr); + } + return; + } + + // The sub vector type for current instruction. + VectorType *SubVT = VectorType::get(ScalarTy, VF); + + // Vectorize the interleaved store group. + for (unsigned Part = 0; Part < UF; Part++) { + // Collect the stored vector from each member. + SmallVector<Value *, 4> StoredVecs; + for (unsigned i = 0; i < InterleaveFactor; i++) { + // Interleaved store group doesn't allow a gap, so each index has a member + Instruction *Member = Group->getMember(i); + assert(Member && "Fail to get a member from an interleaved store group"); + + Value *StoredVec = + getVectorValue(dyn_cast<StoreInst>(Member)->getValueOperand())[Part]; + if (Group->isReverse()) + StoredVec = reverseVector(StoredVec); + + // If this member has different type, cast it to an unified type. + if (StoredVec->getType() != SubVT) + StoredVec = Builder.CreateBitOrPointerCast(StoredVec, SubVT); + + StoredVecs.push_back(StoredVec); + } + + // Concatenate all vectors into a wide vector. + Value *WideVec = ConcatenateVectors(Builder, StoredVecs); + + // Interleave the elements in the wide vector. + Constant *IMask = getInterleavedMask(Builder, VF, InterleaveFactor); + Value *IVec = Builder.CreateShuffleVector(WideVec, UndefVec, IMask, + "interleaved.vec"); + + Instruction *NewStoreInstr = + Builder.CreateAlignedStore(IVec, NewPtrs[Part], Group->getAlignment()); + propagateMetadata(NewStoreInstr, Instr); + } +} + +void InnerLoopVectorizer::vectorizeMemoryInstruction(Instruction *Instr) { + // Attempt to issue a wide load. + LoadInst *LI = dyn_cast<LoadInst>(Instr); + StoreInst *SI = dyn_cast<StoreInst>(Instr); + + assert((LI || SI) && "Invalid Load/Store instruction"); + + // Try to vectorize the interleave group if this access is interleaved. + if (Legal->isAccessInterleaved(Instr)) + return vectorizeInterleaveGroup(Instr); + + Type *ScalarDataTy = LI ? LI->getType() : SI->getValueOperand()->getType(); + Type *DataTy = VectorType::get(ScalarDataTy, VF); + Value *Ptr = LI ? LI->getPointerOperand() : SI->getPointerOperand(); + unsigned Alignment = LI ? LI->getAlignment() : SI->getAlignment(); + // An alignment of 0 means target abi alignment. We need to use the scalar's + // target abi alignment in such a case. + const DataLayout &DL = Instr->getModule()->getDataLayout(); + if (!Alignment) + Alignment = DL.getABITypeAlignment(ScalarDataTy); + unsigned AddressSpace = Ptr->getType()->getPointerAddressSpace(); + unsigned ScalarAllocatedSize = DL.getTypeAllocSize(ScalarDataTy); + unsigned VectorElementSize = DL.getTypeStoreSize(DataTy) / VF; + + if (SI && Legal->blockNeedsPredication(SI->getParent()) && + !Legal->isMaskRequired(SI)) + return scalarizeInstruction(Instr, true); + + if (ScalarAllocatedSize != VectorElementSize) + return scalarizeInstruction(Instr); + + // If the pointer is loop invariant or if it is non-consecutive, + // scalarize the load. + int ConsecutiveStride = Legal->isConsecutivePtr(Ptr); + bool Reverse = ConsecutiveStride < 0; + bool UniformLoad = LI && Legal->isUniform(Ptr); + if (!ConsecutiveStride || UniformLoad) + return scalarizeInstruction(Instr); + + Constant *Zero = Builder.getInt32(0); + VectorParts &Entry = WidenMap.get(Instr); + + // Handle consecutive loads/stores. + GetElementPtrInst *Gep = getGEPInstruction(Ptr); + if (Gep && Legal->isInductionVariable(Gep->getPointerOperand())) { + setDebugLocFromInst(Builder, Gep); + Value *PtrOperand = Gep->getPointerOperand(); + Value *FirstBasePtr = getVectorValue(PtrOperand)[0]; + FirstBasePtr = Builder.CreateExtractElement(FirstBasePtr, Zero); + + // Create the new GEP with the new induction variable. + GetElementPtrInst *Gep2 = cast<GetElementPtrInst>(Gep->clone()); + Gep2->setOperand(0, FirstBasePtr); + Gep2->setName("gep.indvar.base"); + Ptr = Builder.Insert(Gep2); + } else if (Gep) { + setDebugLocFromInst(Builder, Gep); + assert(PSE.getSE()->isLoopInvariant(PSE.getSCEV(Gep->getPointerOperand()), + OrigLoop) && + "Base ptr must be invariant"); + + // The last index does not have to be the induction. It can be + // consecutive and be a function of the index. For example A[I+1]; + unsigned NumOperands = Gep->getNumOperands(); + unsigned InductionOperand = getGEPInductionOperand(Gep); + // Create the new GEP with the new induction variable. + GetElementPtrInst *Gep2 = cast<GetElementPtrInst>(Gep->clone()); + + for (unsigned i = 0; i < NumOperands; ++i) { + Value *GepOperand = Gep->getOperand(i); + Instruction *GepOperandInst = dyn_cast<Instruction>(GepOperand); + + // Update last index or loop invariant instruction anchored in loop. + if (i == InductionOperand || + (GepOperandInst && OrigLoop->contains(GepOperandInst))) { + assert((i == InductionOperand || + PSE.getSE()->isLoopInvariant(PSE.getSCEV(GepOperandInst), + OrigLoop)) && + "Must be last index or loop invariant"); + + VectorParts &GEPParts = getVectorValue(GepOperand); + Value *Index = GEPParts[0]; + Index = Builder.CreateExtractElement(Index, Zero); + Gep2->setOperand(i, Index); + Gep2->setName("gep.indvar.idx"); + } + } + Ptr = Builder.Insert(Gep2); + } else { + // Use the induction element ptr. + assert(isa<PHINode>(Ptr) && "Invalid induction ptr"); + setDebugLocFromInst(Builder, Ptr); + VectorParts &PtrVal = getVectorValue(Ptr); + Ptr = Builder.CreateExtractElement(PtrVal[0], Zero); + } + + VectorParts Mask = createBlockInMask(Instr->getParent()); + // Handle Stores: + if (SI) { + assert(!Legal->isUniform(SI->getPointerOperand()) && + "We do not allow storing to uniform addresses"); + setDebugLocFromInst(Builder, SI); + // We don't want to update the value in the map as it might be used in + // another expression. So don't use a reference type for "StoredVal". + VectorParts StoredVal = getVectorValue(SI->getValueOperand()); + + for (unsigned Part = 0; Part < UF; ++Part) { + // Calculate the pointer for the specific unroll-part. + Value *PartPtr = + Builder.CreateGEP(nullptr, Ptr, Builder.getInt32(Part * VF)); + + if (Reverse) { + // If we store to reverse consecutive memory locations, then we need + // to reverse the order of elements in the stored value. + StoredVal[Part] = reverseVector(StoredVal[Part]); + // If the address is consecutive but reversed, then the + // wide store needs to start at the last vector element. + PartPtr = Builder.CreateGEP(nullptr, Ptr, Builder.getInt32(-Part * VF)); + PartPtr = Builder.CreateGEP(nullptr, PartPtr, Builder.getInt32(1 - VF)); + Mask[Part] = reverseVector(Mask[Part]); + } + + Value *VecPtr = Builder.CreateBitCast(PartPtr, + DataTy->getPointerTo(AddressSpace)); + + Instruction *NewSI; + if (Legal->isMaskRequired(SI)) + NewSI = Builder.CreateMaskedStore(StoredVal[Part], VecPtr, Alignment, + Mask[Part]); + else + NewSI = Builder.CreateAlignedStore(StoredVal[Part], VecPtr, Alignment); + propagateMetadata(NewSI, SI); + } + return; + } + + // Handle loads. + assert(LI && "Must have a load instruction"); + setDebugLocFromInst(Builder, LI); + for (unsigned Part = 0; Part < UF; ++Part) { + // Calculate the pointer for the specific unroll-part. + Value *PartPtr = + Builder.CreateGEP(nullptr, Ptr, Builder.getInt32(Part * VF)); + + if (Reverse) { + // If the address is consecutive but reversed, then the + // wide load needs to start at the last vector element. + PartPtr = Builder.CreateGEP(nullptr, Ptr, Builder.getInt32(-Part * VF)); + PartPtr = Builder.CreateGEP(nullptr, PartPtr, Builder.getInt32(1 - VF)); + Mask[Part] = reverseVector(Mask[Part]); + } + + Instruction* NewLI; + Value *VecPtr = Builder.CreateBitCast(PartPtr, + DataTy->getPointerTo(AddressSpace)); + if (Legal->isMaskRequired(LI)) + NewLI = Builder.CreateMaskedLoad(VecPtr, Alignment, Mask[Part], + UndefValue::get(DataTy), + "wide.masked.load"); + else + NewLI = Builder.CreateAlignedLoad(VecPtr, Alignment, "wide.load"); + propagateMetadata(NewLI, LI); + Entry[Part] = Reverse ? reverseVector(NewLI) : NewLI; + } +} + +void InnerLoopVectorizer::scalarizeInstruction(Instruction *Instr, + bool IfPredicateStore) { + assert(!Instr->getType()->isAggregateType() && "Can't handle vectors"); + // Holds vector parameters or scalars, in case of uniform vals. + SmallVector<VectorParts, 4> Params; + + setDebugLocFromInst(Builder, Instr); + + // Find all of the vectorized parameters. + for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) { + Value *SrcOp = Instr->getOperand(op); + + // If we are accessing the old induction variable, use the new one. + if (SrcOp == OldInduction) { + Params.push_back(getVectorValue(SrcOp)); + continue; + } + + // Try using previously calculated values. + Instruction *SrcInst = dyn_cast<Instruction>(SrcOp); + + // If the src is an instruction that appeared earlier in the basic block, + // then it should already be vectorized. + if (SrcInst && OrigLoop->contains(SrcInst)) { + assert(WidenMap.has(SrcInst) && "Source operand is unavailable"); + // The parameter is a vector value from earlier. + Params.push_back(WidenMap.get(SrcInst)); + } else { + // The parameter is a scalar from outside the loop. Maybe even a constant. + VectorParts Scalars; + Scalars.append(UF, SrcOp); + Params.push_back(Scalars); + } + } + + assert(Params.size() == Instr->getNumOperands() && + "Invalid number of operands"); + + // Does this instruction return a value ? + bool IsVoidRetTy = Instr->getType()->isVoidTy(); + + Value *UndefVec = IsVoidRetTy ? nullptr : + UndefValue::get(VectorType::get(Instr->getType(), VF)); + // Create a new entry in the WidenMap and initialize it to Undef or Null. + VectorParts &VecResults = WidenMap.splat(Instr, UndefVec); + + VectorParts Cond; + if (IfPredicateStore) { + assert(Instr->getParent()->getSinglePredecessor() && + "Only support single predecessor blocks"); + Cond = createEdgeMask(Instr->getParent()->getSinglePredecessor(), + Instr->getParent()); + } + + // For each vector unroll 'part': + for (unsigned Part = 0; Part < UF; ++Part) { + // For each scalar that we create: + for (unsigned Width = 0; Width < VF; ++Width) { + + // Start if-block. + Value *Cmp = nullptr; + if (IfPredicateStore) { + Cmp = Builder.CreateExtractElement(Cond[Part], Builder.getInt32(Width)); + Cmp = Builder.CreateICmp(ICmpInst::ICMP_EQ, Cmp, + ConstantInt::get(Cmp->getType(), 1)); + } + + Instruction *Cloned = Instr->clone(); + if (!IsVoidRetTy) + Cloned->setName(Instr->getName() + ".cloned"); + // Replace the operands of the cloned instructions with extracted scalars. + for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) { + Value *Op = Params[op][Part]; + // Param is a vector. Need to extract the right lane. + if (Op->getType()->isVectorTy()) + Op = Builder.CreateExtractElement(Op, Builder.getInt32(Width)); + Cloned->setOperand(op, Op); + } + + // Place the cloned scalar in the new loop. + Builder.Insert(Cloned); + + // If the original scalar returns a value we need to place it in a vector + // so that future users will be able to use it. + if (!IsVoidRetTy) + VecResults[Part] = Builder.CreateInsertElement(VecResults[Part], Cloned, + Builder.getInt32(Width)); + // End if-block. + if (IfPredicateStore) + PredicatedStores.push_back(std::make_pair(cast<StoreInst>(Cloned), + Cmp)); + } + } +} + +PHINode *InnerLoopVectorizer::createInductionVariable(Loop *L, Value *Start, + Value *End, Value *Step, + Instruction *DL) { + BasicBlock *Header = L->getHeader(); + BasicBlock *Latch = L->getLoopLatch(); + // As we're just creating this loop, it's possible no latch exists + // yet. If so, use the header as this will be a single block loop. + if (!Latch) + Latch = Header; + + IRBuilder<> Builder(&*Header->getFirstInsertionPt()); + setDebugLocFromInst(Builder, getDebugLocFromInstOrOperands(OldInduction)); + auto *Induction = Builder.CreatePHI(Start->getType(), 2, "index"); + + Builder.SetInsertPoint(Latch->getTerminator()); + + // Create i+1 and fill the PHINode. + Value *Next = Builder.CreateAdd(Induction, Step, "index.next"); + Induction->addIncoming(Start, L->getLoopPreheader()); + Induction->addIncoming(Next, Latch); + // Create the compare. + Value *ICmp = Builder.CreateICmpEQ(Next, End); + Builder.CreateCondBr(ICmp, L->getExitBlock(), Header); + + // Now we have two terminators. Remove the old one from the block. + Latch->getTerminator()->eraseFromParent(); + + return Induction; +} + +Value *InnerLoopVectorizer::getOrCreateTripCount(Loop *L) { + if (TripCount) + return TripCount; + + IRBuilder<> Builder(L->getLoopPreheader()->getTerminator()); + // Find the loop boundaries. + ScalarEvolution *SE = PSE.getSE(); + const SCEV *BackedgeTakenCount = SE->getBackedgeTakenCount(OrigLoop); + assert(BackedgeTakenCount != SE->getCouldNotCompute() && + "Invalid loop count"); + + Type *IdxTy = Legal->getWidestInductionType(); + + // The exit count might have the type of i64 while the phi is i32. This can + // happen if we have an induction variable that is sign extended before the + // compare. The only way that we get a backedge taken count is that the + // induction variable was signed and as such will not overflow. In such a case + // truncation is legal. + if (BackedgeTakenCount->getType()->getPrimitiveSizeInBits() > + IdxTy->getPrimitiveSizeInBits()) + BackedgeTakenCount = SE->getTruncateOrNoop(BackedgeTakenCount, IdxTy); + BackedgeTakenCount = SE->getNoopOrZeroExtend(BackedgeTakenCount, IdxTy); + + // Get the total trip count from the count by adding 1. + const SCEV *ExitCount = SE->getAddExpr( + BackedgeTakenCount, SE->getOne(BackedgeTakenCount->getType())); + + const DataLayout &DL = L->getHeader()->getModule()->getDataLayout(); + + // Expand the trip count and place the new instructions in the preheader. + // Notice that the pre-header does not change, only the loop body. + SCEVExpander Exp(*SE, DL, "induction"); + + // Count holds the overall loop count (N). + TripCount = Exp.expandCodeFor(ExitCount, ExitCount->getType(), + L->getLoopPreheader()->getTerminator()); + + if (TripCount->getType()->isPointerTy()) + TripCount = + CastInst::CreatePointerCast(TripCount, IdxTy, + "exitcount.ptrcnt.to.int", + L->getLoopPreheader()->getTerminator()); + + return TripCount; +} + +Value *InnerLoopVectorizer::getOrCreateVectorTripCount(Loop *L) { + if (VectorTripCount) + return VectorTripCount; + + Value *TC = getOrCreateTripCount(L); + IRBuilder<> Builder(L->getLoopPreheader()->getTerminator()); + + // Now we need to generate the expression for N - (N % VF), which is + // the part that the vectorized body will execute. + // The loop step is equal to the vectorization factor (num of SIMD elements) + // times the unroll factor (num of SIMD instructions). + Constant *Step = ConstantInt::get(TC->getType(), VF * UF); + Value *R = Builder.CreateURem(TC, Step, "n.mod.vf"); + VectorTripCount = Builder.CreateSub(TC, R, "n.vec"); + + return VectorTripCount; +} + +void InnerLoopVectorizer::emitMinimumIterationCountCheck(Loop *L, + BasicBlock *Bypass) { + Value *Count = getOrCreateTripCount(L); + BasicBlock *BB = L->getLoopPreheader(); + IRBuilder<> Builder(BB->getTerminator()); + + // Generate code to check that the loop's trip count that we computed by + // adding one to the backedge-taken count will not overflow. + Value *CheckMinIters = + Builder.CreateICmpULT(Count, + ConstantInt::get(Count->getType(), VF * UF), + "min.iters.check"); + + BasicBlock *NewBB = BB->splitBasicBlock(BB->getTerminator(), + "min.iters.checked"); + if (L->getParentLoop()) + L->getParentLoop()->addBasicBlockToLoop(NewBB, *LI); + ReplaceInstWithInst(BB->getTerminator(), + BranchInst::Create(Bypass, NewBB, CheckMinIters)); + LoopBypassBlocks.push_back(BB); +} + +void InnerLoopVectorizer::emitVectorLoopEnteredCheck(Loop *L, + BasicBlock *Bypass) { + Value *TC = getOrCreateVectorTripCount(L); + BasicBlock *BB = L->getLoopPreheader(); + IRBuilder<> Builder(BB->getTerminator()); + + // Now, compare the new count to zero. If it is zero skip the vector loop and + // jump to the scalar loop. + Value *Cmp = Builder.CreateICmpEQ(TC, Constant::getNullValue(TC->getType()), + "cmp.zero"); + + // Generate code to check that the loop's trip count that we computed by + // adding one to the backedge-taken count will not overflow. + BasicBlock *NewBB = BB->splitBasicBlock(BB->getTerminator(), + "vector.ph"); + if (L->getParentLoop()) + L->getParentLoop()->addBasicBlockToLoop(NewBB, *LI); + ReplaceInstWithInst(BB->getTerminator(), + BranchInst::Create(Bypass, NewBB, Cmp)); + LoopBypassBlocks.push_back(BB); +} + +void InnerLoopVectorizer::emitSCEVChecks(Loop *L, BasicBlock *Bypass) { + BasicBlock *BB = L->getLoopPreheader(); + + // Generate the code to check that the SCEV assumptions that we made. + // We want the new basic block to start at the first instruction in a + // sequence of instructions that form a check. + SCEVExpander Exp(*PSE.getSE(), Bypass->getModule()->getDataLayout(), + "scev.check"); + Value *SCEVCheck = + Exp.expandCodeForPredicate(&PSE.getUnionPredicate(), BB->getTerminator()); + + if (auto *C = dyn_cast<ConstantInt>(SCEVCheck)) + if (C->isZero()) + return; + + // Create a new block containing the stride check. + BB->setName("vector.scevcheck"); + auto *NewBB = BB->splitBasicBlock(BB->getTerminator(), "vector.ph"); + if (L->getParentLoop()) + L->getParentLoop()->addBasicBlockToLoop(NewBB, *LI); + ReplaceInstWithInst(BB->getTerminator(), + BranchInst::Create(Bypass, NewBB, SCEVCheck)); + LoopBypassBlocks.push_back(BB); + AddedSafetyChecks = true; +} + +void InnerLoopVectorizer::emitMemRuntimeChecks(Loop *L, + BasicBlock *Bypass) { + BasicBlock *BB = L->getLoopPreheader(); + + // Generate the code that checks in runtime if arrays overlap. We put the + // checks into a separate block to make the more common case of few elements + // faster. + Instruction *FirstCheckInst; + Instruction *MemRuntimeCheck; + std::tie(FirstCheckInst, MemRuntimeCheck) = + Legal->getLAI()->addRuntimeChecks(BB->getTerminator()); + if (!MemRuntimeCheck) + return; + + // Create a new block containing the memory check. + BB->setName("vector.memcheck"); + auto *NewBB = BB->splitBasicBlock(BB->getTerminator(), "vector.ph"); + if (L->getParentLoop()) + L->getParentLoop()->addBasicBlockToLoop(NewBB, *LI); + ReplaceInstWithInst(BB->getTerminator(), + BranchInst::Create(Bypass, NewBB, MemRuntimeCheck)); + LoopBypassBlocks.push_back(BB); + AddedSafetyChecks = true; +} + + +void InnerLoopVectorizer::createEmptyLoop() { + /* + In this function we generate a new loop. The new loop will contain + the vectorized instructions while the old loop will continue to run the + scalar remainder. + + [ ] <-- loop iteration number check. + / | + / v + | [ ] <-- vector loop bypass (may consist of multiple blocks). + | / | + | / v + || [ ] <-- vector pre header. + |/ | + | v + | [ ] \ + | [ ]_| <-- vector loop. + | | + | v + | -[ ] <--- middle-block. + | / | + | / v + -|- >[ ] <--- new preheader. + | | + | v + | [ ] \ + | [ ]_| <-- old scalar loop to handle remainder. + \ | + \ v + >[ ] <-- exit block. + ... + */ + + BasicBlock *OldBasicBlock = OrigLoop->getHeader(); + BasicBlock *VectorPH = OrigLoop->getLoopPreheader(); + BasicBlock *ExitBlock = OrigLoop->getExitBlock(); + assert(VectorPH && "Invalid loop structure"); + assert(ExitBlock && "Must have an exit block"); + + // Some loops have a single integer induction variable, while other loops + // don't. One example is c++ iterators that often have multiple pointer + // induction variables. In the code below we also support a case where we + // don't have a single induction variable. + // + // We try to obtain an induction variable from the original loop as hard + // as possible. However if we don't find one that: + // - is an integer + // - counts from zero, stepping by one + // - is the size of the widest induction variable type + // then we create a new one. + OldInduction = Legal->getInduction(); + Type *IdxTy = Legal->getWidestInductionType(); + + // Split the single block loop into the two loop structure described above. + BasicBlock *VecBody = + VectorPH->splitBasicBlock(VectorPH->getTerminator(), "vector.body"); + BasicBlock *MiddleBlock = + VecBody->splitBasicBlock(VecBody->getTerminator(), "middle.block"); + BasicBlock *ScalarPH = + MiddleBlock->splitBasicBlock(MiddleBlock->getTerminator(), "scalar.ph"); + + // Create and register the new vector loop. + Loop* Lp = new Loop(); + Loop *ParentLoop = OrigLoop->getParentLoop(); + + // 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 (ParentLoop) { + ParentLoop->addChildLoop(Lp); + ParentLoop->addBasicBlockToLoop(ScalarPH, *LI); + ParentLoop->addBasicBlockToLoop(MiddleBlock, *LI); + } else { + LI->addTopLevelLoop(Lp); + } + Lp->addBasicBlockToLoop(VecBody, *LI); + + // Find the loop boundaries. + Value *Count = getOrCreateTripCount(Lp); + + Value *StartIdx = ConstantInt::get(IdxTy, 0); + + // We need to test whether the backedge-taken count is uint##_max. Adding one + // to it will cause overflow and an incorrect loop trip count in the vector + // body. In case of overflow we want to directly jump to the scalar remainder + // loop. + emitMinimumIterationCountCheck(Lp, ScalarPH); + // Now, compare the new count to zero. If it is zero skip the vector loop and + // jump to the scalar loop. + emitVectorLoopEnteredCheck(Lp, ScalarPH); + // Generate the code to check any assumptions that we've made for SCEV + // expressions. + emitSCEVChecks(Lp, ScalarPH); + + // Generate the code that checks in runtime if arrays overlap. We put the + // checks into a separate block to make the more common case of few elements + // faster. + emitMemRuntimeChecks(Lp, ScalarPH); + + // Generate the induction variable. + // The loop step is equal to the vectorization factor (num of SIMD elements) + // times the unroll factor (num of SIMD instructions). + Value *CountRoundDown = getOrCreateVectorTripCount(Lp); + Constant *Step = ConstantInt::get(IdxTy, VF * UF); + Induction = + createInductionVariable(Lp, StartIdx, CountRoundDown, Step, + getDebugLocFromInstOrOperands(OldInduction)); + + // We are going to resume the execution of the scalar loop. + // Go over all of the induction variables that we found and fix the + // PHIs that are left in the scalar version of the loop. + // The starting values of PHI nodes depend on the counter of the last + // iteration in the vectorized loop. + // If we come from a bypass edge then we need to start from the original + // start value. + + // This variable saves the new starting index for the scalar loop. It is used + // to test if there are any tail iterations left once the vector loop has + // completed. + LoopVectorizationLegality::InductionList::iterator I, E; + LoopVectorizationLegality::InductionList *List = Legal->getInductionVars(); + for (I = List->begin(), E = List->end(); I != E; ++I) { + PHINode *OrigPhi = I->first; + InductionDescriptor II = I->second; + + // Create phi nodes to merge from the backedge-taken check block. + PHINode *BCResumeVal = PHINode::Create(OrigPhi->getType(), 3, + "bc.resume.val", + ScalarPH->getTerminator()); + Value *EndValue; + if (OrigPhi == OldInduction) { + // We know what the end value is. + EndValue = CountRoundDown; + } else { + IRBuilder<> B(LoopBypassBlocks.back()->getTerminator()); + Value *CRD = B.CreateSExtOrTrunc(CountRoundDown, + II.getStepValue()->getType(), + "cast.crd"); + EndValue = II.transform(B, CRD); + EndValue->setName("ind.end"); + } + + // The new PHI merges the original incoming value, in case of a bypass, + // or the value at the end of the vectorized loop. + BCResumeVal->addIncoming(EndValue, MiddleBlock); + + // Fix the scalar body counter (PHI node). + unsigned BlockIdx = OrigPhi->getBasicBlockIndex(ScalarPH); + + // The old induction's phi node in the scalar body needs the truncated + // value. + for (unsigned I = 0, E = LoopBypassBlocks.size(); I != E; ++I) + BCResumeVal->addIncoming(II.getStartValue(), LoopBypassBlocks[I]); + OrigPhi->setIncomingValue(BlockIdx, BCResumeVal); + } + + // Add a check in the middle block to see if we have completed + // all of the iterations in the first vector loop. + // If (N - N%VF) == N, then we *don't* need to run the remainder. + Value *CmpN = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_EQ, Count, + CountRoundDown, "cmp.n", + MiddleBlock->getTerminator()); + ReplaceInstWithInst(MiddleBlock->getTerminator(), + BranchInst::Create(ExitBlock, ScalarPH, CmpN)); + + // Get ready to start creating new instructions into the vectorized body. + Builder.SetInsertPoint(&*VecBody->getFirstInsertionPt()); + + // Save the state. + LoopVectorPreHeader = Lp->getLoopPreheader(); + LoopScalarPreHeader = ScalarPH; + LoopMiddleBlock = MiddleBlock; + LoopExitBlock = ExitBlock; + LoopVectorBody.push_back(VecBody); + LoopScalarBody = OldBasicBlock; + + LoopVectorizeHints Hints(Lp, true); + Hints.setAlreadyVectorized(); +} + +namespace { +struct CSEDenseMapInfo { + static bool canHandle(Instruction *I) { + return isa<InsertElementInst>(I) || isa<ExtractElementInst>(I) || + isa<ShuffleVectorInst>(I) || isa<GetElementPtrInst>(I); + } + static inline Instruction *getEmptyKey() { + return DenseMapInfo<Instruction *>::getEmptyKey(); + } + static inline Instruction *getTombstoneKey() { + return DenseMapInfo<Instruction *>::getTombstoneKey(); + } + static unsigned getHashValue(Instruction *I) { + assert(canHandle(I) && "Unknown instruction!"); + return hash_combine(I->getOpcode(), hash_combine_range(I->value_op_begin(), + I->value_op_end())); + } + static bool isEqual(Instruction *LHS, Instruction *RHS) { + if (LHS == getEmptyKey() || RHS == getEmptyKey() || + LHS == getTombstoneKey() || RHS == getTombstoneKey()) + return LHS == RHS; + return LHS->isIdenticalTo(RHS); + } +}; +} + +/// \brief Check whether this block is a predicated block. +/// Due to if predication of stores we might create a sequence of "if(pred) a[i] +/// = ...; " blocks. We start with one vectorized basic block. For every +/// conditional block we split this vectorized block. Therefore, every second +/// block will be a predicated one. +static bool isPredicatedBlock(unsigned BlockNum) { + return BlockNum % 2; +} + +///\brief Perform cse of induction variable instructions. +static void cse(SmallVector<BasicBlock *, 4> &BBs) { + // Perform simple cse. + SmallDenseMap<Instruction *, Instruction *, 4, CSEDenseMapInfo> CSEMap; + for (unsigned i = 0, e = BBs.size(); i != e; ++i) { + BasicBlock *BB = BBs[i]; + for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E;) { + Instruction *In = &*I++; + + if (!CSEDenseMapInfo::canHandle(In)) + continue; + + // Check if we can replace this instruction with any of the + // visited instructions. + if (Instruction *V = CSEMap.lookup(In)) { + In->replaceAllUsesWith(V); + In->eraseFromParent(); + continue; + } + // Ignore instructions in conditional blocks. We create "if (pred) a[i] = + // ...;" blocks for predicated stores. Every second block is a predicated + // block. + if (isPredicatedBlock(i)) + continue; + + CSEMap[In] = In; + } + } +} + +/// \brief Adds a 'fast' flag to floating point operations. +static Value *addFastMathFlag(Value *V) { + if (isa<FPMathOperator>(V)){ + FastMathFlags Flags; + Flags.setUnsafeAlgebra(); + cast<Instruction>(V)->setFastMathFlags(Flags); + } + return V; +} + +/// Estimate the overhead of scalarizing a value. Insert and Extract are set if +/// the result needs to be inserted and/or extracted from vectors. +static unsigned getScalarizationOverhead(Type *Ty, bool Insert, bool Extract, + const TargetTransformInfo &TTI) { + if (Ty->isVoidTy()) + return 0; + + assert(Ty->isVectorTy() && "Can only scalarize vectors"); + unsigned Cost = 0; + + for (int i = 0, e = Ty->getVectorNumElements(); i < e; ++i) { + if (Insert) + Cost += TTI.getVectorInstrCost(Instruction::InsertElement, Ty, i); + if (Extract) + Cost += TTI.getVectorInstrCost(Instruction::ExtractElement, Ty, i); + } + + return Cost; +} + +// Estimate cost of a call instruction CI if it were vectorized with factor VF. +// Return the cost of the instruction, including scalarization overhead if it's +// needed. The flag NeedToScalarize shows if the call needs to be scalarized - +// i.e. either vector version isn't available, or is too expensive. +static unsigned getVectorCallCost(CallInst *CI, unsigned VF, + const TargetTransformInfo &TTI, + const TargetLibraryInfo *TLI, + bool &NeedToScalarize) { + Function *F = CI->getCalledFunction(); + StringRef FnName = CI->getCalledFunction()->getName(); + Type *ScalarRetTy = CI->getType(); + SmallVector<Type *, 4> Tys, ScalarTys; + for (auto &ArgOp : CI->arg_operands()) + ScalarTys.push_back(ArgOp->getType()); + + // Estimate cost of scalarized vector call. The source operands are assumed + // to be vectors, so we need to extract individual elements from there, + // execute VF scalar calls, and then gather the result into the vector return + // value. + unsigned ScalarCallCost = TTI.getCallInstrCost(F, ScalarRetTy, ScalarTys); + if (VF == 1) + return ScalarCallCost; + + // Compute corresponding vector type for return value and arguments. + Type *RetTy = ToVectorTy(ScalarRetTy, VF); + for (unsigned i = 0, ie = ScalarTys.size(); i != ie; ++i) + Tys.push_back(ToVectorTy(ScalarTys[i], VF)); + + // Compute costs of unpacking argument values for the scalar calls and + // packing the return values to a vector. + unsigned ScalarizationCost = + getScalarizationOverhead(RetTy, true, false, TTI); + for (unsigned i = 0, ie = Tys.size(); i != ie; ++i) + ScalarizationCost += getScalarizationOverhead(Tys[i], false, true, TTI); + + unsigned Cost = ScalarCallCost * VF + ScalarizationCost; + + // If we can't emit a vector call for this function, then the currently found + // cost is the cost we need to return. + NeedToScalarize = true; + if (!TLI || !TLI->isFunctionVectorizable(FnName, VF) || CI->isNoBuiltin()) + return Cost; + + // If the corresponding vector cost is cheaper, return its cost. + unsigned VectorCallCost = TTI.getCallInstrCost(nullptr, RetTy, Tys); + if (VectorCallCost < Cost) { + NeedToScalarize = false; + return VectorCallCost; + } + return Cost; +} + +// Estimate cost of an intrinsic call instruction CI if it were vectorized with +// factor VF. Return the cost of the instruction, including scalarization +// overhead if it's needed. +static unsigned getVectorIntrinsicCost(CallInst *CI, unsigned VF, + const TargetTransformInfo &TTI, + const TargetLibraryInfo *TLI) { + Intrinsic::ID ID = getIntrinsicIDForCall(CI, TLI); + assert(ID && "Expected intrinsic call!"); + + Type *RetTy = ToVectorTy(CI->getType(), VF); + SmallVector<Type *, 4> Tys; + for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) + Tys.push_back(ToVectorTy(CI->getArgOperand(i)->getType(), VF)); + + return TTI.getIntrinsicInstrCost(ID, RetTy, Tys); +} + +static Type *smallestIntegerVectorType(Type *T1, Type *T2) { + IntegerType *I1 = cast<IntegerType>(T1->getVectorElementType()); + IntegerType *I2 = cast<IntegerType>(T2->getVectorElementType()); + return I1->getBitWidth() < I2->getBitWidth() ? T1 : T2; +} +static Type *largestIntegerVectorType(Type *T1, Type *T2) { + IntegerType *I1 = cast<IntegerType>(T1->getVectorElementType()); + IntegerType *I2 = cast<IntegerType>(T2->getVectorElementType()); + return I1->getBitWidth() > I2->getBitWidth() ? T1 : T2; +} + +void InnerLoopVectorizer::truncateToMinimalBitwidths() { + // For every instruction `I` in MinBWs, truncate the operands, create a + // truncated version of `I` and reextend its result. InstCombine runs + // later and will remove any ext/trunc pairs. + // + for (auto &KV : MinBWs) { + VectorParts &Parts = WidenMap.get(KV.first); + for (Value *&I : Parts) { + if (I->use_empty()) + continue; + Type *OriginalTy = I->getType(); + Type *ScalarTruncatedTy = IntegerType::get(OriginalTy->getContext(), + KV.second); + Type *TruncatedTy = VectorType::get(ScalarTruncatedTy, + OriginalTy->getVectorNumElements()); + if (TruncatedTy == OriginalTy) + continue; + + IRBuilder<> B(cast<Instruction>(I)); + auto ShrinkOperand = [&](Value *V) -> Value* { + if (auto *ZI = dyn_cast<ZExtInst>(V)) + if (ZI->getSrcTy() == TruncatedTy) + return ZI->getOperand(0); + return B.CreateZExtOrTrunc(V, TruncatedTy); + }; + + // The actual instruction modification depends on the instruction type, + // unfortunately. + Value *NewI = nullptr; + if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I)) { + NewI = B.CreateBinOp(BO->getOpcode(), + ShrinkOperand(BO->getOperand(0)), + ShrinkOperand(BO->getOperand(1))); + cast<BinaryOperator>(NewI)->copyIRFlags(I); + } else if (ICmpInst *CI = dyn_cast<ICmpInst>(I)) { + NewI = B.CreateICmp(CI->getPredicate(), + ShrinkOperand(CI->getOperand(0)), + ShrinkOperand(CI->getOperand(1))); + } else if (SelectInst *SI = dyn_cast<SelectInst>(I)) { + NewI = B.CreateSelect(SI->getCondition(), + ShrinkOperand(SI->getTrueValue()), + ShrinkOperand(SI->getFalseValue())); + } else if (CastInst *CI = dyn_cast<CastInst>(I)) { + switch (CI->getOpcode()) { + default: llvm_unreachable("Unhandled cast!"); + case Instruction::Trunc: + NewI = ShrinkOperand(CI->getOperand(0)); + break; + case Instruction::SExt: + NewI = B.CreateSExtOrTrunc(CI->getOperand(0), + smallestIntegerVectorType(OriginalTy, + TruncatedTy)); + break; + case Instruction::ZExt: + NewI = B.CreateZExtOrTrunc(CI->getOperand(0), + smallestIntegerVectorType(OriginalTy, + TruncatedTy)); + break; + } + } else if (ShuffleVectorInst *SI = dyn_cast<ShuffleVectorInst>(I)) { + auto Elements0 = SI->getOperand(0)->getType()->getVectorNumElements(); + auto *O0 = + B.CreateZExtOrTrunc(SI->getOperand(0), + VectorType::get(ScalarTruncatedTy, Elements0)); + auto Elements1 = SI->getOperand(1)->getType()->getVectorNumElements(); + auto *O1 = + B.CreateZExtOrTrunc(SI->getOperand(1), + VectorType::get(ScalarTruncatedTy, Elements1)); + + NewI = B.CreateShuffleVector(O0, O1, SI->getMask()); + } else if (isa<LoadInst>(I)) { + // Don't do anything with the operands, just extend the result. + continue; + } else { + llvm_unreachable("Unhandled instruction type!"); + } + + // Lastly, extend the result. + NewI->takeName(cast<Instruction>(I)); + Value *Res = B.CreateZExtOrTrunc(NewI, OriginalTy); + I->replaceAllUsesWith(Res); + cast<Instruction>(I)->eraseFromParent(); + I = Res; + } + } + + // We'll have created a bunch of ZExts that are now parentless. Clean up. + for (auto &KV : MinBWs) { + VectorParts &Parts = WidenMap.get(KV.first); + for (Value *&I : Parts) { + ZExtInst *Inst = dyn_cast<ZExtInst>(I); + if (Inst && Inst->use_empty()) { + Value *NewI = Inst->getOperand(0); + Inst->eraseFromParent(); + I = NewI; + } + } + } +} + +void InnerLoopVectorizer::vectorizeLoop() { + //===------------------------------------------------===// + // + // Notice: any optimization or new instruction that go + // into the code below should be also be implemented in + // the cost-model. + // + //===------------------------------------------------===// + Constant *Zero = Builder.getInt32(0); + + // In order to support reduction variables we need to be able to vectorize + // Phi nodes. Phi nodes have cycles, so we need to vectorize them in two + // stages. First, we create a new vector PHI node with no incoming edges. + // We use this value when we vectorize all of the instructions that use the + // PHI. Next, after all of the instructions in the block are complete we + // add the new incoming edges to the PHI. At this point all of the + // instructions in the basic block are vectorized, so we can use them to + // construct the PHI. + PhiVector RdxPHIsToFix; + + // Scan the loop in a topological order to ensure that defs are vectorized + // before users. + LoopBlocksDFS DFS(OrigLoop); + DFS.perform(LI); + + // Vectorize all of the blocks in the original loop. + for (LoopBlocksDFS::RPOIterator bb = DFS.beginRPO(), + be = DFS.endRPO(); bb != be; ++bb) + vectorizeBlockInLoop(*bb, &RdxPHIsToFix); + + // Insert truncates and extends for any truncated instructions as hints to + // InstCombine. + if (VF > 1) + truncateToMinimalBitwidths(); + + // At this point every instruction in the original loop is widened to + // a vector form. We are almost done. Now, we need to fix the PHI nodes + // that we vectorized. The PHI nodes are currently empty because we did + // not want to introduce cycles. Notice that the remaining PHI nodes + // that we need to fix are reduction variables. + + // Create the 'reduced' values for each of the induction vars. + // The reduced values are the vector values that we scalarize and combine + // after the loop is finished. + for (PhiVector::iterator it = RdxPHIsToFix.begin(), e = RdxPHIsToFix.end(); + it != e; ++it) { + PHINode *RdxPhi = *it; + assert(RdxPhi && "Unable to recover vectorized PHI"); + + // Find the reduction variable descriptor. + assert(Legal->isReductionVariable(RdxPhi) && + "Unable to find the reduction variable"); + RecurrenceDescriptor RdxDesc = (*Legal->getReductionVars())[RdxPhi]; + + RecurrenceDescriptor::RecurrenceKind RK = RdxDesc.getRecurrenceKind(); + TrackingVH<Value> ReductionStartValue = RdxDesc.getRecurrenceStartValue(); + Instruction *LoopExitInst = RdxDesc.getLoopExitInstr(); + RecurrenceDescriptor::MinMaxRecurrenceKind MinMaxKind = + RdxDesc.getMinMaxRecurrenceKind(); + setDebugLocFromInst(Builder, ReductionStartValue); + + // We need to generate a reduction vector from the incoming scalar. + // To do so, we need to generate the 'identity' vector and override + // one of the elements with the incoming scalar reduction. We need + // to do it in the vector-loop preheader. + Builder.SetInsertPoint(LoopBypassBlocks[1]->getTerminator()); + + // This is the vector-clone of the value that leaves the loop. + VectorParts &VectorExit = getVectorValue(LoopExitInst); + Type *VecTy = VectorExit[0]->getType(); + + // Find the reduction identity variable. Zero for addition, or, xor, + // one for multiplication, -1 for And. + Value *Identity; + Value *VectorStart; + if (RK == RecurrenceDescriptor::RK_IntegerMinMax || + RK == RecurrenceDescriptor::RK_FloatMinMax) { + // MinMax reduction have the start value as their identify. + if (VF == 1) { + VectorStart = Identity = ReductionStartValue; + } else { + VectorStart = Identity = + Builder.CreateVectorSplat(VF, ReductionStartValue, "minmax.ident"); + } + } else { + // Handle other reduction kinds: + Constant *Iden = RecurrenceDescriptor::getRecurrenceIdentity( + RK, VecTy->getScalarType()); + if (VF == 1) { + Identity = Iden; + // This vector is the Identity vector where the first element is the + // incoming scalar reduction. + VectorStart = ReductionStartValue; + } else { + Identity = ConstantVector::getSplat(VF, Iden); + + // This vector is the Identity vector where the first element is the + // incoming scalar reduction. + VectorStart = + Builder.CreateInsertElement(Identity, ReductionStartValue, Zero); + } + } + + // Fix the vector-loop phi. + + // Reductions do not have to start at zero. They can start with + // any loop invariant values. + VectorParts &VecRdxPhi = WidenMap.get(RdxPhi); + BasicBlock *Latch = OrigLoop->getLoopLatch(); + Value *LoopVal = RdxPhi->getIncomingValueForBlock(Latch); + VectorParts &Val = getVectorValue(LoopVal); + for (unsigned part = 0; part < UF; ++part) { + // Make sure to add the reduction stat value only to the + // first unroll part. + Value *StartVal = (part == 0) ? VectorStart : Identity; + cast<PHINode>(VecRdxPhi[part])->addIncoming(StartVal, + LoopVectorPreHeader); + cast<PHINode>(VecRdxPhi[part])->addIncoming(Val[part], + LoopVectorBody.back()); + } + + // Before each round, move the insertion point right between + // the PHIs and the values we are going to write. + // This allows us to write both PHINodes and the extractelement + // instructions. + Builder.SetInsertPoint(&*LoopMiddleBlock->getFirstInsertionPt()); + + VectorParts RdxParts = getVectorValue(LoopExitInst); + setDebugLocFromInst(Builder, LoopExitInst); + + // If the vector reduction can be performed in a smaller type, we truncate + // then extend the loop exit value to enable InstCombine to evaluate the + // entire expression in the smaller type. + if (VF > 1 && RdxPhi->getType() != RdxDesc.getRecurrenceType()) { + Type *RdxVecTy = VectorType::get(RdxDesc.getRecurrenceType(), VF); + Builder.SetInsertPoint(LoopVectorBody.back()->getTerminator()); + for (unsigned part = 0; part < UF; ++part) { + Value *Trunc = Builder.CreateTrunc(RdxParts[part], RdxVecTy); + Value *Extnd = RdxDesc.isSigned() ? Builder.CreateSExt(Trunc, VecTy) + : Builder.CreateZExt(Trunc, VecTy); + for (Value::user_iterator UI = RdxParts[part]->user_begin(); + UI != RdxParts[part]->user_end();) + if (*UI != Trunc) { + (*UI++)->replaceUsesOfWith(RdxParts[part], Extnd); + RdxParts[part] = Extnd; + } else { + ++UI; + } + } + Builder.SetInsertPoint(&*LoopMiddleBlock->getFirstInsertionPt()); + for (unsigned part = 0; part < UF; ++part) + RdxParts[part] = Builder.CreateTrunc(RdxParts[part], RdxVecTy); + } + + // Reduce all of the unrolled parts into a single vector. + Value *ReducedPartRdx = RdxParts[0]; + unsigned Op = RecurrenceDescriptor::getRecurrenceBinOp(RK); + setDebugLocFromInst(Builder, ReducedPartRdx); + for (unsigned part = 1; part < UF; ++part) { + if (Op != Instruction::ICmp && Op != Instruction::FCmp) + // Floating point operations had to be 'fast' to enable the reduction. + ReducedPartRdx = addFastMathFlag( + Builder.CreateBinOp((Instruction::BinaryOps)Op, RdxParts[part], + ReducedPartRdx, "bin.rdx")); + else + ReducedPartRdx = RecurrenceDescriptor::createMinMaxOp( + Builder, MinMaxKind, ReducedPartRdx, RdxParts[part]); + } + + if (VF > 1) { + // VF is a power of 2 so we can emit the reduction using log2(VF) shuffles + // and vector ops, reducing the set of values being computed by half each + // round. + assert(isPowerOf2_32(VF) && + "Reduction emission only supported for pow2 vectors!"); + Value *TmpVec = ReducedPartRdx; + SmallVector<Constant*, 32> ShuffleMask(VF, nullptr); + for (unsigned i = VF; i != 1; i >>= 1) { + // Move the upper half of the vector to the lower half. + for (unsigned j = 0; j != i/2; ++j) + ShuffleMask[j] = Builder.getInt32(i/2 + j); + + // Fill the rest of the mask with undef. + std::fill(&ShuffleMask[i/2], ShuffleMask.end(), + UndefValue::get(Builder.getInt32Ty())); + + Value *Shuf = + Builder.CreateShuffleVector(TmpVec, + UndefValue::get(TmpVec->getType()), + ConstantVector::get(ShuffleMask), + "rdx.shuf"); + + if (Op != Instruction::ICmp && Op != Instruction::FCmp) + // Floating point operations had to be 'fast' to enable the reduction. + TmpVec = addFastMathFlag(Builder.CreateBinOp( + (Instruction::BinaryOps)Op, TmpVec, Shuf, "bin.rdx")); + else + TmpVec = RecurrenceDescriptor::createMinMaxOp(Builder, MinMaxKind, + TmpVec, Shuf); + } + + // The result is in the first element of the vector. + ReducedPartRdx = Builder.CreateExtractElement(TmpVec, + Builder.getInt32(0)); + + // If the reduction can be performed in a smaller type, we need to extend + // the reduction to the wider type before we branch to the original loop. + if (RdxPhi->getType() != RdxDesc.getRecurrenceType()) + ReducedPartRdx = + RdxDesc.isSigned() + ? Builder.CreateSExt(ReducedPartRdx, RdxPhi->getType()) + : Builder.CreateZExt(ReducedPartRdx, RdxPhi->getType()); + } + + // Create a phi node that merges control-flow from the backedge-taken check + // block and the middle block. + PHINode *BCBlockPhi = PHINode::Create(RdxPhi->getType(), 2, "bc.merge.rdx", + LoopScalarPreHeader->getTerminator()); + for (unsigned I = 0, E = LoopBypassBlocks.size(); I != E; ++I) + BCBlockPhi->addIncoming(ReductionStartValue, LoopBypassBlocks[I]); + BCBlockPhi->addIncoming(ReducedPartRdx, LoopMiddleBlock); + + // Now, we need to fix the users of the reduction variable + // inside and outside of the scalar remainder loop. + // We know that the loop is in LCSSA form. We need to update the + // PHI nodes in the exit blocks. + for (BasicBlock::iterator LEI = LoopExitBlock->begin(), + LEE = LoopExitBlock->end(); LEI != LEE; ++LEI) { + PHINode *LCSSAPhi = dyn_cast<PHINode>(LEI); + if (!LCSSAPhi) break; + + // All PHINodes need to have a single entry edge, or two if + // we already fixed them. + assert(LCSSAPhi->getNumIncomingValues() < 3 && "Invalid LCSSA PHI"); + + // We found our reduction value exit-PHI. Update it with the + // incoming bypass edge. + if (LCSSAPhi->getIncomingValue(0) == LoopExitInst) { + // Add an edge coming from the bypass. + LCSSAPhi->addIncoming(ReducedPartRdx, LoopMiddleBlock); + break; + } + }// end of the LCSSA phi scan. + + // Fix the scalar loop reduction variable with the incoming reduction sum + // from the vector body and from the backedge value. + int IncomingEdgeBlockIdx = + (RdxPhi)->getBasicBlockIndex(OrigLoop->getLoopLatch()); + assert(IncomingEdgeBlockIdx >= 0 && "Invalid block index"); + // Pick the other block. + int SelfEdgeBlockIdx = (IncomingEdgeBlockIdx ? 0 : 1); + (RdxPhi)->setIncomingValue(SelfEdgeBlockIdx, BCBlockPhi); + (RdxPhi)->setIncomingValue(IncomingEdgeBlockIdx, LoopExitInst); + }// end of for each redux variable. + + fixLCSSAPHIs(); + + // Make sure DomTree is updated. + updateAnalysis(); + + // Predicate any stores. + for (auto KV : PredicatedStores) { + BasicBlock::iterator I(KV.first); + auto *BB = SplitBlock(I->getParent(), &*std::next(I), DT, LI); + auto *T = SplitBlockAndInsertIfThen(KV.second, &*I, /*Unreachable=*/false, + /*BranchWeights=*/nullptr, DT); + I->moveBefore(T); + I->getParent()->setName("pred.store.if"); + BB->setName("pred.store.continue"); + } + DEBUG(DT->verifyDomTree()); + // Remove redundant induction instructions. + cse(LoopVectorBody); +} + +void InnerLoopVectorizer::fixLCSSAPHIs() { + for (BasicBlock::iterator LEI = LoopExitBlock->begin(), + LEE = LoopExitBlock->end(); LEI != LEE; ++LEI) { + PHINode *LCSSAPhi = dyn_cast<PHINode>(LEI); + if (!LCSSAPhi) break; + if (LCSSAPhi->getNumIncomingValues() == 1) + LCSSAPhi->addIncoming(UndefValue::get(LCSSAPhi->getType()), + LoopMiddleBlock); + } +} + +InnerLoopVectorizer::VectorParts +InnerLoopVectorizer::createEdgeMask(BasicBlock *Src, BasicBlock *Dst) { + assert(std::find(pred_begin(Dst), pred_end(Dst), Src) != pred_end(Dst) && + "Invalid edge"); + + // Look for cached value. + std::pair<BasicBlock*, BasicBlock*> Edge(Src, Dst); + EdgeMaskCache::iterator ECEntryIt = MaskCache.find(Edge); + if (ECEntryIt != MaskCache.end()) + return ECEntryIt->second; + + VectorParts SrcMask = createBlockInMask(Src); + + // The terminator has to be a branch inst! + BranchInst *BI = dyn_cast<BranchInst>(Src->getTerminator()); + assert(BI && "Unexpected terminator found"); + + if (BI->isConditional()) { + VectorParts EdgeMask = getVectorValue(BI->getCondition()); + + if (BI->getSuccessor(0) != Dst) + for (unsigned part = 0; part < UF; ++part) + EdgeMask[part] = Builder.CreateNot(EdgeMask[part]); + + for (unsigned part = 0; part < UF; ++part) + EdgeMask[part] = Builder.CreateAnd(EdgeMask[part], SrcMask[part]); + + MaskCache[Edge] = EdgeMask; + return EdgeMask; + } + + MaskCache[Edge] = SrcMask; + return SrcMask; +} + +InnerLoopVectorizer::VectorParts +InnerLoopVectorizer::createBlockInMask(BasicBlock *BB) { + assert(OrigLoop->contains(BB) && "Block is not a part of a loop"); + + // Loop incoming mask is all-one. + if (OrigLoop->getHeader() == BB) { + Value *C = ConstantInt::get(IntegerType::getInt1Ty(BB->getContext()), 1); + return getVectorValue(C); + } + + // This is the block mask. We OR all incoming edges, and with zero. + Value *Zero = ConstantInt::get(IntegerType::getInt1Ty(BB->getContext()), 0); + VectorParts BlockMask = getVectorValue(Zero); + + // For each pred: + for (pred_iterator it = pred_begin(BB), e = pred_end(BB); it != e; ++it) { + VectorParts EM = createEdgeMask(*it, BB); + for (unsigned part = 0; part < UF; ++part) + BlockMask[part] = Builder.CreateOr(BlockMask[part], EM[part]); + } + + return BlockMask; +} + +void InnerLoopVectorizer::widenPHIInstruction( + Instruction *PN, InnerLoopVectorizer::VectorParts &Entry, unsigned UF, + unsigned VF, PhiVector *PV) { + PHINode* P = cast<PHINode>(PN); + // Handle reduction variables: + if (Legal->isReductionVariable(P)) { + for (unsigned part = 0; part < UF; ++part) { + // This is phase one of vectorizing PHIs. + Type *VecTy = (VF == 1) ? PN->getType() : + VectorType::get(PN->getType(), VF); + Entry[part] = PHINode::Create( + VecTy, 2, "vec.phi", &*LoopVectorBody.back()->getFirstInsertionPt()); + } + PV->push_back(P); + return; + } + + setDebugLocFromInst(Builder, P); + // Check for PHI nodes that are lowered to vector selects. + if (P->getParent() != OrigLoop->getHeader()) { + // We know that all PHIs in non-header blocks are converted into + // selects, so we don't have to worry about the insertion order and we + // can just use the builder. + // At this point we generate the predication tree. There may be + // duplications since this is a simple recursive scan, but future + // optimizations will clean it up. + + unsigned NumIncoming = P->getNumIncomingValues(); + + // Generate a sequence of selects of the form: + // SELECT(Mask3, In3, + // SELECT(Mask2, In2, + // ( ...))) + for (unsigned In = 0; In < NumIncoming; In++) { + VectorParts Cond = createEdgeMask(P->getIncomingBlock(In), + P->getParent()); + VectorParts &In0 = getVectorValue(P->getIncomingValue(In)); + + for (unsigned part = 0; part < UF; ++part) { + // We might have single edge PHIs (blocks) - use an identity + // 'select' for the first PHI operand. + if (In == 0) + Entry[part] = Builder.CreateSelect(Cond[part], In0[part], + In0[part]); + else + // Select between the current value and the previous incoming edge + // based on the incoming mask. + Entry[part] = Builder.CreateSelect(Cond[part], In0[part], + Entry[part], "predphi"); + } + } + return; + } + + // This PHINode must be an induction variable. + // Make sure that we know about it. + assert(Legal->getInductionVars()->count(P) && + "Not an induction variable"); + + InductionDescriptor II = Legal->getInductionVars()->lookup(P); + + // FIXME: The newly created binary instructions should contain nsw/nuw flags, + // which can be found from the original scalar operations. + switch (II.getKind()) { + case InductionDescriptor::IK_NoInduction: + llvm_unreachable("Unknown induction"); + case InductionDescriptor::IK_IntInduction: { + assert(P->getType() == II.getStartValue()->getType() && + "Types must match"); + // Handle other induction variables that are now based on the + // canonical one. + Value *V = Induction; + if (P != OldInduction) { + V = Builder.CreateSExtOrTrunc(Induction, P->getType()); + V = II.transform(Builder, V); + V->setName("offset.idx"); + } + Value *Broadcasted = getBroadcastInstrs(V); + // After broadcasting the induction variable we need to make the vector + // consecutive by adding 0, 1, 2, etc. + for (unsigned part = 0; part < UF; ++part) + Entry[part] = getStepVector(Broadcasted, VF * part, II.getStepValue()); + return; + } + case InductionDescriptor::IK_PtrInduction: + // Handle the pointer induction variable case. + assert(P->getType()->isPointerTy() && "Unexpected type."); + // This is the normalized GEP that starts counting at zero. + Value *PtrInd = Induction; + PtrInd = Builder.CreateSExtOrTrunc(PtrInd, II.getStepValue()->getType()); + // This is the vector of results. Notice that we don't generate + // vector geps because scalar geps result in better code. + for (unsigned part = 0; part < UF; ++part) { + if (VF == 1) { + int EltIndex = part; + Constant *Idx = ConstantInt::get(PtrInd->getType(), EltIndex); + Value *GlobalIdx = Builder.CreateAdd(PtrInd, Idx); + Value *SclrGep = II.transform(Builder, GlobalIdx); + SclrGep->setName("next.gep"); + Entry[part] = SclrGep; + continue; + } + + Value *VecVal = UndefValue::get(VectorType::get(P->getType(), VF)); + for (unsigned int i = 0; i < VF; ++i) { + int EltIndex = i + part * VF; + Constant *Idx = ConstantInt::get(PtrInd->getType(), EltIndex); + Value *GlobalIdx = Builder.CreateAdd(PtrInd, Idx); + Value *SclrGep = II.transform(Builder, GlobalIdx); + SclrGep->setName("next.gep"); + VecVal = Builder.CreateInsertElement(VecVal, SclrGep, + Builder.getInt32(i), + "insert.gep"); + } + Entry[part] = VecVal; + } + return; + } +} + +void InnerLoopVectorizer::vectorizeBlockInLoop(BasicBlock *BB, PhiVector *PV) { + // For each instruction in the old loop. + for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) { + VectorParts &Entry = WidenMap.get(&*it); + + switch (it->getOpcode()) { + case Instruction::Br: + // Nothing to do for PHIs and BR, since we already took care of the + // loop control flow instructions. + continue; + case Instruction::PHI: { + // Vectorize PHINodes. + widenPHIInstruction(&*it, Entry, UF, VF, PV); + continue; + }// End of PHI. + + case Instruction::Add: + case Instruction::FAdd: + case Instruction::Sub: + case Instruction::FSub: + case Instruction::Mul: + case Instruction::FMul: + case Instruction::UDiv: + case Instruction::SDiv: + case Instruction::FDiv: + case Instruction::URem: + case Instruction::SRem: + case Instruction::FRem: + case Instruction::Shl: + case Instruction::LShr: + case Instruction::AShr: + case Instruction::And: + case Instruction::Or: + case Instruction::Xor: { + // Just widen binops. + BinaryOperator *BinOp = dyn_cast<BinaryOperator>(it); + setDebugLocFromInst(Builder, BinOp); + VectorParts &A = getVectorValue(it->getOperand(0)); + VectorParts &B = getVectorValue(it->getOperand(1)); + + // Use this vector value for all users of the original instruction. + for (unsigned Part = 0; Part < UF; ++Part) { + Value *V = Builder.CreateBinOp(BinOp->getOpcode(), A[Part], B[Part]); + + if (BinaryOperator *VecOp = dyn_cast<BinaryOperator>(V)) + VecOp->copyIRFlags(BinOp); + + Entry[Part] = V; + } + + propagateMetadata(Entry, &*it); + break; + } + case Instruction::Select: { + // Widen selects. + // If the selector is loop invariant we can create a select + // instruction with a scalar condition. Otherwise, use vector-select. + auto *SE = PSE.getSE(); + bool InvariantCond = + SE->isLoopInvariant(PSE.getSCEV(it->getOperand(0)), OrigLoop); + setDebugLocFromInst(Builder, &*it); + + // The condition can be loop invariant but still defined inside the + // loop. This means that we can't just use the original 'cond' value. + // We have to take the 'vectorized' value and pick the first lane. + // Instcombine will make this a no-op. + VectorParts &Cond = getVectorValue(it->getOperand(0)); + VectorParts &Op0 = getVectorValue(it->getOperand(1)); + VectorParts &Op1 = getVectorValue(it->getOperand(2)); + + Value *ScalarCond = (VF == 1) ? Cond[0] : + Builder.CreateExtractElement(Cond[0], Builder.getInt32(0)); + + for (unsigned Part = 0; Part < UF; ++Part) { + Entry[Part] = Builder.CreateSelect( + InvariantCond ? ScalarCond : Cond[Part], + Op0[Part], + Op1[Part]); + } + + propagateMetadata(Entry, &*it); + break; + } + + case Instruction::ICmp: + case Instruction::FCmp: { + // Widen compares. Generate vector compares. + bool FCmp = (it->getOpcode() == Instruction::FCmp); + CmpInst *Cmp = dyn_cast<CmpInst>(it); + setDebugLocFromInst(Builder, &*it); + VectorParts &A = getVectorValue(it->getOperand(0)); + VectorParts &B = getVectorValue(it->getOperand(1)); + for (unsigned Part = 0; Part < UF; ++Part) { + Value *C = nullptr; + if (FCmp) { + C = Builder.CreateFCmp(Cmp->getPredicate(), A[Part], B[Part]); + cast<FCmpInst>(C)->copyFastMathFlags(&*it); + } else { + C = Builder.CreateICmp(Cmp->getPredicate(), A[Part], B[Part]); + } + Entry[Part] = C; + } + + propagateMetadata(Entry, &*it); + break; + } + + case Instruction::Store: + case Instruction::Load: + vectorizeMemoryInstruction(&*it); + break; + case Instruction::ZExt: + case Instruction::SExt: + case Instruction::FPToUI: + case Instruction::FPToSI: + case Instruction::FPExt: + case Instruction::PtrToInt: + case Instruction::IntToPtr: + case Instruction::SIToFP: + case Instruction::UIToFP: + case Instruction::Trunc: + case Instruction::FPTrunc: + case Instruction::BitCast: { + CastInst *CI = dyn_cast<CastInst>(it); + setDebugLocFromInst(Builder, &*it); + /// Optimize the special case where the source is the induction + /// variable. Notice that we can only optimize the 'trunc' case + /// because: a. FP conversions lose precision, b. sext/zext may wrap, + /// c. other casts depend on pointer size. + if (CI->getOperand(0) == OldInduction && + it->getOpcode() == Instruction::Trunc) { + Value *ScalarCast = Builder.CreateCast(CI->getOpcode(), Induction, + CI->getType()); + Value *Broadcasted = getBroadcastInstrs(ScalarCast); + InductionDescriptor II = + Legal->getInductionVars()->lookup(OldInduction); + Constant *Step = ConstantInt::getSigned( + CI->getType(), II.getStepValue()->getSExtValue()); + for (unsigned Part = 0; Part < UF; ++Part) + Entry[Part] = getStepVector(Broadcasted, VF * Part, Step); + propagateMetadata(Entry, &*it); + break; + } + /// Vectorize casts. + Type *DestTy = (VF == 1) ? CI->getType() : + VectorType::get(CI->getType(), VF); + + VectorParts &A = getVectorValue(it->getOperand(0)); + for (unsigned Part = 0; Part < UF; ++Part) + Entry[Part] = Builder.CreateCast(CI->getOpcode(), A[Part], DestTy); + propagateMetadata(Entry, &*it); + break; + } + + case Instruction::Call: { + // Ignore dbg intrinsics. + if (isa<DbgInfoIntrinsic>(it)) + break; + setDebugLocFromInst(Builder, &*it); + + Module *M = BB->getParent()->getParent(); + CallInst *CI = cast<CallInst>(it); + + StringRef FnName = CI->getCalledFunction()->getName(); + Function *F = CI->getCalledFunction(); + Type *RetTy = ToVectorTy(CI->getType(), VF); + SmallVector<Type *, 4> Tys; + for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) + Tys.push_back(ToVectorTy(CI->getArgOperand(i)->getType(), VF)); + + Intrinsic::ID ID = getIntrinsicIDForCall(CI, TLI); + if (ID && + (ID == Intrinsic::assume || ID == Intrinsic::lifetime_end || + ID == Intrinsic::lifetime_start)) { + scalarizeInstruction(&*it); + break; + } + // The flag shows whether we use Intrinsic or a usual Call for vectorized + // version of the instruction. + // Is it beneficial to perform intrinsic call compared to lib call? + bool NeedToScalarize; + unsigned CallCost = getVectorCallCost(CI, VF, *TTI, TLI, NeedToScalarize); + bool UseVectorIntrinsic = + ID && getVectorIntrinsicCost(CI, VF, *TTI, TLI) <= CallCost; + if (!UseVectorIntrinsic && NeedToScalarize) { + scalarizeInstruction(&*it); + break; + } + + for (unsigned Part = 0; Part < UF; ++Part) { + SmallVector<Value *, 4> Args; + for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) { + Value *Arg = CI->getArgOperand(i); + // Some intrinsics have a scalar argument - don't replace it with a + // vector. + if (!UseVectorIntrinsic || !hasVectorInstrinsicScalarOpd(ID, i)) { + VectorParts &VectorArg = getVectorValue(CI->getArgOperand(i)); + Arg = VectorArg[Part]; + } + Args.push_back(Arg); + } + + Function *VectorF; + if (UseVectorIntrinsic) { + // Use vector version of the intrinsic. + Type *TysForDecl[] = {CI->getType()}; + if (VF > 1) + TysForDecl[0] = VectorType::get(CI->getType()->getScalarType(), VF); + VectorF = Intrinsic::getDeclaration(M, ID, TysForDecl); + } else { + // Use vector version of the library call. + StringRef VFnName = TLI->getVectorizedFunction(FnName, VF); + assert(!VFnName.empty() && "Vector function name is empty."); + VectorF = M->getFunction(VFnName); + if (!VectorF) { + // Generate a declaration + FunctionType *FTy = FunctionType::get(RetTy, Tys, false); + VectorF = + Function::Create(FTy, Function::ExternalLinkage, VFnName, M); + VectorF->copyAttributesFrom(F); + } + } + assert(VectorF && "Can't create vector function."); + Entry[Part] = Builder.CreateCall(VectorF, Args); + } + + propagateMetadata(Entry, &*it); + break; + } + + default: + // All other instructions are unsupported. Scalarize them. + scalarizeInstruction(&*it); + break; + }// end of switch. + }// end of for_each instr. +} + +void InnerLoopVectorizer::updateAnalysis() { + // Forget the original basic block. + PSE.getSE()->forgetLoop(OrigLoop); + + // Update the dominator tree information. + assert(DT->properlyDominates(LoopBypassBlocks.front(), LoopExitBlock) && + "Entry does not dominate exit."); + + for (unsigned I = 1, E = LoopBypassBlocks.size(); I != E; ++I) + DT->addNewBlock(LoopBypassBlocks[I], LoopBypassBlocks[I-1]); + DT->addNewBlock(LoopVectorPreHeader, LoopBypassBlocks.back()); + + // We don't predicate stores by this point, so the vector body should be a + // single loop. + assert(LoopVectorBody.size() == 1 && "Expected single block loop!"); + DT->addNewBlock(LoopVectorBody[0], LoopVectorPreHeader); + + DT->addNewBlock(LoopMiddleBlock, LoopVectorBody.back()); + DT->addNewBlock(LoopScalarPreHeader, LoopBypassBlocks[0]); + DT->changeImmediateDominator(LoopScalarBody, LoopScalarPreHeader); + DT->changeImmediateDominator(LoopExitBlock, LoopBypassBlocks[0]); + + DEBUG(DT->verifyDomTree()); +} + +/// \brief Check whether it is safe to if-convert this phi node. +/// +/// Phi nodes with constant expressions that can trap are not safe to if +/// convert. +static bool canIfConvertPHINodes(BasicBlock *BB) { + for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) { + PHINode *Phi = dyn_cast<PHINode>(I); + if (!Phi) + return true; + for (unsigned p = 0, e = Phi->getNumIncomingValues(); p != e; ++p) + if (Constant *C = dyn_cast<Constant>(Phi->getIncomingValue(p))) + if (C->canTrap()) + return false; + } + return true; +} + +bool LoopVectorizationLegality::canVectorizeWithIfConvert() { + if (!EnableIfConversion) { + emitAnalysis(VectorizationReport() << "if-conversion is disabled"); + return false; + } + + assert(TheLoop->getNumBlocks() > 1 && "Single block loops are vectorizable"); + + // A list of pointers that we can safely read and write to. + SmallPtrSet<Value *, 8> SafePointes; + + // Collect safe addresses. + for (Loop::block_iterator BI = TheLoop->block_begin(), + BE = TheLoop->block_end(); BI != BE; ++BI) { + BasicBlock *BB = *BI; + + if (blockNeedsPredication(BB)) + continue; + + for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) { + if (LoadInst *LI = dyn_cast<LoadInst>(I)) + SafePointes.insert(LI->getPointerOperand()); + else if (StoreInst *SI = dyn_cast<StoreInst>(I)) + SafePointes.insert(SI->getPointerOperand()); + } + } + + // Collect the blocks that need predication. + BasicBlock *Header = TheLoop->getHeader(); + for (Loop::block_iterator BI = TheLoop->block_begin(), + BE = TheLoop->block_end(); BI != BE; ++BI) { + BasicBlock *BB = *BI; + + // We don't support switch statements inside loops. + if (!isa<BranchInst>(BB->getTerminator())) { + emitAnalysis(VectorizationReport(BB->getTerminator()) + << "loop contains a switch statement"); + return false; + } + + // We must be able to predicate all blocks that need to be predicated. + if (blockNeedsPredication(BB)) { + if (!blockCanBePredicated(BB, SafePointes)) { + emitAnalysis(VectorizationReport(BB->getTerminator()) + << "control flow cannot be substituted for a select"); + return false; + } + } else if (BB != Header && !canIfConvertPHINodes(BB)) { + emitAnalysis(VectorizationReport(BB->getTerminator()) + << "control flow cannot be substituted for a select"); + return false; + } + } + + // We can if-convert this loop. + return true; +} + +bool LoopVectorizationLegality::canVectorize() { + // We must have a loop in canonical form. Loops with indirectbr in them cannot + // be canonicalized. + if (!TheLoop->getLoopPreheader()) { + emitAnalysis( + VectorizationReport() << + "loop control flow is not understood by vectorizer"); + return false; + } + + // We can only vectorize innermost loops. + if (!TheLoop->empty()) { + emitAnalysis(VectorizationReport() << "loop is not the innermost loop"); + return false; + } + + // We must have a single backedge. + if (TheLoop->getNumBackEdges() != 1) { + emitAnalysis( + VectorizationReport() << + "loop control flow is not understood by vectorizer"); + return false; + } + + // We must have a single exiting block. + if (!TheLoop->getExitingBlock()) { + emitAnalysis( + VectorizationReport() << + "loop control flow is not understood by vectorizer"); + return false; + } + + // We only handle bottom-tested loops, i.e. loop in which the condition is + // checked at the end of each iteration. With that we can assume that all + // instructions in the loop are executed the same number of times. + if (TheLoop->getExitingBlock() != TheLoop->getLoopLatch()) { + emitAnalysis( + VectorizationReport() << + "loop control flow is not understood by vectorizer"); + return false; + } + + // We need to have a loop header. + DEBUG(dbgs() << "LV: Found a loop: " << + TheLoop->getHeader()->getName() << '\n'); + + // Check if we can if-convert non-single-bb loops. + unsigned NumBlocks = TheLoop->getNumBlocks(); + if (NumBlocks != 1 && !canVectorizeWithIfConvert()) { + DEBUG(dbgs() << "LV: Can't if-convert the loop.\n"); + return false; + } + + // ScalarEvolution needs to be able to find the exit count. + const SCEV *ExitCount = PSE.getSE()->getBackedgeTakenCount(TheLoop); + if (ExitCount == PSE.getSE()->getCouldNotCompute()) { + emitAnalysis(VectorizationReport() + << "could not determine number of loop iterations"); + DEBUG(dbgs() << "LV: SCEV could not compute the loop exit count.\n"); + return false; + } + + // Check if we can vectorize the instructions and CFG in this loop. + if (!canVectorizeInstrs()) { + DEBUG(dbgs() << "LV: Can't vectorize the instructions or CFG\n"); + return false; + } + + // Go over each instruction and look at memory deps. + if (!canVectorizeMemory()) { + DEBUG(dbgs() << "LV: Can't vectorize due to memory conflicts\n"); + return false; + } + + // Collect all of the variables that remain uniform after vectorization. + collectLoopUniforms(); + + DEBUG(dbgs() << "LV: We can vectorize this loop" + << (LAI->getRuntimePointerChecking()->Need + ? " (with a runtime bound check)" + : "") + << "!\n"); + + bool UseInterleaved = TTI->enableInterleavedAccessVectorization(); + + // If an override option has been passed in for interleaved accesses, use it. + if (EnableInterleavedMemAccesses.getNumOccurrences() > 0) + UseInterleaved = EnableInterleavedMemAccesses; + + // Analyze interleaved memory accesses. + if (UseInterleaved) + InterleaveInfo.analyzeInterleaving(Strides); + + unsigned SCEVThreshold = VectorizeSCEVCheckThreshold; + if (Hints->getForce() == LoopVectorizeHints::FK_Enabled) + SCEVThreshold = PragmaVectorizeSCEVCheckThreshold; + + if (PSE.getUnionPredicate().getComplexity() > SCEVThreshold) { + emitAnalysis(VectorizationReport() + << "Too many SCEV assumptions need to be made and checked " + << "at runtime"); + DEBUG(dbgs() << "LV: Too many SCEV checks needed.\n"); + return false; + } + + // Okay! We can vectorize. At this point we don't have any other mem analysis + // which may limit our maximum vectorization factor, so just return true with + // no restrictions. + return true; +} + +static Type *convertPointerToIntegerType(const DataLayout &DL, Type *Ty) { + if (Ty->isPointerTy()) + return DL.getIntPtrType(Ty); + + // It is possible that char's or short's overflow when we ask for the loop's + // trip count, work around this by changing the type size. + if (Ty->getScalarSizeInBits() < 32) + return Type::getInt32Ty(Ty->getContext()); + + return Ty; +} + +static Type* getWiderType(const DataLayout &DL, Type *Ty0, Type *Ty1) { + Ty0 = convertPointerToIntegerType(DL, Ty0); + Ty1 = convertPointerToIntegerType(DL, Ty1); + if (Ty0->getScalarSizeInBits() > Ty1->getScalarSizeInBits()) + return Ty0; + return Ty1; +} + +/// \brief Check that the instruction has outside loop users and is not an +/// identified reduction variable. +static bool hasOutsideLoopUser(const Loop *TheLoop, Instruction *Inst, + SmallPtrSetImpl<Value *> &Reductions) { + // Reduction instructions are allowed to have exit users. All other + // instructions must not have external users. + if (!Reductions.count(Inst)) + //Check that all of the users of the loop are inside the BB. + for (User *U : Inst->users()) { + Instruction *UI = cast<Instruction>(U); + // This user may be a reduction exit value. + if (!TheLoop->contains(UI)) { + DEBUG(dbgs() << "LV: Found an outside user for : " << *UI << '\n'); + return true; + } + } + return false; +} + +bool LoopVectorizationLegality::canVectorizeInstrs() { + BasicBlock *Header = TheLoop->getHeader(); + + // Look for the attribute signaling the absence of NaNs. + Function &F = *Header->getParent(); + const DataLayout &DL = F.getParent()->getDataLayout(); + if (F.hasFnAttribute("no-nans-fp-math")) + HasFunNoNaNAttr = + F.getFnAttribute("no-nans-fp-math").getValueAsString() == "true"; + + // For each block in the loop. + for (Loop::block_iterator bb = TheLoop->block_begin(), + be = TheLoop->block_end(); bb != be; ++bb) { + + // Scan the instructions in the block and look for hazards. + for (BasicBlock::iterator it = (*bb)->begin(), e = (*bb)->end(); it != e; + ++it) { + + if (PHINode *Phi = dyn_cast<PHINode>(it)) { + Type *PhiTy = Phi->getType(); + // Check that this PHI type is allowed. + if (!PhiTy->isIntegerTy() && + !PhiTy->isFloatingPointTy() && + !PhiTy->isPointerTy()) { + emitAnalysis(VectorizationReport(&*it) + << "loop control flow is not understood by vectorizer"); + DEBUG(dbgs() << "LV: Found an non-int non-pointer PHI.\n"); + return false; + } + + // If this PHINode is not in the header block, then we know that we + // can convert it to select during if-conversion. No need to check if + // the PHIs in this block are induction or reduction variables. + if (*bb != Header) { + // Check that this instruction has no outside users or is an + // identified reduction value with an outside user. + if (!hasOutsideLoopUser(TheLoop, &*it, AllowedExit)) + continue; + emitAnalysis(VectorizationReport(&*it) << + "value could not be identified as " + "an induction or reduction variable"); + return false; + } + + // We only allow if-converted PHIs with exactly two incoming values. + if (Phi->getNumIncomingValues() != 2) { + emitAnalysis(VectorizationReport(&*it) + << "control flow not understood by vectorizer"); + DEBUG(dbgs() << "LV: Found an invalid PHI.\n"); + return false; + } + + InductionDescriptor ID; + if (InductionDescriptor::isInductionPHI(Phi, PSE.getSE(), ID)) { + Inductions[Phi] = ID; + // Get the widest type. + if (!WidestIndTy) + WidestIndTy = convertPointerToIntegerType(DL, PhiTy); + else + WidestIndTy = getWiderType(DL, PhiTy, WidestIndTy); + + // Int inductions are special because we only allow one IV. + if (ID.getKind() == InductionDescriptor::IK_IntInduction && + ID.getStepValue()->isOne() && + isa<Constant>(ID.getStartValue()) && + cast<Constant>(ID.getStartValue())->isNullValue()) { + // Use the phi node with the widest type as induction. Use the last + // one if there are multiple (no good reason for doing this other + // than it is expedient). We've checked that it begins at zero and + // steps by one, so this is a canonical induction variable. + if (!Induction || PhiTy == WidestIndTy) + Induction = Phi; + } + + DEBUG(dbgs() << "LV: Found an induction variable.\n"); + + // Until we explicitly handle the case of an induction variable with + // an outside loop user we have to give up vectorizing this loop. + if (hasOutsideLoopUser(TheLoop, &*it, AllowedExit)) { + emitAnalysis(VectorizationReport(&*it) << + "use of induction value outside of the " + "loop is not handled by vectorizer"); + return false; + } + + continue; + } + + RecurrenceDescriptor RedDes; + if (RecurrenceDescriptor::isReductionPHI(Phi, TheLoop, RedDes)) { + if (RedDes.hasUnsafeAlgebra()) + Requirements->addUnsafeAlgebraInst(RedDes.getUnsafeAlgebraInst()); + AllowedExit.insert(RedDes.getLoopExitInstr()); + Reductions[Phi] = RedDes; + continue; + } + + emitAnalysis(VectorizationReport(&*it) << + "value that could not be identified as " + "reduction is used outside the loop"); + DEBUG(dbgs() << "LV: Found an unidentified PHI."<< *Phi <<"\n"); + return false; + }// end of PHI handling + + // We handle calls that: + // * Are debug info intrinsics. + // * Have a mapping to an IR intrinsic. + // * Have a vector version available. + CallInst *CI = dyn_cast<CallInst>(it); + if (CI && !getIntrinsicIDForCall(CI, TLI) && !isa<DbgInfoIntrinsic>(CI) && + !(CI->getCalledFunction() && TLI && + TLI->isFunctionVectorizable(CI->getCalledFunction()->getName()))) { + emitAnalysis(VectorizationReport(&*it) + << "call instruction cannot be vectorized"); + DEBUG(dbgs() << "LV: Found a non-intrinsic, non-libfunc callsite.\n"); + return false; + } + + // Intrinsics such as powi,cttz and ctlz are legal to vectorize if the + // second argument is the same (i.e. loop invariant) + if (CI && + hasVectorInstrinsicScalarOpd(getIntrinsicIDForCall(CI, TLI), 1)) { + auto *SE = PSE.getSE(); + if (!SE->isLoopInvariant(PSE.getSCEV(CI->getOperand(1)), TheLoop)) { + emitAnalysis(VectorizationReport(&*it) + << "intrinsic instruction cannot be vectorized"); + DEBUG(dbgs() << "LV: Found unvectorizable intrinsic " << *CI << "\n"); + return false; + } + } + + // Check that the instruction return type is vectorizable. + // Also, we can't vectorize extractelement instructions. + if ((!VectorType::isValidElementType(it->getType()) && + !it->getType()->isVoidTy()) || isa<ExtractElementInst>(it)) { + emitAnalysis(VectorizationReport(&*it) + << "instruction return type cannot be vectorized"); + DEBUG(dbgs() << "LV: Found unvectorizable type.\n"); + return false; + } + + // Check that the stored type is vectorizable. + if (StoreInst *ST = dyn_cast<StoreInst>(it)) { + Type *T = ST->getValueOperand()->getType(); + if (!VectorType::isValidElementType(T)) { + emitAnalysis(VectorizationReport(ST) << + "store instruction cannot be vectorized"); + return false; + } + if (EnableMemAccessVersioning) + collectStridedAccess(ST); + } + + if (EnableMemAccessVersioning) + if (LoadInst *LI = dyn_cast<LoadInst>(it)) + collectStridedAccess(LI); + + // Reduction instructions are allowed to have exit users. + // All other instructions must not have external users. + if (hasOutsideLoopUser(TheLoop, &*it, AllowedExit)) { + emitAnalysis(VectorizationReport(&*it) << + "value cannot be used outside the loop"); + return false; + } + + } // next instr. + + } + + if (!Induction) { + DEBUG(dbgs() << "LV: Did not find one integer induction var.\n"); + if (Inductions.empty()) { + emitAnalysis(VectorizationReport() + << "loop induction variable could not be identified"); + return false; + } + } + + // Now we know the widest induction type, check if our found induction + // is the same size. If it's not, unset it here and InnerLoopVectorizer + // will create another. + if (Induction && WidestIndTy != Induction->getType()) + Induction = nullptr; + + return true; +} + +void LoopVectorizationLegality::collectStridedAccess(Value *MemAccess) { + Value *Ptr = nullptr; + if (LoadInst *LI = dyn_cast<LoadInst>(MemAccess)) + Ptr = LI->getPointerOperand(); + else if (StoreInst *SI = dyn_cast<StoreInst>(MemAccess)) + Ptr = SI->getPointerOperand(); + else + return; + + Value *Stride = getStrideFromPointer(Ptr, PSE.getSE(), TheLoop); + if (!Stride) + return; + + DEBUG(dbgs() << "LV: Found a strided access that we can version"); + DEBUG(dbgs() << " Ptr: " << *Ptr << " Stride: " << *Stride << "\n"); + Strides[Ptr] = Stride; + StrideSet.insert(Stride); +} + +void LoopVectorizationLegality::collectLoopUniforms() { + // We now know that the loop is vectorizable! + // Collect variables that will remain uniform after vectorization. + std::vector<Value*> Worklist; + BasicBlock *Latch = TheLoop->getLoopLatch(); + + // Start with the conditional branch and walk up the block. + Worklist.push_back(Latch->getTerminator()->getOperand(0)); + + // Also add all consecutive pointer values; these values will be uniform + // after vectorization (and subsequent cleanup) and, until revectorization is + // supported, all dependencies must also be uniform. + for (Loop::block_iterator B = TheLoop->block_begin(), + BE = TheLoop->block_end(); B != BE; ++B) + for (BasicBlock::iterator I = (*B)->begin(), IE = (*B)->end(); + I != IE; ++I) + if (I->getType()->isPointerTy() && isConsecutivePtr(&*I)) + Worklist.insert(Worklist.end(), I->op_begin(), I->op_end()); + + while (!Worklist.empty()) { + Instruction *I = dyn_cast<Instruction>(Worklist.back()); + Worklist.pop_back(); + + // Look at instructions inside this loop. + // Stop when reaching PHI nodes. + // TODO: we need to follow values all over the loop, not only in this block. + if (!I || !TheLoop->contains(I) || isa<PHINode>(I)) + continue; + + // This is a known uniform. + Uniforms.insert(I); + + // Insert all operands. + Worklist.insert(Worklist.end(), I->op_begin(), I->op_end()); + } +} + +bool LoopVectorizationLegality::canVectorizeMemory() { + LAI = &LAA->getInfo(TheLoop, Strides); + auto &OptionalReport = LAI->getReport(); + if (OptionalReport) + emitAnalysis(VectorizationReport(*OptionalReport)); + if (!LAI->canVectorizeMemory()) + return false; + + if (LAI->hasStoreToLoopInvariantAddress()) { + emitAnalysis( + VectorizationReport() + << "write to a loop invariant address could not be vectorized"); + DEBUG(dbgs() << "LV: We don't allow storing to uniform addresses\n"); + return false; + } + + Requirements->addRuntimePointerChecks(LAI->getNumRuntimePointerChecks()); + PSE.addPredicate(LAI->PSE.getUnionPredicate()); + + return true; +} + +bool LoopVectorizationLegality::isInductionVariable(const Value *V) { + Value *In0 = const_cast<Value*>(V); + PHINode *PN = dyn_cast_or_null<PHINode>(In0); + if (!PN) + return false; + + return Inductions.count(PN); +} + +bool LoopVectorizationLegality::blockNeedsPredication(BasicBlock *BB) { + return LoopAccessInfo::blockNeedsPredication(BB, TheLoop, DT); +} + +bool LoopVectorizationLegality::blockCanBePredicated(BasicBlock *BB, + SmallPtrSetImpl<Value *> &SafePtrs) { + + for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) { + // Check that we don't have a constant expression that can trap as operand. + for (Instruction::op_iterator OI = it->op_begin(), OE = it->op_end(); + OI != OE; ++OI) { + if (Constant *C = dyn_cast<Constant>(*OI)) + if (C->canTrap()) + return false; + } + // We might be able to hoist the load. + if (it->mayReadFromMemory()) { + LoadInst *LI = dyn_cast<LoadInst>(it); + if (!LI) + return false; + if (!SafePtrs.count(LI->getPointerOperand())) { + if (isLegalMaskedLoad(LI->getType(), LI->getPointerOperand())) { + MaskedOp.insert(LI); + continue; + } + return false; + } + } + + // We don't predicate stores at the moment. + if (it->mayWriteToMemory()) { + StoreInst *SI = dyn_cast<StoreInst>(it); + // We only support predication of stores in basic blocks with one + // predecessor. + if (!SI) + return false; + + bool isSafePtr = (SafePtrs.count(SI->getPointerOperand()) != 0); + bool isSinglePredecessor = SI->getParent()->getSinglePredecessor(); + + if (++NumPredStores > NumberOfStoresToPredicate || !isSafePtr || + !isSinglePredecessor) { + // Build a masked store if it is legal for the target, otherwise + // scalarize the block. + bool isLegalMaskedOp = + isLegalMaskedStore(SI->getValueOperand()->getType(), + SI->getPointerOperand()); + if (isLegalMaskedOp) { + --NumPredStores; + MaskedOp.insert(SI); + continue; + } + return false; + } + } + if (it->mayThrow()) + return false; + + // The instructions below can trap. + switch (it->getOpcode()) { + default: continue; + case Instruction::UDiv: + case Instruction::SDiv: + case Instruction::URem: + case Instruction::SRem: + return false; + } + } + + return true; +} + +void InterleavedAccessInfo::collectConstStridedAccesses( + MapVector<Instruction *, StrideDescriptor> &StrideAccesses, + const ValueToValueMap &Strides) { + // Holds load/store instructions in program order. + SmallVector<Instruction *, 16> AccessList; + + for (auto *BB : TheLoop->getBlocks()) { + bool IsPred = LoopAccessInfo::blockNeedsPredication(BB, TheLoop, DT); + + for (auto &I : *BB) { + if (!isa<LoadInst>(&I) && !isa<StoreInst>(&I)) + continue; + // FIXME: Currently we can't handle mixed accesses and predicated accesses + if (IsPred) + return; + + AccessList.push_back(&I); + } + } + + if (AccessList.empty()) + return; + + auto &DL = TheLoop->getHeader()->getModule()->getDataLayout(); + for (auto I : AccessList) { + LoadInst *LI = dyn_cast<LoadInst>(I); + StoreInst *SI = dyn_cast<StoreInst>(I); + + Value *Ptr = LI ? LI->getPointerOperand() : SI->getPointerOperand(); + int Stride = isStridedPtr(PSE, Ptr, TheLoop, Strides); + + // The factor of the corresponding interleave group. + unsigned Factor = std::abs(Stride); + + // Ignore the access if the factor is too small or too large. + if (Factor < 2 || Factor > MaxInterleaveGroupFactor) + continue; + + const SCEV *Scev = replaceSymbolicStrideSCEV(PSE, Strides, Ptr); + PointerType *PtrTy = dyn_cast<PointerType>(Ptr->getType()); + unsigned Size = DL.getTypeAllocSize(PtrTy->getElementType()); + + // An alignment of 0 means target ABI alignment. + unsigned Align = LI ? LI->getAlignment() : SI->getAlignment(); + if (!Align) + Align = DL.getABITypeAlignment(PtrTy->getElementType()); + + StrideAccesses[I] = StrideDescriptor(Stride, Scev, Size, Align); + } +} + +// Analyze interleaved accesses and collect them into interleave groups. +// +// Notice that the vectorization on interleaved groups will change instruction +// orders and may break dependences. But the memory dependence check guarantees +// that there is no overlap between two pointers of different strides, element +// sizes or underlying bases. +// +// For pointers sharing the same stride, element size and underlying base, no +// need to worry about Read-After-Write dependences and Write-After-Read +// dependences. +// +// E.g. The RAW dependence: A[i] = a; +// b = A[i]; +// This won't exist as it is a store-load forwarding conflict, which has +// already been checked and forbidden in the dependence check. +// +// E.g. The WAR dependence: a = A[i]; // (1) +// A[i] = b; // (2) +// The store group of (2) is always inserted at or below (2), and the load group +// of (1) is always inserted at or above (1). The dependence is safe. +void InterleavedAccessInfo::analyzeInterleaving( + const ValueToValueMap &Strides) { + DEBUG(dbgs() << "LV: Analyzing interleaved accesses...\n"); + + // Holds all the stride accesses. + MapVector<Instruction *, StrideDescriptor> StrideAccesses; + collectConstStridedAccesses(StrideAccesses, Strides); + + if (StrideAccesses.empty()) + return; + + // Holds all interleaved store groups temporarily. + SmallSetVector<InterleaveGroup *, 4> StoreGroups; + + // Search the load-load/write-write pair B-A in bottom-up order and try to + // insert B into the interleave group of A according to 3 rules: + // 1. A and B have the same stride. + // 2. A and B have the same memory object size. + // 3. B belongs to the group according to the distance. + // + // The bottom-up order can avoid breaking the Write-After-Write dependences + // between two pointers of the same base. + // E.g. A[i] = a; (1) + // A[i] = b; (2) + // A[i+1] = c (3) + // We form the group (2)+(3) in front, so (1) has to form groups with accesses + // above (1), which guarantees that (1) is always above (2). + for (auto I = StrideAccesses.rbegin(), E = StrideAccesses.rend(); I != E; + ++I) { + Instruction *A = I->first; + StrideDescriptor DesA = I->second; + + InterleaveGroup *Group = getInterleaveGroup(A); + if (!Group) { + DEBUG(dbgs() << "LV: Creating an interleave group with:" << *A << '\n'); + Group = createInterleaveGroup(A, DesA.Stride, DesA.Align); + } + + if (A->mayWriteToMemory()) + StoreGroups.insert(Group); + + for (auto II = std::next(I); II != E; ++II) { + Instruction *B = II->first; + StrideDescriptor DesB = II->second; + + // Ignore if B is already in a group or B is a different memory operation. + if (isInterleaved(B) || A->mayReadFromMemory() != B->mayReadFromMemory()) + continue; + + // Check the rule 1 and 2. + if (DesB.Stride != DesA.Stride || DesB.Size != DesA.Size) + continue; + + // Calculate the distance and prepare for the rule 3. + const SCEVConstant *DistToA = dyn_cast<SCEVConstant>( + PSE.getSE()->getMinusSCEV(DesB.Scev, DesA.Scev)); + if (!DistToA) + continue; + + int DistanceToA = DistToA->getAPInt().getSExtValue(); + + // Skip if the distance is not multiple of size as they are not in the + // same group. + if (DistanceToA % static_cast<int>(DesA.Size)) + continue; + + // The index of B is the index of A plus the related index to A. + int IndexB = + Group->getIndex(A) + DistanceToA / static_cast<int>(DesA.Size); + + // Try to insert B into the group. + if (Group->insertMember(B, IndexB, DesB.Align)) { + DEBUG(dbgs() << "LV: Inserted:" << *B << '\n' + << " into the interleave group with" << *A << '\n'); + InterleaveGroupMap[B] = Group; + + // Set the first load in program order as the insert position. + if (B->mayReadFromMemory()) + Group->setInsertPos(B); + } + } // Iteration on instruction B + } // Iteration on instruction A + + // Remove interleaved store groups with gaps. + for (InterleaveGroup *Group : StoreGroups) + if (Group->getNumMembers() != Group->getFactor()) + releaseGroup(Group); +} + +LoopVectorizationCostModel::VectorizationFactor +LoopVectorizationCostModel::selectVectorizationFactor(bool OptForSize) { + // Width 1 means no vectorize + VectorizationFactor Factor = { 1U, 0U }; + if (OptForSize && Legal->getRuntimePointerChecking()->Need) { + emitAnalysis(VectorizationReport() << + "runtime pointer checks needed. Enable vectorization of this " + "loop with '#pragma clang loop vectorize(enable)' when " + "compiling with -Os/-Oz"); + DEBUG(dbgs() << + "LV: Aborting. Runtime ptr check is required with -Os/-Oz.\n"); + return Factor; + } + + if (!EnableCondStoresVectorization && Legal->getNumPredStores()) { + emitAnalysis(VectorizationReport() << + "store that is conditionally executed prevents vectorization"); + DEBUG(dbgs() << "LV: No vectorization. There are conditional stores.\n"); + return Factor; + } + + // Find the trip count. + unsigned TC = PSE.getSE()->getSmallConstantTripCount(TheLoop); + DEBUG(dbgs() << "LV: Found trip count: " << TC << '\n'); + + MinBWs = computeMinimumValueSizes(TheLoop->getBlocks(), *DB, &TTI); + unsigned SmallestType, WidestType; + std::tie(SmallestType, WidestType) = getSmallestAndWidestTypes(); + unsigned WidestRegister = TTI.getRegisterBitWidth(true); + unsigned MaxSafeDepDist = -1U; + if (Legal->getMaxSafeDepDistBytes() != -1U) + MaxSafeDepDist = Legal->getMaxSafeDepDistBytes() * 8; + WidestRegister = ((WidestRegister < MaxSafeDepDist) ? + WidestRegister : MaxSafeDepDist); + unsigned MaxVectorSize = WidestRegister / WidestType; + + DEBUG(dbgs() << "LV: The Smallest and Widest types: " << SmallestType << " / " + << WidestType << " bits.\n"); + DEBUG(dbgs() << "LV: The Widest register is: " + << WidestRegister << " bits.\n"); + + if (MaxVectorSize == 0) { + DEBUG(dbgs() << "LV: The target has no vector registers.\n"); + MaxVectorSize = 1; + } + + assert(MaxVectorSize <= 64 && "Did not expect to pack so many elements" + " into one vector!"); + + unsigned VF = MaxVectorSize; + if (MaximizeBandwidth && !OptForSize) { + // Collect all viable vectorization factors. + SmallVector<unsigned, 8> VFs; + unsigned NewMaxVectorSize = WidestRegister / SmallestType; + for (unsigned VS = MaxVectorSize; VS <= NewMaxVectorSize; VS *= 2) + VFs.push_back(VS); + + // For each VF calculate its register usage. + auto RUs = calculateRegisterUsage(VFs); + + // Select the largest VF which doesn't require more registers than existing + // ones. + unsigned TargetNumRegisters = TTI.getNumberOfRegisters(true); + for (int i = RUs.size() - 1; i >= 0; --i) { + if (RUs[i].MaxLocalUsers <= TargetNumRegisters) { + VF = VFs[i]; + break; + } + } + } + + // If we optimize the program for size, avoid creating the tail loop. + if (OptForSize) { + // If we are unable to calculate the trip count then don't try to vectorize. + if (TC < 2) { + emitAnalysis + (VectorizationReport() << + "unable to calculate the loop count due to complex control flow"); + DEBUG(dbgs() << "LV: Aborting. A tail loop is required with -Os/-Oz.\n"); + return Factor; + } + + // Find the maximum SIMD width that can fit within the trip count. + VF = TC % MaxVectorSize; + + if (VF == 0) + VF = MaxVectorSize; + else { + // If the trip count that we found modulo the vectorization factor is not + // zero then we require a tail. + emitAnalysis(VectorizationReport() << + "cannot optimize for size and vectorize at the " + "same time. Enable vectorization of this loop " + "with '#pragma clang loop vectorize(enable)' " + "when compiling with -Os/-Oz"); + DEBUG(dbgs() << "LV: Aborting. A tail loop is required with -Os/-Oz.\n"); + return Factor; + } + } + + int UserVF = Hints->getWidth(); + if (UserVF != 0) { + assert(isPowerOf2_32(UserVF) && "VF needs to be a power of two"); + DEBUG(dbgs() << "LV: Using user VF " << UserVF << ".\n"); + + Factor.Width = UserVF; + return Factor; + } + + float Cost = expectedCost(1); +#ifndef NDEBUG + const float ScalarCost = Cost; +#endif /* NDEBUG */ + unsigned Width = 1; + DEBUG(dbgs() << "LV: Scalar loop costs: " << (int)ScalarCost << ".\n"); + + bool ForceVectorization = Hints->getForce() == LoopVectorizeHints::FK_Enabled; + // Ignore scalar width, because the user explicitly wants vectorization. + if (ForceVectorization && VF > 1) { + Width = 2; + Cost = expectedCost(Width) / (float)Width; + } + + for (unsigned i=2; i <= VF; i*=2) { + // Notice that the vector loop needs to be executed less times, so + // we need to divide the cost of the vector loops by the width of + // the vector elements. + float VectorCost = expectedCost(i) / (float)i; + DEBUG(dbgs() << "LV: Vector loop of width " << i << " costs: " << + (int)VectorCost << ".\n"); + if (VectorCost < Cost) { + Cost = VectorCost; + Width = i; + } + } + + DEBUG(if (ForceVectorization && Width > 1 && Cost >= ScalarCost) dbgs() + << "LV: Vectorization seems to be not beneficial, " + << "but was forced by a user.\n"); + DEBUG(dbgs() << "LV: Selecting VF: "<< Width << ".\n"); + Factor.Width = Width; + Factor.Cost = Width * Cost; + return Factor; +} + +std::pair<unsigned, unsigned> +LoopVectorizationCostModel::getSmallestAndWidestTypes() { + unsigned MinWidth = -1U; + unsigned MaxWidth = 8; + const DataLayout &DL = TheFunction->getParent()->getDataLayout(); + + // For each block. + for (Loop::block_iterator bb = TheLoop->block_begin(), + be = TheLoop->block_end(); bb != be; ++bb) { + BasicBlock *BB = *bb; + + // For each instruction in the loop. + for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) { + Type *T = it->getType(); + + // Skip ignored values. + if (ValuesToIgnore.count(&*it)) + continue; + + // Only examine Loads, Stores and PHINodes. + if (!isa<LoadInst>(it) && !isa<StoreInst>(it) && !isa<PHINode>(it)) + continue; + + // Examine PHI nodes that are reduction variables. Update the type to + // account for the recurrence type. + if (PHINode *PN = dyn_cast<PHINode>(it)) { + if (!Legal->isReductionVariable(PN)) + continue; + RecurrenceDescriptor RdxDesc = (*Legal->getReductionVars())[PN]; + T = RdxDesc.getRecurrenceType(); + } + + // Examine the stored values. + if (StoreInst *ST = dyn_cast<StoreInst>(it)) + T = ST->getValueOperand()->getType(); + + // Ignore loaded pointer types and stored pointer types that are not + // consecutive. However, we do want to take consecutive stores/loads of + // pointer vectors into account. + if (T->isPointerTy() && !isConsecutiveLoadOrStore(&*it)) + continue; + + MinWidth = std::min(MinWidth, + (unsigned)DL.getTypeSizeInBits(T->getScalarType())); + MaxWidth = std::max(MaxWidth, + (unsigned)DL.getTypeSizeInBits(T->getScalarType())); + } + } + + return {MinWidth, MaxWidth}; +} + +unsigned LoopVectorizationCostModel::selectInterleaveCount(bool OptForSize, + unsigned VF, + unsigned LoopCost) { + + // -- The interleave heuristics -- + // We interleave the loop in order to expose ILP and reduce the loop overhead. + // There are many micro-architectural considerations that we can't predict + // at this level. For example, frontend pressure (on decode or fetch) due to + // code size, or the number and capabilities of the execution ports. + // + // We use the following heuristics to select the interleave count: + // 1. If the code has reductions, then we interleave to break the cross + // iteration dependency. + // 2. If the loop is really small, then we interleave to reduce the loop + // overhead. + // 3. We don't interleave if we think that we will spill registers to memory + // due to the increased register pressure. + + // When we optimize for size, we don't interleave. + if (OptForSize) + return 1; + + // We used the distance for the interleave count. + if (Legal->getMaxSafeDepDistBytes() != -1U) + return 1; + + // Do not interleave loops with a relatively small trip count. + unsigned TC = PSE.getSE()->getSmallConstantTripCount(TheLoop); + if (TC > 1 && TC < TinyTripCountInterleaveThreshold) + return 1; + + unsigned TargetNumRegisters = TTI.getNumberOfRegisters(VF > 1); + DEBUG(dbgs() << "LV: The target has " << TargetNumRegisters << + " registers\n"); + + if (VF == 1) { + if (ForceTargetNumScalarRegs.getNumOccurrences() > 0) + TargetNumRegisters = ForceTargetNumScalarRegs; + } else { + if (ForceTargetNumVectorRegs.getNumOccurrences() > 0) + TargetNumRegisters = ForceTargetNumVectorRegs; + } + + RegisterUsage R = calculateRegisterUsage({VF})[0]; + // We divide by these constants so assume that we have at least one + // instruction that uses at least one register. + R.MaxLocalUsers = std::max(R.MaxLocalUsers, 1U); + R.NumInstructions = std::max(R.NumInstructions, 1U); + + // We calculate the interleave count using the following formula. + // Subtract the number of loop invariants from the number of available + // registers. These registers are used by all of the interleaved instances. + // Next, divide the remaining registers by the number of registers that is + // required by the loop, in order to estimate how many parallel instances + // fit without causing spills. All of this is rounded down if necessary to be + // a power of two. We want power of two interleave count to simplify any + // addressing operations or alignment considerations. + unsigned IC = PowerOf2Floor((TargetNumRegisters - R.LoopInvariantRegs) / + R.MaxLocalUsers); + + // Don't count the induction variable as interleaved. + if (EnableIndVarRegisterHeur) + IC = PowerOf2Floor((TargetNumRegisters - R.LoopInvariantRegs - 1) / + std::max(1U, (R.MaxLocalUsers - 1))); + + // Clamp the interleave ranges to reasonable counts. + unsigned MaxInterleaveCount = TTI.getMaxInterleaveFactor(VF); + + // Check if the user has overridden the max. + if (VF == 1) { + if (ForceTargetMaxScalarInterleaveFactor.getNumOccurrences() > 0) + MaxInterleaveCount = ForceTargetMaxScalarInterleaveFactor; + } else { + if (ForceTargetMaxVectorInterleaveFactor.getNumOccurrences() > 0) + MaxInterleaveCount = ForceTargetMaxVectorInterleaveFactor; + } + + // If we did not calculate the cost for VF (because the user selected the VF) + // then we calculate the cost of VF here. + if (LoopCost == 0) + LoopCost = expectedCost(VF); + + // Clamp the calculated IC to be between the 1 and the max interleave count + // that the target allows. + if (IC > MaxInterleaveCount) + IC = MaxInterleaveCount; + else if (IC < 1) + IC = 1; + + // Interleave if we vectorized this loop and there is a reduction that could + // benefit from interleaving. + if (VF > 1 && Legal->getReductionVars()->size()) { + DEBUG(dbgs() << "LV: Interleaving because of reductions.\n"); + return IC; + } + + // Note that if we've already vectorized the loop we will have done the + // runtime check and so interleaving won't require further checks. + bool InterleavingRequiresRuntimePointerCheck = + (VF == 1 && Legal->getRuntimePointerChecking()->Need); + + // We want to interleave small loops in order to reduce the loop overhead and + // potentially expose ILP opportunities. + DEBUG(dbgs() << "LV: Loop cost is " << LoopCost << '\n'); + if (!InterleavingRequiresRuntimePointerCheck && LoopCost < SmallLoopCost) { + // We assume that the cost overhead is 1 and we use the cost model + // to estimate the cost of the loop and interleave until the cost of the + // loop overhead is about 5% of the cost of the loop. + unsigned SmallIC = + std::min(IC, (unsigned)PowerOf2Floor(SmallLoopCost / LoopCost)); + + // Interleave until store/load ports (estimated by max interleave count) are + // saturated. + unsigned NumStores = Legal->getNumStores(); + unsigned NumLoads = Legal->getNumLoads(); + unsigned StoresIC = IC / (NumStores ? NumStores : 1); + unsigned LoadsIC = IC / (NumLoads ? NumLoads : 1); + + // If we have a scalar reduction (vector reductions are already dealt with + // by this point), we can increase the critical path length if the loop + // we're interleaving is inside another loop. Limit, by default to 2, so the + // critical path only gets increased by one reduction operation. + if (Legal->getReductionVars()->size() && + TheLoop->getLoopDepth() > 1) { + unsigned F = static_cast<unsigned>(MaxNestedScalarReductionIC); + SmallIC = std::min(SmallIC, F); + StoresIC = std::min(StoresIC, F); + LoadsIC = std::min(LoadsIC, F); + } + + if (EnableLoadStoreRuntimeInterleave && + std::max(StoresIC, LoadsIC) > SmallIC) { + DEBUG(dbgs() << "LV: Interleaving to saturate store or load ports.\n"); + return std::max(StoresIC, LoadsIC); + } + + DEBUG(dbgs() << "LV: Interleaving to reduce branch cost.\n"); + return SmallIC; + } + + // Interleave if this is a large loop (small loops are already dealt with by + // this point) that could benefit from interleaving. + bool HasReductions = (Legal->getReductionVars()->size() > 0); + if (TTI.enableAggressiveInterleaving(HasReductions)) { + DEBUG(dbgs() << "LV: Interleaving to expose ILP.\n"); + return IC; + } + + DEBUG(dbgs() << "LV: Not Interleaving.\n"); + return 1; +} + +SmallVector<LoopVectorizationCostModel::RegisterUsage, 8> +LoopVectorizationCostModel::calculateRegisterUsage( + const SmallVector<unsigned, 8> &VFs) { + // This function calculates the register usage by measuring the highest number + // of values that are alive at a single location. Obviously, this is a very + // rough estimation. We scan the loop in a topological order in order and + // assign a number to each instruction. We use RPO to ensure that defs are + // met before their users. We assume that each instruction that has in-loop + // users starts an interval. We record every time that an in-loop value is + // used, so we have a list of the first and last occurrences of each + // instruction. Next, we transpose this data structure into a multi map that + // holds the list of intervals that *end* at a specific location. This multi + // map allows us to perform a linear search. We scan the instructions linearly + // and record each time that a new interval starts, by placing it in a set. + // If we find this value in the multi-map then we remove it from the set. + // The max register usage is the maximum size of the set. + // We also search for instructions that are defined outside the loop, but are + // used inside the loop. We need this number separately from the max-interval + // usage number because when we unroll, loop-invariant values do not take + // more register. + LoopBlocksDFS DFS(TheLoop); + DFS.perform(LI); + + RegisterUsage RU; + RU.NumInstructions = 0; + + // Each 'key' in the map opens a new interval. The values + // of the map are the index of the 'last seen' usage of the + // instruction that is the key. + typedef DenseMap<Instruction*, unsigned> IntervalMap; + // Maps instruction to its index. + DenseMap<unsigned, Instruction*> IdxToInstr; + // Marks the end of each interval. + IntervalMap EndPoint; + // Saves the list of instruction indices that are used in the loop. + SmallSet<Instruction*, 8> Ends; + // Saves the list of values that are used in the loop but are + // defined outside the loop, such as arguments and constants. + SmallPtrSet<Value*, 8> LoopInvariants; + + unsigned Index = 0; + for (LoopBlocksDFS::RPOIterator bb = DFS.beginRPO(), + be = DFS.endRPO(); bb != be; ++bb) { + RU.NumInstructions += (*bb)->size(); + for (Instruction &I : **bb) { + IdxToInstr[Index++] = &I; + + // Save the end location of each USE. + for (unsigned i = 0; i < I.getNumOperands(); ++i) { + Value *U = I.getOperand(i); + Instruction *Instr = dyn_cast<Instruction>(U); + + // Ignore non-instruction values such as arguments, constants, etc. + if (!Instr) continue; + + // If this instruction is outside the loop then record it and continue. + if (!TheLoop->contains(Instr)) { + LoopInvariants.insert(Instr); + continue; + } + + // Overwrite previous end points. + EndPoint[Instr] = Index; + Ends.insert(Instr); + } + } + } + + // Saves the list of intervals that end with the index in 'key'. + typedef SmallVector<Instruction*, 2> InstrList; + DenseMap<unsigned, InstrList> TransposeEnds; + + // Transpose the EndPoints to a list of values that end at each index. + for (IntervalMap::iterator it = EndPoint.begin(), e = EndPoint.end(); + it != e; ++it) + TransposeEnds[it->second].push_back(it->first); + + SmallSet<Instruction*, 8> OpenIntervals; + + // Get the size of the widest register. + unsigned MaxSafeDepDist = -1U; + if (Legal->getMaxSafeDepDistBytes() != -1U) + MaxSafeDepDist = Legal->getMaxSafeDepDistBytes() * 8; + unsigned WidestRegister = + std::min(TTI.getRegisterBitWidth(true), MaxSafeDepDist); + const DataLayout &DL = TheFunction->getParent()->getDataLayout(); + + SmallVector<RegisterUsage, 8> RUs(VFs.size()); + SmallVector<unsigned, 8> MaxUsages(VFs.size(), 0); + + DEBUG(dbgs() << "LV(REG): Calculating max register usage:\n"); + + // A lambda that gets the register usage for the given type and VF. + auto GetRegUsage = [&DL, WidestRegister](Type *Ty, unsigned VF) { + unsigned TypeSize = DL.getTypeSizeInBits(Ty->getScalarType()); + return std::max<unsigned>(1, VF * TypeSize / WidestRegister); + }; + + for (unsigned int i = 0; i < Index; ++i) { + Instruction *I = IdxToInstr[i]; + // Ignore instructions that are never used within the loop. + if (!Ends.count(I)) continue; + + // Remove all of the instructions that end at this location. + InstrList &List = TransposeEnds[i]; + for (unsigned int j = 0, e = List.size(); j < e; ++j) + OpenIntervals.erase(List[j]); + + // Skip ignored values. + if (ValuesToIgnore.count(I)) + continue; + + // For each VF find the maximum usage of registers. + for (unsigned j = 0, e = VFs.size(); j < e; ++j) { + if (VFs[j] == 1) { + MaxUsages[j] = std::max(MaxUsages[j], OpenIntervals.size()); + continue; + } + + // Count the number of live intervals. + unsigned RegUsage = 0; + for (auto Inst : OpenIntervals) { + // Skip ignored values for VF > 1. + if (VecValuesToIgnore.count(Inst)) + continue; + RegUsage += GetRegUsage(Inst->getType(), VFs[j]); + } + MaxUsages[j] = std::max(MaxUsages[j], RegUsage); + } + + DEBUG(dbgs() << "LV(REG): At #" << i << " Interval # " + << OpenIntervals.size() << '\n'); + + // Add the current instruction to the list of open intervals. + OpenIntervals.insert(I); + } + + for (unsigned i = 0, e = VFs.size(); i < e; ++i) { + unsigned Invariant = 0; + if (VFs[i] == 1) + Invariant = LoopInvariants.size(); + else { + for (auto Inst : LoopInvariants) + Invariant += GetRegUsage(Inst->getType(), VFs[i]); + } + + DEBUG(dbgs() << "LV(REG): VF = " << VFs[i] << '\n'); + DEBUG(dbgs() << "LV(REG): Found max usage: " << MaxUsages[i] << '\n'); + DEBUG(dbgs() << "LV(REG): Found invariant usage: " << Invariant << '\n'); + DEBUG(dbgs() << "LV(REG): LoopSize: " << RU.NumInstructions << '\n'); + + RU.LoopInvariantRegs = Invariant; + RU.MaxLocalUsers = MaxUsages[i]; + RUs[i] = RU; + } + + return RUs; +} + +unsigned LoopVectorizationCostModel::expectedCost(unsigned VF) { + unsigned Cost = 0; + + // For each block. + for (Loop::block_iterator bb = TheLoop->block_begin(), + be = TheLoop->block_end(); bb != be; ++bb) { + unsigned BlockCost = 0; + BasicBlock *BB = *bb; + + // For each instruction in the old loop. + for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) { + // Skip dbg intrinsics. + if (isa<DbgInfoIntrinsic>(it)) + continue; + + // Skip ignored values. + if (ValuesToIgnore.count(&*it)) + continue; + + unsigned C = getInstructionCost(&*it, VF); + + // Check if we should override the cost. + if (ForceTargetInstructionCost.getNumOccurrences() > 0) + C = ForceTargetInstructionCost; + + BlockCost += C; + DEBUG(dbgs() << "LV: Found an estimated cost of " << C << " for VF " << + VF << " For instruction: " << *it << '\n'); + } + + // We assume that if-converted blocks have a 50% chance of being executed. + // When the code is scalar then some of the blocks are avoided due to CF. + // When the code is vectorized we execute all code paths. + if (VF == 1 && Legal->blockNeedsPredication(*bb)) + BlockCost /= 2; + + Cost += BlockCost; + } + + return Cost; +} + +/// \brief Check whether the address computation for a non-consecutive memory +/// access looks like an unlikely candidate for being merged into the indexing +/// mode. +/// +/// We look for a GEP which has one index that is an induction variable and all +/// other indices are loop invariant. If the stride of this access is also +/// within a small bound we decide that this address computation can likely be +/// merged into the addressing mode. +/// In all other cases, we identify the address computation as complex. +static bool isLikelyComplexAddressComputation(Value *Ptr, + LoopVectorizationLegality *Legal, + ScalarEvolution *SE, + const Loop *TheLoop) { + GetElementPtrInst *Gep = dyn_cast<GetElementPtrInst>(Ptr); + if (!Gep) + return true; + + // We are looking for a gep with all loop invariant indices except for one + // which should be an induction variable. + unsigned NumOperands = Gep->getNumOperands(); + for (unsigned i = 1; i < NumOperands; ++i) { + Value *Opd = Gep->getOperand(i); + if (!SE->isLoopInvariant(SE->getSCEV(Opd), TheLoop) && + !Legal->isInductionVariable(Opd)) + return true; + } + + // Now we know we have a GEP ptr, %inv, %ind, %inv. Make sure that the step + // can likely be merged into the address computation. + unsigned MaxMergeDistance = 64; + + const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(SE->getSCEV(Ptr)); + if (!AddRec) + return true; + + // Check the step is constant. + const SCEV *Step = AddRec->getStepRecurrence(*SE); + // Calculate the pointer stride and check if it is consecutive. + const SCEVConstant *C = dyn_cast<SCEVConstant>(Step); + if (!C) + return true; + + const APInt &APStepVal = C->getAPInt(); + + // Huge step value - give up. + if (APStepVal.getBitWidth() > 64) + return true; + + int64_t StepVal = APStepVal.getSExtValue(); + + return StepVal > MaxMergeDistance; +} + +static bool isStrideMul(Instruction *I, LoopVectorizationLegality *Legal) { + return Legal->hasStride(I->getOperand(0)) || + Legal->hasStride(I->getOperand(1)); +} + +unsigned +LoopVectorizationCostModel::getInstructionCost(Instruction *I, unsigned VF) { + // If we know that this instruction will remain uniform, check the cost of + // the scalar version. + if (Legal->isUniformAfterVectorization(I)) + VF = 1; + + Type *RetTy = I->getType(); + if (VF > 1 && MinBWs.count(I)) + RetTy = IntegerType::get(RetTy->getContext(), MinBWs[I]); + Type *VectorTy = ToVectorTy(RetTy, VF); + auto SE = PSE.getSE(); + + // TODO: We need to estimate the cost of intrinsic calls. + switch (I->getOpcode()) { + case Instruction::GetElementPtr: + // We mark this instruction as zero-cost because the cost of GEPs in + // vectorized code depends on whether the corresponding memory instruction + // is scalarized or not. Therefore, we handle GEPs with the memory + // instruction cost. + return 0; + case Instruction::Br: { + return TTI.getCFInstrCost(I->getOpcode()); + } + case Instruction::PHI: + //TODO: IF-converted IFs become selects. + return 0; + case Instruction::Add: + case Instruction::FAdd: + case Instruction::Sub: + case Instruction::FSub: + case Instruction::Mul: + case Instruction::FMul: + case Instruction::UDiv: + case Instruction::SDiv: + case Instruction::FDiv: + case Instruction::URem: + case Instruction::SRem: + case Instruction::FRem: + case Instruction::Shl: + case Instruction::LShr: + case Instruction::AShr: + case Instruction::And: + case Instruction::Or: + case Instruction::Xor: { + // Since we will replace the stride by 1 the multiplication should go away. + if (I->getOpcode() == Instruction::Mul && isStrideMul(I, Legal)) + return 0; + // Certain instructions can be cheaper to vectorize if they have a constant + // second vector operand. One example of this are shifts on x86. + TargetTransformInfo::OperandValueKind Op1VK = + TargetTransformInfo::OK_AnyValue; + TargetTransformInfo::OperandValueKind Op2VK = + TargetTransformInfo::OK_AnyValue; + TargetTransformInfo::OperandValueProperties Op1VP = + TargetTransformInfo::OP_None; + TargetTransformInfo::OperandValueProperties Op2VP = + TargetTransformInfo::OP_None; + Value *Op2 = I->getOperand(1); + + // Check for a splat of a constant or for a non uniform vector of constants. + if (isa<ConstantInt>(Op2)) { + ConstantInt *CInt = cast<ConstantInt>(Op2); + if (CInt && CInt->getValue().isPowerOf2()) + Op2VP = TargetTransformInfo::OP_PowerOf2; + Op2VK = TargetTransformInfo::OK_UniformConstantValue; + } else if (isa<ConstantVector>(Op2) || isa<ConstantDataVector>(Op2)) { + Op2VK = TargetTransformInfo::OK_NonUniformConstantValue; + Constant *SplatValue = cast<Constant>(Op2)->getSplatValue(); + if (SplatValue) { + ConstantInt *CInt = dyn_cast<ConstantInt>(SplatValue); + if (CInt && CInt->getValue().isPowerOf2()) + Op2VP = TargetTransformInfo::OP_PowerOf2; + Op2VK = TargetTransformInfo::OK_UniformConstantValue; + } + } + + return TTI.getArithmeticInstrCost(I->getOpcode(), VectorTy, Op1VK, Op2VK, + Op1VP, Op2VP); + } + case Instruction::Select: { + SelectInst *SI = cast<SelectInst>(I); + const SCEV *CondSCEV = SE->getSCEV(SI->getCondition()); + bool ScalarCond = (SE->isLoopInvariant(CondSCEV, TheLoop)); + Type *CondTy = SI->getCondition()->getType(); + if (!ScalarCond) + CondTy = VectorType::get(CondTy, VF); + + return TTI.getCmpSelInstrCost(I->getOpcode(), VectorTy, CondTy); + } + case Instruction::ICmp: + case Instruction::FCmp: { + Type *ValTy = I->getOperand(0)->getType(); + Instruction *Op0AsInstruction = dyn_cast<Instruction>(I->getOperand(0)); + auto It = MinBWs.find(Op0AsInstruction); + if (VF > 1 && It != MinBWs.end()) + ValTy = IntegerType::get(ValTy->getContext(), It->second); + VectorTy = ToVectorTy(ValTy, VF); + return TTI.getCmpSelInstrCost(I->getOpcode(), VectorTy); + } + case Instruction::Store: + case Instruction::Load: { + StoreInst *SI = dyn_cast<StoreInst>(I); + LoadInst *LI = dyn_cast<LoadInst>(I); + Type *ValTy = (SI ? SI->getValueOperand()->getType() : + LI->getType()); + VectorTy = ToVectorTy(ValTy, VF); + + unsigned Alignment = SI ? SI->getAlignment() : LI->getAlignment(); + unsigned AS = SI ? SI->getPointerAddressSpace() : + LI->getPointerAddressSpace(); + Value *Ptr = SI ? SI->getPointerOperand() : LI->getPointerOperand(); + // We add the cost of address computation here instead of with the gep + // instruction because only here we know whether the operation is + // scalarized. + if (VF == 1) + return TTI.getAddressComputationCost(VectorTy) + + TTI.getMemoryOpCost(I->getOpcode(), VectorTy, Alignment, AS); + + // For an interleaved access, calculate the total cost of the whole + // interleave group. + if (Legal->isAccessInterleaved(I)) { + auto Group = Legal->getInterleavedAccessGroup(I); + assert(Group && "Fail to get an interleaved access group."); + + // Only calculate the cost once at the insert position. + if (Group->getInsertPos() != I) + return 0; + + unsigned InterleaveFactor = Group->getFactor(); + Type *WideVecTy = + VectorType::get(VectorTy->getVectorElementType(), + VectorTy->getVectorNumElements() * InterleaveFactor); + + // Holds the indices of existing members in an interleaved load group. + // An interleaved store group doesn't need this as it dones't allow gaps. + SmallVector<unsigned, 4> Indices; + if (LI) { + for (unsigned i = 0; i < InterleaveFactor; i++) + if (Group->getMember(i)) + Indices.push_back(i); + } + + // Calculate the cost of the whole interleaved group. + unsigned Cost = TTI.getInterleavedMemoryOpCost( + I->getOpcode(), WideVecTy, Group->getFactor(), Indices, + Group->getAlignment(), AS); + + if (Group->isReverse()) + Cost += + Group->getNumMembers() * + TTI.getShuffleCost(TargetTransformInfo::SK_Reverse, VectorTy, 0); + + // FIXME: The interleaved load group with a huge gap could be even more + // expensive than scalar operations. Then we could ignore such group and + // use scalar operations instead. + return Cost; + } + + // Scalarized loads/stores. + int ConsecutiveStride = Legal->isConsecutivePtr(Ptr); + bool Reverse = ConsecutiveStride < 0; + const DataLayout &DL = I->getModule()->getDataLayout(); + unsigned ScalarAllocatedSize = DL.getTypeAllocSize(ValTy); + unsigned VectorElementSize = DL.getTypeStoreSize(VectorTy) / VF; + if (!ConsecutiveStride || ScalarAllocatedSize != VectorElementSize) { + bool IsComplexComputation = + isLikelyComplexAddressComputation(Ptr, Legal, SE, TheLoop); + unsigned Cost = 0; + // The cost of extracting from the value vector and pointer vector. + Type *PtrTy = ToVectorTy(Ptr->getType(), VF); + for (unsigned i = 0; i < VF; ++i) { + // The cost of extracting the pointer operand. + Cost += TTI.getVectorInstrCost(Instruction::ExtractElement, PtrTy, i); + // In case of STORE, the cost of ExtractElement from the vector. + // In case of LOAD, the cost of InsertElement into the returned + // vector. + Cost += TTI.getVectorInstrCost(SI ? Instruction::ExtractElement : + Instruction::InsertElement, + VectorTy, i); + } + + // The cost of the scalar loads/stores. + Cost += VF * TTI.getAddressComputationCost(PtrTy, IsComplexComputation); + Cost += VF * TTI.getMemoryOpCost(I->getOpcode(), ValTy->getScalarType(), + Alignment, AS); + return Cost; + } + + // Wide load/stores. + unsigned Cost = TTI.getAddressComputationCost(VectorTy); + if (Legal->isMaskRequired(I)) + Cost += TTI.getMaskedMemoryOpCost(I->getOpcode(), VectorTy, Alignment, + AS); + else + Cost += TTI.getMemoryOpCost(I->getOpcode(), VectorTy, Alignment, AS); + + if (Reverse) + Cost += TTI.getShuffleCost(TargetTransformInfo::SK_Reverse, + VectorTy, 0); + return Cost; + } + case Instruction::ZExt: + case Instruction::SExt: + case Instruction::FPToUI: + case Instruction::FPToSI: + case Instruction::FPExt: + case Instruction::PtrToInt: + case Instruction::IntToPtr: + case Instruction::SIToFP: + case Instruction::UIToFP: + case Instruction::Trunc: + case Instruction::FPTrunc: + case Instruction::BitCast: { + // We optimize the truncation of induction variable. + // The cost of these is the same as the scalar operation. + if (I->getOpcode() == Instruction::Trunc && + Legal->isInductionVariable(I->getOperand(0))) + return TTI.getCastInstrCost(I->getOpcode(), I->getType(), + I->getOperand(0)->getType()); + + Type *SrcScalarTy = I->getOperand(0)->getType(); + Type *SrcVecTy = ToVectorTy(SrcScalarTy, VF); + if (VF > 1 && MinBWs.count(I)) { + // This cast is going to be shrunk. This may remove the cast or it might + // turn it into slightly different cast. For example, if MinBW == 16, + // "zext i8 %1 to i32" becomes "zext i8 %1 to i16". + // + // Calculate the modified src and dest types. + Type *MinVecTy = VectorTy; + if (I->getOpcode() == Instruction::Trunc) { + SrcVecTy = smallestIntegerVectorType(SrcVecTy, MinVecTy); + VectorTy = largestIntegerVectorType(ToVectorTy(I->getType(), VF), + MinVecTy); + } else if (I->getOpcode() == Instruction::ZExt || + I->getOpcode() == Instruction::SExt) { + SrcVecTy = largestIntegerVectorType(SrcVecTy, MinVecTy); + VectorTy = smallestIntegerVectorType(ToVectorTy(I->getType(), VF), + MinVecTy); + } + } + + return TTI.getCastInstrCost(I->getOpcode(), VectorTy, SrcVecTy); + } + case Instruction::Call: { + bool NeedToScalarize; + CallInst *CI = cast<CallInst>(I); + unsigned CallCost = getVectorCallCost(CI, VF, TTI, TLI, NeedToScalarize); + if (getIntrinsicIDForCall(CI, TLI)) + return std::min(CallCost, getVectorIntrinsicCost(CI, VF, TTI, TLI)); + return CallCost; + } + default: { + // We are scalarizing the instruction. Return the cost of the scalar + // instruction, plus the cost of insert and extract into vector + // elements, times the vector width. + unsigned Cost = 0; + + if (!RetTy->isVoidTy() && VF != 1) { + unsigned InsCost = TTI.getVectorInstrCost(Instruction::InsertElement, + VectorTy); + unsigned ExtCost = TTI.getVectorInstrCost(Instruction::ExtractElement, + VectorTy); + + // The cost of inserting the results plus extracting each one of the + // operands. + Cost += VF * (InsCost + ExtCost * I->getNumOperands()); + } + + // The cost of executing VF copies of the scalar instruction. This opcode + // is unknown. Assume that it is the same as 'mul'. + Cost += VF * TTI.getArithmeticInstrCost(Instruction::Mul, VectorTy); + return Cost; + } + }// end of switch. +} + +char LoopVectorize::ID = 0; +static const char lv_name[] = "Loop Vectorization"; +INITIALIZE_PASS_BEGIN(LoopVectorize, LV_NAME, lv_name, false, false) +INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass) +INITIALIZE_PASS_DEPENDENCY(BasicAAWrapperPass) +INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass) +INITIALIZE_PASS_DEPENDENCY(GlobalsAAWrapperPass) +INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker) +INITIALIZE_PASS_DEPENDENCY(BlockFrequencyInfoWrapperPass) +INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass) +INITIALIZE_PASS_DEPENDENCY(ScalarEvolutionWrapperPass) +INITIALIZE_PASS_DEPENDENCY(LCSSA) +INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass) +INITIALIZE_PASS_DEPENDENCY(LoopSimplify) +INITIALIZE_PASS_DEPENDENCY(LoopAccessAnalysis) +INITIALIZE_PASS_DEPENDENCY(DemandedBits) +INITIALIZE_PASS_END(LoopVectorize, LV_NAME, lv_name, false, false) + +namespace llvm { + Pass *createLoopVectorizePass(bool NoUnrolling, bool AlwaysVectorize) { + return new LoopVectorize(NoUnrolling, AlwaysVectorize); + } +} + +bool LoopVectorizationCostModel::isConsecutiveLoadOrStore(Instruction *Inst) { + // Check for a store. + if (StoreInst *ST = dyn_cast<StoreInst>(Inst)) + return Legal->isConsecutivePtr(ST->getPointerOperand()) != 0; + + // Check for a load. + if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) + return Legal->isConsecutivePtr(LI->getPointerOperand()) != 0; + + return false; +} + +void LoopVectorizationCostModel::collectValuesToIgnore() { + // Ignore ephemeral values. + CodeMetrics::collectEphemeralValues(TheLoop, AC, ValuesToIgnore); + + // Ignore type-promoting instructions we identified during reduction + // detection. + for (auto &Reduction : *Legal->getReductionVars()) { + RecurrenceDescriptor &RedDes = Reduction.second; + SmallPtrSetImpl<Instruction *> &Casts = RedDes.getCastInsts(); + VecValuesToIgnore.insert(Casts.begin(), Casts.end()); + } + + // Ignore induction phis that are only used in either GetElementPtr or ICmp + // instruction to exit loop. Induction variables usually have large types and + // can have big impact when estimating register usage. + // This is for when VF > 1. + for (auto &Induction : *Legal->getInductionVars()) { + auto *PN = Induction.first; + auto *UpdateV = PN->getIncomingValueForBlock(TheLoop->getLoopLatch()); + + // Check that the PHI is only used by the induction increment (UpdateV) or + // by GEPs. Then check that UpdateV is only used by a compare instruction or + // the loop header PHI. + // FIXME: Need precise def-use analysis to determine if this instruction + // variable will be vectorized. + if (std::all_of(PN->user_begin(), PN->user_end(), + [&](const User *U) -> bool { + return U == UpdateV || isa<GetElementPtrInst>(U); + }) && + std::all_of(UpdateV->user_begin(), UpdateV->user_end(), + [&](const User *U) -> bool { + return U == PN || isa<ICmpInst>(U); + })) { + VecValuesToIgnore.insert(PN); + VecValuesToIgnore.insert(UpdateV); + } + } + + // Ignore instructions that will not be vectorized. + // This is for when VF > 1. + for (auto bb = TheLoop->block_begin(), be = TheLoop->block_end(); bb != be; + ++bb) { + for (auto &Inst : **bb) { + switch (Inst.getOpcode()) { + case Instruction::GetElementPtr: { + // Ignore GEP if its last operand is an induction variable so that it is + // a consecutive load/store and won't be vectorized as scatter/gather + // pattern. + + GetElementPtrInst *Gep = cast<GetElementPtrInst>(&Inst); + unsigned NumOperands = Gep->getNumOperands(); + unsigned InductionOperand = getGEPInductionOperand(Gep); + bool GepToIgnore = true; + + // Check that all of the gep indices are uniform except for the + // induction operand. + for (unsigned i = 0; i != NumOperands; ++i) { + if (i != InductionOperand && + !PSE.getSE()->isLoopInvariant(PSE.getSCEV(Gep->getOperand(i)), + TheLoop)) { + GepToIgnore = false; + break; + } + } + + if (GepToIgnore) + VecValuesToIgnore.insert(&Inst); + break; + } + } + } + } +} + +void InnerLoopUnroller::scalarizeInstruction(Instruction *Instr, + bool IfPredicateStore) { + assert(!Instr->getType()->isAggregateType() && "Can't handle vectors"); + // Holds vector parameters or scalars, in case of uniform vals. + SmallVector<VectorParts, 4> Params; + + setDebugLocFromInst(Builder, Instr); + + // Find all of the vectorized parameters. + for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) { + Value *SrcOp = Instr->getOperand(op); + + // If we are accessing the old induction variable, use the new one. + if (SrcOp == OldInduction) { + Params.push_back(getVectorValue(SrcOp)); + continue; + } + + // Try using previously calculated values. + Instruction *SrcInst = dyn_cast<Instruction>(SrcOp); + + // If the src is an instruction that appeared earlier in the basic block + // then it should already be vectorized. + if (SrcInst && OrigLoop->contains(SrcInst)) { + assert(WidenMap.has(SrcInst) && "Source operand is unavailable"); + // The parameter is a vector value from earlier. + Params.push_back(WidenMap.get(SrcInst)); + } else { + // The parameter is a scalar from outside the loop. Maybe even a constant. + VectorParts Scalars; + Scalars.append(UF, SrcOp); + Params.push_back(Scalars); + } + } + + assert(Params.size() == Instr->getNumOperands() && + "Invalid number of operands"); + + // Does this instruction return a value ? + bool IsVoidRetTy = Instr->getType()->isVoidTy(); + + Value *UndefVec = IsVoidRetTy ? nullptr : + UndefValue::get(Instr->getType()); + // Create a new entry in the WidenMap and initialize it to Undef or Null. + VectorParts &VecResults = WidenMap.splat(Instr, UndefVec); + + VectorParts Cond; + if (IfPredicateStore) { + assert(Instr->getParent()->getSinglePredecessor() && + "Only support single predecessor blocks"); + Cond = createEdgeMask(Instr->getParent()->getSinglePredecessor(), + Instr->getParent()); + } + + // For each vector unroll 'part': + for (unsigned Part = 0; Part < UF; ++Part) { + // For each scalar that we create: + + // Start an "if (pred) a[i] = ..." block. + Value *Cmp = nullptr; + if (IfPredicateStore) { + if (Cond[Part]->getType()->isVectorTy()) + Cond[Part] = + Builder.CreateExtractElement(Cond[Part], Builder.getInt32(0)); + Cmp = Builder.CreateICmp(ICmpInst::ICMP_EQ, Cond[Part], + ConstantInt::get(Cond[Part]->getType(), 1)); + } + + Instruction *Cloned = Instr->clone(); + if (!IsVoidRetTy) + Cloned->setName(Instr->getName() + ".cloned"); + // Replace the operands of the cloned instructions with extracted scalars. + for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) { + Value *Op = Params[op][Part]; + Cloned->setOperand(op, Op); + } + + // Place the cloned scalar in the new loop. + Builder.Insert(Cloned); + + // If the original scalar returns a value we need to place it in a vector + // so that future users will be able to use it. + if (!IsVoidRetTy) + VecResults[Part] = Cloned; + + // End if-block. + if (IfPredicateStore) + PredicatedStores.push_back(std::make_pair(cast<StoreInst>(Cloned), + Cmp)); + } +} + +void InnerLoopUnroller::vectorizeMemoryInstruction(Instruction *Instr) { + StoreInst *SI = dyn_cast<StoreInst>(Instr); + bool IfPredicateStore = (SI && Legal->blockNeedsPredication(SI->getParent())); + + return scalarizeInstruction(Instr, IfPredicateStore); +} + +Value *InnerLoopUnroller::reverseVector(Value *Vec) { + return Vec; +} + +Value *InnerLoopUnroller::getBroadcastInstrs(Value *V) { + return V; +} + +Value *InnerLoopUnroller::getStepVector(Value *Val, int StartIdx, Value *Step) { + // When unrolling and the VF is 1, we only need to add a simple scalar. + Type *ITy = Val->getType(); + assert(!ITy->isVectorTy() && "Val must be a scalar"); + Constant *C = ConstantInt::get(ITy, StartIdx); + return Builder.CreateAdd(Val, Builder.CreateMul(C, Step), "induction"); +} diff --git a/contrib/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/contrib/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp new file mode 100644 index 0000000..27d3337 --- /dev/null +++ b/contrib/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp @@ -0,0 +1,4222 @@ +//===- SLPVectorizer.cpp - A bottom up SLP Vectorizer ---------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// This pass implements the Bottom Up SLP vectorizer. It detects consecutive +// stores that can be put together into vector-stores. Next, it attempts to +// construct vectorizable tree using the use-def chains. If a profitable tree +// was found, the SLP vectorizer performs vectorization on the tree. +// +// The pass is inspired by the work described in the paper: +// "Loop-Aware SLP in GCC" by Ira Rosen, Dorit Nuzman, Ayal Zaks. +// +//===----------------------------------------------------------------------===// +#include "llvm/Transforms/Vectorize.h" +#include "llvm/ADT/MapVector.h" +#include "llvm/ADT/Optional.h" +#include "llvm/ADT/PostOrderIterator.h" +#include "llvm/ADT/SetVector.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/Analysis/AliasAnalysis.h" +#include "llvm/Analysis/GlobalsModRef.h" +#include "llvm/Analysis/AssumptionCache.h" +#include "llvm/Analysis/CodeMetrics.h" +#include "llvm/Analysis/LoopInfo.h" +#include "llvm/Analysis/ScalarEvolution.h" +#include "llvm/Analysis/ScalarEvolutionExpressions.h" +#include "llvm/Analysis/TargetTransformInfo.h" +#include "llvm/Analysis/ValueTracking.h" +#include "llvm/IR/DataLayout.h" +#include "llvm/IR/Dominators.h" +#include "llvm/IR/IRBuilder.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/Module.h" +#include "llvm/IR/NoFolder.h" +#include "llvm/IR/Type.h" +#include "llvm/IR/Value.h" +#include "llvm/IR/Verifier.h" +#include "llvm/Pass.h" +#include "llvm/Support/CommandLine.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/raw_ostream.h" +#include "llvm/Analysis/VectorUtils.h" +#include <algorithm> +#include <map> +#include <memory> + +using namespace llvm; + +#define SV_NAME "slp-vectorizer" +#define DEBUG_TYPE "SLP" + +STATISTIC(NumVectorInstructions, "Number of vector instructions generated"); + +static cl::opt<int> + SLPCostThreshold("slp-threshold", cl::init(0), cl::Hidden, + cl::desc("Only vectorize if you gain more than this " + "number ")); + +static cl::opt<bool> +ShouldVectorizeHor("slp-vectorize-hor", cl::init(true), cl::Hidden, + cl::desc("Attempt to vectorize horizontal reductions")); + +static cl::opt<bool> ShouldStartVectorizeHorAtStore( + "slp-vectorize-hor-store", cl::init(false), cl::Hidden, + cl::desc( + "Attempt to vectorize horizontal reductions feeding into a store")); + +static cl::opt<int> +MaxVectorRegSizeOption("slp-max-reg-size", cl::init(128), cl::Hidden, + cl::desc("Attempt to vectorize for this register size in bits")); + +/// Limits the size of scheduling regions in a block. +/// It avoid long compile times for _very_ large blocks where vector +/// instructions are spread over a wide range. +/// This limit is way higher than needed by real-world functions. +static cl::opt<int> +ScheduleRegionSizeBudget("slp-schedule-budget", cl::init(100000), cl::Hidden, + cl::desc("Limit the size of the SLP scheduling region per block")); + +namespace { + +// FIXME: Set this via cl::opt to allow overriding. +static const unsigned MinVecRegSize = 128; + +static const unsigned RecursionMaxDepth = 12; + +// Limit the number of alias checks. The limit is chosen so that +// it has no negative effect on the llvm benchmarks. +static const unsigned AliasedCheckLimit = 10; + +// Another limit for the alias checks: The maximum distance between load/store +// instructions where alias checks are done. +// This limit is useful for very large basic blocks. +static const unsigned MaxMemDepDistance = 160; + +/// If the ScheduleRegionSizeBudget is exhausted, we allow small scheduling +/// regions to be handled. +static const int MinScheduleRegionSize = 16; + +/// \brief Predicate for the element types that the SLP vectorizer supports. +/// +/// The most important thing to filter here are types which are invalid in LLVM +/// vectors. We also filter target specific types which have absolutely no +/// meaningful vectorization path such as x86_fp80 and ppc_f128. This just +/// avoids spending time checking the cost model and realizing that they will +/// be inevitably scalarized. +static bool isValidElementType(Type *Ty) { + return VectorType::isValidElementType(Ty) && !Ty->isX86_FP80Ty() && + !Ty->isPPC_FP128Ty(); +} + +/// \returns the parent basic block if all of the instructions in \p VL +/// are in the same block or null otherwise. +static BasicBlock *getSameBlock(ArrayRef<Value *> VL) { + Instruction *I0 = dyn_cast<Instruction>(VL[0]); + if (!I0) + return nullptr; + BasicBlock *BB = I0->getParent(); + for (int i = 1, e = VL.size(); i < e; i++) { + Instruction *I = dyn_cast<Instruction>(VL[i]); + if (!I) + return nullptr; + + if (BB != I->getParent()) + return nullptr; + } + return BB; +} + +/// \returns True if all of the values in \p VL are constants. +static bool allConstant(ArrayRef<Value *> VL) { + for (unsigned i = 0, e = VL.size(); i < e; ++i) + if (!isa<Constant>(VL[i])) + return false; + return true; +} + +/// \returns True if all of the values in \p VL are identical. +static bool isSplat(ArrayRef<Value *> VL) { + for (unsigned i = 1, e = VL.size(); i < e; ++i) + if (VL[i] != VL[0]) + return false; + return true; +} + +///\returns Opcode that can be clubbed with \p Op to create an alternate +/// sequence which can later be merged as a ShuffleVector instruction. +static unsigned getAltOpcode(unsigned Op) { + switch (Op) { + case Instruction::FAdd: + return Instruction::FSub; + case Instruction::FSub: + return Instruction::FAdd; + case Instruction::Add: + return Instruction::Sub; + case Instruction::Sub: + return Instruction::Add; + default: + return 0; + } +} + +///\returns bool representing if Opcode \p Op can be part +/// of an alternate sequence which can later be merged as +/// a ShuffleVector instruction. +static bool canCombineAsAltInst(unsigned Op) { + return Op == Instruction::FAdd || Op == Instruction::FSub || + Op == Instruction::Sub || Op == Instruction::Add; +} + +/// \returns ShuffleVector instruction if instructions in \p VL have +/// alternate fadd,fsub / fsub,fadd/add,sub/sub,add sequence. +/// (i.e. e.g. opcodes of fadd,fsub,fadd,fsub...) +static unsigned isAltInst(ArrayRef<Value *> VL) { + Instruction *I0 = dyn_cast<Instruction>(VL[0]); + unsigned Opcode = I0->getOpcode(); + unsigned AltOpcode = getAltOpcode(Opcode); + for (int i = 1, e = VL.size(); i < e; i++) { + Instruction *I = dyn_cast<Instruction>(VL[i]); + if (!I || I->getOpcode() != ((i & 1) ? AltOpcode : Opcode)) + return 0; + } + return Instruction::ShuffleVector; +} + +/// \returns The opcode if all of the Instructions in \p VL have the same +/// opcode, or zero. +static unsigned getSameOpcode(ArrayRef<Value *> VL) { + Instruction *I0 = dyn_cast<Instruction>(VL[0]); + if (!I0) + return 0; + unsigned Opcode = I0->getOpcode(); + for (int i = 1, e = VL.size(); i < e; i++) { + Instruction *I = dyn_cast<Instruction>(VL[i]); + if (!I || Opcode != I->getOpcode()) { + if (canCombineAsAltInst(Opcode) && i == 1) + return isAltInst(VL); + return 0; + } + } + return Opcode; +} + +/// Get the intersection (logical and) of all of the potential IR flags +/// of each scalar operation (VL) that will be converted into a vector (I). +/// Flag set: NSW, NUW, exact, and all of fast-math. +static void propagateIRFlags(Value *I, ArrayRef<Value *> VL) { + if (auto *VecOp = dyn_cast<BinaryOperator>(I)) { + if (auto *Intersection = dyn_cast<BinaryOperator>(VL[0])) { + // Intersection is initialized to the 0th scalar, + // so start counting from index '1'. + for (int i = 1, e = VL.size(); i < e; ++i) { + if (auto *Scalar = dyn_cast<BinaryOperator>(VL[i])) + Intersection->andIRFlags(Scalar); + } + VecOp->copyIRFlags(Intersection); + } + } +} + +/// \returns \p I after propagating metadata from \p VL. +static Instruction *propagateMetadata(Instruction *I, ArrayRef<Value *> VL) { + Instruction *I0 = cast<Instruction>(VL[0]); + SmallVector<std::pair<unsigned, MDNode *>, 4> Metadata; + I0->getAllMetadataOtherThanDebugLoc(Metadata); + + for (unsigned i = 0, n = Metadata.size(); i != n; ++i) { + unsigned Kind = Metadata[i].first; + MDNode *MD = Metadata[i].second; + + for (int i = 1, e = VL.size(); MD && i != e; i++) { + Instruction *I = cast<Instruction>(VL[i]); + MDNode *IMD = I->getMetadata(Kind); + + switch (Kind) { + default: + MD = nullptr; // Remove unknown metadata + break; + case LLVMContext::MD_tbaa: + MD = MDNode::getMostGenericTBAA(MD, IMD); + break; + case LLVMContext::MD_alias_scope: + MD = MDNode::getMostGenericAliasScope(MD, IMD); + break; + case LLVMContext::MD_noalias: + MD = MDNode::intersect(MD, IMD); + break; + case LLVMContext::MD_fpmath: + MD = MDNode::getMostGenericFPMath(MD, IMD); + break; + case LLVMContext::MD_nontemporal: + MD = MDNode::intersect(MD, IMD); + break; + } + } + I->setMetadata(Kind, MD); + } + return I; +} + +/// \returns The type that all of the values in \p VL have or null if there +/// are different types. +static Type* getSameType(ArrayRef<Value *> VL) { + Type *Ty = VL[0]->getType(); + for (int i = 1, e = VL.size(); i < e; i++) + if (VL[i]->getType() != Ty) + return nullptr; + + return Ty; +} + +/// \returns True if the ExtractElement instructions in VL can be vectorized +/// to use the original vector. +static bool CanReuseExtract(ArrayRef<Value *> VL) { + assert(Instruction::ExtractElement == getSameOpcode(VL) && "Invalid opcode"); + // Check if all of the extracts come from the same vector and from the + // correct offset. + Value *VL0 = VL[0]; + ExtractElementInst *E0 = cast<ExtractElementInst>(VL0); + Value *Vec = E0->getOperand(0); + + // We have to extract from the same vector type. + unsigned NElts = Vec->getType()->getVectorNumElements(); + + if (NElts != VL.size()) + return false; + + // Check that all of the indices extract from the correct offset. + ConstantInt *CI = dyn_cast<ConstantInt>(E0->getOperand(1)); + if (!CI || CI->getZExtValue()) + return false; + + for (unsigned i = 1, e = VL.size(); i < e; ++i) { + ExtractElementInst *E = cast<ExtractElementInst>(VL[i]); + ConstantInt *CI = dyn_cast<ConstantInt>(E->getOperand(1)); + + if (!CI || CI->getZExtValue() != i || E->getOperand(0) != Vec) + return false; + } + + return true; +} + +/// \returns True if in-tree use also needs extract. This refers to +/// possible scalar operand in vectorized instruction. +static bool InTreeUserNeedToExtract(Value *Scalar, Instruction *UserInst, + TargetLibraryInfo *TLI) { + + unsigned Opcode = UserInst->getOpcode(); + switch (Opcode) { + case Instruction::Load: { + LoadInst *LI = cast<LoadInst>(UserInst); + return (LI->getPointerOperand() == Scalar); + } + case Instruction::Store: { + StoreInst *SI = cast<StoreInst>(UserInst); + return (SI->getPointerOperand() == Scalar); + } + case Instruction::Call: { + CallInst *CI = cast<CallInst>(UserInst); + Intrinsic::ID ID = getIntrinsicIDForCall(CI, TLI); + if (hasVectorInstrinsicScalarOpd(ID, 1)) { + return (CI->getArgOperand(1) == Scalar); + } + } + default: + return false; + } +} + +/// \returns the AA location that is being access by the instruction. +static MemoryLocation getLocation(Instruction *I, AliasAnalysis *AA) { + if (StoreInst *SI = dyn_cast<StoreInst>(I)) + return MemoryLocation::get(SI); + if (LoadInst *LI = dyn_cast<LoadInst>(I)) + return MemoryLocation::get(LI); + return MemoryLocation(); +} + +/// \returns True if the instruction is not a volatile or atomic load/store. +static bool isSimple(Instruction *I) { + if (LoadInst *LI = dyn_cast<LoadInst>(I)) + return LI->isSimple(); + if (StoreInst *SI = dyn_cast<StoreInst>(I)) + return SI->isSimple(); + if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(I)) + return !MI->isVolatile(); + return true; +} + +/// Bottom Up SLP Vectorizer. +class BoUpSLP { +public: + typedef SmallVector<Value *, 8> ValueList; + typedef SmallVector<Instruction *, 16> InstrList; + typedef SmallPtrSet<Value *, 16> ValueSet; + typedef SmallVector<StoreInst *, 8> StoreList; + + BoUpSLP(Function *Func, ScalarEvolution *Se, TargetTransformInfo *Tti, + TargetLibraryInfo *TLi, AliasAnalysis *Aa, LoopInfo *Li, + DominatorTree *Dt, AssumptionCache *AC) + : NumLoadsWantToKeepOrder(0), NumLoadsWantToChangeOrder(0), F(Func), + SE(Se), TTI(Tti), TLI(TLi), AA(Aa), LI(Li), DT(Dt), + Builder(Se->getContext()) { + CodeMetrics::collectEphemeralValues(F, AC, EphValues); + } + + /// \brief Vectorize the tree that starts with the elements in \p VL. + /// Returns the vectorized root. + Value *vectorizeTree(); + + /// \returns the cost incurred by unwanted spills and fills, caused by + /// holding live values over call sites. + int getSpillCost(); + + /// \returns the vectorization cost of the subtree that starts at \p VL. + /// A negative number means that this is profitable. + int getTreeCost(); + + /// Construct a vectorizable tree that starts at \p Roots, ignoring users for + /// the purpose of scheduling and extraction in the \p UserIgnoreLst. + void buildTree(ArrayRef<Value *> Roots, + ArrayRef<Value *> UserIgnoreLst = None); + + /// Clear the internal data structures that are created by 'buildTree'. + void deleteTree() { + VectorizableTree.clear(); + ScalarToTreeEntry.clear(); + MustGather.clear(); + ExternalUses.clear(); + NumLoadsWantToKeepOrder = 0; + NumLoadsWantToChangeOrder = 0; + for (auto &Iter : BlocksSchedules) { + BlockScheduling *BS = Iter.second.get(); + BS->clear(); + } + } + + /// \returns true if the memory operations A and B are consecutive. + bool isConsecutiveAccess(Value *A, Value *B, const DataLayout &DL); + + /// \brief Perform LICM and CSE on the newly generated gather sequences. + void optimizeGatherSequence(); + + /// \returns true if it is beneficial to reverse the vector order. + bool shouldReorder() const { + return NumLoadsWantToChangeOrder > NumLoadsWantToKeepOrder; + } + +private: + struct TreeEntry; + + /// \returns the cost of the vectorizable entry. + int getEntryCost(TreeEntry *E); + + /// This is the recursive part of buildTree. + void buildTree_rec(ArrayRef<Value *> Roots, unsigned Depth); + + /// Vectorize a single entry in the tree. + Value *vectorizeTree(TreeEntry *E); + + /// Vectorize a single entry in the tree, starting in \p VL. + Value *vectorizeTree(ArrayRef<Value *> VL); + + /// \returns the pointer to the vectorized value if \p VL is already + /// vectorized, or NULL. They may happen in cycles. + Value *alreadyVectorized(ArrayRef<Value *> VL) const; + + /// \brief Take the pointer operand from the Load/Store instruction. + /// \returns NULL if this is not a valid Load/Store instruction. + static Value *getPointerOperand(Value *I); + + /// \brief Take the address space operand from the Load/Store instruction. + /// \returns -1 if this is not a valid Load/Store instruction. + static unsigned getAddressSpaceOperand(Value *I); + + /// \returns the scalarization cost for this type. Scalarization in this + /// context means the creation of vectors from a group of scalars. + int getGatherCost(Type *Ty); + + /// \returns the scalarization cost for this list of values. Assuming that + /// this subtree gets vectorized, we may need to extract the values from the + /// roots. This method calculates the cost of extracting the values. + int getGatherCost(ArrayRef<Value *> VL); + + /// \brief Set the Builder insert point to one after the last instruction in + /// the bundle + void setInsertPointAfterBundle(ArrayRef<Value *> VL); + + /// \returns a vector from a collection of scalars in \p VL. + Value *Gather(ArrayRef<Value *> VL, VectorType *Ty); + + /// \returns whether the VectorizableTree is fully vectorizable and will + /// be beneficial even the tree height is tiny. + bool isFullyVectorizableTinyTree(); + + /// \reorder commutative operands in alt shuffle if they result in + /// vectorized code. + void reorderAltShuffleOperands(ArrayRef<Value *> VL, + SmallVectorImpl<Value *> &Left, + SmallVectorImpl<Value *> &Right); + /// \reorder commutative operands to get better probability of + /// generating vectorized code. + void reorderInputsAccordingToOpcode(ArrayRef<Value *> VL, + SmallVectorImpl<Value *> &Left, + SmallVectorImpl<Value *> &Right); + struct TreeEntry { + TreeEntry() : Scalars(), VectorizedValue(nullptr), + NeedToGather(0) {} + + /// \returns true if the scalars in VL are equal to this entry. + bool isSame(ArrayRef<Value *> VL) const { + assert(VL.size() == Scalars.size() && "Invalid size"); + return std::equal(VL.begin(), VL.end(), Scalars.begin()); + } + + /// A vector of scalars. + ValueList Scalars; + + /// The Scalars are vectorized into this value. It is initialized to Null. + Value *VectorizedValue; + + /// Do we need to gather this sequence ? + bool NeedToGather; + }; + + /// Create a new VectorizableTree entry. + TreeEntry *newTreeEntry(ArrayRef<Value *> VL, bool Vectorized) { + VectorizableTree.emplace_back(); + int idx = VectorizableTree.size() - 1; + TreeEntry *Last = &VectorizableTree[idx]; + Last->Scalars.insert(Last->Scalars.begin(), VL.begin(), VL.end()); + Last->NeedToGather = !Vectorized; + if (Vectorized) { + for (int i = 0, e = VL.size(); i != e; ++i) { + assert(!ScalarToTreeEntry.count(VL[i]) && "Scalar already in tree!"); + ScalarToTreeEntry[VL[i]] = idx; + } + } else { + MustGather.insert(VL.begin(), VL.end()); + } + return Last; + } + + /// -- Vectorization State -- + /// Holds all of the tree entries. + std::vector<TreeEntry> VectorizableTree; + + /// Maps a specific scalar to its tree entry. + SmallDenseMap<Value*, int> ScalarToTreeEntry; + + /// A list of scalars that we found that we need to keep as scalars. + ValueSet MustGather; + + /// This POD struct describes one external user in the vectorized tree. + struct ExternalUser { + ExternalUser (Value *S, llvm::User *U, int L) : + Scalar(S), User(U), Lane(L){} + // Which scalar in our function. + Value *Scalar; + // Which user that uses the scalar. + llvm::User *User; + // Which lane does the scalar belong to. + int Lane; + }; + typedef SmallVector<ExternalUser, 16> UserList; + + /// Checks if two instructions may access the same memory. + /// + /// \p Loc1 is the location of \p Inst1. It is passed explicitly because it + /// is invariant in the calling loop. + bool isAliased(const MemoryLocation &Loc1, Instruction *Inst1, + Instruction *Inst2) { + + // First check if the result is already in the cache. + AliasCacheKey key = std::make_pair(Inst1, Inst2); + Optional<bool> &result = AliasCache[key]; + if (result.hasValue()) { + return result.getValue(); + } + MemoryLocation Loc2 = getLocation(Inst2, AA); + bool aliased = true; + if (Loc1.Ptr && Loc2.Ptr && isSimple(Inst1) && isSimple(Inst2)) { + // Do the alias check. + aliased = AA->alias(Loc1, Loc2); + } + // Store the result in the cache. + result = aliased; + return aliased; + } + + typedef std::pair<Instruction *, Instruction *> AliasCacheKey; + + /// Cache for alias results. + /// TODO: consider moving this to the AliasAnalysis itself. + DenseMap<AliasCacheKey, Optional<bool>> AliasCache; + + /// Removes an instruction from its block and eventually deletes it. + /// It's like Instruction::eraseFromParent() except that the actual deletion + /// is delayed until BoUpSLP is destructed. + /// This is required to ensure that there are no incorrect collisions in the + /// AliasCache, which can happen if a new instruction is allocated at the + /// same address as a previously deleted instruction. + void eraseInstruction(Instruction *I) { + I->removeFromParent(); + I->dropAllReferences(); + DeletedInstructions.push_back(std::unique_ptr<Instruction>(I)); + } + + /// Temporary store for deleted instructions. Instructions will be deleted + /// eventually when the BoUpSLP is destructed. + SmallVector<std::unique_ptr<Instruction>, 8> DeletedInstructions; + + /// A list of values that need to extracted out of the tree. + /// This list holds pairs of (Internal Scalar : External User). + UserList ExternalUses; + + /// Values used only by @llvm.assume calls. + SmallPtrSet<const Value *, 32> EphValues; + + /// Holds all of the instructions that we gathered. + SetVector<Instruction *> GatherSeq; + /// A list of blocks that we are going to CSE. + SetVector<BasicBlock *> CSEBlocks; + + /// Contains all scheduling relevant data for an instruction. + /// A ScheduleData either represents a single instruction or a member of an + /// instruction bundle (= a group of instructions which is combined into a + /// vector instruction). + struct ScheduleData { + + // The initial value for the dependency counters. It means that the + // dependencies are not calculated yet. + enum { InvalidDeps = -1 }; + + ScheduleData() + : Inst(nullptr), FirstInBundle(nullptr), NextInBundle(nullptr), + NextLoadStore(nullptr), SchedulingRegionID(0), SchedulingPriority(0), + Dependencies(InvalidDeps), UnscheduledDeps(InvalidDeps), + UnscheduledDepsInBundle(InvalidDeps), IsScheduled(false) {} + + void init(int BlockSchedulingRegionID) { + FirstInBundle = this; + NextInBundle = nullptr; + NextLoadStore = nullptr; + IsScheduled = false; + SchedulingRegionID = BlockSchedulingRegionID; + UnscheduledDepsInBundle = UnscheduledDeps; + clearDependencies(); + } + + /// Returns true if the dependency information has been calculated. + bool hasValidDependencies() const { return Dependencies != InvalidDeps; } + + /// Returns true for single instructions and for bundle representatives + /// (= the head of a bundle). + bool isSchedulingEntity() const { return FirstInBundle == this; } + + /// Returns true if it represents an instruction bundle and not only a + /// single instruction. + bool isPartOfBundle() const { + return NextInBundle != nullptr || FirstInBundle != this; + } + + /// Returns true if it is ready for scheduling, i.e. it has no more + /// unscheduled depending instructions/bundles. + bool isReady() const { + assert(isSchedulingEntity() && + "can't consider non-scheduling entity for ready list"); + return UnscheduledDepsInBundle == 0 && !IsScheduled; + } + + /// Modifies the number of unscheduled dependencies, also updating it for + /// the whole bundle. + int incrementUnscheduledDeps(int Incr) { + UnscheduledDeps += Incr; + return FirstInBundle->UnscheduledDepsInBundle += Incr; + } + + /// Sets the number of unscheduled dependencies to the number of + /// dependencies. + void resetUnscheduledDeps() { + incrementUnscheduledDeps(Dependencies - UnscheduledDeps); + } + + /// Clears all dependency information. + void clearDependencies() { + Dependencies = InvalidDeps; + resetUnscheduledDeps(); + MemoryDependencies.clear(); + } + + void dump(raw_ostream &os) const { + if (!isSchedulingEntity()) { + os << "/ " << *Inst; + } else if (NextInBundle) { + os << '[' << *Inst; + ScheduleData *SD = NextInBundle; + while (SD) { + os << ';' << *SD->Inst; + SD = SD->NextInBundle; + } + os << ']'; + } else { + os << *Inst; + } + } + + Instruction *Inst; + + /// Points to the head in an instruction bundle (and always to this for + /// single instructions). + ScheduleData *FirstInBundle; + + /// Single linked list of all instructions in a bundle. Null if it is a + /// single instruction. + ScheduleData *NextInBundle; + + /// Single linked list of all memory instructions (e.g. load, store, call) + /// in the block - until the end of the scheduling region. + ScheduleData *NextLoadStore; + + /// The dependent memory instructions. + /// This list is derived on demand in calculateDependencies(). + SmallVector<ScheduleData *, 4> MemoryDependencies; + + /// This ScheduleData is in the current scheduling region if this matches + /// the current SchedulingRegionID of BlockScheduling. + int SchedulingRegionID; + + /// Used for getting a "good" final ordering of instructions. + int SchedulingPriority; + + /// The number of dependencies. Constitutes of the number of users of the + /// instruction plus the number of dependent memory instructions (if any). + /// This value is calculated on demand. + /// If InvalidDeps, the number of dependencies is not calculated yet. + /// + int Dependencies; + + /// The number of dependencies minus the number of dependencies of scheduled + /// instructions. As soon as this is zero, the instruction/bundle gets ready + /// for scheduling. + /// Note that this is negative as long as Dependencies is not calculated. + int UnscheduledDeps; + + /// The sum of UnscheduledDeps in a bundle. Equals to UnscheduledDeps for + /// single instructions. + int UnscheduledDepsInBundle; + + /// True if this instruction is scheduled (or considered as scheduled in the + /// dry-run). + bool IsScheduled; + }; + +#ifndef NDEBUG + friend raw_ostream &operator<<(raw_ostream &os, + const BoUpSLP::ScheduleData &SD); +#endif + + /// Contains all scheduling data for a basic block. + /// + struct BlockScheduling { + + BlockScheduling(BasicBlock *BB) + : BB(BB), ChunkSize(BB->size()), ChunkPos(ChunkSize), + ScheduleStart(nullptr), ScheduleEnd(nullptr), + FirstLoadStoreInRegion(nullptr), LastLoadStoreInRegion(nullptr), + ScheduleRegionSize(0), + ScheduleRegionSizeLimit(ScheduleRegionSizeBudget), + // Make sure that the initial SchedulingRegionID is greater than the + // initial SchedulingRegionID in ScheduleData (which is 0). + SchedulingRegionID(1) {} + + void clear() { + ReadyInsts.clear(); + ScheduleStart = nullptr; + ScheduleEnd = nullptr; + FirstLoadStoreInRegion = nullptr; + LastLoadStoreInRegion = nullptr; + + // Reduce the maximum schedule region size by the size of the + // previous scheduling run. + ScheduleRegionSizeLimit -= ScheduleRegionSize; + if (ScheduleRegionSizeLimit < MinScheduleRegionSize) + ScheduleRegionSizeLimit = MinScheduleRegionSize; + ScheduleRegionSize = 0; + + // Make a new scheduling region, i.e. all existing ScheduleData is not + // in the new region yet. + ++SchedulingRegionID; + } + + ScheduleData *getScheduleData(Value *V) { + ScheduleData *SD = ScheduleDataMap[V]; + if (SD && SD->SchedulingRegionID == SchedulingRegionID) + return SD; + return nullptr; + } + + bool isInSchedulingRegion(ScheduleData *SD) { + return SD->SchedulingRegionID == SchedulingRegionID; + } + + /// Marks an instruction as scheduled and puts all dependent ready + /// instructions into the ready-list. + template <typename ReadyListType> + void schedule(ScheduleData *SD, ReadyListType &ReadyList) { + SD->IsScheduled = true; + DEBUG(dbgs() << "SLP: schedule " << *SD << "\n"); + + ScheduleData *BundleMember = SD; + while (BundleMember) { + // Handle the def-use chain dependencies. + for (Use &U : BundleMember->Inst->operands()) { + ScheduleData *OpDef = getScheduleData(U.get()); + if (OpDef && OpDef->hasValidDependencies() && + OpDef->incrementUnscheduledDeps(-1) == 0) { + // There are no more unscheduled dependencies after decrementing, + // so we can put the dependent instruction into the ready list. + ScheduleData *DepBundle = OpDef->FirstInBundle; + assert(!DepBundle->IsScheduled && + "already scheduled bundle gets ready"); + ReadyList.insert(DepBundle); + DEBUG(dbgs() << "SLP: gets ready (def): " << *DepBundle << "\n"); + } + } + // Handle the memory dependencies. + for (ScheduleData *MemoryDepSD : BundleMember->MemoryDependencies) { + if (MemoryDepSD->incrementUnscheduledDeps(-1) == 0) { + // There are no more unscheduled dependencies after decrementing, + // so we can put the dependent instruction into the ready list. + ScheduleData *DepBundle = MemoryDepSD->FirstInBundle; + assert(!DepBundle->IsScheduled && + "already scheduled bundle gets ready"); + ReadyList.insert(DepBundle); + DEBUG(dbgs() << "SLP: gets ready (mem): " << *DepBundle << "\n"); + } + } + BundleMember = BundleMember->NextInBundle; + } + } + + /// Put all instructions into the ReadyList which are ready for scheduling. + template <typename ReadyListType> + void initialFillReadyList(ReadyListType &ReadyList) { + for (auto *I = ScheduleStart; I != ScheduleEnd; I = I->getNextNode()) { + ScheduleData *SD = getScheduleData(I); + if (SD->isSchedulingEntity() && SD->isReady()) { + ReadyList.insert(SD); + DEBUG(dbgs() << "SLP: initially in ready list: " << *I << "\n"); + } + } + } + + /// Checks if a bundle of instructions can be scheduled, i.e. has no + /// cyclic dependencies. This is only a dry-run, no instructions are + /// actually moved at this stage. + bool tryScheduleBundle(ArrayRef<Value *> VL, BoUpSLP *SLP); + + /// Un-bundles a group of instructions. + void cancelScheduling(ArrayRef<Value *> VL); + + /// Extends the scheduling region so that V is inside the region. + /// \returns true if the region size is within the limit. + bool extendSchedulingRegion(Value *V); + + /// Initialize the ScheduleData structures for new instructions in the + /// scheduling region. + void initScheduleData(Instruction *FromI, Instruction *ToI, + ScheduleData *PrevLoadStore, + ScheduleData *NextLoadStore); + + /// Updates the dependency information of a bundle and of all instructions/ + /// bundles which depend on the original bundle. + void calculateDependencies(ScheduleData *SD, bool InsertInReadyList, + BoUpSLP *SLP); + + /// Sets all instruction in the scheduling region to un-scheduled. + void resetSchedule(); + + BasicBlock *BB; + + /// Simple memory allocation for ScheduleData. + std::vector<std::unique_ptr<ScheduleData[]>> ScheduleDataChunks; + + /// The size of a ScheduleData array in ScheduleDataChunks. + int ChunkSize; + + /// The allocator position in the current chunk, which is the last entry + /// of ScheduleDataChunks. + int ChunkPos; + + /// Attaches ScheduleData to Instruction. + /// Note that the mapping survives during all vectorization iterations, i.e. + /// ScheduleData structures are recycled. + DenseMap<Value *, ScheduleData *> ScheduleDataMap; + + struct ReadyList : SmallVector<ScheduleData *, 8> { + void insert(ScheduleData *SD) { push_back(SD); } + }; + + /// The ready-list for scheduling (only used for the dry-run). + ReadyList ReadyInsts; + + /// The first instruction of the scheduling region. + Instruction *ScheduleStart; + + /// The first instruction _after_ the scheduling region. + Instruction *ScheduleEnd; + + /// The first memory accessing instruction in the scheduling region + /// (can be null). + ScheduleData *FirstLoadStoreInRegion; + + /// The last memory accessing instruction in the scheduling region + /// (can be null). + ScheduleData *LastLoadStoreInRegion; + + /// The current size of the scheduling region. + int ScheduleRegionSize; + + /// The maximum size allowed for the scheduling region. + int ScheduleRegionSizeLimit; + + /// The ID of the scheduling region. For a new vectorization iteration this + /// is incremented which "removes" all ScheduleData from the region. + int SchedulingRegionID; + }; + + /// Attaches the BlockScheduling structures to basic blocks. + MapVector<BasicBlock *, std::unique_ptr<BlockScheduling>> BlocksSchedules; + + /// Performs the "real" scheduling. Done before vectorization is actually + /// performed in a basic block. + void scheduleBlock(BlockScheduling *BS); + + /// List of users to ignore during scheduling and that don't need extracting. + ArrayRef<Value *> UserIgnoreList; + + // Number of load-bundles, which contain consecutive loads. + int NumLoadsWantToKeepOrder; + + // Number of load-bundles of size 2, which are consecutive loads if reversed. + int NumLoadsWantToChangeOrder; + + // Analysis and block reference. + Function *F; + ScalarEvolution *SE; + TargetTransformInfo *TTI; + TargetLibraryInfo *TLI; + AliasAnalysis *AA; + LoopInfo *LI; + DominatorTree *DT; + /// Instruction builder to construct the vectorized tree. + IRBuilder<> Builder; +}; + +#ifndef NDEBUG +raw_ostream &operator<<(raw_ostream &os, const BoUpSLP::ScheduleData &SD) { + SD.dump(os); + return os; +} +#endif + +void BoUpSLP::buildTree(ArrayRef<Value *> Roots, + ArrayRef<Value *> UserIgnoreLst) { + deleteTree(); + UserIgnoreList = UserIgnoreLst; + if (!getSameType(Roots)) + return; + buildTree_rec(Roots, 0); + + // Collect the values that we need to extract from the tree. + for (int EIdx = 0, EE = VectorizableTree.size(); EIdx < EE; ++EIdx) { + TreeEntry *Entry = &VectorizableTree[EIdx]; + + // For each lane: + for (int Lane = 0, LE = Entry->Scalars.size(); Lane != LE; ++Lane) { + Value *Scalar = Entry->Scalars[Lane]; + + // No need to handle users of gathered values. + if (Entry->NeedToGather) + continue; + + for (User *U : Scalar->users()) { + DEBUG(dbgs() << "SLP: Checking user:" << *U << ".\n"); + + Instruction *UserInst = dyn_cast<Instruction>(U); + if (!UserInst) + continue; + + // Skip in-tree scalars that become vectors + if (ScalarToTreeEntry.count(U)) { + int Idx = ScalarToTreeEntry[U]; + TreeEntry *UseEntry = &VectorizableTree[Idx]; + Value *UseScalar = UseEntry->Scalars[0]; + // Some in-tree scalars will remain as scalar in vectorized + // instructions. If that is the case, the one in Lane 0 will + // be used. + if (UseScalar != U || + !InTreeUserNeedToExtract(Scalar, UserInst, TLI)) { + DEBUG(dbgs() << "SLP: \tInternal user will be removed:" << *U + << ".\n"); + assert(!VectorizableTree[Idx].NeedToGather && "Bad state"); + continue; + } + } + + // Ignore users in the user ignore list. + if (std::find(UserIgnoreList.begin(), UserIgnoreList.end(), UserInst) != + UserIgnoreList.end()) + continue; + + DEBUG(dbgs() << "SLP: Need to extract:" << *U << " from lane " << + Lane << " from " << *Scalar << ".\n"); + ExternalUses.push_back(ExternalUser(Scalar, U, Lane)); + } + } + } +} + + +void BoUpSLP::buildTree_rec(ArrayRef<Value *> VL, unsigned Depth) { + bool SameTy = getSameType(VL); (void)SameTy; + bool isAltShuffle = false; + assert(SameTy && "Invalid types!"); + + if (Depth == RecursionMaxDepth) { + DEBUG(dbgs() << "SLP: Gathering due to max recursion depth.\n"); + newTreeEntry(VL, false); + return; + } + + // Don't handle vectors. + if (VL[0]->getType()->isVectorTy()) { + DEBUG(dbgs() << "SLP: Gathering due to vector type.\n"); + newTreeEntry(VL, false); + return; + } + + if (StoreInst *SI = dyn_cast<StoreInst>(VL[0])) + if (SI->getValueOperand()->getType()->isVectorTy()) { + DEBUG(dbgs() << "SLP: Gathering due to store vector type.\n"); + newTreeEntry(VL, false); + return; + } + unsigned Opcode = getSameOpcode(VL); + + // Check that this shuffle vector refers to the alternate + // sequence of opcodes. + if (Opcode == Instruction::ShuffleVector) { + Instruction *I0 = dyn_cast<Instruction>(VL[0]); + unsigned Op = I0->getOpcode(); + if (Op != Instruction::ShuffleVector) + isAltShuffle = true; + } + + // If all of the operands are identical or constant we have a simple solution. + if (allConstant(VL) || isSplat(VL) || !getSameBlock(VL) || !Opcode) { + DEBUG(dbgs() << "SLP: Gathering due to C,S,B,O. \n"); + newTreeEntry(VL, false); + return; + } + + // We now know that this is a vector of instructions of the same type from + // the same block. + + // Don't vectorize ephemeral values. + for (unsigned i = 0, e = VL.size(); i != e; ++i) { + if (EphValues.count(VL[i])) { + DEBUG(dbgs() << "SLP: The instruction (" << *VL[i] << + ") is ephemeral.\n"); + newTreeEntry(VL, false); + return; + } + } + + // Check if this is a duplicate of another entry. + if (ScalarToTreeEntry.count(VL[0])) { + int Idx = ScalarToTreeEntry[VL[0]]; + TreeEntry *E = &VectorizableTree[Idx]; + for (unsigned i = 0, e = VL.size(); i != e; ++i) { + DEBUG(dbgs() << "SLP: \tChecking bundle: " << *VL[i] << ".\n"); + if (E->Scalars[i] != VL[i]) { + DEBUG(dbgs() << "SLP: Gathering due to partial overlap.\n"); + newTreeEntry(VL, false); + return; + } + } + DEBUG(dbgs() << "SLP: Perfect diamond merge at " << *VL[0] << ".\n"); + return; + } + + // Check that none of the instructions in the bundle are already in the tree. + for (unsigned i = 0, e = VL.size(); i != e; ++i) { + if (ScalarToTreeEntry.count(VL[i])) { + DEBUG(dbgs() << "SLP: The instruction (" << *VL[i] << + ") is already in tree.\n"); + newTreeEntry(VL, false); + return; + } + } + + // If any of the scalars is marked as a value that needs to stay scalar then + // we need to gather the scalars. + for (unsigned i = 0, e = VL.size(); i != e; ++i) { + if (MustGather.count(VL[i])) { + DEBUG(dbgs() << "SLP: Gathering due to gathered scalar.\n"); + newTreeEntry(VL, false); + return; + } + } + + // Check that all of the users of the scalars that we want to vectorize are + // schedulable. + Instruction *VL0 = cast<Instruction>(VL[0]); + BasicBlock *BB = cast<Instruction>(VL0)->getParent(); + + if (!DT->isReachableFromEntry(BB)) { + // Don't go into unreachable blocks. They may contain instructions with + // dependency cycles which confuse the final scheduling. + DEBUG(dbgs() << "SLP: bundle in unreachable block.\n"); + newTreeEntry(VL, false); + return; + } + + // Check that every instructions appears once in this bundle. + for (unsigned i = 0, e = VL.size(); i < e; ++i) + for (unsigned j = i+1; j < e; ++j) + if (VL[i] == VL[j]) { + DEBUG(dbgs() << "SLP: Scalar used twice in bundle.\n"); + newTreeEntry(VL, false); + return; + } + + auto &BSRef = BlocksSchedules[BB]; + if (!BSRef) { + BSRef = llvm::make_unique<BlockScheduling>(BB); + } + BlockScheduling &BS = *BSRef.get(); + + if (!BS.tryScheduleBundle(VL, this)) { + DEBUG(dbgs() << "SLP: We are not able to schedule this bundle!\n"); + assert((!BS.getScheduleData(VL[0]) || + !BS.getScheduleData(VL[0])->isPartOfBundle()) && + "tryScheduleBundle should cancelScheduling on failure"); + newTreeEntry(VL, false); + return; + } + DEBUG(dbgs() << "SLP: We are able to schedule this bundle.\n"); + + switch (Opcode) { + case Instruction::PHI: { + PHINode *PH = dyn_cast<PHINode>(VL0); + + // Check for terminator values (e.g. invoke). + for (unsigned j = 0; j < VL.size(); ++j) + for (unsigned i = 0, e = PH->getNumIncomingValues(); i < e; ++i) { + TerminatorInst *Term = dyn_cast<TerminatorInst>( + cast<PHINode>(VL[j])->getIncomingValueForBlock(PH->getIncomingBlock(i))); + if (Term) { + DEBUG(dbgs() << "SLP: Need to swizzle PHINodes (TerminatorInst use).\n"); + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + return; + } + } + + newTreeEntry(VL, true); + DEBUG(dbgs() << "SLP: added a vector of PHINodes.\n"); + + for (unsigned i = 0, e = PH->getNumIncomingValues(); i < e; ++i) { + ValueList Operands; + // Prepare the operand vector. + for (unsigned j = 0; j < VL.size(); ++j) + Operands.push_back(cast<PHINode>(VL[j])->getIncomingValueForBlock( + PH->getIncomingBlock(i))); + + buildTree_rec(Operands, Depth + 1); + } + return; + } + case Instruction::ExtractElement: { + bool Reuse = CanReuseExtract(VL); + if (Reuse) { + DEBUG(dbgs() << "SLP: Reusing extract sequence.\n"); + } else { + BS.cancelScheduling(VL); + } + newTreeEntry(VL, Reuse); + return; + } + case Instruction::Load: { + // Check that a vectorized load would load the same memory as a scalar + // load. + // For example we don't want vectorize loads that are smaller than 8 bit. + // Even though we have a packed struct {<i2, i2, i2, i2>} LLVM treats + // loading/storing it as an i8 struct. If we vectorize loads/stores from + // such a struct we read/write packed bits disagreeing with the + // unvectorized version. + const DataLayout &DL = F->getParent()->getDataLayout(); + Type *ScalarTy = VL[0]->getType(); + + if (DL.getTypeSizeInBits(ScalarTy) != + DL.getTypeAllocSizeInBits(ScalarTy)) { + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + DEBUG(dbgs() << "SLP: Gathering loads of non-packed type.\n"); + return; + } + // Check if the loads are consecutive or of we need to swizzle them. + for (unsigned i = 0, e = VL.size() - 1; i < e; ++i) { + LoadInst *L = cast<LoadInst>(VL[i]); + if (!L->isSimple()) { + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + DEBUG(dbgs() << "SLP: Gathering non-simple loads.\n"); + return; + } + + if (!isConsecutiveAccess(VL[i], VL[i + 1], DL)) { + if (VL.size() == 2 && isConsecutiveAccess(VL[1], VL[0], DL)) { + ++NumLoadsWantToChangeOrder; + } + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + DEBUG(dbgs() << "SLP: Gathering non-consecutive loads.\n"); + return; + } + } + ++NumLoadsWantToKeepOrder; + newTreeEntry(VL, true); + DEBUG(dbgs() << "SLP: added a vector of loads.\n"); + return; + } + case Instruction::ZExt: + case Instruction::SExt: + case Instruction::FPToUI: + case Instruction::FPToSI: + case Instruction::FPExt: + case Instruction::PtrToInt: + case Instruction::IntToPtr: + case Instruction::SIToFP: + case Instruction::UIToFP: + case Instruction::Trunc: + case Instruction::FPTrunc: + case Instruction::BitCast: { + Type *SrcTy = VL0->getOperand(0)->getType(); + for (unsigned i = 0; i < VL.size(); ++i) { + Type *Ty = cast<Instruction>(VL[i])->getOperand(0)->getType(); + if (Ty != SrcTy || !isValidElementType(Ty)) { + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + DEBUG(dbgs() << "SLP: Gathering casts with different src types.\n"); + return; + } + } + newTreeEntry(VL, true); + DEBUG(dbgs() << "SLP: added a vector of casts.\n"); + + for (unsigned i = 0, e = VL0->getNumOperands(); i < e; ++i) { + ValueList Operands; + // Prepare the operand vector. + for (unsigned j = 0; j < VL.size(); ++j) + Operands.push_back(cast<Instruction>(VL[j])->getOperand(i)); + + buildTree_rec(Operands, Depth+1); + } + return; + } + case Instruction::ICmp: + case Instruction::FCmp: { + // Check that all of the compares have the same predicate. + CmpInst::Predicate P0 = cast<CmpInst>(VL0)->getPredicate(); + Type *ComparedTy = cast<Instruction>(VL[0])->getOperand(0)->getType(); + for (unsigned i = 1, e = VL.size(); i < e; ++i) { + CmpInst *Cmp = cast<CmpInst>(VL[i]); + if (Cmp->getPredicate() != P0 || + Cmp->getOperand(0)->getType() != ComparedTy) { + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + DEBUG(dbgs() << "SLP: Gathering cmp with different predicate.\n"); + return; + } + } + + newTreeEntry(VL, true); + DEBUG(dbgs() << "SLP: added a vector of compares.\n"); + + for (unsigned i = 0, e = VL0->getNumOperands(); i < e; ++i) { + ValueList Operands; + // Prepare the operand vector. + for (unsigned j = 0; j < VL.size(); ++j) + Operands.push_back(cast<Instruction>(VL[j])->getOperand(i)); + + buildTree_rec(Operands, Depth+1); + } + return; + } + case Instruction::Select: + case Instruction::Add: + case Instruction::FAdd: + case Instruction::Sub: + case Instruction::FSub: + case Instruction::Mul: + case Instruction::FMul: + case Instruction::UDiv: + case Instruction::SDiv: + case Instruction::FDiv: + case Instruction::URem: + case Instruction::SRem: + case Instruction::FRem: + case Instruction::Shl: + case Instruction::LShr: + case Instruction::AShr: + case Instruction::And: + case Instruction::Or: + case Instruction::Xor: { + newTreeEntry(VL, true); + DEBUG(dbgs() << "SLP: added a vector of bin op.\n"); + + // Sort operands of the instructions so that each side is more likely to + // have the same opcode. + if (isa<BinaryOperator>(VL0) && VL0->isCommutative()) { + ValueList Left, Right; + reorderInputsAccordingToOpcode(VL, Left, Right); + buildTree_rec(Left, Depth + 1); + buildTree_rec(Right, Depth + 1); + return; + } + + for (unsigned i = 0, e = VL0->getNumOperands(); i < e; ++i) { + ValueList Operands; + // Prepare the operand vector. + for (unsigned j = 0; j < VL.size(); ++j) + Operands.push_back(cast<Instruction>(VL[j])->getOperand(i)); + + buildTree_rec(Operands, Depth+1); + } + return; + } + case Instruction::GetElementPtr: { + // We don't combine GEPs with complicated (nested) indexing. + for (unsigned j = 0; j < VL.size(); ++j) { + if (cast<Instruction>(VL[j])->getNumOperands() != 2) { + DEBUG(dbgs() << "SLP: not-vectorizable GEP (nested indexes).\n"); + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + return; + } + } + + // We can't combine several GEPs into one vector if they operate on + // different types. + Type *Ty0 = cast<Instruction>(VL0)->getOperand(0)->getType(); + for (unsigned j = 0; j < VL.size(); ++j) { + Type *CurTy = cast<Instruction>(VL[j])->getOperand(0)->getType(); + if (Ty0 != CurTy) { + DEBUG(dbgs() << "SLP: not-vectorizable GEP (different types).\n"); + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + return; + } + } + + // We don't combine GEPs with non-constant indexes. + for (unsigned j = 0; j < VL.size(); ++j) { + auto Op = cast<Instruction>(VL[j])->getOperand(1); + if (!isa<ConstantInt>(Op)) { + DEBUG( + dbgs() << "SLP: not-vectorizable GEP (non-constant indexes).\n"); + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + return; + } + } + + newTreeEntry(VL, true); + DEBUG(dbgs() << "SLP: added a vector of GEPs.\n"); + for (unsigned i = 0, e = 2; i < e; ++i) { + ValueList Operands; + // Prepare the operand vector. + for (unsigned j = 0; j < VL.size(); ++j) + Operands.push_back(cast<Instruction>(VL[j])->getOperand(i)); + + buildTree_rec(Operands, Depth + 1); + } + return; + } + case Instruction::Store: { + const DataLayout &DL = F->getParent()->getDataLayout(); + // Check if the stores are consecutive or of we need to swizzle them. + for (unsigned i = 0, e = VL.size() - 1; i < e; ++i) + if (!isConsecutiveAccess(VL[i], VL[i + 1], DL)) { + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + DEBUG(dbgs() << "SLP: Non-consecutive store.\n"); + return; + } + + newTreeEntry(VL, true); + DEBUG(dbgs() << "SLP: added a vector of stores.\n"); + + ValueList Operands; + for (unsigned j = 0; j < VL.size(); ++j) + Operands.push_back(cast<Instruction>(VL[j])->getOperand(0)); + + buildTree_rec(Operands, Depth + 1); + return; + } + case Instruction::Call: { + // Check if the calls are all to the same vectorizable intrinsic. + CallInst *CI = cast<CallInst>(VL[0]); + // Check if this is an Intrinsic call or something that can be + // represented by an intrinsic call + Intrinsic::ID ID = getIntrinsicIDForCall(CI, TLI); + if (!isTriviallyVectorizable(ID)) { + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + DEBUG(dbgs() << "SLP: Non-vectorizable call.\n"); + return; + } + Function *Int = CI->getCalledFunction(); + Value *A1I = nullptr; + if (hasVectorInstrinsicScalarOpd(ID, 1)) + A1I = CI->getArgOperand(1); + for (unsigned i = 1, e = VL.size(); i != e; ++i) { + CallInst *CI2 = dyn_cast<CallInst>(VL[i]); + if (!CI2 || CI2->getCalledFunction() != Int || + getIntrinsicIDForCall(CI2, TLI) != ID) { + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + DEBUG(dbgs() << "SLP: mismatched calls:" << *CI << "!=" << *VL[i] + << "\n"); + return; + } + // ctlz,cttz and powi are special intrinsics whose second argument + // should be same in order for them to be vectorized. + if (hasVectorInstrinsicScalarOpd(ID, 1)) { + Value *A1J = CI2->getArgOperand(1); + if (A1I != A1J) { + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + DEBUG(dbgs() << "SLP: mismatched arguments in call:" << *CI + << " argument "<< A1I<<"!=" << A1J + << "\n"); + return; + } + } + } + + newTreeEntry(VL, true); + for (unsigned i = 0, e = CI->getNumArgOperands(); i != e; ++i) { + ValueList Operands; + // Prepare the operand vector. + for (unsigned j = 0; j < VL.size(); ++j) { + CallInst *CI2 = dyn_cast<CallInst>(VL[j]); + Operands.push_back(CI2->getArgOperand(i)); + } + buildTree_rec(Operands, Depth + 1); + } + return; + } + case Instruction::ShuffleVector: { + // If this is not an alternate sequence of opcode like add-sub + // then do not vectorize this instruction. + if (!isAltShuffle) { + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + DEBUG(dbgs() << "SLP: ShuffleVector are not vectorized.\n"); + return; + } + newTreeEntry(VL, true); + DEBUG(dbgs() << "SLP: added a ShuffleVector op.\n"); + + // Reorder operands if reordering would enable vectorization. + if (isa<BinaryOperator>(VL0)) { + ValueList Left, Right; + reorderAltShuffleOperands(VL, Left, Right); + buildTree_rec(Left, Depth + 1); + buildTree_rec(Right, Depth + 1); + return; + } + + for (unsigned i = 0, e = VL0->getNumOperands(); i < e; ++i) { + ValueList Operands; + // Prepare the operand vector. + for (unsigned j = 0; j < VL.size(); ++j) + Operands.push_back(cast<Instruction>(VL[j])->getOperand(i)); + + buildTree_rec(Operands, Depth + 1); + } + return; + } + default: + BS.cancelScheduling(VL); + newTreeEntry(VL, false); + DEBUG(dbgs() << "SLP: Gathering unknown instruction.\n"); + return; + } +} + +int BoUpSLP::getEntryCost(TreeEntry *E) { + ArrayRef<Value*> VL = E->Scalars; + + Type *ScalarTy = VL[0]->getType(); + if (StoreInst *SI = dyn_cast<StoreInst>(VL[0])) + ScalarTy = SI->getValueOperand()->getType(); + VectorType *VecTy = VectorType::get(ScalarTy, VL.size()); + + if (E->NeedToGather) { + if (allConstant(VL)) + return 0; + if (isSplat(VL)) { + return TTI->getShuffleCost(TargetTransformInfo::SK_Broadcast, VecTy, 0); + } + return getGatherCost(E->Scalars); + } + unsigned Opcode = getSameOpcode(VL); + assert(Opcode && getSameType(VL) && getSameBlock(VL) && "Invalid VL"); + Instruction *VL0 = cast<Instruction>(VL[0]); + switch (Opcode) { + case Instruction::PHI: { + return 0; + } + case Instruction::ExtractElement: { + if (CanReuseExtract(VL)) { + int DeadCost = 0; + for (unsigned i = 0, e = VL.size(); i < e; ++i) { + ExtractElementInst *E = cast<ExtractElementInst>(VL[i]); + if (E->hasOneUse()) + // Take credit for instruction that will become dead. + DeadCost += + TTI->getVectorInstrCost(Instruction::ExtractElement, VecTy, i); + } + return -DeadCost; + } + return getGatherCost(VecTy); + } + case Instruction::ZExt: + case Instruction::SExt: + case Instruction::FPToUI: + case Instruction::FPToSI: + case Instruction::FPExt: + case Instruction::PtrToInt: + case Instruction::IntToPtr: + case Instruction::SIToFP: + case Instruction::UIToFP: + case Instruction::Trunc: + case Instruction::FPTrunc: + case Instruction::BitCast: { + Type *SrcTy = VL0->getOperand(0)->getType(); + + // Calculate the cost of this instruction. + int ScalarCost = VL.size() * TTI->getCastInstrCost(VL0->getOpcode(), + VL0->getType(), SrcTy); + + VectorType *SrcVecTy = VectorType::get(SrcTy, VL.size()); + int VecCost = TTI->getCastInstrCost(VL0->getOpcode(), VecTy, SrcVecTy); + return VecCost - ScalarCost; + } + case Instruction::FCmp: + case Instruction::ICmp: + case Instruction::Select: + case Instruction::Add: + case Instruction::FAdd: + case Instruction::Sub: + case Instruction::FSub: + case Instruction::Mul: + case Instruction::FMul: + case Instruction::UDiv: + case Instruction::SDiv: + case Instruction::FDiv: + case Instruction::URem: + case Instruction::SRem: + case Instruction::FRem: + case Instruction::Shl: + case Instruction::LShr: + case Instruction::AShr: + case Instruction::And: + case Instruction::Or: + case Instruction::Xor: { + // Calculate the cost of this instruction. + int ScalarCost = 0; + int VecCost = 0; + if (Opcode == Instruction::FCmp || Opcode == Instruction::ICmp || + Opcode == Instruction::Select) { + VectorType *MaskTy = VectorType::get(Builder.getInt1Ty(), VL.size()); + ScalarCost = VecTy->getNumElements() * + TTI->getCmpSelInstrCost(Opcode, ScalarTy, Builder.getInt1Ty()); + VecCost = TTI->getCmpSelInstrCost(Opcode, VecTy, MaskTy); + } else { + // Certain instructions can be cheaper to vectorize if they have a + // constant second vector operand. + TargetTransformInfo::OperandValueKind Op1VK = + TargetTransformInfo::OK_AnyValue; + TargetTransformInfo::OperandValueKind Op2VK = + TargetTransformInfo::OK_UniformConstantValue; + TargetTransformInfo::OperandValueProperties Op1VP = + TargetTransformInfo::OP_None; + TargetTransformInfo::OperandValueProperties Op2VP = + TargetTransformInfo::OP_None; + + // If all operands are exactly the same ConstantInt then set the + // operand kind to OK_UniformConstantValue. + // If instead not all operands are constants, then set the operand kind + // to OK_AnyValue. If all operands are constants but not the same, + // then set the operand kind to OK_NonUniformConstantValue. + ConstantInt *CInt = nullptr; + for (unsigned i = 0; i < VL.size(); ++i) { + const Instruction *I = cast<Instruction>(VL[i]); + if (!isa<ConstantInt>(I->getOperand(1))) { + Op2VK = TargetTransformInfo::OK_AnyValue; + break; + } + if (i == 0) { + CInt = cast<ConstantInt>(I->getOperand(1)); + continue; + } + if (Op2VK == TargetTransformInfo::OK_UniformConstantValue && + CInt != cast<ConstantInt>(I->getOperand(1))) + Op2VK = TargetTransformInfo::OK_NonUniformConstantValue; + } + // FIXME: Currently cost of model modification for division by + // power of 2 is handled only for X86. Add support for other targets. + if (Op2VK == TargetTransformInfo::OK_UniformConstantValue && CInt && + CInt->getValue().isPowerOf2()) + Op2VP = TargetTransformInfo::OP_PowerOf2; + + ScalarCost = VecTy->getNumElements() * + TTI->getArithmeticInstrCost(Opcode, ScalarTy, Op1VK, Op2VK, + Op1VP, Op2VP); + VecCost = TTI->getArithmeticInstrCost(Opcode, VecTy, Op1VK, Op2VK, + Op1VP, Op2VP); + } + return VecCost - ScalarCost; + } + case Instruction::GetElementPtr: { + TargetTransformInfo::OperandValueKind Op1VK = + TargetTransformInfo::OK_AnyValue; + TargetTransformInfo::OperandValueKind Op2VK = + TargetTransformInfo::OK_UniformConstantValue; + + int ScalarCost = + VecTy->getNumElements() * + TTI->getArithmeticInstrCost(Instruction::Add, ScalarTy, Op1VK, Op2VK); + int VecCost = + TTI->getArithmeticInstrCost(Instruction::Add, VecTy, Op1VK, Op2VK); + + return VecCost - ScalarCost; + } + case Instruction::Load: { + // Cost of wide load - cost of scalar loads. + int ScalarLdCost = VecTy->getNumElements() * + TTI->getMemoryOpCost(Instruction::Load, ScalarTy, 1, 0); + int VecLdCost = TTI->getMemoryOpCost(Instruction::Load, VecTy, 1, 0); + return VecLdCost - ScalarLdCost; + } + case Instruction::Store: { + // We know that we can merge the stores. Calculate the cost. + int ScalarStCost = VecTy->getNumElements() * + TTI->getMemoryOpCost(Instruction::Store, ScalarTy, 1, 0); + int VecStCost = TTI->getMemoryOpCost(Instruction::Store, VecTy, 1, 0); + return VecStCost - ScalarStCost; + } + case Instruction::Call: { + CallInst *CI = cast<CallInst>(VL0); + Intrinsic::ID ID = getIntrinsicIDForCall(CI, TLI); + + // Calculate the cost of the scalar and vector calls. + SmallVector<Type*, 4> ScalarTys, VecTys; + for (unsigned op = 0, opc = CI->getNumArgOperands(); op!= opc; ++op) { + ScalarTys.push_back(CI->getArgOperand(op)->getType()); + VecTys.push_back(VectorType::get(CI->getArgOperand(op)->getType(), + VecTy->getNumElements())); + } + + int ScalarCallCost = VecTy->getNumElements() * + TTI->getIntrinsicInstrCost(ID, ScalarTy, ScalarTys); + + int VecCallCost = TTI->getIntrinsicInstrCost(ID, VecTy, VecTys); + + DEBUG(dbgs() << "SLP: Call cost "<< VecCallCost - ScalarCallCost + << " (" << VecCallCost << "-" << ScalarCallCost << ")" + << " for " << *CI << "\n"); + + return VecCallCost - ScalarCallCost; + } + case Instruction::ShuffleVector: { + TargetTransformInfo::OperandValueKind Op1VK = + TargetTransformInfo::OK_AnyValue; + TargetTransformInfo::OperandValueKind Op2VK = + TargetTransformInfo::OK_AnyValue; + int ScalarCost = 0; + int VecCost = 0; + for (unsigned i = 0; i < VL.size(); ++i) { + Instruction *I = cast<Instruction>(VL[i]); + if (!I) + break; + ScalarCost += + TTI->getArithmeticInstrCost(I->getOpcode(), ScalarTy, Op1VK, Op2VK); + } + // VecCost is equal to sum of the cost of creating 2 vectors + // and the cost of creating shuffle. + Instruction *I0 = cast<Instruction>(VL[0]); + VecCost = + TTI->getArithmeticInstrCost(I0->getOpcode(), VecTy, Op1VK, Op2VK); + Instruction *I1 = cast<Instruction>(VL[1]); + VecCost += + TTI->getArithmeticInstrCost(I1->getOpcode(), VecTy, Op1VK, Op2VK); + VecCost += + TTI->getShuffleCost(TargetTransformInfo::SK_Alternate, VecTy, 0); + return VecCost - ScalarCost; + } + default: + llvm_unreachable("Unknown instruction"); + } +} + +bool BoUpSLP::isFullyVectorizableTinyTree() { + DEBUG(dbgs() << "SLP: Check whether the tree with height " << + VectorizableTree.size() << " is fully vectorizable .\n"); + + // We only handle trees of height 2. + if (VectorizableTree.size() != 2) + return false; + + // Handle splat stores. + if (!VectorizableTree[0].NeedToGather && isSplat(VectorizableTree[1].Scalars)) + return true; + + // Gathering cost would be too much for tiny trees. + if (VectorizableTree[0].NeedToGather || VectorizableTree[1].NeedToGather) + return false; + + return true; +} + +int BoUpSLP::getSpillCost() { + // Walk from the bottom of the tree to the top, tracking which values are + // live. When we see a call instruction that is not part of our tree, + // query TTI to see if there is a cost to keeping values live over it + // (for example, if spills and fills are required). + unsigned BundleWidth = VectorizableTree.front().Scalars.size(); + int Cost = 0; + + SmallPtrSet<Instruction*, 4> LiveValues; + Instruction *PrevInst = nullptr; + + for (unsigned N = 0; N < VectorizableTree.size(); ++N) { + Instruction *Inst = dyn_cast<Instruction>(VectorizableTree[N].Scalars[0]); + if (!Inst) + continue; + + if (!PrevInst) { + PrevInst = Inst; + continue; + } + + DEBUG( + dbgs() << "SLP: #LV: " << LiveValues.size(); + for (auto *X : LiveValues) + dbgs() << " " << X->getName(); + dbgs() << ", Looking at "; + Inst->dump(); + ); + + // Update LiveValues. + LiveValues.erase(PrevInst); + for (auto &J : PrevInst->operands()) { + if (isa<Instruction>(&*J) && ScalarToTreeEntry.count(&*J)) + LiveValues.insert(cast<Instruction>(&*J)); + } + + // Now find the sequence of instructions between PrevInst and Inst. + BasicBlock::reverse_iterator InstIt(Inst->getIterator()), + PrevInstIt(PrevInst->getIterator()); + --PrevInstIt; + while (InstIt != PrevInstIt) { + if (PrevInstIt == PrevInst->getParent()->rend()) { + PrevInstIt = Inst->getParent()->rbegin(); + continue; + } + + if (isa<CallInst>(&*PrevInstIt) && &*PrevInstIt != PrevInst) { + SmallVector<Type*, 4> V; + for (auto *II : LiveValues) + V.push_back(VectorType::get(II->getType(), BundleWidth)); + Cost += TTI->getCostOfKeepingLiveOverCall(V); + } + + ++PrevInstIt; + } + + PrevInst = Inst; + } + + DEBUG(dbgs() << "SLP: SpillCost=" << Cost << "\n"); + return Cost; +} + +int BoUpSLP::getTreeCost() { + int Cost = 0; + DEBUG(dbgs() << "SLP: Calculating cost for tree of size " << + VectorizableTree.size() << ".\n"); + + // We only vectorize tiny trees if it is fully vectorizable. + if (VectorizableTree.size() < 3 && !isFullyVectorizableTinyTree()) { + if (VectorizableTree.empty()) { + assert(!ExternalUses.size() && "We should not have any external users"); + } + return INT_MAX; + } + + unsigned BundleWidth = VectorizableTree[0].Scalars.size(); + + for (TreeEntry &TE : VectorizableTree) { + int C = getEntryCost(&TE); + DEBUG(dbgs() << "SLP: Adding cost " << C << " for bundle that starts with " + << TE.Scalars[0] << " .\n"); + Cost += C; + } + + SmallSet<Value *, 16> ExtractCostCalculated; + int ExtractCost = 0; + for (ExternalUser &EU : ExternalUses) { + // We only add extract cost once for the same scalar. + if (!ExtractCostCalculated.insert(EU.Scalar).second) + continue; + + // Uses by ephemeral values are free (because the ephemeral value will be + // removed prior to code generation, and so the extraction will be + // removed as well). + if (EphValues.count(EU.User)) + continue; + + VectorType *VecTy = VectorType::get(EU.Scalar->getType(), BundleWidth); + ExtractCost += TTI->getVectorInstrCost(Instruction::ExtractElement, VecTy, + EU.Lane); + } + + Cost += getSpillCost(); + + DEBUG(dbgs() << "SLP: Total Cost " << Cost + ExtractCost<< ".\n"); + return Cost + ExtractCost; +} + +int BoUpSLP::getGatherCost(Type *Ty) { + int Cost = 0; + for (unsigned i = 0, e = cast<VectorType>(Ty)->getNumElements(); i < e; ++i) + Cost += TTI->getVectorInstrCost(Instruction::InsertElement, Ty, i); + return Cost; +} + +int BoUpSLP::getGatherCost(ArrayRef<Value *> VL) { + // Find the type of the operands in VL. + Type *ScalarTy = VL[0]->getType(); + if (StoreInst *SI = dyn_cast<StoreInst>(VL[0])) + ScalarTy = SI->getValueOperand()->getType(); + VectorType *VecTy = VectorType::get(ScalarTy, VL.size()); + // Find the cost of inserting/extracting values from the vector. + return getGatherCost(VecTy); +} + +Value *BoUpSLP::getPointerOperand(Value *I) { + if (LoadInst *LI = dyn_cast<LoadInst>(I)) + return LI->getPointerOperand(); + if (StoreInst *SI = dyn_cast<StoreInst>(I)) + return SI->getPointerOperand(); + return nullptr; +} + +unsigned BoUpSLP::getAddressSpaceOperand(Value *I) { + if (LoadInst *L = dyn_cast<LoadInst>(I)) + return L->getPointerAddressSpace(); + if (StoreInst *S = dyn_cast<StoreInst>(I)) + return S->getPointerAddressSpace(); + return -1; +} + +bool BoUpSLP::isConsecutiveAccess(Value *A, Value *B, const DataLayout &DL) { + Value *PtrA = getPointerOperand(A); + Value *PtrB = getPointerOperand(B); + unsigned ASA = getAddressSpaceOperand(A); + unsigned ASB = getAddressSpaceOperand(B); + + // Check that the address spaces match and that the pointers are valid. + if (!PtrA || !PtrB || (ASA != ASB)) + return false; + + // Make sure that A and B are different pointers of the same type. + if (PtrA == PtrB || PtrA->getType() != PtrB->getType()) + return false; + + unsigned PtrBitWidth = DL.getPointerSizeInBits(ASA); + Type *Ty = cast<PointerType>(PtrA->getType())->getElementType(); + APInt Size(PtrBitWidth, DL.getTypeStoreSize(Ty)); + + APInt OffsetA(PtrBitWidth, 0), OffsetB(PtrBitWidth, 0); + PtrA = PtrA->stripAndAccumulateInBoundsConstantOffsets(DL, OffsetA); + PtrB = PtrB->stripAndAccumulateInBoundsConstantOffsets(DL, OffsetB); + + APInt OffsetDelta = OffsetB - OffsetA; + + // Check if they are based on the same pointer. That makes the offsets + // sufficient. + if (PtrA == PtrB) + return OffsetDelta == Size; + + // Compute the necessary base pointer delta to have the necessary final delta + // equal to the size. + APInt BaseDelta = Size - OffsetDelta; + + // Otherwise compute the distance with SCEV between the base pointers. + const SCEV *PtrSCEVA = SE->getSCEV(PtrA); + const SCEV *PtrSCEVB = SE->getSCEV(PtrB); + const SCEV *C = SE->getConstant(BaseDelta); + const SCEV *X = SE->getAddExpr(PtrSCEVA, C); + return X == PtrSCEVB; +} + +// Reorder commutative operations in alternate shuffle if the resulting vectors +// are consecutive loads. This would allow us to vectorize the tree. +// If we have something like- +// load a[0] - load b[0] +// load b[1] + load a[1] +// load a[2] - load b[2] +// load a[3] + load b[3] +// Reordering the second load b[1] load a[1] would allow us to vectorize this +// code. +void BoUpSLP::reorderAltShuffleOperands(ArrayRef<Value *> VL, + SmallVectorImpl<Value *> &Left, + SmallVectorImpl<Value *> &Right) { + const DataLayout &DL = F->getParent()->getDataLayout(); + + // Push left and right operands of binary operation into Left and Right + for (unsigned i = 0, e = VL.size(); i < e; ++i) { + Left.push_back(cast<Instruction>(VL[i])->getOperand(0)); + Right.push_back(cast<Instruction>(VL[i])->getOperand(1)); + } + + // Reorder if we have a commutative operation and consecutive access + // are on either side of the alternate instructions. + for (unsigned j = 0; j < VL.size() - 1; ++j) { + if (LoadInst *L = dyn_cast<LoadInst>(Left[j])) { + if (LoadInst *L1 = dyn_cast<LoadInst>(Right[j + 1])) { + Instruction *VL1 = cast<Instruction>(VL[j]); + Instruction *VL2 = cast<Instruction>(VL[j + 1]); + if (isConsecutiveAccess(L, L1, DL) && VL1->isCommutative()) { + std::swap(Left[j], Right[j]); + continue; + } else if (isConsecutiveAccess(L, L1, DL) && VL2->isCommutative()) { + std::swap(Left[j + 1], Right[j + 1]); + continue; + } + // else unchanged + } + } + if (LoadInst *L = dyn_cast<LoadInst>(Right[j])) { + if (LoadInst *L1 = dyn_cast<LoadInst>(Left[j + 1])) { + Instruction *VL1 = cast<Instruction>(VL[j]); + Instruction *VL2 = cast<Instruction>(VL[j + 1]); + if (isConsecutiveAccess(L, L1, DL) && VL1->isCommutative()) { + std::swap(Left[j], Right[j]); + continue; + } else if (isConsecutiveAccess(L, L1, DL) && VL2->isCommutative()) { + std::swap(Left[j + 1], Right[j + 1]); + continue; + } + // else unchanged + } + } + } +} + +// Return true if I should be commuted before adding it's left and right +// operands to the arrays Left and Right. +// +// The vectorizer is trying to either have all elements one side being +// instruction with the same opcode to enable further vectorization, or having +// a splat to lower the vectorizing cost. +static bool shouldReorderOperands(int i, Instruction &I, + SmallVectorImpl<Value *> &Left, + SmallVectorImpl<Value *> &Right, + bool AllSameOpcodeLeft, + bool AllSameOpcodeRight, bool SplatLeft, + bool SplatRight) { + Value *VLeft = I.getOperand(0); + Value *VRight = I.getOperand(1); + // If we have "SplatRight", try to see if commuting is needed to preserve it. + if (SplatRight) { + if (VRight == Right[i - 1]) + // Preserve SplatRight + return false; + if (VLeft == Right[i - 1]) { + // Commuting would preserve SplatRight, but we don't want to break + // SplatLeft either, i.e. preserve the original order if possible. + // (FIXME: why do we care?) + if (SplatLeft && VLeft == Left[i - 1]) + return false; + return true; + } + } + // Symmetrically handle Right side. + if (SplatLeft) { + if (VLeft == Left[i - 1]) + // Preserve SplatLeft + return false; + if (VRight == Left[i - 1]) + return true; + } + + Instruction *ILeft = dyn_cast<Instruction>(VLeft); + Instruction *IRight = dyn_cast<Instruction>(VRight); + + // If we have "AllSameOpcodeRight", try to see if the left operands preserves + // it and not the right, in this case we want to commute. + if (AllSameOpcodeRight) { + unsigned RightPrevOpcode = cast<Instruction>(Right[i - 1])->getOpcode(); + if (IRight && RightPrevOpcode == IRight->getOpcode()) + // Do not commute, a match on the right preserves AllSameOpcodeRight + return false; + if (ILeft && RightPrevOpcode == ILeft->getOpcode()) { + // We have a match and may want to commute, but first check if there is + // not also a match on the existing operands on the Left to preserve + // AllSameOpcodeLeft, i.e. preserve the original order if possible. + // (FIXME: why do we care?) + if (AllSameOpcodeLeft && ILeft && + cast<Instruction>(Left[i - 1])->getOpcode() == ILeft->getOpcode()) + return false; + return true; + } + } + // Symmetrically handle Left side. + if (AllSameOpcodeLeft) { + unsigned LeftPrevOpcode = cast<Instruction>(Left[i - 1])->getOpcode(); + if (ILeft && LeftPrevOpcode == ILeft->getOpcode()) + return false; + if (IRight && LeftPrevOpcode == IRight->getOpcode()) + return true; + } + return false; +} + +void BoUpSLP::reorderInputsAccordingToOpcode(ArrayRef<Value *> VL, + SmallVectorImpl<Value *> &Left, + SmallVectorImpl<Value *> &Right) { + + if (VL.size()) { + // Peel the first iteration out of the loop since there's nothing + // interesting to do anyway and it simplifies the checks in the loop. + auto VLeft = cast<Instruction>(VL[0])->getOperand(0); + auto VRight = cast<Instruction>(VL[0])->getOperand(1); + if (!isa<Instruction>(VRight) && isa<Instruction>(VLeft)) + // Favor having instruction to the right. FIXME: why? + std::swap(VLeft, VRight); + Left.push_back(VLeft); + Right.push_back(VRight); + } + + // Keep track if we have instructions with all the same opcode on one side. + bool AllSameOpcodeLeft = isa<Instruction>(Left[0]); + bool AllSameOpcodeRight = isa<Instruction>(Right[0]); + // Keep track if we have one side with all the same value (broadcast). + bool SplatLeft = true; + bool SplatRight = true; + + for (unsigned i = 1, e = VL.size(); i != e; ++i) { + Instruction *I = cast<Instruction>(VL[i]); + assert(I->isCommutative() && "Can only process commutative instruction"); + // Commute to favor either a splat or maximizing having the same opcodes on + // one side. + if (shouldReorderOperands(i, *I, Left, Right, AllSameOpcodeLeft, + AllSameOpcodeRight, SplatLeft, SplatRight)) { + Left.push_back(I->getOperand(1)); + Right.push_back(I->getOperand(0)); + } else { + Left.push_back(I->getOperand(0)); + Right.push_back(I->getOperand(1)); + } + // Update Splat* and AllSameOpcode* after the insertion. + SplatRight = SplatRight && (Right[i - 1] == Right[i]); + SplatLeft = SplatLeft && (Left[i - 1] == Left[i]); + AllSameOpcodeLeft = AllSameOpcodeLeft && isa<Instruction>(Left[i]) && + (cast<Instruction>(Left[i - 1])->getOpcode() == + cast<Instruction>(Left[i])->getOpcode()); + AllSameOpcodeRight = AllSameOpcodeRight && isa<Instruction>(Right[i]) && + (cast<Instruction>(Right[i - 1])->getOpcode() == + cast<Instruction>(Right[i])->getOpcode()); + } + + // If one operand end up being broadcast, return this operand order. + if (SplatRight || SplatLeft) + return; + + const DataLayout &DL = F->getParent()->getDataLayout(); + + // Finally check if we can get longer vectorizable chain by reordering + // without breaking the good operand order detected above. + // E.g. If we have something like- + // load a[0] load b[0] + // load b[1] load a[1] + // load a[2] load b[2] + // load a[3] load b[3] + // Reordering the second load b[1] load a[1] would allow us to vectorize + // this code and we still retain AllSameOpcode property. + // FIXME: This load reordering might break AllSameOpcode in some rare cases + // such as- + // add a[0],c[0] load b[0] + // add a[1],c[2] load b[1] + // b[2] load b[2] + // add a[3],c[3] load b[3] + for (unsigned j = 0; j < VL.size() - 1; ++j) { + if (LoadInst *L = dyn_cast<LoadInst>(Left[j])) { + if (LoadInst *L1 = dyn_cast<LoadInst>(Right[j + 1])) { + if (isConsecutiveAccess(L, L1, DL)) { + std::swap(Left[j + 1], Right[j + 1]); + continue; + } + } + } + if (LoadInst *L = dyn_cast<LoadInst>(Right[j])) { + if (LoadInst *L1 = dyn_cast<LoadInst>(Left[j + 1])) { + if (isConsecutiveAccess(L, L1, DL)) { + std::swap(Left[j + 1], Right[j + 1]); + continue; + } + } + } + // else unchanged + } +} + +void BoUpSLP::setInsertPointAfterBundle(ArrayRef<Value *> VL) { + Instruction *VL0 = cast<Instruction>(VL[0]); + BasicBlock::iterator NextInst(VL0); + ++NextInst; + Builder.SetInsertPoint(VL0->getParent(), NextInst); + Builder.SetCurrentDebugLocation(VL0->getDebugLoc()); +} + +Value *BoUpSLP::Gather(ArrayRef<Value *> VL, VectorType *Ty) { + Value *Vec = UndefValue::get(Ty); + // Generate the 'InsertElement' instruction. + for (unsigned i = 0; i < Ty->getNumElements(); ++i) { + Vec = Builder.CreateInsertElement(Vec, VL[i], Builder.getInt32(i)); + if (Instruction *Insrt = dyn_cast<Instruction>(Vec)) { + GatherSeq.insert(Insrt); + CSEBlocks.insert(Insrt->getParent()); + + // Add to our 'need-to-extract' list. + if (ScalarToTreeEntry.count(VL[i])) { + int Idx = ScalarToTreeEntry[VL[i]]; + TreeEntry *E = &VectorizableTree[Idx]; + // Find which lane we need to extract. + int FoundLane = -1; + for (unsigned Lane = 0, LE = VL.size(); Lane != LE; ++Lane) { + // Is this the lane of the scalar that we are looking for ? + if (E->Scalars[Lane] == VL[i]) { + FoundLane = Lane; + break; + } + } + assert(FoundLane >= 0 && "Could not find the correct lane"); + ExternalUses.push_back(ExternalUser(VL[i], Insrt, FoundLane)); + } + } + } + + return Vec; +} + +Value *BoUpSLP::alreadyVectorized(ArrayRef<Value *> VL) const { + SmallDenseMap<Value*, int>::const_iterator Entry + = ScalarToTreeEntry.find(VL[0]); + if (Entry != ScalarToTreeEntry.end()) { + int Idx = Entry->second; + const TreeEntry *En = &VectorizableTree[Idx]; + if (En->isSame(VL) && En->VectorizedValue) + return En->VectorizedValue; + } + return nullptr; +} + +Value *BoUpSLP::vectorizeTree(ArrayRef<Value *> VL) { + if (ScalarToTreeEntry.count(VL[0])) { + int Idx = ScalarToTreeEntry[VL[0]]; + TreeEntry *E = &VectorizableTree[Idx]; + if (E->isSame(VL)) + return vectorizeTree(E); + } + + Type *ScalarTy = VL[0]->getType(); + if (StoreInst *SI = dyn_cast<StoreInst>(VL[0])) + ScalarTy = SI->getValueOperand()->getType(); + VectorType *VecTy = VectorType::get(ScalarTy, VL.size()); + + return Gather(VL, VecTy); +} + +Value *BoUpSLP::vectorizeTree(TreeEntry *E) { + IRBuilder<>::InsertPointGuard Guard(Builder); + + if (E->VectorizedValue) { + DEBUG(dbgs() << "SLP: Diamond merged for " << *E->Scalars[0] << ".\n"); + return E->VectorizedValue; + } + + Instruction *VL0 = cast<Instruction>(E->Scalars[0]); + Type *ScalarTy = VL0->getType(); + if (StoreInst *SI = dyn_cast<StoreInst>(VL0)) + ScalarTy = SI->getValueOperand()->getType(); + VectorType *VecTy = VectorType::get(ScalarTy, E->Scalars.size()); + + if (E->NeedToGather) { + setInsertPointAfterBundle(E->Scalars); + return Gather(E->Scalars, VecTy); + } + + const DataLayout &DL = F->getParent()->getDataLayout(); + unsigned Opcode = getSameOpcode(E->Scalars); + + switch (Opcode) { + case Instruction::PHI: { + PHINode *PH = dyn_cast<PHINode>(VL0); + Builder.SetInsertPoint(PH->getParent()->getFirstNonPHI()); + Builder.SetCurrentDebugLocation(PH->getDebugLoc()); + PHINode *NewPhi = Builder.CreatePHI(VecTy, PH->getNumIncomingValues()); + E->VectorizedValue = NewPhi; + + // PHINodes may have multiple entries from the same block. We want to + // visit every block once. + SmallSet<BasicBlock*, 4> VisitedBBs; + + for (unsigned i = 0, e = PH->getNumIncomingValues(); i < e; ++i) { + ValueList Operands; + BasicBlock *IBB = PH->getIncomingBlock(i); + + if (!VisitedBBs.insert(IBB).second) { + NewPhi->addIncoming(NewPhi->getIncomingValueForBlock(IBB), IBB); + continue; + } + + // Prepare the operand vector. + for (Value *V : E->Scalars) + Operands.push_back(cast<PHINode>(V)->getIncomingValueForBlock(IBB)); + + Builder.SetInsertPoint(IBB->getTerminator()); + Builder.SetCurrentDebugLocation(PH->getDebugLoc()); + Value *Vec = vectorizeTree(Operands); + NewPhi->addIncoming(Vec, IBB); + } + + assert(NewPhi->getNumIncomingValues() == PH->getNumIncomingValues() && + "Invalid number of incoming values"); + return NewPhi; + } + + case Instruction::ExtractElement: { + if (CanReuseExtract(E->Scalars)) { + Value *V = VL0->getOperand(0); + E->VectorizedValue = V; + return V; + } + return Gather(E->Scalars, VecTy); + } + case Instruction::ZExt: + case Instruction::SExt: + case Instruction::FPToUI: + case Instruction::FPToSI: + case Instruction::FPExt: + case Instruction::PtrToInt: + case Instruction::IntToPtr: + case Instruction::SIToFP: + case Instruction::UIToFP: + case Instruction::Trunc: + case Instruction::FPTrunc: + case Instruction::BitCast: { + ValueList INVL; + for (Value *V : E->Scalars) + INVL.push_back(cast<Instruction>(V)->getOperand(0)); + + setInsertPointAfterBundle(E->Scalars); + + Value *InVec = vectorizeTree(INVL); + + if (Value *V = alreadyVectorized(E->Scalars)) + return V; + + CastInst *CI = dyn_cast<CastInst>(VL0); + Value *V = Builder.CreateCast(CI->getOpcode(), InVec, VecTy); + E->VectorizedValue = V; + ++NumVectorInstructions; + return V; + } + case Instruction::FCmp: + case Instruction::ICmp: { + ValueList LHSV, RHSV; + for (Value *V : E->Scalars) { + LHSV.push_back(cast<Instruction>(V)->getOperand(0)); + RHSV.push_back(cast<Instruction>(V)->getOperand(1)); + } + + setInsertPointAfterBundle(E->Scalars); + + Value *L = vectorizeTree(LHSV); + Value *R = vectorizeTree(RHSV); + + if (Value *V = alreadyVectorized(E->Scalars)) + return V; + + CmpInst::Predicate P0 = cast<CmpInst>(VL0)->getPredicate(); + Value *V; + if (Opcode == Instruction::FCmp) + V = Builder.CreateFCmp(P0, L, R); + else + V = Builder.CreateICmp(P0, L, R); + + E->VectorizedValue = V; + ++NumVectorInstructions; + return V; + } + case Instruction::Select: { + ValueList TrueVec, FalseVec, CondVec; + for (Value *V : E->Scalars) { + CondVec.push_back(cast<Instruction>(V)->getOperand(0)); + TrueVec.push_back(cast<Instruction>(V)->getOperand(1)); + FalseVec.push_back(cast<Instruction>(V)->getOperand(2)); + } + + setInsertPointAfterBundle(E->Scalars); + + Value *Cond = vectorizeTree(CondVec); + Value *True = vectorizeTree(TrueVec); + Value *False = vectorizeTree(FalseVec); + + if (Value *V = alreadyVectorized(E->Scalars)) + return V; + + Value *V = Builder.CreateSelect(Cond, True, False); + E->VectorizedValue = V; + ++NumVectorInstructions; + return V; + } + case Instruction::Add: + case Instruction::FAdd: + case Instruction::Sub: + case Instruction::FSub: + case Instruction::Mul: + case Instruction::FMul: + case Instruction::UDiv: + case Instruction::SDiv: + case Instruction::FDiv: + case Instruction::URem: + case Instruction::SRem: + case Instruction::FRem: + case Instruction::Shl: + case Instruction::LShr: + case Instruction::AShr: + case Instruction::And: + case Instruction::Or: + case Instruction::Xor: { + ValueList LHSVL, RHSVL; + if (isa<BinaryOperator>(VL0) && VL0->isCommutative()) + reorderInputsAccordingToOpcode(E->Scalars, LHSVL, RHSVL); + else + for (Value *V : E->Scalars) { + LHSVL.push_back(cast<Instruction>(V)->getOperand(0)); + RHSVL.push_back(cast<Instruction>(V)->getOperand(1)); + } + + setInsertPointAfterBundle(E->Scalars); + + Value *LHS = vectorizeTree(LHSVL); + Value *RHS = vectorizeTree(RHSVL); + + if (LHS == RHS && isa<Instruction>(LHS)) { + assert((VL0->getOperand(0) == VL0->getOperand(1)) && "Invalid order"); + } + + if (Value *V = alreadyVectorized(E->Scalars)) + return V; + + BinaryOperator *BinOp = cast<BinaryOperator>(VL0); + Value *V = Builder.CreateBinOp(BinOp->getOpcode(), LHS, RHS); + E->VectorizedValue = V; + propagateIRFlags(E->VectorizedValue, E->Scalars); + ++NumVectorInstructions; + + if (Instruction *I = dyn_cast<Instruction>(V)) + return propagateMetadata(I, E->Scalars); + + return V; + } + case Instruction::Load: { + // Loads are inserted at the head of the tree because we don't want to + // sink them all the way down past store instructions. + setInsertPointAfterBundle(E->Scalars); + + LoadInst *LI = cast<LoadInst>(VL0); + Type *ScalarLoadTy = LI->getType(); + unsigned AS = LI->getPointerAddressSpace(); + + Value *VecPtr = Builder.CreateBitCast(LI->getPointerOperand(), + VecTy->getPointerTo(AS)); + + // The pointer operand uses an in-tree scalar so we add the new BitCast to + // ExternalUses list to make sure that an extract will be generated in the + // future. + if (ScalarToTreeEntry.count(LI->getPointerOperand())) + ExternalUses.push_back( + ExternalUser(LI->getPointerOperand(), cast<User>(VecPtr), 0)); + + unsigned Alignment = LI->getAlignment(); + LI = Builder.CreateLoad(VecPtr); + if (!Alignment) { + Alignment = DL.getABITypeAlignment(ScalarLoadTy); + } + LI->setAlignment(Alignment); + E->VectorizedValue = LI; + ++NumVectorInstructions; + return propagateMetadata(LI, E->Scalars); + } + case Instruction::Store: { + StoreInst *SI = cast<StoreInst>(VL0); + unsigned Alignment = SI->getAlignment(); + unsigned AS = SI->getPointerAddressSpace(); + + ValueList ValueOp; + for (Value *V : E->Scalars) + ValueOp.push_back(cast<StoreInst>(V)->getValueOperand()); + + setInsertPointAfterBundle(E->Scalars); + + Value *VecValue = vectorizeTree(ValueOp); + Value *VecPtr = Builder.CreateBitCast(SI->getPointerOperand(), + VecTy->getPointerTo(AS)); + StoreInst *S = Builder.CreateStore(VecValue, VecPtr); + + // The pointer operand uses an in-tree scalar so we add the new BitCast to + // ExternalUses list to make sure that an extract will be generated in the + // future. + if (ScalarToTreeEntry.count(SI->getPointerOperand())) + ExternalUses.push_back( + ExternalUser(SI->getPointerOperand(), cast<User>(VecPtr), 0)); + + if (!Alignment) { + Alignment = DL.getABITypeAlignment(SI->getValueOperand()->getType()); + } + S->setAlignment(Alignment); + E->VectorizedValue = S; + ++NumVectorInstructions; + return propagateMetadata(S, E->Scalars); + } + case Instruction::GetElementPtr: { + setInsertPointAfterBundle(E->Scalars); + + ValueList Op0VL; + for (Value *V : E->Scalars) + Op0VL.push_back(cast<GetElementPtrInst>(V)->getOperand(0)); + + Value *Op0 = vectorizeTree(Op0VL); + + std::vector<Value *> OpVecs; + for (int j = 1, e = cast<GetElementPtrInst>(VL0)->getNumOperands(); j < e; + ++j) { + ValueList OpVL; + for (Value *V : E->Scalars) + OpVL.push_back(cast<GetElementPtrInst>(V)->getOperand(j)); + + Value *OpVec = vectorizeTree(OpVL); + OpVecs.push_back(OpVec); + } + + Value *V = Builder.CreateGEP( + cast<GetElementPtrInst>(VL0)->getSourceElementType(), Op0, OpVecs); + E->VectorizedValue = V; + ++NumVectorInstructions; + + if (Instruction *I = dyn_cast<Instruction>(V)) + return propagateMetadata(I, E->Scalars); + + return V; + } + case Instruction::Call: { + CallInst *CI = cast<CallInst>(VL0); + setInsertPointAfterBundle(E->Scalars); + Function *FI; + Intrinsic::ID IID = Intrinsic::not_intrinsic; + Value *ScalarArg = nullptr; + if (CI && (FI = CI->getCalledFunction())) { + IID = FI->getIntrinsicID(); + } + std::vector<Value *> OpVecs; + for (int j = 0, e = CI->getNumArgOperands(); j < e; ++j) { + ValueList OpVL; + // ctlz,cttz and powi are special intrinsics whose second argument is + // a scalar. This argument should not be vectorized. + if (hasVectorInstrinsicScalarOpd(IID, 1) && j == 1) { + CallInst *CEI = cast<CallInst>(E->Scalars[0]); + ScalarArg = CEI->getArgOperand(j); + OpVecs.push_back(CEI->getArgOperand(j)); + continue; + } + for (Value *V : E->Scalars) { + CallInst *CEI = cast<CallInst>(V); + OpVL.push_back(CEI->getArgOperand(j)); + } + + Value *OpVec = vectorizeTree(OpVL); + DEBUG(dbgs() << "SLP: OpVec[" << j << "]: " << *OpVec << "\n"); + OpVecs.push_back(OpVec); + } + + Module *M = F->getParent(); + Intrinsic::ID ID = getIntrinsicIDForCall(CI, TLI); + Type *Tys[] = { VectorType::get(CI->getType(), E->Scalars.size()) }; + Function *CF = Intrinsic::getDeclaration(M, ID, Tys); + Value *V = Builder.CreateCall(CF, OpVecs); + + // The scalar argument uses an in-tree scalar so we add the new vectorized + // call to ExternalUses list to make sure that an extract will be + // generated in the future. + if (ScalarArg && ScalarToTreeEntry.count(ScalarArg)) + ExternalUses.push_back(ExternalUser(ScalarArg, cast<User>(V), 0)); + + E->VectorizedValue = V; + ++NumVectorInstructions; + return V; + } + case Instruction::ShuffleVector: { + ValueList LHSVL, RHSVL; + assert(isa<BinaryOperator>(VL0) && "Invalid Shuffle Vector Operand"); + reorderAltShuffleOperands(E->Scalars, LHSVL, RHSVL); + setInsertPointAfterBundle(E->Scalars); + + Value *LHS = vectorizeTree(LHSVL); + Value *RHS = vectorizeTree(RHSVL); + + if (Value *V = alreadyVectorized(E->Scalars)) + return V; + + // Create a vector of LHS op1 RHS + BinaryOperator *BinOp0 = cast<BinaryOperator>(VL0); + Value *V0 = Builder.CreateBinOp(BinOp0->getOpcode(), LHS, RHS); + + // Create a vector of LHS op2 RHS + Instruction *VL1 = cast<Instruction>(E->Scalars[1]); + BinaryOperator *BinOp1 = cast<BinaryOperator>(VL1); + Value *V1 = Builder.CreateBinOp(BinOp1->getOpcode(), LHS, RHS); + + // Create shuffle to take alternate operations from the vector. + // Also, gather up odd and even scalar ops to propagate IR flags to + // each vector operation. + ValueList OddScalars, EvenScalars; + unsigned e = E->Scalars.size(); + SmallVector<Constant *, 8> Mask(e); + for (unsigned i = 0; i < e; ++i) { + if (i & 1) { + Mask[i] = Builder.getInt32(e + i); + OddScalars.push_back(E->Scalars[i]); + } else { + Mask[i] = Builder.getInt32(i); + EvenScalars.push_back(E->Scalars[i]); + } + } + + Value *ShuffleMask = ConstantVector::get(Mask); + propagateIRFlags(V0, EvenScalars); + propagateIRFlags(V1, OddScalars); + + Value *V = Builder.CreateShuffleVector(V0, V1, ShuffleMask); + E->VectorizedValue = V; + ++NumVectorInstructions; + if (Instruction *I = dyn_cast<Instruction>(V)) + return propagateMetadata(I, E->Scalars); + + return V; + } + default: + llvm_unreachable("unknown inst"); + } + return nullptr; +} + +Value *BoUpSLP::vectorizeTree() { + + // All blocks must be scheduled before any instructions are inserted. + for (auto &BSIter : BlocksSchedules) { + scheduleBlock(BSIter.second.get()); + } + + Builder.SetInsertPoint(&F->getEntryBlock().front()); + vectorizeTree(&VectorizableTree[0]); + + DEBUG(dbgs() << "SLP: Extracting " << ExternalUses.size() << " values .\n"); + + // Extract all of the elements with the external uses. + for (UserList::iterator it = ExternalUses.begin(), e = ExternalUses.end(); + it != e; ++it) { + Value *Scalar = it->Scalar; + llvm::User *User = it->User; + + // Skip users that we already RAUW. This happens when one instruction + // has multiple uses of the same value. + if (std::find(Scalar->user_begin(), Scalar->user_end(), User) == + Scalar->user_end()) + continue; + assert(ScalarToTreeEntry.count(Scalar) && "Invalid scalar"); + + int Idx = ScalarToTreeEntry[Scalar]; + TreeEntry *E = &VectorizableTree[Idx]; + assert(!E->NeedToGather && "Extracting from a gather list"); + + Value *Vec = E->VectorizedValue; + assert(Vec && "Can't find vectorizable value"); + + Value *Lane = Builder.getInt32(it->Lane); + // Generate extracts for out-of-tree users. + // Find the insertion point for the extractelement lane. + if (isa<Instruction>(Vec)){ + if (PHINode *PH = dyn_cast<PHINode>(User)) { + for (int i = 0, e = PH->getNumIncomingValues(); i != e; ++i) { + if (PH->getIncomingValue(i) == Scalar) { + Builder.SetInsertPoint(PH->getIncomingBlock(i)->getTerminator()); + Value *Ex = Builder.CreateExtractElement(Vec, Lane); + CSEBlocks.insert(PH->getIncomingBlock(i)); + PH->setOperand(i, Ex); + } + } + } else { + Builder.SetInsertPoint(cast<Instruction>(User)); + Value *Ex = Builder.CreateExtractElement(Vec, Lane); + CSEBlocks.insert(cast<Instruction>(User)->getParent()); + User->replaceUsesOfWith(Scalar, Ex); + } + } else { + Builder.SetInsertPoint(&F->getEntryBlock().front()); + Value *Ex = Builder.CreateExtractElement(Vec, Lane); + CSEBlocks.insert(&F->getEntryBlock()); + User->replaceUsesOfWith(Scalar, Ex); + } + + DEBUG(dbgs() << "SLP: Replaced:" << *User << ".\n"); + } + + // For each vectorized value: + for (int EIdx = 0, EE = VectorizableTree.size(); EIdx < EE; ++EIdx) { + TreeEntry *Entry = &VectorizableTree[EIdx]; + + // For each lane: + for (int Lane = 0, LE = Entry->Scalars.size(); Lane != LE; ++Lane) { + Value *Scalar = Entry->Scalars[Lane]; + // No need to handle users of gathered values. + if (Entry->NeedToGather) + continue; + + assert(Entry->VectorizedValue && "Can't find vectorizable value"); + + Type *Ty = Scalar->getType(); + if (!Ty->isVoidTy()) { +#ifndef NDEBUG + for (User *U : Scalar->users()) { + DEBUG(dbgs() << "SLP: \tvalidating user:" << *U << ".\n"); + + assert((ScalarToTreeEntry.count(U) || + // It is legal to replace users in the ignorelist by undef. + (std::find(UserIgnoreList.begin(), UserIgnoreList.end(), U) != + UserIgnoreList.end())) && + "Replacing out-of-tree value with undef"); + } +#endif + Value *Undef = UndefValue::get(Ty); + Scalar->replaceAllUsesWith(Undef); + } + DEBUG(dbgs() << "SLP: \tErasing scalar:" << *Scalar << ".\n"); + eraseInstruction(cast<Instruction>(Scalar)); + } + } + + Builder.ClearInsertionPoint(); + + return VectorizableTree[0].VectorizedValue; +} + +void BoUpSLP::optimizeGatherSequence() { + DEBUG(dbgs() << "SLP: Optimizing " << GatherSeq.size() + << " gather sequences instructions.\n"); + // LICM InsertElementInst sequences. + for (SetVector<Instruction *>::iterator it = GatherSeq.begin(), + e = GatherSeq.end(); it != e; ++it) { + InsertElementInst *Insert = dyn_cast<InsertElementInst>(*it); + + if (!Insert) + continue; + + // Check if this block is inside a loop. + Loop *L = LI->getLoopFor(Insert->getParent()); + if (!L) + continue; + + // Check if it has a preheader. + BasicBlock *PreHeader = L->getLoopPreheader(); + if (!PreHeader) + continue; + + // If the vector or the element that we insert into it are + // instructions that are defined in this basic block then we can't + // hoist this instruction. + Instruction *CurrVec = dyn_cast<Instruction>(Insert->getOperand(0)); + Instruction *NewElem = dyn_cast<Instruction>(Insert->getOperand(1)); + if (CurrVec && L->contains(CurrVec)) + continue; + if (NewElem && L->contains(NewElem)) + continue; + + // We can hoist this instruction. Move it to the pre-header. + Insert->moveBefore(PreHeader->getTerminator()); + } + + // Make a list of all reachable blocks in our CSE queue. + SmallVector<const DomTreeNode *, 8> CSEWorkList; + CSEWorkList.reserve(CSEBlocks.size()); + for (BasicBlock *BB : CSEBlocks) + if (DomTreeNode *N = DT->getNode(BB)) { + assert(DT->isReachableFromEntry(N)); + CSEWorkList.push_back(N); + } + + // Sort blocks by domination. This ensures we visit a block after all blocks + // dominating it are visited. + std::stable_sort(CSEWorkList.begin(), CSEWorkList.end(), + [this](const DomTreeNode *A, const DomTreeNode *B) { + return DT->properlyDominates(A, B); + }); + + // Perform O(N^2) search over the gather sequences and merge identical + // instructions. TODO: We can further optimize this scan if we split the + // instructions into different buckets based on the insert lane. + SmallVector<Instruction *, 16> Visited; + for (auto I = CSEWorkList.begin(), E = CSEWorkList.end(); I != E; ++I) { + assert((I == CSEWorkList.begin() || !DT->dominates(*I, *std::prev(I))) && + "Worklist not sorted properly!"); + BasicBlock *BB = (*I)->getBlock(); + // For all instructions in blocks containing gather sequences: + for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e;) { + Instruction *In = &*it++; + if (!isa<InsertElementInst>(In) && !isa<ExtractElementInst>(In)) + continue; + + // Check if we can replace this instruction with any of the + // visited instructions. + for (SmallVectorImpl<Instruction *>::iterator v = Visited.begin(), + ve = Visited.end(); + v != ve; ++v) { + if (In->isIdenticalTo(*v) && + DT->dominates((*v)->getParent(), In->getParent())) { + In->replaceAllUsesWith(*v); + eraseInstruction(In); + In = nullptr; + break; + } + } + if (In) { + assert(std::find(Visited.begin(), Visited.end(), In) == Visited.end()); + Visited.push_back(In); + } + } + } + CSEBlocks.clear(); + GatherSeq.clear(); +} + +// Groups the instructions to a bundle (which is then a single scheduling entity) +// and schedules instructions until the bundle gets ready. +bool BoUpSLP::BlockScheduling::tryScheduleBundle(ArrayRef<Value *> VL, + BoUpSLP *SLP) { + if (isa<PHINode>(VL[0])) + return true; + + // Initialize the instruction bundle. + Instruction *OldScheduleEnd = ScheduleEnd; + ScheduleData *PrevInBundle = nullptr; + ScheduleData *Bundle = nullptr; + bool ReSchedule = false; + DEBUG(dbgs() << "SLP: bundle: " << *VL[0] << "\n"); + + // Make sure that the scheduling region contains all + // instructions of the bundle. + for (Value *V : VL) { + if (!extendSchedulingRegion(V)) + return false; + } + + for (Value *V : VL) { + ScheduleData *BundleMember = getScheduleData(V); + assert(BundleMember && + "no ScheduleData for bundle member (maybe not in same basic block)"); + if (BundleMember->IsScheduled) { + // A bundle member was scheduled as single instruction before and now + // needs to be scheduled as part of the bundle. We just get rid of the + // existing schedule. + DEBUG(dbgs() << "SLP: reset schedule because " << *BundleMember + << " was already scheduled\n"); + ReSchedule = true; + } + assert(BundleMember->isSchedulingEntity() && + "bundle member already part of other bundle"); + if (PrevInBundle) { + PrevInBundle->NextInBundle = BundleMember; + } else { + Bundle = BundleMember; + } + BundleMember->UnscheduledDepsInBundle = 0; + Bundle->UnscheduledDepsInBundle += BundleMember->UnscheduledDeps; + + // Group the instructions to a bundle. + BundleMember->FirstInBundle = Bundle; + PrevInBundle = BundleMember; + } + if (ScheduleEnd != OldScheduleEnd) { + // The scheduling region got new instructions at the lower end (or it is a + // new region for the first bundle). This makes it necessary to + // recalculate all dependencies. + // It is seldom that this needs to be done a second time after adding the + // initial bundle to the region. + for (auto *I = ScheduleStart; I != ScheduleEnd; I = I->getNextNode()) { + ScheduleData *SD = getScheduleData(I); + SD->clearDependencies(); + } + ReSchedule = true; + } + if (ReSchedule) { + resetSchedule(); + initialFillReadyList(ReadyInsts); + } + + DEBUG(dbgs() << "SLP: try schedule bundle " << *Bundle << " in block " + << BB->getName() << "\n"); + + calculateDependencies(Bundle, true, SLP); + + // Now try to schedule the new bundle. As soon as the bundle is "ready" it + // means that there are no cyclic dependencies and we can schedule it. + // Note that's important that we don't "schedule" the bundle yet (see + // cancelScheduling). + while (!Bundle->isReady() && !ReadyInsts.empty()) { + + ScheduleData *pickedSD = ReadyInsts.back(); + ReadyInsts.pop_back(); + + if (pickedSD->isSchedulingEntity() && pickedSD->isReady()) { + schedule(pickedSD, ReadyInsts); + } + } + if (!Bundle->isReady()) { + cancelScheduling(VL); + return false; + } + return true; +} + +void BoUpSLP::BlockScheduling::cancelScheduling(ArrayRef<Value *> VL) { + if (isa<PHINode>(VL[0])) + return; + + ScheduleData *Bundle = getScheduleData(VL[0]); + DEBUG(dbgs() << "SLP: cancel scheduling of " << *Bundle << "\n"); + assert(!Bundle->IsScheduled && + "Can't cancel bundle which is already scheduled"); + assert(Bundle->isSchedulingEntity() && Bundle->isPartOfBundle() && + "tried to unbundle something which is not a bundle"); + + // Un-bundle: make single instructions out of the bundle. + ScheduleData *BundleMember = Bundle; + while (BundleMember) { + assert(BundleMember->FirstInBundle == Bundle && "corrupt bundle links"); + BundleMember->FirstInBundle = BundleMember; + ScheduleData *Next = BundleMember->NextInBundle; + BundleMember->NextInBundle = nullptr; + BundleMember->UnscheduledDepsInBundle = BundleMember->UnscheduledDeps; + if (BundleMember->UnscheduledDepsInBundle == 0) { + ReadyInsts.insert(BundleMember); + } + BundleMember = Next; + } +} + +bool BoUpSLP::BlockScheduling::extendSchedulingRegion(Value *V) { + if (getScheduleData(V)) + return true; + Instruction *I = dyn_cast<Instruction>(V); + assert(I && "bundle member must be an instruction"); + assert(!isa<PHINode>(I) && "phi nodes don't need to be scheduled"); + if (!ScheduleStart) { + // It's the first instruction in the new region. + initScheduleData(I, I->getNextNode(), nullptr, nullptr); + ScheduleStart = I; + ScheduleEnd = I->getNextNode(); + assert(ScheduleEnd && "tried to vectorize a TerminatorInst?"); + DEBUG(dbgs() << "SLP: initialize schedule region to " << *I << "\n"); + return true; + } + // Search up and down at the same time, because we don't know if the new + // instruction is above or below the existing scheduling region. + BasicBlock::reverse_iterator UpIter(ScheduleStart->getIterator()); + BasicBlock::reverse_iterator UpperEnd = BB->rend(); + BasicBlock::iterator DownIter(ScheduleEnd); + BasicBlock::iterator LowerEnd = BB->end(); + for (;;) { + if (++ScheduleRegionSize > ScheduleRegionSizeLimit) { + DEBUG(dbgs() << "SLP: exceeded schedule region size limit\n"); + return false; + } + + if (UpIter != UpperEnd) { + if (&*UpIter == I) { + initScheduleData(I, ScheduleStart, nullptr, FirstLoadStoreInRegion); + ScheduleStart = I; + DEBUG(dbgs() << "SLP: extend schedule region start to " << *I << "\n"); + return true; + } + UpIter++; + } + if (DownIter != LowerEnd) { + if (&*DownIter == I) { + initScheduleData(ScheduleEnd, I->getNextNode(), LastLoadStoreInRegion, + nullptr); + ScheduleEnd = I->getNextNode(); + assert(ScheduleEnd && "tried to vectorize a TerminatorInst?"); + DEBUG(dbgs() << "SLP: extend schedule region end to " << *I << "\n"); + return true; + } + DownIter++; + } + assert((UpIter != UpperEnd || DownIter != LowerEnd) && + "instruction not found in block"); + } + return true; +} + +void BoUpSLP::BlockScheduling::initScheduleData(Instruction *FromI, + Instruction *ToI, + ScheduleData *PrevLoadStore, + ScheduleData *NextLoadStore) { + ScheduleData *CurrentLoadStore = PrevLoadStore; + for (Instruction *I = FromI; I != ToI; I = I->getNextNode()) { + ScheduleData *SD = ScheduleDataMap[I]; + if (!SD) { + // Allocate a new ScheduleData for the instruction. + if (ChunkPos >= ChunkSize) { + ScheduleDataChunks.push_back( + llvm::make_unique<ScheduleData[]>(ChunkSize)); + ChunkPos = 0; + } + SD = &(ScheduleDataChunks.back()[ChunkPos++]); + ScheduleDataMap[I] = SD; + SD->Inst = I; + } + assert(!isInSchedulingRegion(SD) && + "new ScheduleData already in scheduling region"); + SD->init(SchedulingRegionID); + + if (I->mayReadOrWriteMemory()) { + // Update the linked list of memory accessing instructions. + if (CurrentLoadStore) { + CurrentLoadStore->NextLoadStore = SD; + } else { + FirstLoadStoreInRegion = SD; + } + CurrentLoadStore = SD; + } + } + if (NextLoadStore) { + if (CurrentLoadStore) + CurrentLoadStore->NextLoadStore = NextLoadStore; + } else { + LastLoadStoreInRegion = CurrentLoadStore; + } +} + +void BoUpSLP::BlockScheduling::calculateDependencies(ScheduleData *SD, + bool InsertInReadyList, + BoUpSLP *SLP) { + assert(SD->isSchedulingEntity()); + + SmallVector<ScheduleData *, 10> WorkList; + WorkList.push_back(SD); + + while (!WorkList.empty()) { + ScheduleData *SD = WorkList.back(); + WorkList.pop_back(); + + ScheduleData *BundleMember = SD; + while (BundleMember) { + assert(isInSchedulingRegion(BundleMember)); + if (!BundleMember->hasValidDependencies()) { + + DEBUG(dbgs() << "SLP: update deps of " << *BundleMember << "\n"); + BundleMember->Dependencies = 0; + BundleMember->resetUnscheduledDeps(); + + // Handle def-use chain dependencies. + for (User *U : BundleMember->Inst->users()) { + if (isa<Instruction>(U)) { + ScheduleData *UseSD = getScheduleData(U); + if (UseSD && isInSchedulingRegion(UseSD->FirstInBundle)) { + BundleMember->Dependencies++; + ScheduleData *DestBundle = UseSD->FirstInBundle; + if (!DestBundle->IsScheduled) { + BundleMember->incrementUnscheduledDeps(1); + } + if (!DestBundle->hasValidDependencies()) { + WorkList.push_back(DestBundle); + } + } + } else { + // I'm not sure if this can ever happen. But we need to be safe. + // This lets the instruction/bundle never be scheduled and + // eventually disable vectorization. + BundleMember->Dependencies++; + BundleMember->incrementUnscheduledDeps(1); + } + } + + // Handle the memory dependencies. + ScheduleData *DepDest = BundleMember->NextLoadStore; + if (DepDest) { + Instruction *SrcInst = BundleMember->Inst; + MemoryLocation SrcLoc = getLocation(SrcInst, SLP->AA); + bool SrcMayWrite = BundleMember->Inst->mayWriteToMemory(); + unsigned numAliased = 0; + unsigned DistToSrc = 1; + + while (DepDest) { + assert(isInSchedulingRegion(DepDest)); + + // We have two limits to reduce the complexity: + // 1) AliasedCheckLimit: It's a small limit to reduce calls to + // SLP->isAliased (which is the expensive part in this loop). + // 2) MaxMemDepDistance: It's for very large blocks and it aborts + // the whole loop (even if the loop is fast, it's quadratic). + // It's important for the loop break condition (see below) to + // check this limit even between two read-only instructions. + if (DistToSrc >= MaxMemDepDistance || + ((SrcMayWrite || DepDest->Inst->mayWriteToMemory()) && + (numAliased >= AliasedCheckLimit || + SLP->isAliased(SrcLoc, SrcInst, DepDest->Inst)))) { + + // We increment the counter only if the locations are aliased + // (instead of counting all alias checks). This gives a better + // balance between reduced runtime and accurate dependencies. + numAliased++; + + DepDest->MemoryDependencies.push_back(BundleMember); + BundleMember->Dependencies++; + ScheduleData *DestBundle = DepDest->FirstInBundle; + if (!DestBundle->IsScheduled) { + BundleMember->incrementUnscheduledDeps(1); + } + if (!DestBundle->hasValidDependencies()) { + WorkList.push_back(DestBundle); + } + } + DepDest = DepDest->NextLoadStore; + + // Example, explaining the loop break condition: Let's assume our + // starting instruction is i0 and MaxMemDepDistance = 3. + // + // +--------v--v--v + // i0,i1,i2,i3,i4,i5,i6,i7,i8 + // +--------^--^--^ + // + // MaxMemDepDistance let us stop alias-checking at i3 and we add + // dependencies from i0 to i3,i4,.. (even if they are not aliased). + // Previously we already added dependencies from i3 to i6,i7,i8 + // (because of MaxMemDepDistance). As we added a dependency from + // i0 to i3, we have transitive dependencies from i0 to i6,i7,i8 + // and we can abort this loop at i6. + if (DistToSrc >= 2 * MaxMemDepDistance) + break; + DistToSrc++; + } + } + } + BundleMember = BundleMember->NextInBundle; + } + if (InsertInReadyList && SD->isReady()) { + ReadyInsts.push_back(SD); + DEBUG(dbgs() << "SLP: gets ready on update: " << *SD->Inst << "\n"); + } + } +} + +void BoUpSLP::BlockScheduling::resetSchedule() { + assert(ScheduleStart && + "tried to reset schedule on block which has not been scheduled"); + for (Instruction *I = ScheduleStart; I != ScheduleEnd; I = I->getNextNode()) { + ScheduleData *SD = getScheduleData(I); + assert(isInSchedulingRegion(SD)); + SD->IsScheduled = false; + SD->resetUnscheduledDeps(); + } + ReadyInsts.clear(); +} + +void BoUpSLP::scheduleBlock(BlockScheduling *BS) { + + if (!BS->ScheduleStart) + return; + + DEBUG(dbgs() << "SLP: schedule block " << BS->BB->getName() << "\n"); + + BS->resetSchedule(); + + // For the real scheduling we use a more sophisticated ready-list: it is + // sorted by the original instruction location. This lets the final schedule + // be as close as possible to the original instruction order. + struct ScheduleDataCompare { + bool operator()(ScheduleData *SD1, ScheduleData *SD2) { + return SD2->SchedulingPriority < SD1->SchedulingPriority; + } + }; + std::set<ScheduleData *, ScheduleDataCompare> ReadyInsts; + + // Ensure that all dependency data is updated and fill the ready-list with + // initial instructions. + int Idx = 0; + int NumToSchedule = 0; + for (auto *I = BS->ScheduleStart; I != BS->ScheduleEnd; + I = I->getNextNode()) { + ScheduleData *SD = BS->getScheduleData(I); + assert( + SD->isPartOfBundle() == (ScalarToTreeEntry.count(SD->Inst) != 0) && + "scheduler and vectorizer have different opinion on what is a bundle"); + SD->FirstInBundle->SchedulingPriority = Idx++; + if (SD->isSchedulingEntity()) { + BS->calculateDependencies(SD, false, this); + NumToSchedule++; + } + } + BS->initialFillReadyList(ReadyInsts); + + Instruction *LastScheduledInst = BS->ScheduleEnd; + + // Do the "real" scheduling. + while (!ReadyInsts.empty()) { + ScheduleData *picked = *ReadyInsts.begin(); + ReadyInsts.erase(ReadyInsts.begin()); + + // Move the scheduled instruction(s) to their dedicated places, if not + // there yet. + ScheduleData *BundleMember = picked; + while (BundleMember) { + Instruction *pickedInst = BundleMember->Inst; + if (LastScheduledInst->getNextNode() != pickedInst) { + BS->BB->getInstList().remove(pickedInst); + BS->BB->getInstList().insert(LastScheduledInst->getIterator(), + pickedInst); + } + LastScheduledInst = pickedInst; + BundleMember = BundleMember->NextInBundle; + } + + BS->schedule(picked, ReadyInsts); + NumToSchedule--; + } + assert(NumToSchedule == 0 && "could not schedule all instructions"); + + // Avoid duplicate scheduling of the block. + BS->ScheduleStart = nullptr; +} + +/// The SLPVectorizer Pass. +struct SLPVectorizer : public FunctionPass { + typedef SmallVector<StoreInst *, 8> StoreList; + typedef MapVector<Value *, StoreList> StoreListMap; + + /// Pass identification, replacement for typeid + static char ID; + + explicit SLPVectorizer() : FunctionPass(ID) { + initializeSLPVectorizerPass(*PassRegistry::getPassRegistry()); + } + + ScalarEvolution *SE; + TargetTransformInfo *TTI; + TargetLibraryInfo *TLI; + AliasAnalysis *AA; + LoopInfo *LI; + DominatorTree *DT; + AssumptionCache *AC; + + bool runOnFunction(Function &F) override { + if (skipOptnoneFunction(F)) + return false; + + SE = &getAnalysis<ScalarEvolutionWrapperPass>().getSE(); + TTI = &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F); + auto *TLIP = getAnalysisIfAvailable<TargetLibraryInfoWrapperPass>(); + TLI = TLIP ? &TLIP->getTLI() : nullptr; + AA = &getAnalysis<AAResultsWrapperPass>().getAAResults(); + LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo(); + DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree(); + AC = &getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F); + + StoreRefs.clear(); + bool Changed = false; + + // If the target claims to have no vector registers don't attempt + // vectorization. + if (!TTI->getNumberOfRegisters(true)) + return false; + + // Use the vector register size specified by the target unless overridden + // by a command-line option. + // TODO: It would be better to limit the vectorization factor based on + // data type rather than just register size. For example, x86 AVX has + // 256-bit registers, but it does not support integer operations + // at that width (that requires AVX2). + if (MaxVectorRegSizeOption.getNumOccurrences()) + MaxVecRegSize = MaxVectorRegSizeOption; + else + MaxVecRegSize = TTI->getRegisterBitWidth(true); + + // Don't vectorize when the attribute NoImplicitFloat is used. + if (F.hasFnAttribute(Attribute::NoImplicitFloat)) + return false; + + DEBUG(dbgs() << "SLP: Analyzing blocks in " << F.getName() << ".\n"); + + // Use the bottom up slp vectorizer to construct chains that start with + // store instructions. + BoUpSLP R(&F, SE, TTI, TLI, AA, LI, DT, AC); + + // A general note: the vectorizer must use BoUpSLP::eraseInstruction() to + // delete instructions. + + // Scan the blocks in the function in post order. + for (auto BB : post_order(&F.getEntryBlock())) { + // Vectorize trees that end at stores. + if (unsigned count = collectStores(BB, R)) { + (void)count; + DEBUG(dbgs() << "SLP: Found " << count << " stores to vectorize.\n"); + Changed |= vectorizeStoreChains(R); + } + + // Vectorize trees that end at reductions. + Changed |= vectorizeChainsInBlock(BB, R); + } + + if (Changed) { + R.optimizeGatherSequence(); + DEBUG(dbgs() << "SLP: vectorized \"" << F.getName() << "\"\n"); + DEBUG(verifyFunction(F)); + } + return Changed; + } + + void getAnalysisUsage(AnalysisUsage &AU) const override { + FunctionPass::getAnalysisUsage(AU); + AU.addRequired<AssumptionCacheTracker>(); + AU.addRequired<ScalarEvolutionWrapperPass>(); + AU.addRequired<AAResultsWrapperPass>(); + AU.addRequired<TargetTransformInfoWrapperPass>(); + AU.addRequired<LoopInfoWrapperPass>(); + AU.addRequired<DominatorTreeWrapperPass>(); + AU.addPreserved<LoopInfoWrapperPass>(); + AU.addPreserved<DominatorTreeWrapperPass>(); + AU.addPreserved<AAResultsWrapperPass>(); + AU.addPreserved<GlobalsAAWrapperPass>(); + AU.setPreservesCFG(); + } + +private: + + /// \brief Collect memory references and sort them according to their base + /// object. We sort the stores to their base objects to reduce the cost of the + /// quadratic search on the stores. TODO: We can further reduce this cost + /// if we flush the chain creation every time we run into a memory barrier. + unsigned collectStores(BasicBlock *BB, BoUpSLP &R); + + /// \brief Try to vectorize a chain that starts at two arithmetic instrs. + bool tryToVectorizePair(Value *A, Value *B, BoUpSLP &R); + + /// \brief Try to vectorize a list of operands. + /// \@param BuildVector A list of users to ignore for the purpose of + /// scheduling and that don't need extracting. + /// \returns true if a value was vectorized. + bool tryToVectorizeList(ArrayRef<Value *> VL, BoUpSLP &R, + ArrayRef<Value *> BuildVector = None, + bool allowReorder = false); + + /// \brief Try to vectorize a chain that may start at the operands of \V; + bool tryToVectorize(BinaryOperator *V, BoUpSLP &R); + + /// \brief Vectorize the stores that were collected in StoreRefs. + bool vectorizeStoreChains(BoUpSLP &R); + + /// \brief Scan the basic block and look for patterns that are likely to start + /// a vectorization chain. + bool vectorizeChainsInBlock(BasicBlock *BB, BoUpSLP &R); + + bool vectorizeStoreChain(ArrayRef<Value *> Chain, int CostThreshold, + BoUpSLP &R, unsigned VecRegSize); + + bool vectorizeStores(ArrayRef<StoreInst *> Stores, int costThreshold, + BoUpSLP &R); +private: + StoreListMap StoreRefs; + unsigned MaxVecRegSize; // This is set by TTI or overridden by cl::opt. +}; + +/// \brief Check that the Values in the slice in VL array are still existent in +/// the WeakVH array. +/// Vectorization of part of the VL array may cause later values in the VL array +/// to become invalid. We track when this has happened in the WeakVH array. +static bool hasValueBeenRAUWed(ArrayRef<Value *> VL, ArrayRef<WeakVH> VH, + unsigned SliceBegin, unsigned SliceSize) { + VL = VL.slice(SliceBegin, SliceSize); + VH = VH.slice(SliceBegin, SliceSize); + return !std::equal(VL.begin(), VL.end(), VH.begin()); +} + +bool SLPVectorizer::vectorizeStoreChain(ArrayRef<Value *> Chain, + int CostThreshold, BoUpSLP &R, + unsigned VecRegSize) { + unsigned ChainLen = Chain.size(); + DEBUG(dbgs() << "SLP: Analyzing a store chain of length " << ChainLen + << "\n"); + Type *StoreTy = cast<StoreInst>(Chain[0])->getValueOperand()->getType(); + auto &DL = cast<StoreInst>(Chain[0])->getModule()->getDataLayout(); + unsigned Sz = DL.getTypeSizeInBits(StoreTy); + unsigned VF = VecRegSize / Sz; + + if (!isPowerOf2_32(Sz) || VF < 2) + return false; + + // Keep track of values that were deleted by vectorizing in the loop below. + SmallVector<WeakVH, 8> TrackValues(Chain.begin(), Chain.end()); + + bool Changed = false; + // Look for profitable vectorizable trees at all offsets, starting at zero. + for (unsigned i = 0, e = ChainLen; i < e; ++i) { + if (i + VF > e) + break; + + // Check that a previous iteration of this loop did not delete the Value. + if (hasValueBeenRAUWed(Chain, TrackValues, i, VF)) + continue; + + DEBUG(dbgs() << "SLP: Analyzing " << VF << " stores at offset " << i + << "\n"); + ArrayRef<Value *> Operands = Chain.slice(i, VF); + + R.buildTree(Operands); + + int Cost = R.getTreeCost(); + + DEBUG(dbgs() << "SLP: Found cost=" << Cost << " for VF=" << VF << "\n"); + if (Cost < CostThreshold) { + DEBUG(dbgs() << "SLP: Decided to vectorize cost=" << Cost << "\n"); + R.vectorizeTree(); + + // Move to the next bundle. + i += VF - 1; + Changed = true; + } + } + + return Changed; +} + +bool SLPVectorizer::vectorizeStores(ArrayRef<StoreInst *> Stores, + int costThreshold, BoUpSLP &R) { + SetVector<StoreInst *> Heads, Tails; + SmallDenseMap<StoreInst *, StoreInst *> ConsecutiveChain; + + // We may run into multiple chains that merge into a single chain. We mark the + // stores that we vectorized so that we don't visit the same store twice. + BoUpSLP::ValueSet VectorizedStores; + bool Changed = false; + + // Do a quadratic search on all of the given stores and find + // all of the pairs of stores that follow each other. + SmallVector<unsigned, 16> IndexQueue; + for (unsigned i = 0, e = Stores.size(); i < e; ++i) { + const DataLayout &DL = Stores[i]->getModule()->getDataLayout(); + IndexQueue.clear(); + // If a store has multiple consecutive store candidates, search Stores + // array according to the sequence: from i+1 to e, then from i-1 to 0. + // This is because usually pairing with immediate succeeding or preceding + // candidate create the best chance to find slp vectorization opportunity. + unsigned j = 0; + for (j = i + 1; j < e; ++j) + IndexQueue.push_back(j); + for (j = i; j > 0; --j) + IndexQueue.push_back(j - 1); + + for (auto &k : IndexQueue) { + if (R.isConsecutiveAccess(Stores[i], Stores[k], DL)) { + Tails.insert(Stores[k]); + Heads.insert(Stores[i]); + ConsecutiveChain[Stores[i]] = Stores[k]; + break; + } + } + } + + // For stores that start but don't end a link in the chain: + for (SetVector<StoreInst *>::iterator it = Heads.begin(), e = Heads.end(); + it != e; ++it) { + if (Tails.count(*it)) + continue; + + // We found a store instr that starts a chain. Now follow the chain and try + // to vectorize it. + BoUpSLP::ValueList Operands; + StoreInst *I = *it; + // Collect the chain into a list. + while (Tails.count(I) || Heads.count(I)) { + if (VectorizedStores.count(I)) + break; + Operands.push_back(I); + // Move to the next value in the chain. + I = ConsecutiveChain[I]; + } + + // FIXME: Is division-by-2 the correct step? Should we assert that the + // register size is a power-of-2? + for (unsigned Size = MaxVecRegSize; Size >= MinVecRegSize; Size /= 2) { + if (vectorizeStoreChain(Operands, costThreshold, R, Size)) { + // Mark the vectorized stores so that we don't vectorize them again. + VectorizedStores.insert(Operands.begin(), Operands.end()); + Changed = true; + break; + } + } + } + + return Changed; +} + + +unsigned SLPVectorizer::collectStores(BasicBlock *BB, BoUpSLP &R) { + unsigned count = 0; + StoreRefs.clear(); + const DataLayout &DL = BB->getModule()->getDataLayout(); + for (Instruction &I : *BB) { + StoreInst *SI = dyn_cast<StoreInst>(&I); + if (!SI) + continue; + + // Don't touch volatile stores. + if (!SI->isSimple()) + continue; + + // Check that the pointer points to scalars. + Type *Ty = SI->getValueOperand()->getType(); + if (!isValidElementType(Ty)) + continue; + + // Find the base pointer. + Value *Ptr = GetUnderlyingObject(SI->getPointerOperand(), DL); + + // Save the store locations. + StoreRefs[Ptr].push_back(SI); + count++; + } + return count; +} + +bool SLPVectorizer::tryToVectorizePair(Value *A, Value *B, BoUpSLP &R) { + if (!A || !B) + return false; + Value *VL[] = { A, B }; + return tryToVectorizeList(VL, R, None, true); +} + +bool SLPVectorizer::tryToVectorizeList(ArrayRef<Value *> VL, BoUpSLP &R, + ArrayRef<Value *> BuildVector, + bool allowReorder) { + if (VL.size() < 2) + return false; + + DEBUG(dbgs() << "SLP: Vectorizing a list of length = " << VL.size() << ".\n"); + + // Check that all of the parts are scalar instructions of the same type. + Instruction *I0 = dyn_cast<Instruction>(VL[0]); + if (!I0) + return false; + + unsigned Opcode0 = I0->getOpcode(); + const DataLayout &DL = I0->getModule()->getDataLayout(); + + Type *Ty0 = I0->getType(); + unsigned Sz = DL.getTypeSizeInBits(Ty0); + // FIXME: Register size should be a parameter to this function, so we can + // try different vectorization factors. + unsigned VF = MinVecRegSize / Sz; + + for (Value *V : VL) { + Type *Ty = V->getType(); + if (!isValidElementType(Ty)) + return false; + Instruction *Inst = dyn_cast<Instruction>(V); + if (!Inst || Inst->getOpcode() != Opcode0) + return false; + } + + bool Changed = false; + + // Keep track of values that were deleted by vectorizing in the loop below. + SmallVector<WeakVH, 8> TrackValues(VL.begin(), VL.end()); + + for (unsigned i = 0, e = VL.size(); i < e; ++i) { + unsigned OpsWidth = 0; + + if (i + VF > e) + OpsWidth = e - i; + else + OpsWidth = VF; + + if (!isPowerOf2_32(OpsWidth) || OpsWidth < 2) + break; + + // Check that a previous iteration of this loop did not delete the Value. + if (hasValueBeenRAUWed(VL, TrackValues, i, OpsWidth)) + continue; + + DEBUG(dbgs() << "SLP: Analyzing " << OpsWidth << " operations " + << "\n"); + ArrayRef<Value *> Ops = VL.slice(i, OpsWidth); + + ArrayRef<Value *> BuildVectorSlice; + if (!BuildVector.empty()) + BuildVectorSlice = BuildVector.slice(i, OpsWidth); + + R.buildTree(Ops, BuildVectorSlice); + // TODO: check if we can allow reordering also for other cases than + // tryToVectorizePair() + if (allowReorder && R.shouldReorder()) { + assert(Ops.size() == 2); + assert(BuildVectorSlice.empty()); + Value *ReorderedOps[] = { Ops[1], Ops[0] }; + R.buildTree(ReorderedOps, None); + } + int Cost = R.getTreeCost(); + + if (Cost < -SLPCostThreshold) { + DEBUG(dbgs() << "SLP: Vectorizing list at cost:" << Cost << ".\n"); + Value *VectorizedRoot = R.vectorizeTree(); + + // Reconstruct the build vector by extracting the vectorized root. This + // way we handle the case where some elements of the vector are undefined. + // (return (inserelt <4 xi32> (insertelt undef (opd0) 0) (opd1) 2)) + if (!BuildVectorSlice.empty()) { + // The insert point is the last build vector instruction. The vectorized + // root will precede it. This guarantees that we get an instruction. The + // vectorized tree could have been constant folded. + Instruction *InsertAfter = cast<Instruction>(BuildVectorSlice.back()); + unsigned VecIdx = 0; + for (auto &V : BuildVectorSlice) { + IRBuilder<true, NoFolder> Builder( + InsertAfter->getParent(), ++BasicBlock::iterator(InsertAfter)); + InsertElementInst *IE = cast<InsertElementInst>(V); + Instruction *Extract = cast<Instruction>(Builder.CreateExtractElement( + VectorizedRoot, Builder.getInt32(VecIdx++))); + IE->setOperand(1, Extract); + IE->removeFromParent(); + IE->insertAfter(Extract); + InsertAfter = IE; + } + } + // Move to the next bundle. + i += VF - 1; + Changed = true; + } + } + + return Changed; +} + +bool SLPVectorizer::tryToVectorize(BinaryOperator *V, BoUpSLP &R) { + if (!V) + return false; + + // Try to vectorize V. + if (tryToVectorizePair(V->getOperand(0), V->getOperand(1), R)) + return true; + + BinaryOperator *A = dyn_cast<BinaryOperator>(V->getOperand(0)); + BinaryOperator *B = dyn_cast<BinaryOperator>(V->getOperand(1)); + // Try to skip B. + if (B && B->hasOneUse()) { + BinaryOperator *B0 = dyn_cast<BinaryOperator>(B->getOperand(0)); + BinaryOperator *B1 = dyn_cast<BinaryOperator>(B->getOperand(1)); + if (tryToVectorizePair(A, B0, R)) { + return true; + } + if (tryToVectorizePair(A, B1, R)) { + return true; + } + } + + // Try to skip A. + if (A && A->hasOneUse()) { + BinaryOperator *A0 = dyn_cast<BinaryOperator>(A->getOperand(0)); + BinaryOperator *A1 = dyn_cast<BinaryOperator>(A->getOperand(1)); + if (tryToVectorizePair(A0, B, R)) { + return true; + } + if (tryToVectorizePair(A1, B, R)) { + return true; + } + } + return 0; +} + +/// \brief Generate a shuffle mask to be used in a reduction tree. +/// +/// \param VecLen The length of the vector to be reduced. +/// \param NumEltsToRdx The number of elements that should be reduced in the +/// vector. +/// \param IsPairwise Whether the reduction is a pairwise or splitting +/// reduction. A pairwise reduction will generate a mask of +/// <0,2,...> or <1,3,..> while a splitting reduction will generate +/// <2,3, undef,undef> for a vector of 4 and NumElts = 2. +/// \param IsLeft True will generate a mask of even elements, odd otherwise. +static Value *createRdxShuffleMask(unsigned VecLen, unsigned NumEltsToRdx, + bool IsPairwise, bool IsLeft, + IRBuilder<> &Builder) { + assert((IsPairwise || !IsLeft) && "Don't support a <0,1,undef,...> mask"); + + SmallVector<Constant *, 32> ShuffleMask( + VecLen, UndefValue::get(Builder.getInt32Ty())); + + if (IsPairwise) + // Build a mask of 0, 2, ... (left) or 1, 3, ... (right). + for (unsigned i = 0; i != NumEltsToRdx; ++i) + ShuffleMask[i] = Builder.getInt32(2 * i + !IsLeft); + else + // Move the upper half of the vector to the lower half. + for (unsigned i = 0; i != NumEltsToRdx; ++i) + ShuffleMask[i] = Builder.getInt32(NumEltsToRdx + i); + + return ConstantVector::get(ShuffleMask); +} + + +/// Model horizontal reductions. +/// +/// A horizontal reduction is a tree of reduction operations (currently add and +/// fadd) that has operations that can be put into a vector as its leaf. +/// For example, this tree: +/// +/// mul mul mul mul +/// \ / \ / +/// + + +/// \ / +/// + +/// This tree has "mul" as its reduced values and "+" as its reduction +/// operations. A reduction might be feeding into a store or a binary operation +/// feeding a phi. +/// ... +/// \ / +/// + +/// | +/// phi += +/// +/// Or: +/// ... +/// \ / +/// + +/// | +/// *p = +/// +class HorizontalReduction { + SmallVector<Value *, 16> ReductionOps; + SmallVector<Value *, 32> ReducedVals; + + BinaryOperator *ReductionRoot; + PHINode *ReductionPHI; + + /// The opcode of the reduction. + unsigned ReductionOpcode; + /// The opcode of the values we perform a reduction on. + unsigned ReducedValueOpcode; + /// Should we model this reduction as a pairwise reduction tree or a tree that + /// splits the vector in halves and adds those halves. + bool IsPairwiseReduction; + +public: + /// The width of one full horizontal reduction operation. + unsigned ReduxWidth; + + HorizontalReduction() + : ReductionRoot(nullptr), ReductionPHI(nullptr), ReductionOpcode(0), + ReducedValueOpcode(0), IsPairwiseReduction(false), ReduxWidth(0) {} + + /// \brief Try to find a reduction tree. + bool matchAssociativeReduction(PHINode *Phi, BinaryOperator *B) { + assert((!Phi || + std::find(Phi->op_begin(), Phi->op_end(), B) != Phi->op_end()) && + "Thi phi needs to use the binary operator"); + + // We could have a initial reductions that is not an add. + // r *= v1 + v2 + v3 + v4 + // In such a case start looking for a tree rooted in the first '+'. + if (Phi) { + if (B->getOperand(0) == Phi) { + Phi = nullptr; + B = dyn_cast<BinaryOperator>(B->getOperand(1)); + } else if (B->getOperand(1) == Phi) { + Phi = nullptr; + B = dyn_cast<BinaryOperator>(B->getOperand(0)); + } + } + + if (!B) + return false; + + Type *Ty = B->getType(); + if (!isValidElementType(Ty)) + return false; + + const DataLayout &DL = B->getModule()->getDataLayout(); + ReductionOpcode = B->getOpcode(); + ReducedValueOpcode = 0; + // FIXME: Register size should be a parameter to this function, so we can + // try different vectorization factors. + ReduxWidth = MinVecRegSize / DL.getTypeSizeInBits(Ty); + ReductionRoot = B; + ReductionPHI = Phi; + + if (ReduxWidth < 4) + return false; + + // We currently only support adds. + if (ReductionOpcode != Instruction::Add && + ReductionOpcode != Instruction::FAdd) + return false; + + // Post order traverse the reduction tree starting at B. We only handle true + // trees containing only binary operators or selects. + SmallVector<std::pair<Instruction *, unsigned>, 32> Stack; + Stack.push_back(std::make_pair(B, 0)); + while (!Stack.empty()) { + Instruction *TreeN = Stack.back().first; + unsigned EdgeToVist = Stack.back().second++; + bool IsReducedValue = TreeN->getOpcode() != ReductionOpcode; + + // Only handle trees in the current basic block. + if (TreeN->getParent() != B->getParent()) + return false; + + // Each tree node needs to have one user except for the ultimate + // reduction. + if (!TreeN->hasOneUse() && TreeN != B) + return false; + + // Postorder vist. + if (EdgeToVist == 2 || IsReducedValue) { + if (IsReducedValue) { + // Make sure that the opcodes of the operations that we are going to + // reduce match. + if (!ReducedValueOpcode) + ReducedValueOpcode = TreeN->getOpcode(); + else if (ReducedValueOpcode != TreeN->getOpcode()) + return false; + ReducedVals.push_back(TreeN); + } else { + // We need to be able to reassociate the adds. + if (!TreeN->isAssociative()) + return false; + ReductionOps.push_back(TreeN); + } + // Retract. + Stack.pop_back(); + continue; + } + + // Visit left or right. + Value *NextV = TreeN->getOperand(EdgeToVist); + // We currently only allow BinaryOperator's and SelectInst's as reduction + // values in our tree. + if (isa<BinaryOperator>(NextV) || isa<SelectInst>(NextV)) + Stack.push_back(std::make_pair(cast<Instruction>(NextV), 0)); + else if (NextV != Phi) + return false; + } + return true; + } + + /// \brief Attempt to vectorize the tree found by + /// matchAssociativeReduction. + bool tryToReduce(BoUpSLP &V, TargetTransformInfo *TTI) { + if (ReducedVals.empty()) + return false; + + unsigned NumReducedVals = ReducedVals.size(); + if (NumReducedVals < ReduxWidth) + return false; + + Value *VectorizedTree = nullptr; + IRBuilder<> Builder(ReductionRoot); + FastMathFlags Unsafe; + Unsafe.setUnsafeAlgebra(); + Builder.setFastMathFlags(Unsafe); + unsigned i = 0; + + for (; i < NumReducedVals - ReduxWidth + 1; i += ReduxWidth) { + V.buildTree(makeArrayRef(&ReducedVals[i], ReduxWidth), ReductionOps); + + // Estimate cost. + int Cost = V.getTreeCost() + getReductionCost(TTI, ReducedVals[i]); + if (Cost >= -SLPCostThreshold) + break; + + DEBUG(dbgs() << "SLP: Vectorizing horizontal reduction at cost:" << Cost + << ". (HorRdx)\n"); + + // Vectorize a tree. + DebugLoc Loc = cast<Instruction>(ReducedVals[i])->getDebugLoc(); + Value *VectorizedRoot = V.vectorizeTree(); + + // Emit a reduction. + Value *ReducedSubTree = emitReduction(VectorizedRoot, Builder); + if (VectorizedTree) { + Builder.SetCurrentDebugLocation(Loc); + VectorizedTree = createBinOp(Builder, ReductionOpcode, VectorizedTree, + ReducedSubTree, "bin.rdx"); + } else + VectorizedTree = ReducedSubTree; + } + + if (VectorizedTree) { + // Finish the reduction. + for (; i < NumReducedVals; ++i) { + Builder.SetCurrentDebugLocation( + cast<Instruction>(ReducedVals[i])->getDebugLoc()); + VectorizedTree = createBinOp(Builder, ReductionOpcode, VectorizedTree, + ReducedVals[i]); + } + // Update users. + if (ReductionPHI) { + assert(ReductionRoot && "Need a reduction operation"); + ReductionRoot->setOperand(0, VectorizedTree); + ReductionRoot->setOperand(1, ReductionPHI); + } else + ReductionRoot->replaceAllUsesWith(VectorizedTree); + } + return VectorizedTree != nullptr; + } + + unsigned numReductionValues() const { + return ReducedVals.size(); + } + +private: + /// \brief Calculate the cost of a reduction. + int getReductionCost(TargetTransformInfo *TTI, Value *FirstReducedVal) { + Type *ScalarTy = FirstReducedVal->getType(); + Type *VecTy = VectorType::get(ScalarTy, ReduxWidth); + + int PairwiseRdxCost = TTI->getReductionCost(ReductionOpcode, VecTy, true); + int SplittingRdxCost = TTI->getReductionCost(ReductionOpcode, VecTy, false); + + IsPairwiseReduction = PairwiseRdxCost < SplittingRdxCost; + int VecReduxCost = IsPairwiseReduction ? PairwiseRdxCost : SplittingRdxCost; + + int ScalarReduxCost = + ReduxWidth * TTI->getArithmeticInstrCost(ReductionOpcode, VecTy); + + DEBUG(dbgs() << "SLP: Adding cost " << VecReduxCost - ScalarReduxCost + << " for reduction that starts with " << *FirstReducedVal + << " (It is a " + << (IsPairwiseReduction ? "pairwise" : "splitting") + << " reduction)\n"); + + return VecReduxCost - ScalarReduxCost; + } + + static Value *createBinOp(IRBuilder<> &Builder, unsigned Opcode, Value *L, + Value *R, const Twine &Name = "") { + if (Opcode == Instruction::FAdd) + return Builder.CreateFAdd(L, R, Name); + return Builder.CreateBinOp((Instruction::BinaryOps)Opcode, L, R, Name); + } + + /// \brief Emit a horizontal reduction of the vectorized value. + Value *emitReduction(Value *VectorizedValue, IRBuilder<> &Builder) { + assert(VectorizedValue && "Need to have a vectorized tree node"); + assert(isPowerOf2_32(ReduxWidth) && + "We only handle power-of-two reductions for now"); + + Value *TmpVec = VectorizedValue; + for (unsigned i = ReduxWidth / 2; i != 0; i >>= 1) { + if (IsPairwiseReduction) { + Value *LeftMask = + createRdxShuffleMask(ReduxWidth, i, true, true, Builder); + Value *RightMask = + createRdxShuffleMask(ReduxWidth, i, true, false, Builder); + + Value *LeftShuf = Builder.CreateShuffleVector( + TmpVec, UndefValue::get(TmpVec->getType()), LeftMask, "rdx.shuf.l"); + Value *RightShuf = Builder.CreateShuffleVector( + TmpVec, UndefValue::get(TmpVec->getType()), (RightMask), + "rdx.shuf.r"); + TmpVec = createBinOp(Builder, ReductionOpcode, LeftShuf, RightShuf, + "bin.rdx"); + } else { + Value *UpperHalf = + createRdxShuffleMask(ReduxWidth, i, false, false, Builder); + Value *Shuf = Builder.CreateShuffleVector( + TmpVec, UndefValue::get(TmpVec->getType()), UpperHalf, "rdx.shuf"); + TmpVec = createBinOp(Builder, ReductionOpcode, TmpVec, Shuf, "bin.rdx"); + } + } + + // The result is in the first element of the vector. + return Builder.CreateExtractElement(TmpVec, Builder.getInt32(0)); + } +}; + +/// \brief Recognize construction of vectors like +/// %ra = insertelement <4 x float> undef, float %s0, i32 0 +/// %rb = insertelement <4 x float> %ra, float %s1, i32 1 +/// %rc = insertelement <4 x float> %rb, float %s2, i32 2 +/// %rd = insertelement <4 x float> %rc, float %s3, i32 3 +/// +/// Returns true if it matches +/// +static bool findBuildVector(InsertElementInst *FirstInsertElem, + SmallVectorImpl<Value *> &BuildVector, + SmallVectorImpl<Value *> &BuildVectorOpds) { + if (!isa<UndefValue>(FirstInsertElem->getOperand(0))) + return false; + + InsertElementInst *IE = FirstInsertElem; + while (true) { + BuildVector.push_back(IE); + BuildVectorOpds.push_back(IE->getOperand(1)); + + if (IE->use_empty()) + return false; + + InsertElementInst *NextUse = dyn_cast<InsertElementInst>(IE->user_back()); + if (!NextUse) + return true; + + // If this isn't the final use, make sure the next insertelement is the only + // use. It's OK if the final constructed vector is used multiple times + if (!IE->hasOneUse()) + return false; + + IE = NextUse; + } + + return false; +} + +static bool PhiTypeSorterFunc(Value *V, Value *V2) { + return V->getType() < V2->getType(); +} + +/// \brief Try and get a reduction value from a phi node. +/// +/// Given a phi node \p P in a block \p ParentBB, consider possible reductions +/// if they come from either \p ParentBB or a containing loop latch. +/// +/// \returns A candidate reduction value if possible, or \code nullptr \endcode +/// if not possible. +static Value *getReductionValue(const DominatorTree *DT, PHINode *P, + BasicBlock *ParentBB, LoopInfo *LI) { + // There are situations where the reduction value is not dominated by the + // reduction phi. Vectorizing such cases has been reported to cause + // miscompiles. See PR25787. + auto DominatedReduxValue = [&](Value *R) { + return ( + dyn_cast<Instruction>(R) && + DT->dominates(P->getParent(), dyn_cast<Instruction>(R)->getParent())); + }; + + Value *Rdx = nullptr; + + // Return the incoming value if it comes from the same BB as the phi node. + if (P->getIncomingBlock(0) == ParentBB) { + Rdx = P->getIncomingValue(0); + } else if (P->getIncomingBlock(1) == ParentBB) { + Rdx = P->getIncomingValue(1); + } + + if (Rdx && DominatedReduxValue(Rdx)) + return Rdx; + + // Otherwise, check whether we have a loop latch to look at. + Loop *BBL = LI->getLoopFor(ParentBB); + if (!BBL) + return nullptr; + BasicBlock *BBLatch = BBL->getLoopLatch(); + if (!BBLatch) + return nullptr; + + // There is a loop latch, return the incoming value if it comes from + // that. This reduction pattern occassionaly turns up. + if (P->getIncomingBlock(0) == BBLatch) { + Rdx = P->getIncomingValue(0); + } else if (P->getIncomingBlock(1) == BBLatch) { + Rdx = P->getIncomingValue(1); + } + + if (Rdx && DominatedReduxValue(Rdx)) + return Rdx; + + return nullptr; +} + +/// \brief Attempt to reduce a horizontal reduction. +/// If it is legal to match a horizontal reduction feeding +/// the phi node P with reduction operators BI, then check if it +/// can be done. +/// \returns true if a horizontal reduction was matched and reduced. +/// \returns false if a horizontal reduction was not matched. +static bool canMatchHorizontalReduction(PHINode *P, BinaryOperator *BI, + BoUpSLP &R, TargetTransformInfo *TTI) { + if (!ShouldVectorizeHor) + return false; + + HorizontalReduction HorRdx; + if (!HorRdx.matchAssociativeReduction(P, BI)) + return false; + + // If there is a sufficient number of reduction values, reduce + // to a nearby power-of-2. Can safely generate oversized + // vectors and rely on the backend to split them to legal sizes. + HorRdx.ReduxWidth = + std::max((uint64_t)4, PowerOf2Floor(HorRdx.numReductionValues())); + + return HorRdx.tryToReduce(R, TTI); +} + +bool SLPVectorizer::vectorizeChainsInBlock(BasicBlock *BB, BoUpSLP &R) { + bool Changed = false; + SmallVector<Value *, 4> Incoming; + SmallSet<Value *, 16> VisitedInstrs; + + bool HaveVectorizedPhiNodes = true; + while (HaveVectorizedPhiNodes) { + HaveVectorizedPhiNodes = false; + + // Collect the incoming values from the PHIs. + Incoming.clear(); + for (Instruction &I : *BB) { + PHINode *P = dyn_cast<PHINode>(&I); + if (!P) + break; + + if (!VisitedInstrs.count(P)) + Incoming.push_back(P); + } + + // Sort by type. + std::stable_sort(Incoming.begin(), Incoming.end(), PhiTypeSorterFunc); + + // Try to vectorize elements base on their type. + for (SmallVector<Value *, 4>::iterator IncIt = Incoming.begin(), + E = Incoming.end(); + IncIt != E;) { + + // Look for the next elements with the same type. + SmallVector<Value *, 4>::iterator SameTypeIt = IncIt; + while (SameTypeIt != E && + (*SameTypeIt)->getType() == (*IncIt)->getType()) { + VisitedInstrs.insert(*SameTypeIt); + ++SameTypeIt; + } + + // Try to vectorize them. + unsigned NumElts = (SameTypeIt - IncIt); + DEBUG(errs() << "SLP: Trying to vectorize starting at PHIs (" << NumElts << ")\n"); + if (NumElts > 1 && tryToVectorizeList(makeArrayRef(IncIt, NumElts), R)) { + // Success start over because instructions might have been changed. + HaveVectorizedPhiNodes = true; + Changed = true; + break; + } + + // Start over at the next instruction of a different type (or the end). + IncIt = SameTypeIt; + } + } + + VisitedInstrs.clear(); + + for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; it++) { + // We may go through BB multiple times so skip the one we have checked. + if (!VisitedInstrs.insert(&*it).second) + continue; + + if (isa<DbgInfoIntrinsic>(it)) + continue; + + // Try to vectorize reductions that use PHINodes. + if (PHINode *P = dyn_cast<PHINode>(it)) { + // Check that the PHI is a reduction PHI. + if (P->getNumIncomingValues() != 2) + return Changed; + + Value *Rdx = getReductionValue(DT, P, BB, LI); + + // Check if this is a Binary Operator. + BinaryOperator *BI = dyn_cast_or_null<BinaryOperator>(Rdx); + if (!BI) + continue; + + // Try to match and vectorize a horizontal reduction. + if (canMatchHorizontalReduction(P, BI, R, TTI)) { + Changed = true; + it = BB->begin(); + e = BB->end(); + continue; + } + + Value *Inst = BI->getOperand(0); + if (Inst == P) + Inst = BI->getOperand(1); + + if (tryToVectorize(dyn_cast<BinaryOperator>(Inst), R)) { + // We would like to start over since some instructions are deleted + // and the iterator may become invalid value. + Changed = true; + it = BB->begin(); + e = BB->end(); + continue; + } + + continue; + } + + if (ShouldStartVectorizeHorAtStore) + if (StoreInst *SI = dyn_cast<StoreInst>(it)) + if (BinaryOperator *BinOp = + dyn_cast<BinaryOperator>(SI->getValueOperand())) { + if (canMatchHorizontalReduction(nullptr, BinOp, R, TTI) || + tryToVectorize(BinOp, R)) { + Changed = true; + it = BB->begin(); + e = BB->end(); + continue; + } + } + + // Try to vectorize horizontal reductions feeding into a return. + if (ReturnInst *RI = dyn_cast<ReturnInst>(it)) + if (RI->getNumOperands() != 0) + if (BinaryOperator *BinOp = + dyn_cast<BinaryOperator>(RI->getOperand(0))) { + DEBUG(dbgs() << "SLP: Found a return to vectorize.\n"); + if (tryToVectorizePair(BinOp->getOperand(0), + BinOp->getOperand(1), R)) { + Changed = true; + it = BB->begin(); + e = BB->end(); + continue; + } + } + + // Try to vectorize trees that start at compare instructions. + if (CmpInst *CI = dyn_cast<CmpInst>(it)) { + if (tryToVectorizePair(CI->getOperand(0), CI->getOperand(1), R)) { + Changed = true; + // We would like to start over since some instructions are deleted + // and the iterator may become invalid value. + it = BB->begin(); + e = BB->end(); + continue; + } + + for (int i = 0; i < 2; ++i) { + if (BinaryOperator *BI = dyn_cast<BinaryOperator>(CI->getOperand(i))) { + if (tryToVectorizePair(BI->getOperand(0), BI->getOperand(1), R)) { + Changed = true; + // We would like to start over since some instructions are deleted + // and the iterator may become invalid value. + it = BB->begin(); + e = BB->end(); + break; + } + } + } + continue; + } + + // Try to vectorize trees that start at insertelement instructions. + if (InsertElementInst *FirstInsertElem = dyn_cast<InsertElementInst>(it)) { + SmallVector<Value *, 16> BuildVector; + SmallVector<Value *, 16> BuildVectorOpds; + if (!findBuildVector(FirstInsertElem, BuildVector, BuildVectorOpds)) + continue; + + // Vectorize starting with the build vector operands ignoring the + // BuildVector instructions for the purpose of scheduling and user + // extraction. + if (tryToVectorizeList(BuildVectorOpds, R, BuildVector)) { + Changed = true; + it = BB->begin(); + e = BB->end(); + } + + continue; + } + } + + return Changed; +} + +bool SLPVectorizer::vectorizeStoreChains(BoUpSLP &R) { + bool Changed = false; + // Attempt to sort and vectorize each of the store-groups. + for (StoreListMap::iterator it = StoreRefs.begin(), e = StoreRefs.end(); + it != e; ++it) { + if (it->second.size() < 2) + continue; + + DEBUG(dbgs() << "SLP: Analyzing a store chain of length " + << it->second.size() << ".\n"); + + // Process the stores in chunks of 16. + // TODO: The limit of 16 inhibits greater vectorization factors. + // For example, AVX2 supports v32i8. Increasing this limit, however, + // may cause a significant compile-time increase. + for (unsigned CI = 0, CE = it->second.size(); CI < CE; CI+=16) { + unsigned Len = std::min<unsigned>(CE - CI, 16); + Changed |= vectorizeStores(makeArrayRef(&it->second[CI], Len), + -SLPCostThreshold, R); + } + } + return Changed; +} + +} // end anonymous namespace + +char SLPVectorizer::ID = 0; +static const char lv_name[] = "SLP Vectorizer"; +INITIALIZE_PASS_BEGIN(SLPVectorizer, SV_NAME, lv_name, false, false) +INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass) +INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass) +INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker) +INITIALIZE_PASS_DEPENDENCY(ScalarEvolutionWrapperPass) +INITIALIZE_PASS_DEPENDENCY(LoopSimplify) +INITIALIZE_PASS_END(SLPVectorizer, SV_NAME, lv_name, false, false) + +namespace llvm { +Pass *createSLPVectorizerPass() { return new SLPVectorizer(); } +} diff --git a/contrib/llvm/lib/Transforms/Vectorize/Vectorize.cpp b/contrib/llvm/lib/Transforms/Vectorize/Vectorize.cpp new file mode 100644 index 0000000..6e002fd --- /dev/null +++ b/contrib/llvm/lib/Transforms/Vectorize/Vectorize.cpp @@ -0,0 +1,48 @@ +//===-- Vectorize.cpp -----------------------------------------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file implements common infrastructure for libLLVMVectorizeOpts.a, which +// implements several vectorization transformations over the LLVM intermediate +// representation, including the C bindings for that library. +// +//===----------------------------------------------------------------------===// + +#include "llvm/Transforms/Vectorize.h" +#include "llvm-c/Initialization.h" +#include "llvm-c/Transforms/Vectorize.h" +#include "llvm/Analysis/Passes.h" +#include "llvm/IR/Verifier.h" +#include "llvm/InitializePasses.h" +#include "llvm/IR/LegacyPassManager.h" + +using namespace llvm; + +/// initializeVectorizationPasses - Initialize all passes linked into the +/// Vectorization library. +void llvm::initializeVectorization(PassRegistry &Registry) { + initializeBBVectorizePass(Registry); + initializeLoopVectorizePass(Registry); + initializeSLPVectorizerPass(Registry); +} + +void LLVMInitializeVectorization(LLVMPassRegistryRef R) { + initializeVectorization(*unwrap(R)); +} + +void LLVMAddBBVectorizePass(LLVMPassManagerRef PM) { + unwrap(PM)->add(createBBVectorizePass()); +} + +void LLVMAddLoopVectorizePass(LLVMPassManagerRef PM) { + unwrap(PM)->add(createLoopVectorizePass()); +} + +void LLVMAddSLPVectorizePass(LLVMPassManagerRef PM) { + unwrap(PM)->add(createSLPVectorizerPass()); +} |
