X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTransforms%2FScalar%2FLoopIdiomRecognize.cpp;h=95aadb190751d0a8faff984cc355a7f61c3d35a9;hp=cb20c0334843a63d05a61f48cb6efe93a8ca7423;hb=e6f43815c33626bf11659552634047d7606dc808;hpb=c94da20917cb4dfa750903b366c920210c5265ee diff --git a/lib/Transforms/Scalar/LoopIdiomRecognize.cpp b/lib/Transforms/Scalar/LoopIdiomRecognize.cpp index cb20c033484..95aadb19075 100644 --- a/lib/Transforms/Scalar/LoopIdiomRecognize.cpp +++ b/lib/Transforms/Scalar/LoopIdiomRecognize.cpp @@ -47,6 +47,7 @@ #include "llvm/Analysis/LoopPass.h" #include "llvm/Analysis/ScalarEvolutionExpander.h" #include "llvm/Analysis/ScalarEvolutionExpressions.h" +#include "llvm/Analysis/TargetLibraryInfo.h" #include "llvm/Analysis/TargetTransformInfo.h" #include "llvm/Analysis/ValueTracking.h" #include "llvm/IR/DataLayout.h" @@ -56,7 +57,6 @@ #include "llvm/IR/Module.h" #include "llvm/Support/Debug.h" #include "llvm/Support/raw_ostream.h" -#include "llvm/Analysis/TargetLibraryInfo.h" #include "llvm/Transforms/Utils/Local.h" using namespace llvm; @@ -67,149 +67,143 @@ STATISTIC(NumMemCpy, "Number of memcpy's formed from loop load+stores"); namespace { - class LoopIdiomRecognize; +class LoopIdiomRecognize; + +/// This class is to recoginize idioms of population-count conducted in +/// a noncountable loop. Currently it only recognizes this pattern: +/// \code +/// while(x) {cnt++; ...; x &= x - 1; ...} +/// \endcode +class NclPopcountRecognize { + LoopIdiomRecognize &LIR; + Loop *CurLoop; + BasicBlock *PreCondBB; + + typedef IRBuilder<> IRBuilderTy; + +public: + explicit NclPopcountRecognize(LoopIdiomRecognize &TheLIR); + bool recognize(); + +private: + /// Take a glimpse of the loop to see if we need to go ahead recoginizing + /// the idiom. + bool preliminaryScreen(); + + /// Check if the given conditional branch is based on the comparison + /// between a variable and zero, and if the variable is non-zero, the + /// control yields to the loop entry. If the branch matches the behavior, + /// the variable involved in the comparion is returned. This function will + /// be called to see if the precondition and postcondition of the loop + /// are in desirable form. + Value *matchCondition(BranchInst *Br, BasicBlock *NonZeroTarget) const; + + /// Return true iff the idiom is detected in the loop. and 1) \p CntInst + /// is set to the instruction counting the population bit. 2) \p CntPhi + /// is set to the corresponding phi node. 3) \p Var is set to the value + /// whose population bits are being counted. + bool detectIdiom(Instruction *&CntInst, PHINode *&CntPhi, Value *&Var) const; + + /// Insert ctpop intrinsic function and some obviously dead instructions. + void transform(Instruction *CntInst, PHINode *CntPhi, Value *Var); + + /// Create llvm.ctpop.* intrinsic function. + CallInst *createPopcntIntrinsic(IRBuilderTy &IRB, Value *Val, DebugLoc DL); +}; + +class LoopIdiomRecognize : public LoopPass { + Loop *CurLoop; + DominatorTree *DT; + ScalarEvolution *SE; + TargetLibraryInfo *TLI; + const TargetTransformInfo *TTI; + +public: + static char ID; + explicit LoopIdiomRecognize() : LoopPass(ID) { + initializeLoopIdiomRecognizePass(*PassRegistry::getPassRegistry()); + DT = nullptr; + SE = nullptr; + TLI = nullptr; + TTI = nullptr; + } - /// This class defines some utility functions for loop idiom recognization. - class LIRUtil { - public: - /// Return true iff the block contains nothing but an uncondition branch - /// (aka goto instruction). - static bool isAlmostEmpty(BasicBlock *); + bool runOnLoop(Loop *L, LPPassManager &LPM) override; + + /// This transformation requires natural loop information & requires that + /// loop preheaders be inserted into the CFG. + /// + void getAnalysisUsage(AnalysisUsage &AU) const override { + AU.addRequired(); + AU.addPreserved(); + AU.addRequiredID(LoopSimplifyID); + AU.addPreservedID(LoopSimplifyID); + AU.addRequiredID(LCSSAID); + AU.addPreservedID(LCSSAID); + AU.addRequired(); + AU.addPreserved(); + AU.addRequired(); + AU.addPreserved(); + AU.addPreserved(); + AU.addRequired(); + AU.addRequired(); + AU.addRequired(); + } - static BranchInst *getBranch(BasicBlock *BB) { - return dyn_cast(BB->getTerminator()); - } + DominatorTree *getDominatorTree() { + return DT ? DT + : (DT = &getAnalysis().getDomTree()); + } - /// Derive the precondition block (i.e the block that guards the loop - /// preheader) from the given preheader. - static BasicBlock *getPrecondBb(BasicBlock *PreHead); - }; - - /// This class is to recoginize idioms of population-count conducted in - /// a noncountable loop. Currently it only recognizes this pattern: - /// \code - /// while(x) {cnt++; ...; x &= x - 1; ...} - /// \endcode - class NclPopcountRecognize { - LoopIdiomRecognize &LIR; - Loop *CurLoop; - BasicBlock *PreCondBB; - - typedef IRBuilder<> IRBuilderTy; - - public: - explicit NclPopcountRecognize(LoopIdiomRecognize &TheLIR); - bool recognize(); - - private: - /// Take a glimpse of the loop to see if we need to go ahead recoginizing - /// the idiom. - bool preliminaryScreen(); - - /// Check if the given conditional branch is based on the comparison - /// between a variable and zero, and if the variable is non-zero, the - /// control yields to the loop entry. If the branch matches the behavior, - /// the variable involved in the comparion is returned. This function will - /// be called to see if the precondition and postcondition of the loop - /// are in desirable form. - Value *matchCondition(BranchInst *Br, BasicBlock *NonZeroTarget) const; - - /// Return true iff the idiom is detected in the loop. and 1) \p CntInst - /// is set to the instruction counting the population bit. 2) \p CntPhi - /// is set to the corresponding phi node. 3) \p Var is set to the value - /// whose population bits are being counted. - bool detectIdiom - (Instruction *&CntInst, PHINode *&CntPhi, Value *&Var) const; - - /// Insert ctpop intrinsic function and some obviously dead instructions. - void transform(Instruction *CntInst, PHINode *CntPhi, Value *Var); - - /// Create llvm.ctpop.* intrinsic function. - CallInst *createPopcntIntrinsic(IRBuilderTy &IRB, Value *Val, DebugLoc DL); - }; - - class LoopIdiomRecognize : public LoopPass { - Loop *CurLoop; - DominatorTree *DT; - ScalarEvolution *SE; - TargetLibraryInfo *TLI; - const TargetTransformInfo *TTI; - public: - static char ID; - explicit LoopIdiomRecognize() : LoopPass(ID) { - initializeLoopIdiomRecognizePass(*PassRegistry::getPassRegistry()); - DT = nullptr; - SE = nullptr; - TLI = nullptr; - TTI = nullptr; - } + ScalarEvolution *getScalarEvolution() { + return SE ? SE : (SE = &getAnalysis()); + } - bool runOnLoop(Loop *L, LPPassManager &LPM) override; - bool runOnLoopBlock(BasicBlock *BB, const SCEV *BECount, - SmallVectorImpl &ExitBlocks); - - bool processLoopStore(StoreInst *SI, const SCEV *BECount); - bool processLoopMemSet(MemSetInst *MSI, const SCEV *BECount); - - bool processLoopStridedStore(Value *DestPtr, unsigned StoreSize, - unsigned StoreAlignment, - Value *SplatValue, Instruction *TheStore, - const SCEVAddRecExpr *Ev, - const SCEV *BECount); - bool processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize, - const SCEVAddRecExpr *StoreEv, - const SCEVAddRecExpr *LoadEv, - const SCEV *BECount); - - /// This transformation requires natural loop information & requires that - /// loop preheaders be inserted into the CFG. - /// - void getAnalysisUsage(AnalysisUsage &AU) const override { - AU.addRequired(); - AU.addPreserved(); - AU.addRequiredID(LoopSimplifyID); - AU.addPreservedID(LoopSimplifyID); - AU.addRequiredID(LCSSAID); - AU.addPreservedID(LCSSAID); - AU.addRequired(); - AU.addPreserved(); - AU.addRequired(); - AU.addPreserved(); - AU.addPreserved(); - AU.addRequired(); - AU.addRequired(); - AU.addRequired(); - } + TargetLibraryInfo *getTargetLibraryInfo() { + if (!TLI) + TLI = &getAnalysis().getTLI(); - DominatorTree *getDominatorTree() { - return DT ? DT - : (DT = &getAnalysis().getDomTree()); - } + return TLI; + } - ScalarEvolution *getScalarEvolution() { - return SE ? SE : (SE = &getAnalysis()); - } + const TargetTransformInfo *getTargetTransformInfo() { + return TTI ? TTI + : (TTI = &getAnalysis().getTTI( + *CurLoop->getHeader()->getParent())); + } - TargetLibraryInfo *getTargetLibraryInfo() { - if (!TLI) - TLI = &getAnalysis().getTLI(); + Loop *getLoop() const { return CurLoop; } - return TLI; - } +private: + /// \name Countable Loop Idiom Handling + /// @{ - const TargetTransformInfo *getTargetTransformInfo() { - return TTI ? TTI - : (TTI = &getAnalysis().getTTI( - *CurLoop->getHeader()->getParent())); - } + bool runOnCountableLoop(); + bool runOnLoopBlock(BasicBlock *BB, const SCEV *BECount, + SmallVectorImpl &ExitBlocks); - Loop *getLoop() const { return CurLoop; } + bool processLoopStore(StoreInst *SI, const SCEV *BECount); + bool processLoopMemSet(MemSetInst *MSI, const SCEV *BECount); - private: - bool runOnNoncountableLoop(); - bool runOnCountableLoop(); - }; -} + bool processLoopStridedStore(Value *DestPtr, unsigned StoreSize, + unsigned StoreAlignment, Value *SplatValue, + Instruction *TheStore, const SCEVAddRecExpr *Ev, + const SCEV *BECount); + bool processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize, + const SCEVAddRecExpr *StoreEv, + const SCEVAddRecExpr *LoadEv, + const SCEV *BECount); + + /// @} + /// \name Noncountable Loop Idiom Handling + /// @{ + + bool runOnNoncountableLoop(); + + /// @} +}; + +} // End anonymous namespace. char LoopIdiomRecognize::ID = 0; INITIALIZE_PASS_BEGIN(LoopIdiomRecognize, "loop-idiom", "Recognize loop idioms", @@ -240,41 +234,14 @@ static void deleteDeadInstruction(Instruction *I, RecursivelyDeleteTriviallyDeadInstructions(Op, TLI); } -//===----------------------------------------------------------------------===// -// -// Implementation of LIRUtil -// -//===----------------------------------------------------------------------===// - -// This function will return true iff the given block contains nothing but goto. -// A typical usage of this function is to check if the preheader function is -// "almost" empty such that generated intrinsic functions can be moved across -// the preheader and be placed at the end of the precondition block without -// the concern of breaking data dependence. -bool LIRUtil::isAlmostEmpty(BasicBlock *BB) { - if (BranchInst *Br = getBranch(BB)) { - return Br->isUnconditional() && Br == BB->begin(); - } - return false; -} - -BasicBlock *LIRUtil::getPrecondBb(BasicBlock *PreHead) { - if (BasicBlock *BB = PreHead->getSinglePredecessor()) { - BranchInst *Br = getBranch(BB); - return Br && Br->isConditional() ? BB : nullptr; - } - return nullptr; -} - //===----------------------------------------------------------------------===// // // Implementation of NclPopcountRecognize // //===----------------------------------------------------------------------===// -NclPopcountRecognize::NclPopcountRecognize(LoopIdiomRecognize &TheLIR): - LIR(TheLIR), CurLoop(TheLIR.getLoop()), PreCondBB(nullptr) { -} +NclPopcountRecognize::NclPopcountRecognize(LoopIdiomRecognize &TheLIR) + : LIR(TheLIR), CurLoop(TheLIR.getLoop()), PreCondBB(nullptr) {} bool NclPopcountRecognize::preliminaryScreen() { const TargetTransformInfo *TTI = LIR.getTargetTransformInfo(); @@ -296,16 +263,24 @@ bool NclPopcountRecognize::preliminaryScreen() { return false; } - // It should have a preheader containing nothing but a goto instruction. - BasicBlock *PreHead = CurLoop->getLoopPreheader(); - if (!PreHead || !LIRUtil::isAlmostEmpty(PreHead)) + // It should have a preheader containing nothing but an unconditional branch. + BasicBlock *PH = CurLoop->getLoopPreheader(); + if (!PH) + return false; + if (&PH->front() != PH->getTerminator()) + return false; + auto *EntryBI = dyn_cast(PH->getTerminator()); + if (!EntryBI || EntryBI->isConditional()) return false; // It should have a precondition block where the generated popcount instrinsic - // function will be inserted. - PreCondBB = LIRUtil::getPrecondBb(PreHead); + // function can be inserted. + PreCondBB = PH->getSinglePredecessor(); if (!PreCondBB) return false; + auto *PreCondBI = dyn_cast(PreCondBB->getTerminator()); + if (!PreCondBI || PreCondBI->isUnconditional()) + return false; return true; } @@ -331,8 +306,7 @@ Value *NclPopcountRecognize::matchCondition(BranchInst *Br, return nullptr; } -bool NclPopcountRecognize::detectIdiom(Instruction *&CntInst, - PHINode *&CntPhi, +bool NclPopcountRecognize::detectIdiom(Instruction *&CntInst, PHINode *&CntPhi, Value *&Var) const { // Following code tries to detect this idiom: // @@ -366,7 +340,8 @@ bool NclPopcountRecognize::detectIdiom(Instruction *&CntInst, // step 1: Check if the loop-back branch is in desirable form. { - if (Value *T = matchCondition (LIRUtil::getBranch(LoopEntry), LoopEntry)) + if (Value *T = matchCondition( + dyn_cast(LoopEntry->getTerminator()), LoopEntry)) DefX2 = dyn_cast(T); else return false; @@ -392,7 +367,8 @@ bool NclPopcountRecognize::detectIdiom(Instruction *&CntInst, ConstantInt *Dec = dyn_cast(SubInst->getOperand(1)); if (!Dec || !((SubInst->getOpcode() == Instruction::Sub && Dec->isOne()) || - (SubInst->getOpcode() == Instruction::Add && Dec->isAllOnesValue()))) { + (SubInst->getOpcode() == Instruction::Add && + Dec->isAllOnesValue()))) { return false; } } @@ -410,7 +386,8 @@ bool NclPopcountRecognize::detectIdiom(Instruction *&CntInst, { CountInst = nullptr; for (BasicBlock::iterator Iter = LoopEntry->getFirstNonPHI(), - IterE = LoopEntry->end(); Iter != IterE; Iter++) { + IterE = LoopEntry->end(); + Iter != IterE; Iter++) { Instruction *Inst = Iter; if (Inst->getOpcode() != Instruction::Add) continue; @@ -427,7 +404,8 @@ bool NclPopcountRecognize::detectIdiom(Instruction *&CntInst, bool LiveOutLoop = false; for (User *U : Inst->users()) { if ((cast(U))->getParent() != LoopEntry) { - LiveOutLoop = true; break; + LiveOutLoop = true; + break; } } @@ -445,8 +423,8 @@ bool NclPopcountRecognize::detectIdiom(Instruction *&CntInst, // step 5: check if the precondition is in this form: // "if (x != 0) goto loop-head ; else goto somewhere-we-don't-care;" { - BranchInst *PreCondBr = LIRUtil::getBranch(PreCondBB); - Value *T = matchCondition (PreCondBr, CurLoop->getLoopPreheader()); + auto *PreCondBr = dyn_cast(PreCondBB->getTerminator()); + Value *T = matchCondition(PreCondBr, CurLoop->getLoopPreheader()); if (T != PhiX->getOperand(0) && T != PhiX->getOperand(1)) return false; @@ -458,13 +436,13 @@ bool NclPopcountRecognize::detectIdiom(Instruction *&CntInst, return true; } -void NclPopcountRecognize::transform(Instruction *CntInst, - PHINode *CntPhi, Value *Var) { +void NclPopcountRecognize::transform(Instruction *CntInst, PHINode *CntPhi, + Value *Var) { ScalarEvolution *SE = LIR.getScalarEvolution(); TargetLibraryInfo *TLI = LIR.getTargetLibraryInfo(); BasicBlock *PreHead = CurLoop->getLoopPreheader(); - BranchInst *PreCondBr = LIRUtil::getBranch(PreCondBB); + auto *PreCondBr = dyn_cast(PreCondBB->getTerminator()); const DebugLoc DL = CntInst->getDebugLoc(); // Assuming before transformation, the loop is following: @@ -477,7 +455,7 @@ void NclPopcountRecognize::transform(Instruction *CntInst, { PopCnt = createPopcntIntrinsic(Builder, Var, DL); NewCount = PopCntZext = - Builder.CreateZExtOrTrunc(PopCnt, cast(CntPhi->getType())); + Builder.CreateZExtOrTrunc(PopCnt, cast(CntPhi->getType())); if (NewCount != PopCnt) (cast(NewCount))->setDebugLoc(DL); @@ -506,9 +484,9 @@ void NclPopcountRecognize::transform(Instruction *CntInst, if (PreCond->getOperand(0) != Var) std::swap(Opnd0, Opnd1); - ICmpInst *NewPreCond = - cast(Builder.CreateICmp(PreCond->getPredicate(), Opnd0, Opnd1)); - PreCond->replaceAllUsesWith(NewPreCond); + ICmpInst *NewPreCond = cast( + Builder.CreateICmp(PreCond->getPredicate(), Opnd0, Opnd1)); + PreCondBr->setCondition(NewPreCond); RecursivelyDeleteTriviallyDeadInstructions(PreCond, TLI); } @@ -535,7 +513,7 @@ void NclPopcountRecognize::transform(Instruction *CntInst, // do { cnt++; x &= x-1; t--) } while (t > 0); BasicBlock *Body = *(CurLoop->block_begin()); { - BranchInst *LbBr = LIRUtil::getBranch(Body); + auto *LbBr = dyn_cast(Body->getTerminator()); ICmpInst *LbCond = cast(LbBr->getCondition()); Type *Ty = TripCnt->getType(); @@ -544,14 +522,14 @@ void NclPopcountRecognize::transform(Instruction *CntInst, Builder.SetInsertPoint(LbCond); Value *Opnd1 = cast(TcPhi); Value *Opnd2 = cast(ConstantInt::get(Ty, 1)); - Instruction *TcDec = - cast(Builder.CreateSub(Opnd1, Opnd2, "tcdec", false, true)); + Instruction *TcDec = cast( + Builder.CreateSub(Opnd1, Opnd2, "tcdec", false, true)); TcPhi->addIncoming(TripCnt, PreHead); TcPhi->addIncoming(TcDec, Body); - CmpInst::Predicate Pred = (LbBr->getSuccessor(0) == Body) ? - CmpInst::ICMP_UGT : CmpInst::ICMP_SLE; + CmpInst::Predicate Pred = + (LbBr->getSuccessor(0) == Body) ? CmpInst::ICMP_UGT : CmpInst::ICMP_SLE; LbCond->setPredicate(Pred); LbCond->setOperand(0, TcDec); LbCond->setOperand(1, cast(ConstantInt::get(Ty, 0))); @@ -569,8 +547,8 @@ void NclPopcountRecognize::transform(Instruction *CntInst, CallInst *NclPopcountRecognize::createPopcntIntrinsic(IRBuilderTy &IRBuilder, Value *Val, DebugLoc DL) { - Value *Ops[] = { Val }; - Type *Tys[] = { Val->getType() }; + Value *Ops[] = {Val}; + Type *Tys[] = {Val->getType()}; Module *M = (*(CurLoop->block_begin()))->getParent()->getParent(); Value *Func = Intrinsic::getDeclaration(M, Intrinsic::ctpop, Tys); @@ -584,7 +562,6 @@ CallInst *NclPopcountRecognize::createPopcntIntrinsic(IRBuilderTy &IRBuilder, /// detected, transform the relevant code to popcount intrinsic function /// call, and return true; otherwise, return false. bool NclPopcountRecognize::recognize() { - if (!LIR.getTargetTransformInfo()) return false; @@ -609,9 +586,33 @@ bool NclPopcountRecognize::recognize() { // //===----------------------------------------------------------------------===// +bool LoopIdiomRecognize::runOnLoop(Loop *L, LPPassManager &LPM) { + if (skipOptnoneFunction(L)) + return false; + + CurLoop = L; + + // If the loop could not be converted to canonical form, it must have an + // indirectbr in it, just give up. + if (!L->getLoopPreheader()) + return false; + + // Disable loop idiom recognition if the function's name is a common idiom. + StringRef Name = L->getHeader()->getParent()->getName(); + if (Name == "memset" || Name == "memcpy") + return false; + + SE = &getAnalysis(); + if (SE->hasLoopInvariantBackedgeTakenCount(L)) + return runOnCountableLoop(); + return runOnNoncountableLoop(); +} + bool LoopIdiomRecognize::runOnCountableLoop() { const SCEV *BECount = SE->getBackedgeTakenCount(CurLoop); - if (isa(BECount)) return false; + assert(!isa(BECount) && + "runOnCountableLoop() called on a loop without a predictable" + "backedge-taken count"); // If this loop executes exactly one time, then it should be peeled, not // optimized by this pass. @@ -628,61 +629,31 @@ bool LoopIdiomRecognize::runOnCountableLoop() { // set TLI (void)getTargetLibraryInfo(); - SmallVector ExitBlocks; + SmallVector ExitBlocks; CurLoop->getUniqueExitBlocks(ExitBlocks); DEBUG(dbgs() << "loop-idiom Scanning: F[" - << CurLoop->getHeader()->getParent()->getName() - << "] Loop %" << CurLoop->getHeader()->getName() << "\n"); + << CurLoop->getHeader()->getParent()->getName() << "] Loop %" + << CurLoop->getHeader()->getName() << "\n"); bool MadeChange = false; // Scan all the blocks in the loop that are not in subloops. - for (Loop::block_iterator BI = CurLoop->block_begin(), - E = CurLoop->block_end(); BI != E; ++BI) { + for (auto *BB : CurLoop->getBlocks()) { // Ignore blocks in subloops. - if (LI.getLoopFor(*BI) != CurLoop) + if (LI.getLoopFor(BB) != CurLoop) continue; - MadeChange |= runOnLoopBlock(*BI, BECount, ExitBlocks); + MadeChange |= runOnLoopBlock(BB, BECount, ExitBlocks); } return MadeChange; } -bool LoopIdiomRecognize::runOnNoncountableLoop() { - NclPopcountRecognize Popcount(*this); - if (Popcount.recognize()) - return true; - - return false; -} - -bool LoopIdiomRecognize::runOnLoop(Loop *L, LPPassManager &LPM) { - if (skipOptnoneFunction(L)) - return false; - - CurLoop = L; - - // If the loop could not be converted to canonical form, it must have an - // indirectbr in it, just give up. - if (!L->getLoopPreheader()) - return false; - - // Disable loop idiom recognition if the function's name is a common idiom. - StringRef Name = L->getHeader()->getParent()->getName(); - if (Name == "memset" || Name == "memcpy") - return false; - - SE = &getAnalysis(); - if (SE->hasLoopInvariantBackedgeTakenCount(L)) - return runOnCountableLoop(); - return runOnNoncountableLoop(); -} - /// runOnLoopBlock - Process the specified block, which lives in a counted loop /// with the specified backedge count. This block is known to be in the current /// loop and not in any subloops. -bool LoopIdiomRecognize::runOnLoopBlock(BasicBlock *BB, const SCEV *BECount, - SmallVectorImpl &ExitBlocks) { +bool LoopIdiomRecognize::runOnLoopBlock( + BasicBlock *BB, const SCEV *BECount, + SmallVectorImpl &ExitBlocks) { // We can only promote stores in this block if they are unconditionally // executed in the loop. For a block to be unconditionally executed, it has // to dominate all the exit blocks of the loop. Verify this now. @@ -691,12 +662,13 @@ bool LoopIdiomRecognize::runOnLoopBlock(BasicBlock *BB, const SCEV *BECount, return false; bool MadeChange = false; - for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ) { + for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E;) { Instruction *Inst = I++; // Look for store instructions, which may be optimized to memset/memcpy. - if (StoreInst *SI = dyn_cast(Inst)) { + if (StoreInst *SI = dyn_cast(Inst)) { WeakVH InstPtr(I); - if (!processLoopStore(SI, BECount)) continue; + if (!processLoopStore(SI, BECount)) + continue; MadeChange = true; // If processing the store invalidated our iterator, start over from the @@ -707,9 +679,10 @@ bool LoopIdiomRecognize::runOnLoopBlock(BasicBlock *BB, const SCEV *BECount, } // Look for memset instructions, which may be optimized to a larger memset. - if (MemSetInst *MSI = dyn_cast(Inst)) { + if (MemSetInst *MSI = dyn_cast(Inst)) { WeakVH InstPtr(I); - if (!processLoopMemSet(MSI, BECount)) continue; + if (!processLoopMemSet(MSI, BECount)) + continue; MadeChange = true; // If processing the memset invalidated our iterator, start over from the @@ -723,10 +696,10 @@ bool LoopIdiomRecognize::runOnLoopBlock(BasicBlock *BB, const SCEV *BECount, return MadeChange; } - /// processLoopStore - See if this store can be promoted to a memset or memcpy. bool LoopIdiomRecognize::processLoopStore(StoreInst *SI, const SCEV *BECount) { - if (!SI->isSimple()) return false; + if (!SI->isSimple()) + return false; Value *StoredVal = SI->getValueOperand(); Value *StorePtr = SI->getPointerOperand(); @@ -741,7 +714,7 @@ bool LoopIdiomRecognize::processLoopStore(StoreInst *SI, const SCEV *BECount) { // loop, which indicates a strided store. If we have something else, it's a // random store we can't handle. const SCEVAddRecExpr *StoreEv = - dyn_cast(SE->getSCEV(StorePtr)); + dyn_cast(SE->getSCEV(StorePtr)); if (!StoreEv || StoreEv->getLoop() != CurLoop || !StoreEv->isAffine()) return false; @@ -772,22 +745,23 @@ bool LoopIdiomRecognize::processLoopStore(StoreInst *SI, const SCEV *BECount) { // for (i) A[i] = B[i]; if (LoadInst *LI = dyn_cast(StoredVal)) { const SCEVAddRecExpr *LoadEv = - dyn_cast(SE->getSCEV(LI->getOperand(0))); + dyn_cast(SE->getSCEV(LI->getOperand(0))); if (LoadEv && LoadEv->getLoop() == CurLoop && LoadEv->isAffine() && StoreEv->getOperand(1) == LoadEv->getOperand(1) && LI->isSimple()) if (processLoopStoreOfLoopLoad(SI, StoreSize, StoreEv, LoadEv, BECount)) return true; } - //errs() << "UNHANDLED strided store: " << *StoreEv << " - " << *SI << "\n"; + // errs() << "UNHANDLED strided store: " << *StoreEv << " - " << *SI << "\n"; return false; } /// processLoopMemSet - See if this memset can be promoted to a large memset. -bool LoopIdiomRecognize:: -processLoopMemSet(MemSetInst *MSI, const SCEV *BECount) { +bool LoopIdiomRecognize::processLoopMemSet(MemSetInst *MSI, + const SCEV *BECount) { // We can only handle non-volatile memsets with a constant size. - if (MSI->isVolatile() || !isa(MSI->getLength())) return false; + if (MSI->isVolatile() || !isa(MSI->getLength())) + return false; // If we're not allowed to hack on memset, we fail. if (!TLI->has(LibFunc::memset)) @@ -817,39 +791,37 @@ processLoopMemSet(MemSetInst *MSI, const SCEV *BECount) { return false; return processLoopStridedStore(Pointer, (unsigned)SizeInBytes, - MSI->getAlignment(), MSI->getValue(), - MSI, Ev, BECount); + MSI->getAlignment(), MSI->getValue(), MSI, Ev, + BECount); } - /// mayLoopAccessLocation - Return true if the specified loop might access the /// specified pointer location, which is a loop-strided access. The 'Access' /// argument specifies what the verboten forms of access are (read or write). -static bool mayLoopAccessLocation(Value *Ptr,AliasAnalysis::ModRefResult Access, - Loop *L, const SCEV *BECount, - unsigned StoreSize, AliasAnalysis &AA, +static bool mayLoopAccessLocation(Value *Ptr, ModRefInfo Access, Loop *L, + const SCEV *BECount, unsigned StoreSize, + AliasAnalysis &AA, Instruction *IgnoredStore) { // Get the location that may be stored across the loop. Since the access is // strided positively through memory, we say that the modified location starts // at the pointer and has infinite size. - uint64_t AccessSize = AliasAnalysis::UnknownSize; + uint64_t AccessSize = MemoryLocation::UnknownSize; // If the loop iterates a fixed number of times, we can refine the access size // to be exactly the size of the memset, which is (BECount+1)*StoreSize if (const SCEVConstant *BECst = dyn_cast(BECount)) - AccessSize = (BECst->getValue()->getZExtValue()+1)*StoreSize; + AccessSize = (BECst->getValue()->getZExtValue() + 1) * StoreSize; // TODO: For this to be really effective, we have to dive into the pointer // operand in the store. Store to &A[i] of 100 will always return may alias // with store of &A[100], we need to StoreLoc to be "A" with size of 100, // which will then no-alias a store to &A[100]. - AliasAnalysis::Location StoreLoc(Ptr, AccessSize); + MemoryLocation StoreLoc(Ptr, AccessSize); for (Loop::block_iterator BI = L->block_begin(), E = L->block_end(); BI != E; ++BI) for (BasicBlock::iterator I = (*BI)->begin(), E = (*BI)->end(); I != E; ++I) - if (&*I != IgnoredStore && - (AA.getModRefInfo(I, StoreLoc) & Access)) + if (&*I != IgnoredStore && (AA.getModRefInfo(I, StoreLoc) & Access)) return true; return false; @@ -866,11 +838,12 @@ static Constant *getMemSetPatternValue(Value *V, const DataLayout &DL) { // array. We could theoretically do a store to an alloca or something, but // that doesn't seem worthwhile. Constant *C = dyn_cast(V); - if (!C) return nullptr; + if (!C) + return nullptr; // Only handle simple values that are a power of two bytes in size. uint64_t Size = DL.getTypeSizeInBits(V->getType()); - if (Size == 0 || (Size & 7) || (Size & (Size-1))) + if (Size == 0 || (Size & 7) || (Size & (Size - 1))) return nullptr; // Don't care enough about darwin/ppc to implement this. @@ -882,25 +855,25 @@ static Constant *getMemSetPatternValue(Value *V, const DataLayout &DL) { // TODO: If CI is larger than 16-bytes, we can try slicing it in half to see // if the top and bottom are the same (e.g. for vectors and large integers). - if (Size > 16) return nullptr; + if (Size > 16) + return nullptr; // If the constant is exactly 16 bytes, just use it. - if (Size == 16) return C; + if (Size == 16) + return C; // Otherwise, we'll use an array of the constants. - unsigned ArraySize = 16/Size; + unsigned ArraySize = 16 / Size; ArrayType *AT = ArrayType::get(V->getType(), ArraySize); - return ConstantArray::get(AT, std::vector(ArraySize, C)); + return ConstantArray::get(AT, std::vector(ArraySize, C)); } - /// processLoopStridedStore - We see a strided store of some value. If we can /// transform this into a memset or memset_pattern in the loop preheader, do so. -bool LoopIdiomRecognize:: -processLoopStridedStore(Value *DestPtr, unsigned StoreSize, - unsigned StoreAlignment, Value *StoredVal, - Instruction *TheStore, const SCEVAddRecExpr *Ev, - const SCEV *BECount) { +bool LoopIdiomRecognize::processLoopStridedStore( + Value *DestPtr, unsigned StoreSize, unsigned StoreAlignment, + Value *StoredVal, Instruction *TheStore, const SCEVAddRecExpr *Ev, + const SCEV *BECount) { // If the stored value is a byte-wise value (like i32 -1), then it may be // turned into a memset of i8 -1, assuming that all the consecutive bytes @@ -935,7 +908,7 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize, // header. This allows us to insert code for it in the preheader. BasicBlock *Preheader = CurLoop->getLoopPreheader(); IRBuilder<> Builder(Preheader->getTerminator()); - SCEVExpander Expander(*SE, "loop-idiom"); + SCEVExpander Expander(*SE, DL, "loop-idiom"); Type *DestInt8PtrTy = Builder.getInt8PtrTy(DestAS); @@ -944,13 +917,11 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize, // would be unsafe to do if there is anything else in the loop that may read // or write to the aliased location. Check for any overlap by generating the // base pointer and checking the region. - Value *BasePtr = - Expander.expandCodeFor(Ev->getStart(), DestInt8PtrTy, - Preheader->getTerminator()); + Value *BasePtr = Expander.expandCodeFor(Ev->getStart(), DestInt8PtrTy, + Preheader->getTerminator()); - if (mayLoopAccessLocation(BasePtr, AliasAnalysis::ModRef, - CurLoop, BECount, - StoreSize, getAnalysis(), TheStore)) { + if (mayLoopAccessLocation(BasePtr, MRI_ModRef, CurLoop, BECount, StoreSize, + getAnalysis(), TheStore)) { Expander.clear(); // If we generated new code for the base pointer, clean up. RecursivelyDeleteTriviallyDeadInstructions(BasePtr, TLI); @@ -961,36 +932,31 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize, // The # stored bytes is (BECount+1)*Size. Expand the trip count out to // pointer size if it isn't already. - Type *IntPtr = Builder.getIntPtrTy(&DL, DestAS); + Type *IntPtr = Builder.getIntPtrTy(DL, DestAS); BECount = SE->getTruncateOrZeroExtend(BECount, IntPtr); - const SCEV *NumBytesS = SE->getAddExpr(BECount, SE->getConstant(IntPtr, 1), - SCEV::FlagNUW); + const SCEV *NumBytesS = + SE->getAddExpr(BECount, SE->getConstant(IntPtr, 1), SCEV::FlagNUW); if (StoreSize != 1) { NumBytesS = SE->getMulExpr(NumBytesS, SE->getConstant(IntPtr, StoreSize), SCEV::FlagNUW); } Value *NumBytes = - Expander.expandCodeFor(NumBytesS, IntPtr, Preheader->getTerminator()); + Expander.expandCodeFor(NumBytesS, IntPtr, Preheader->getTerminator()); CallInst *NewCall; if (SplatValue) { - NewCall = Builder.CreateMemSet(BasePtr, - SplatValue, - NumBytes, - StoreAlignment); + NewCall = + Builder.CreateMemSet(BasePtr, SplatValue, NumBytes, StoreAlignment); } else { // Everything is emitted in default address space Type *Int8PtrTy = DestInt8PtrTy; Module *M = TheStore->getParent()->getParent()->getParent(); - Value *MSP = M->getOrInsertFunction("memset_pattern16", - Builder.getVoidTy(), - Int8PtrTy, - Int8PtrTy, - IntPtr, - (void*)nullptr); + Value *MSP = + M->getOrInsertFunction("memset_pattern16", Builder.getVoidTy(), + Int8PtrTy, Int8PtrTy, IntPtr, (void *)nullptr); // Otherwise we should form a memset_pattern16. PatternValue is known to be // an constant array of 16-bytes. Plop the value into a mergable global. @@ -1000,7 +966,7 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize, GV->setUnnamedAddr(true); // Ok to merge these. GV->setAlignment(16); Value *PatternPtr = ConstantExpr::getBitCast(GV, Int8PtrTy); - NewCall = Builder.CreateCall3(MSP, BasePtr, PatternPtr, NumBytes); + NewCall = Builder.CreateCall(MSP, {BasePtr, PatternPtr, NumBytes}); } DEBUG(dbgs() << " Formed memset: " << *NewCall << "\n" @@ -1016,11 +982,9 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize, /// processLoopStoreOfLoopLoad - We see a strided store whose value is a /// same-strided load. -bool LoopIdiomRecognize:: -processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize, - const SCEVAddRecExpr *StoreEv, - const SCEVAddRecExpr *LoadEv, - const SCEV *BECount) { +bool LoopIdiomRecognize::processLoopStoreOfLoopLoad( + StoreInst *SI, unsigned StoreSize, const SCEVAddRecExpr *StoreEv, + const SCEVAddRecExpr *LoadEv, const SCEV *BECount) { // If we're not allowed to form memcpy, we fail. if (!TLI->has(LibFunc::memcpy)) return false; @@ -1032,7 +996,8 @@ processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize, // header. This allows us to insert code for it in the preheader. BasicBlock *Preheader = CurLoop->getLoopPreheader(); IRBuilder<> Builder(Preheader->getTerminator()); - SCEVExpander Expander(*SE, "loop-idiom"); + const DataLayout &DL = Preheader->getModule()->getDataLayout(); + SCEVExpander Expander(*SE, DL, "loop-idiom"); // Okay, we have a strided store "p[i]" of a loaded value. We can turn // this into a memcpy in the loop preheader now if we want. However, this @@ -1040,14 +1005,12 @@ processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize, // or write the memory region we're storing to. This includes the load that // feeds the stores. Check for an alias by generating the base address and // checking everything. - Value *StoreBasePtr = - Expander.expandCodeFor(StoreEv->getStart(), - Builder.getInt8PtrTy(SI->getPointerAddressSpace()), - Preheader->getTerminator()); + Value *StoreBasePtr = Expander.expandCodeFor( + StoreEv->getStart(), Builder.getInt8PtrTy(SI->getPointerAddressSpace()), + Preheader->getTerminator()); - if (mayLoopAccessLocation(StoreBasePtr, AliasAnalysis::ModRef, - CurLoop, BECount, StoreSize, - getAnalysis(), SI)) { + if (mayLoopAccessLocation(StoreBasePtr, MRI_ModRef, CurLoop, BECount, + StoreSize, getAnalysis(), SI)) { Expander.clear(); // If we generated new code for the base pointer, clean up. RecursivelyDeleteTriviallyDeadInstructions(StoreBasePtr, TLI); @@ -1056,13 +1019,12 @@ processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize, // For a memcpy, we have to make sure that the input array is not being // mutated by the loop. - Value *LoadBasePtr = - Expander.expandCodeFor(LoadEv->getStart(), - Builder.getInt8PtrTy(LI->getPointerAddressSpace()), - Preheader->getTerminator()); + Value *LoadBasePtr = Expander.expandCodeFor( + LoadEv->getStart(), Builder.getInt8PtrTy(LI->getPointerAddressSpace()), + Preheader->getTerminator()); - if (mayLoopAccessLocation(LoadBasePtr, AliasAnalysis::Mod, CurLoop, BECount, - StoreSize, getAnalysis(), SI)) { + if (mayLoopAccessLocation(LoadBasePtr, MRI_Mod, CurLoop, BECount, StoreSize, + getAnalysis(), SI)) { Expander.clear(); // If we generated new code for the base pointer, clean up. RecursivelyDeleteTriviallyDeadInstructions(LoadBasePtr, TLI); @@ -1072,35 +1034,40 @@ processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize, // Okay, everything is safe, we can transform this! - // The # stored bytes is (BECount+1)*Size. Expand the trip count out to // pointer size if it isn't already. - auto &DL = CurLoop->getHeader()->getModule()->getDataLayout(); - Type *IntPtrTy = Builder.getIntPtrTy(&DL, SI->getPointerAddressSpace()); + Type *IntPtrTy = Builder.getIntPtrTy(DL, SI->getPointerAddressSpace()); BECount = SE->getTruncateOrZeroExtend(BECount, IntPtrTy); - const SCEV *NumBytesS = SE->getAddExpr(BECount, SE->getConstant(IntPtrTy, 1), - SCEV::FlagNUW); + const SCEV *NumBytesS = + SE->getAddExpr(BECount, SE->getConstant(IntPtrTy, 1), SCEV::FlagNUW); if (StoreSize != 1) NumBytesS = SE->getMulExpr(NumBytesS, SE->getConstant(IntPtrTy, StoreSize), SCEV::FlagNUW); Value *NumBytes = - Expander.expandCodeFor(NumBytesS, IntPtrTy, Preheader->getTerminator()); + Expander.expandCodeFor(NumBytesS, IntPtrTy, Preheader->getTerminator()); CallInst *NewCall = - Builder.CreateMemCpy(StoreBasePtr, LoadBasePtr, NumBytes, - std::min(SI->getAlignment(), LI->getAlignment())); + Builder.CreateMemCpy(StoreBasePtr, LoadBasePtr, NumBytes, + std::min(SI->getAlignment(), LI->getAlignment())); NewCall->setDebugLoc(SI->getDebugLoc()); DEBUG(dbgs() << " Formed memcpy: " << *NewCall << "\n" << " from load ptr=" << *LoadEv << " at: " << *LI << "\n" << " from store ptr=" << *StoreEv << " at: " << *SI << "\n"); - // Okay, the memset has been formed. Zap the original store and anything that // feeds into it. deleteDeadInstruction(SI, TLI); ++NumMemCpy; return true; } + +bool LoopIdiomRecognize::runOnNoncountableLoop() { + NclPopcountRecognize Popcount(*this); + if (Popcount.recognize()) + return true; + + return false; +}