//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "loop-idiom"
#include "llvm/Transforms/Scalar.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.h"
#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/DataLayout.h"
-#include "llvm/IRBuilder.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/Module.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Module.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetLibraryInfo.h"
-#include "llvm/TargetTransformInfo.h"
#include "llvm/Transforms/Utils/Local.h"
using namespace llvm;
+#define DEBUG_TYPE "loop-idiom"
+
STATISTIC(NumMemSet, "Number of memset's formed from loop stores");
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<LoopInfoWrapperPass>();
+ AU.addPreserved<LoopInfoWrapperPass>();
+ AU.addRequiredID(LoopSimplifyID);
+ AU.addPreservedID(LoopSimplifyID);
+ AU.addRequiredID(LCSSAID);
+ AU.addPreservedID(LCSSAID);
+ AU.addRequired<AliasAnalysis>();
+ AU.addPreserved<AliasAnalysis>();
+ AU.addRequired<ScalarEvolution>();
+ AU.addPreserved<ScalarEvolution>();
+ AU.addPreserved<DominatorTreeWrapperPass>();
+ AU.addRequired<DominatorTreeWrapperPass>();
+ AU.addRequired<TargetLibraryInfoWrapperPass>();
+ AU.addRequired<TargetTransformInfoWrapperPass>();
+ }
- static BranchInst *getBranch(BasicBlock *BB) {
- return dyn_cast<BranchInst>(BB->getTerminator());
- }
+ DominatorTree *getDominatorTree() {
+ return DT ? DT
+ : (DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree());
+ }
- /// Return the condition of the branch terminating the given basic block.
- static Value *getBrCondtion(BasicBlock *);
-
- /// 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
- /// beween a variable and zero, and if the variable is non-zero, the
- /// control yeilds 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 pupulation 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;
- const DataLayout *TD;
- DominatorTree *DT;
- ScalarEvolution *SE;
- TargetLibraryInfo *TLI;
- const ScalarTargetTransformInfo *STTI;
- public:
- static char ID;
- explicit LoopIdiomRecognize() : LoopPass(ID) {
- initializeLoopIdiomRecognizePass(*PassRegistry::getPassRegistry());
- TD = 0; DT = 0; SE = 0; TLI = 0; STTI = 0;
- }
+ ScalarEvolution *getScalarEvolution() {
+ return SE ? SE : (SE = &getAnalysis<ScalarEvolution>());
+ }
- bool runOnLoop(Loop *L, LPPassManager &LPM);
- bool runOnLoopBlock(BasicBlock *BB, const SCEV *BECount,
- SmallVectorImpl<BasicBlock*> &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.
- ///
- virtual void getAnalysisUsage(AnalysisUsage &AU) const {
- AU.addRequired<LoopInfo>();
- AU.addPreserved<LoopInfo>();
- AU.addRequiredID(LoopSimplifyID);
- AU.addPreservedID(LoopSimplifyID);
- AU.addRequiredID(LCSSAID);
- AU.addPreservedID(LCSSAID);
- AU.addRequired<AliasAnalysis>();
- AU.addPreserved<AliasAnalysis>();
- AU.addRequired<ScalarEvolution>();
- AU.addPreserved<ScalarEvolution>();
- AU.addPreserved<DominatorTree>();
- AU.addRequired<DominatorTree>();
- AU.addRequired<TargetLibraryInfo>();
- }
+ TargetLibraryInfo *getTargetLibraryInfo() {
+ if (!TLI)
+ TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
- const DataLayout *getDataLayout() {
- return TD ? TD : TD=getAnalysisIfAvailable<DataLayout>();
- }
+ return TLI;
+ }
- DominatorTree *getDominatorTree() {
- return DT ? DT : (DT=&getAnalysis<DominatorTree>());
- }
+ const TargetTransformInfo *getTargetTransformInfo() {
+ return TTI ? TTI
+ : (TTI = &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(
+ *CurLoop->getHeader()->getParent()));
+ }
- ScalarEvolution *getScalarEvolution() {
- return SE ? SE : (SE = &getAnalysis<ScalarEvolution>());
- }
+ Loop *getLoop() const { return CurLoop; }
- TargetLibraryInfo *getTargetLibraryInfo() {
- return TLI ? TLI : (TLI = &getAnalysis<TargetLibraryInfo>());
- }
+private:
+ /// \name Countable Loop Idiom Handling
+ /// @{
- const ScalarTargetTransformInfo *getScalarTargetTransformInfo() {
- if (!STTI) {
- TargetTransformInfo *TTI = getAnalysisIfAvailable<TargetTransformInfo>();
- if (TTI) STTI = TTI->getScalarTargetTransformInfo();
- }
- return STTI;
- }
+ bool runOnCountableLoop();
+ bool runOnLoopBlock(BasicBlock *BB, const SCEV *BECount,
+ SmallVectorImpl<BasicBlock *> &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",
false, false)
-INITIALIZE_PASS_DEPENDENCY(LoopInfo)
-INITIALIZE_PASS_DEPENDENCY(DominatorTree)
+INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)
+INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
INITIALIZE_PASS_DEPENDENCY(LCSSA)
INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
-INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo)
+INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
INITIALIZE_PASS_END(LoopIdiomRecognize, "loop-idiom", "Recognize loop idioms",
false, false)
/// and zero out all the operands of this instruction. If any of them become
/// dead, delete them and the computation tree that feeds them.
///
-static void deleteDeadInstruction(Instruction *I, ScalarEvolution &SE,
+static void deleteDeadInstruction(Instruction *I,
const TargetLibraryInfo *TLI) {
- SmallVector<Instruction*, 32> NowDeadInsts;
-
- NowDeadInsts.push_back(I);
-
- // Before we touch this instruction, remove it from SE!
- do {
- Instruction *DeadInst = NowDeadInsts.pop_back_val();
-
- // This instruction is dead, zap it, in stages. Start by removing it from
- // SCEV.
- SE.forgetValue(DeadInst);
-
- for (unsigned op = 0, e = DeadInst->getNumOperands(); op != e; ++op) {
- Value *Op = DeadInst->getOperand(op);
- DeadInst->setOperand(op, 0);
-
- // If this operand just became dead, add it to the NowDeadInsts list.
- if (!Op->use_empty()) continue;
-
- if (Instruction *OpI = dyn_cast<Instruction>(Op))
- if (isInstructionTriviallyDead(OpI, TLI))
- NowDeadInsts.push_back(OpI);
- }
-
- DeadInst->eraseFromParent();
-
- } while (!NowDeadInsts.empty());
-}
-
-/// deleteIfDeadInstruction - If the specified value is a dead instruction,
-/// delete it and any recursively used instructions.
-static void deleteIfDeadInstruction(Value *V, ScalarEvolution &SE,
- const TargetLibraryInfo *TLI) {
- if (Instruction *I = dyn_cast<Instruction>(V))
- if (isInstructionTriviallyDead(I, TLI))
- deleteDeadInstruction(I, SE, TLI);
-}
-
-//===----------------------------------------------------------------------===//
-//
-// Implementation of LIRUtil
-//
-//===----------------------------------------------------------------------===//
-
-// This fucntion will return true iff the given block contains nothing but goto.
-// A typical usage of this function is to check if the preheader fucntion is
-// "almost" empty such that generated intrinsic function can be moved across
-// preheader and to be placed at the end of the preconditiona block without
-// concerning of breaking data dependence.
-bool LIRUtil::isAlmostEmpty(BasicBlock *BB) {
- if (BranchInst *Br = getBranch(BB)) {
- return Br->isUnconditional() && BB->size() == 1;
- }
- return false;
-}
-
-Value *LIRUtil::getBrCondtion(BasicBlock *BB) {
- BranchInst *Br = getBranch(BB);
- return Br ? Br->getCondition() : 0;
-}
-
-BasicBlock *LIRUtil::getPrecondBb(BasicBlock *PreHead) {
- if (BasicBlock *BB = PreHead->getSinglePredecessor()) {
- BranchInst *Br = getBranch(BB);
- return Br && Br->isConditional() ? BB : 0;
- }
- return 0;
+ SmallVector<Value *, 16> Operands(I->value_op_begin(), I->value_op_end());
+ I->replaceAllUsesWith(UndefValue::get(I->getType()));
+ I->eraseFromParent();
+ for (Value *Op : Operands)
+ RecursivelyDeleteTriviallyDeadInstructions(Op, TLI);
}
//===----------------------------------------------------------------------===//
//
//===----------------------------------------------------------------------===//
-NclPopcountRecognize::NclPopcountRecognize(LoopIdiomRecognize &TheLIR):
- LIR(TheLIR), CurLoop(TheLIR.getLoop()), PreCondBB(0) {
-}
+NclPopcountRecognize::NclPopcountRecognize(LoopIdiomRecognize &TheLIR)
+ : LIR(TheLIR), CurLoop(TheLIR.getLoop()), PreCondBB(nullptr) {}
bool NclPopcountRecognize::preliminaryScreen() {
- const ScalarTargetTransformInfo *STTI = LIR.getScalarTargetTransformInfo();
- if (STTI->getPopcntHwSupport(32) != ScalarTargetTransformInfo::Fast)
+ const TargetTransformInfo *TTI = LIR.getTargetTransformInfo();
+ if (TTI->getPopcntSupport(32) != TargetTransformInfo::PSK_FastHardware)
return false;
- // Counting population are usually conducted by few arithmetic instrutions.
+ // Counting population are usually conducted by few arithmetic instructions.
// Such instructions can be easilly "absorbed" by vacant slots in a
// non-compact loop. Therefore, recognizing popcount idiom only makes sense
// in a compact loop.
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<BranchInst>(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<BranchInst>(PreCondBB->getTerminator());
+ if (!PreCondBI || PreCondBI->isUnconditional())
+ return false;
+
return true;
}
-Value *NclPopcountRecognize::matchCondition (BranchInst *Br,
- BasicBlock *LoopEntry) const {
+Value *NclPopcountRecognize::matchCondition(BranchInst *Br,
+ BasicBlock *LoopEntry) const {
if (!Br || !Br->isConditional())
- return 0;
+ return nullptr;
ICmpInst *Cond = dyn_cast<ICmpInst>(Br->getCondition());
if (!Cond)
- return 0;
+ return nullptr;
ConstantInt *CmpZero = dyn_cast<ConstantInt>(Cond->getOperand(1));
if (!CmpZero || !CmpZero->isZero())
- return 0;
+ return nullptr;
ICmpInst::Predicate Pred = Cond->getPredicate();
if ((Pred == ICmpInst::ICMP_NE && Br->getSuccessor(0) == LoopEntry) ||
(Pred == ICmpInst::ICMP_EQ && Br->getSuccessor(1) == LoopEntry))
return Cond->getOperand(0);
- return 0;
+ 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:
//
Value *VarX1, *VarX0;
PHINode *PhiX, *CountPhi;
- DefX2 = CountInst = 0;
- VarX1 = VarX0 = 0;
- PhiX = CountPhi = 0;
+ DefX2 = CountInst = nullptr;
+ VarX1 = VarX0 = nullptr;
+ PhiX = CountPhi = nullptr;
LoopEntry = *(CurLoop->block_begin());
// 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<BranchInst>(LoopEntry->getTerminator()), LoopEntry))
DefX2 = dyn_cast<Instruction>(T);
else
return false;
// step 2: detect instructions corresponding to "x2 = x1 & (x1 - 1)"
{
- if (DefX2->getOpcode() != Instruction::And)
+ if (!DefX2 || DefX2->getOpcode() != Instruction::And)
return false;
BinaryOperator *SubOneOp;
ConstantInt *Dec = dyn_cast<ConstantInt>(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;
}
}
// step 4: Find the instruction which count the population: cnt2 = cnt1 + 1
{
- CountInst = NULL;
+ 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;
continue;
PHINode *Phi = dyn_cast<PHINode>(Inst->getOperand(0));
- if (!Phi && Phi->getParent() != LoopEntry)
+ if (!Phi || Phi->getParent() != LoopEntry)
continue;
// Check if the result of the instruction is live of the loop.
bool LiveOutLoop = false;
- for (Value::use_iterator I = Inst->use_begin(), E = Inst->use_end();
- I != E; I++) {
- if ((cast<Instruction>(*I))->getParent() != LoopEntry) {
- LiveOutLoop = true; break;
+ for (User *U : Inst->users()) {
+ if ((cast<Instruction>(U))->getParent() != LoopEntry) {
+ LiveOutLoop = true;
+ break;
}
}
// 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<BranchInst>(PreCondBB->getTerminator());
+ Value *T = matchCondition(PreCondBr, CurLoop->getLoopPreheader());
if (T != PhiX->getOperand(0) && T != PhiX->getOperand(1))
return false;
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<BranchInst>(PreCondBB->getTerminator());
const DebugLoc DL = CntInst->getDebugLoc();
// Assuming before transformation, the loop is following:
// if (x) // the precondition
// do { cnt++; x &= x - 1; } while(x);
-
+
// Step 1: Insert the ctpop instruction at the end of the precondition block
IRBuilderTy Builder(PreCondBr);
- Value *PopCnt, *PopCntZext, *NewCount;
+ Value *PopCnt, *PopCntZext, *NewCount, *TripCnt;
{
PopCnt = createPopcntIntrinsic(Builder, Var, DL);
NewCount = PopCntZext =
- Builder.CreateZExtOrTrunc(PopCnt, cast<IntegerType>(CntPhi->getType()));
+ Builder.CreateZExtOrTrunc(PopCnt, cast<IntegerType>(CntPhi->getType()));
if (NewCount != PopCnt)
(cast<Instruction>(NewCount))->setDebugLoc(DL);
- // If the popoulation counter's initial value is not zero, insert Add Inst.
+ // TripCnt is exactly the number of iterations the loop has
+ TripCnt = NewCount;
+
+ // If the population counter's initial value is not zero, insert Add Inst.
Value *CntInitVal = CntPhi->getIncomingValueForBlock(PreHead);
ConstantInt *InitConst = dyn_cast<ConstantInt>(CntInitVal);
if (!InitConst || !InitConst->isZero()) {
- NewCount = Builder.CreateAdd(PopCnt, InitConst);
+ NewCount = Builder.CreateAdd(NewCount, CntInitVal);
(cast<Instruction>(NewCount))->setDebugLoc(DL);
}
}
if (PreCond->getOperand(0) != Var)
std::swap(Opnd0, Opnd1);
- ICmpInst *NewPreCond =
- cast<ICmpInst>(Builder.CreateICmp(PreCond->getPredicate(), Opnd0, Opnd1));
- PreCond->replaceAllUsesWith(NewPreCond);
+ ICmpInst *NewPreCond = cast<ICmpInst>(
+ Builder.CreateICmp(PreCond->getPredicate(), Opnd0, Opnd1));
+ PreCondBr->setCondition(NewPreCond);
- deleteDeadInstruction(PreCond, *SE, TLI);
+ RecursivelyDeleteTriviallyDeadInstructions(PreCond, TLI);
}
// Step 3: Note that the population count is exactly the trip count of the
// do { cnt++; x &= x-1; t--) } while (t > 0);
BasicBlock *Body = *(CurLoop->block_begin());
{
- BranchInst *LbBr = LIRUtil::getBranch(Body);
+ auto *LbBr = dyn_cast<BranchInst>(Body->getTerminator());
ICmpInst *LbCond = cast<ICmpInst>(LbBr->getCondition());
- Type *Ty = NewCount->getType();
+ Type *Ty = TripCnt->getType();
PHINode *TcPhi = PHINode::Create(Ty, 2, "tcphi", Body->begin());
Builder.SetInsertPoint(LbCond);
Value *Opnd1 = cast<Value>(TcPhi);
Value *Opnd2 = cast<Value>(ConstantInt::get(Ty, 1));
- Instruction *TcDec =
- cast<Instruction>(Builder.CreateSub(Opnd1, Opnd2, "tcdec", false, true));
+ Instruction *TcDec = cast<Instruction>(
+ Builder.CreateSub(Opnd1, Opnd2, "tcdec", false, true));
- TcPhi->addIncoming(NewCount, PreHead);
+ 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<Value>(ConstantInt::get(Ty, 0)));
// Step 4: All the references to the original population counter outside
// the loop are replaced with the NewCount -- the value returned from
// __builtin_ctpop().
- {
- SmallVector<Value *, 4> CntUses;
- for (Value::use_iterator I = CntInst->use_begin(), E = CntInst->use_end();
- I != E; I++) {
- if (cast<Instruction>(*I)->getParent() != Body)
- CntUses.push_back(*I);
- }
- for (unsigned Idx = 0; Idx < CntUses.size(); Idx++) {
- (cast<Instruction>(CntUses[Idx]))->replaceUsesOfWith(CntInst, NewCount);
- }
- }
+ CntInst->replaceUsesOutsideBlock(NewCount, Body);
// step 5: Forget the "non-computable" trip-count SCEV associated with the
// loop. The loop would otherwise not be deleted even if it becomes empty.
SE->forgetLoop(CurLoop);
}
-CallInst *NclPopcountRecognize::createPopcntIntrinsic(IRBuilderTy &IRBuilder,
+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);
/// detected, transform the relevant code to popcount intrinsic function
/// call, and return true; otherwise, return false.
bool NclPopcountRecognize::recognize() {
-
- if (!LIR.getScalarTargetTransformInfo())
+ if (!LIR.getTargetTransformInfo())
return false;
LIR.getScalarEvolution();
//
//===----------------------------------------------------------------------===//
+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<ScalarEvolution>();
+ if (SE->hasLoopInvariantBackedgeTakenCount(L))
+ return runOnCountableLoop();
+ return runOnNoncountableLoop();
+}
+
bool LoopIdiomRecognize::runOnCountableLoop() {
const SCEV *BECount = SE->getBackedgeTakenCount(CurLoop);
- if (isa<SCEVCouldNotCompute>(BECount)) return false;
+ assert(!isa<SCEVCouldNotCompute>(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.
if (BECst->getValue()->getValue() == 0)
return false;
- // We require target data for now.
- if (!getDataLayout())
- return false;
-
- getDominatorTree();
+ // set DT
+ (void)getDominatorTree();
- LoopInfo &LI = getAnalysis<LoopInfo>();
- TLI = &getAnalysis<TargetLibraryInfo>();
+ LoopInfo &LI = getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
+ TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
- getTargetLibraryInfo();
+ // set TLI
+ (void)getTargetLibraryInfo();
- SmallVector<BasicBlock*, 8> ExitBlocks;
+ SmallVector<BasicBlock *, 8> 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) {
- 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<ScalarEvolution>();
- 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<BasicBlock*> &ExitBlocks) {
+bool LoopIdiomRecognize::runOnLoopBlock(
+ BasicBlock *BB, const SCEV *BECount,
+ SmallVectorImpl<BasicBlock *> &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.
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<StoreInst>(Inst)) {
+ if (StoreInst *SI = dyn_cast<StoreInst>(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
// top of the block.
- if (InstPtr == 0)
+ if (!InstPtr)
I = BB->begin();
continue;
}
// Look for memset instructions, which may be optimized to a larger memset.
- if (MemSetInst *MSI = dyn_cast<MemSetInst>(Inst)) {
+ if (MemSetInst *MSI = dyn_cast<MemSetInst>(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
// top of the block.
- if (InstPtr == 0)
+ if (!InstPtr)
I = BB->begin();
continue;
}
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();
// Reject stores that are so large that they overflow an unsigned.
- uint64_t SizeInBits = TD->getTypeSizeInBits(StoredVal->getType());
+ auto &DL = CurLoop->getHeader()->getModule()->getDataLayout();
+ uint64_t SizeInBits = DL.getTypeSizeInBits(StoredVal->getType());
if ((SizeInBits & 7) || (SizeInBits >> 32) != 0)
return false;
// 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<SCEVAddRecExpr>(SE->getSCEV(StorePtr));
- if (StoreEv == 0 || StoreEv->getLoop() != CurLoop || !StoreEv->isAffine())
+ dyn_cast<SCEVAddRecExpr>(SE->getSCEV(StorePtr));
+ if (!StoreEv || StoreEv->getLoop() != CurLoop || !StoreEv->isAffine())
return false;
// Check to see if the stride matches the size of the store. If so, then we
unsigned StoreSize = (unsigned)SizeInBits >> 3;
const SCEVConstant *Stride = dyn_cast<SCEVConstant>(StoreEv->getOperand(1));
- if (Stride == 0 || StoreSize != Stride->getValue()->getValue()) {
+ if (!Stride || StoreSize != Stride->getValue()->getValue()) {
// TODO: Could also handle negative stride here someday, that will require
// the validity check in mayLoopAccessLocation to be updated though.
// Enable this to print exact negative strides.
// for (i) A[i] = B[i];
if (LoadInst *LI = dyn_cast<LoadInst>(StoredVal)) {
const SCEVAddRecExpr *LoadEv =
- dyn_cast<SCEVAddRecExpr>(SE->getSCEV(LI->getOperand(0)));
+ dyn_cast<SCEVAddRecExpr>(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<ConstantInt>(MSI->getLength())) return false;
+ if (MSI->isVolatile() || !isa<ConstantInt>(MSI->getLength()))
+ return false;
// If we're not allowed to hack on memset, we fail.
if (!TLI->has(LibFunc::memset))
// loop, which indicates a strided store. If we have something else, it's a
// random store we can't handle.
const SCEVAddRecExpr *Ev = dyn_cast<SCEVAddRecExpr>(SE->getSCEV(Pointer));
- if (Ev == 0 || Ev->getLoop() != CurLoop || !Ev->isAffine())
+ if (!Ev || Ev->getLoop() != CurLoop || !Ev->isAffine())
return false;
// Reject memsets that are so large that they overflow an unsigned.
// TODO: Could also handle negative stride here someday, that will require the
// validity check in mayLoopAccessLocation to be updated though.
- if (Stride == 0 || MSI->getLength() != Stride->getValue())
+ if (!Stride || MSI->getLength() != Stride->getValue())
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<SCEVConstant>(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;
///
/// Note that we don't ever attempt to use memset_pattern8 or 4, because these
/// just replicate their input array and then pass on to memset_pattern16.
-static Constant *getMemSetPatternValue(Value *V, const DataLayout &TD) {
+static Constant *getMemSetPatternValue(Value *V, const DataLayout &DL) {
// If the value isn't a constant, we can't promote it to being in a constant
// array. We could theoretically do a store to an alloca or something, but
// that doesn't seem worthwhile.
Constant *C = dyn_cast<Constant>(V);
- if (C == 0) return 0;
+ if (!C)
+ return nullptr;
// Only handle simple values that are a power of two bytes in size.
- uint64_t Size = TD.getTypeSizeInBits(V->getType());
- if (Size == 0 || (Size & 7) || (Size & (Size-1)))
- return 0;
+ uint64_t Size = DL.getTypeSizeInBits(V->getType());
+ if (Size == 0 || (Size & 7) || (Size & (Size - 1)))
+ return nullptr;
// Don't care enough about darwin/ppc to implement this.
- if (TD.isBigEndian())
- return 0;
+ if (DL.isBigEndian())
+ return nullptr;
// Convert to size in bytes.
Size /= 8;
// 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 0;
+ 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<Constant*>(ArraySize, C));
+ return ConstantArray::get(AT, std::vector<Constant *>(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
// are stored. A store of i32 0x01020304 can never be turned into a memset,
// but it can be turned into memset_pattern if the target supports it.
Value *SplatValue = isBytewiseValue(StoredVal);
- Constant *PatternValue = 0;
+ Constant *PatternValue = nullptr;
+ auto &DL = CurLoop->getHeader()->getModule()->getDataLayout();
+ unsigned DestAS = DestPtr->getType()->getPointerAddressSpace();
// If we're allowed to form a memset, and the stored value would be acceptable
// for memset, use it.
// promote the memset.
CurLoop->isLoopInvariant(SplatValue)) {
// Keep and use SplatValue.
- PatternValue = 0;
- } else if (TLI->has(LibFunc::memset_pattern16) &&
- (PatternValue = getMemSetPatternValue(StoredVal, *TD))) {
+ PatternValue = nullptr;
+ } else if (DestAS == 0 && TLI->has(LibFunc::memset_pattern16) &&
+ (PatternValue = getMemSetPatternValue(StoredVal, DL))) {
+ // Don't create memset_pattern16s with address spaces.
// It looks like we can use PatternValue!
- SplatValue = 0;
+ SplatValue = nullptr;
} else {
// Otherwise, this isn't an idiom we can transform. For example, we can't
// do anything with a 3-byte store.
// 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);
// Okay, we have a strided store "p[i]" of a splattable value. We can turn
// this into a memset in the loop preheader now if we want. However, this
// 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.
- unsigned AddrSpace = cast<PointerType>(DestPtr->getType())->getAddressSpace();
- Value *BasePtr =
- Expander.expandCodeFor(Ev->getStart(), Builder.getInt8PtrTy(AddrSpace),
- Preheader->getTerminator());
+ Value *BasePtr = Expander.expandCodeFor(Ev->getStart(), DestInt8PtrTy,
+ Preheader->getTerminator());
-
- if (mayLoopAccessLocation(BasePtr, AliasAnalysis::ModRef,
- CurLoop, BECount,
- StoreSize, getAnalysis<AliasAnalysis>(), TheStore)){
+ if (mayLoopAccessLocation(BasePtr, MRI_ModRef, CurLoop, BECount, StoreSize,
+ getAnalysis<AliasAnalysis>(), TheStore)) {
Expander.clear();
// If we generated new code for the base pointer, clean up.
- deleteIfDeadInstruction(BasePtr, *SE, TLI);
+ RecursivelyDeleteTriviallyDeadInstructions(BasePtr, TLI);
return false;
}
// The # stored bytes is (BECount+1)*Size. Expand the trip count out to
// pointer size if it isn't already.
- Type *IntPtr = TD->getIntPtrType(DestPtr->getContext());
+ Type *IntPtr = Builder.getIntPtrTy(DL, DestAS);
BECount = SE->getTruncateOrZeroExtend(BECount, IntPtr);
- const SCEV *NumBytesS = SE->getAddExpr(BECount, SE->getConstant(IntPtr, 1),
- SCEV::FlagNUW);
- if (StoreSize != 1)
+ 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);
- else {
+ if (SplatValue) {
+ 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(),
- Builder.getInt8PtrTy(),
- Builder.getInt8PtrTy(), IntPtr,
- (void*)0);
+ 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.
GlobalVariable *GV = new GlobalVariable(*M, PatternValue->getType(), true,
- GlobalValue::InternalLinkage,
+ GlobalValue::PrivateLinkage,
PatternValue, ".memset_pattern");
GV->setUnnamedAddr(true); // Ok to merge these.
GV->setAlignment(16);
- Value *PatternPtr = ConstantExpr::getBitCast(GV, Builder.getInt8PtrTy());
- NewCall = Builder.CreateCall3(MSP, BasePtr, PatternPtr, NumBytes);
+ Value *PatternPtr = ConstantExpr::getBitCast(GV, Int8PtrTy);
+ NewCall = Builder.CreateCall(MSP, {BasePtr, PatternPtr, NumBytes});
}
DEBUG(dbgs() << " Formed memset: " << *NewCall << "\n"
// Okay, the memset has been formed. Zap the original store and anything that
// feeds into it.
- deleteDeadInstruction(TheStore, *SE, TLI);
+ deleteDeadInstruction(TheStore, TLI);
++NumMemSet;
return true;
}
/// 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;
// 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
// 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<AliasAnalysis>(), SI)) {
+ if (mayLoopAccessLocation(StoreBasePtr, MRI_ModRef, CurLoop, BECount,
+ StoreSize, getAnalysis<AliasAnalysis>(), SI)) {
Expander.clear();
// If we generated new code for the base pointer, clean up.
- deleteIfDeadInstruction(StoreBasePtr, *SE, TLI);
+ RecursivelyDeleteTriviallyDeadInstructions(StoreBasePtr, TLI);
return false;
}
// 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<AliasAnalysis>(), SI)) {
+ if (mayLoopAccessLocation(LoadBasePtr, MRI_Mod, CurLoop, BECount, StoreSize,
+ getAnalysis<AliasAnalysis>(), SI)) {
Expander.clear();
// If we generated new code for the base pointer, clean up.
- deleteIfDeadInstruction(LoadBasePtr, *SE, TLI);
- deleteIfDeadInstruction(StoreBasePtr, *SE, TLI);
+ RecursivelyDeleteTriviallyDeadInstructions(LoadBasePtr, TLI);
+ RecursivelyDeleteTriviallyDeadInstructions(StoreBasePtr, TLI);
return false;
}
// 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.
- Type *IntPtr = TD->getIntPtrType(SI->getContext());
- BECount = SE->getTruncateOrZeroExtend(BECount, IntPtr);
+ Type *IntPtrTy = Builder.getIntPtrTy(DL, SI->getPointerAddressSpace());
+ BECount = SE->getTruncateOrZeroExtend(BECount, IntPtrTy);
- const SCEV *NumBytesS = SE->getAddExpr(BECount, SE->getConstant(IntPtr, 1),
- SCEV::FlagNUW);
+ const SCEV *NumBytesS =
+ SE->getAddExpr(BECount, SE->getConstant(IntPtrTy, 1), SCEV::FlagNUW);
if (StoreSize != 1)
- NumBytesS = SE->getMulExpr(NumBytesS, SE->getConstant(IntPtr, StoreSize),
+ NumBytesS = SE->getMulExpr(NumBytesS, SE->getConstant(IntPtrTy, StoreSize),
SCEV::FlagNUW);
Value *NumBytes =
- Expander.expandCodeFor(NumBytesS, IntPtr, 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, *SE, TLI);
+ deleteDeadInstruction(SI, TLI);
++NumMemCpy;
return true;
}
+
+bool LoopIdiomRecognize::runOnNoncountableLoop() {
+ NclPopcountRecognize Popcount(*this);
+ if (Popcount.recognize())
+ return true;
+
+ return false;
+}
projects / oota-llvm.git / blobdiff