bool InductionDescriptor::isInductionPHI(PHINode *Phi, const Loop *TheLoop,
ScalarEvolution *SE,
InductionDescriptor &D,
const SCEV *Expr) {
Type *PhiTy = Phi->getType();
// We only handle integer and pointer inductions variables.
if (!PhiTy->isIntegerTy() && !PhiTy->isPointerTy())
return false;
// Check that the PHI is consecutive.
const SCEV *PhiScev = Expr ? Expr : SE->getSCEV(Phi);
const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(PhiScev);
if (!AR) {
DEBUG(dbgs() << "LV: PHI is not a poly recurrence.\n");
return false;
}
assert(TheLoop->getHeader() == Phi->getParent() &&
"PHI is an AddRec for a different loop?!");
Value *StartValue =
Phi->getIncomingValueForBlock(AR->getLoop()->getLoopPreheader());
const SCEV *Step = AR->getStepRecurrence(*SE);
// Calculate the pointer stride and check if it is consecutive.
// The stride may be a constant or a loop invariant integer value.
const SCEVConstant *ConstStep = dyn_cast<SCEVConstant>(Step);
if (!ConstStep && !SE->isLoopInvariant(Step, TheLoop))
return false;
if (PhiTy->isIntegerTy()) {
D = InductionDescriptor(StartValue, IK_IntInduction, Step);
return true;
}
assert(PhiTy->isPointerTy() && "The PHI must be a pointer");
// Pointer induction should be a constant.
if (!ConstStep)
return false;
ConstantInt *CV = ConstStep->getValue();
Type *PointerElementType = PhiTy->getPointerElementType();
// The pointer stride cannot be determined if the pointer element type is not
// sized.
if (!PointerElementType->isSized())
return false;
const DataLayout &DL = Phi->getModule()->getDataLayout();
int64_t Size = static_cast<int64_t>(DL.getTypeAllocSize(PointerElementType));
if (!Size)
return false;
int64_t CVSize = CV->getSExtValue();
if (CVSize % Size)
return false;
auto *StepValue = SE->getConstant(CV->getType(), CVSize / Size,
true /* signed */);
D = InductionDescriptor(StartValue, IK_PtrInduction, StepValue);
return true;
}
bool llvm::decomposeBitTestICmp(const ICmpInst *I, CmpInst::Predicate &Pred,
Value *&X, Value *&Y, Value *&Z) {
ConstantInt *C = dyn_cast<ConstantInt>(I->getOperand(1));
if (!C)
return false;
switch (I->getPredicate()) {
default:
return false;
case ICmpInst::ICMP_SLT:
// X < 0 is equivalent to (X & SignBit) != 0.
if (!C->isZero())
return false;
Y = ConstantInt::get(I->getContext(), APInt::getSignBit(C->getBitWidth()));
Pred = ICmpInst::ICMP_NE;
break;
case ICmpInst::ICMP_SGT:
// X > -1 is equivalent to (X & SignBit) == 0.
if (!C->isAllOnesValue())
return false;
Y = ConstantInt::get(I->getContext(), APInt::getSignBit(C->getBitWidth()));
Pred = ICmpInst::ICMP_EQ;
break;
case ICmpInst::ICMP_ULT:
// X <u 2^n is equivalent to (X & ~(2^n-1)) == 0.
if (!C->getValue().isPowerOf2())
return false;
Y = ConstantInt::get(I->getContext(), -C->getValue());
Pred = ICmpInst::ICMP_EQ;
break;
case ICmpInst::ICMP_UGT:
// X >u 2^n-1 is equivalent to (X & ~(2^n-1)) != 0.
if (!(C->getValue() + 1).isPowerOf2())
return false;
Y = ConstantInt::get(I->getContext(), ~C->getValue());
Pred = ICmpInst::ICMP_NE;
break;
}
X = I->getOperand(0);
Z = ConstantInt::getNullValue(C->getType());
return true;
}
//
// Method: runOnModule()
//
// Description:
// Entry point for this LLVM pass.
// Find all GEPs, and simplify them.
//
// Inputs:
// M - A reference to the LLVM module to transform
//
// Outputs:
// M - The transformed LLVM module.
//
// Return value:
// true - The module was modified.
// false - The module was not modified.
//
bool SimplifyGEP::runOnModule(Module& M) {
TD = &getAnalysis<TargetData>();
preprocess(M);
for (Module::iterator F = M.begin(); F != M.end(); ++F){
for (Function::iterator B = F->begin(), FE = F->end(); B != FE; ++B) {
for (BasicBlock::iterator I = B->begin(), BE = B->end(); I != BE; I++) {
if(!(isa<GetElementPtrInst>(I)))
continue;
GetElementPtrInst *GEP = cast<GetElementPtrInst>(I);
Value *PtrOp = GEP->getOperand(0);
Value *StrippedPtr = PtrOp->stripPointerCasts();
// Check if the GEP base pointer is enclosed in a cast
if (StrippedPtr != PtrOp) {
const PointerType *StrippedPtrTy =cast<PointerType>(StrippedPtr->getType());
bool HasZeroPointerIndex = false;
if (ConstantInt *C = dyn_cast<ConstantInt>(GEP->getOperand(1)))
HasZeroPointerIndex = C->isZero();
// Transform: GEP (bitcast [10 x i8]* X to [0 x i8]*), i32 0, ...
// into : GEP [10 x i8]* X, i32 0, ...
//
// Likewise, transform: GEP (bitcast i8* X to [0 x i8]*), i32 0, ...
// into : GEP i8* X, ...
//
// This occurs when the program declares an array extern like "int X[];"
if (HasZeroPointerIndex) {
const PointerType *CPTy = cast<PointerType>(PtrOp->getType());
if (const ArrayType *CATy =
dyn_cast<ArrayType>(CPTy->getElementType())) {
// GEP (bitcast i8* X to [0 x i8]*), i32 0, ... ?
if (CATy->getElementType() == StrippedPtrTy->getElementType()) {
// -> GEP i8* X, ...
SmallVector<Value*, 8> Idx(GEP->idx_begin()+1, GEP->idx_end());
GetElementPtrInst *Res =
GetElementPtrInst::Create(StrippedPtr, Idx, GEP->getName(), GEP);
Res->setIsInBounds(GEP->isInBounds());
GEP->replaceAllUsesWith(Res);
continue;
}
if (const ArrayType *XATy =
dyn_cast<ArrayType>(StrippedPtrTy->getElementType())){
// GEP (bitcast [10 x i8]* X to [0 x i8]*), i32 0, ... ?
if (CATy->getElementType() == XATy->getElementType()) {
// -> GEP [10 x i8]* X, i32 0, ...
// At this point, we know that the cast source type is a pointer
// to an array of the same type as the destination pointer
// array. Because the array type is never stepped over (there
// is a leading zero) we can fold the cast into this GEP.
GEP->setOperand(0, StrippedPtr);
continue;
}
}
}
} else if (GEP->getNumOperands() == 2) {
// Transform things like:
// %t = getelementptr i32* bitcast ([2 x i32]* %str to i32*), i32 %V
// into: %t1 = getelementptr [2 x i32]* %str, i32 0, i32 %V; bitcast
Type *SrcElTy = StrippedPtrTy->getElementType();
Type *ResElTy=cast<PointerType>(PtrOp->getType())->getElementType();
if (TD && SrcElTy->isArrayTy() &&
TD->getTypeAllocSize(cast<ArrayType>(SrcElTy)->getElementType()) ==
TD->getTypeAllocSize(ResElTy)) {
Value *Idx[2];
Idx[0] = Constant::getNullValue(Type::getInt32Ty(GEP->getContext()));
Idx[1] = GEP->getOperand(1);
Value *NewGEP = GetElementPtrInst::Create(StrippedPtr, Idx,
GEP->getName(), GEP);
// V and GEP are both pointer types --> BitCast
GEP->replaceAllUsesWith(new BitCastInst(NewGEP, GEP->getType(), GEP->getName(), GEP));
continue;
}
// Transform things like:
// getelementptr i8* bitcast ([100 x double]* X to i8*), i32 %tmp
// (where tmp = 8*tmp2) into:
// getelementptr [100 x double]* %arr, i32 0, i32 %tmp2; bitcast
if (TD && SrcElTy->isArrayTy() && ResElTy->isIntegerTy(8)) {
uint64_t ArrayEltSize =
TD->getTypeAllocSize(cast<ArrayType>(SrcElTy)->getElementType());
// Check to see if "tmp" is a scale by a multiple of ArrayEltSize. We
// allow either a mul, shift, or constant here.
Value *NewIdx = 0;
ConstantInt *Scale = 0;
if (ArrayEltSize == 1) {
NewIdx = GEP->getOperand(1);
Scale = ConstantInt::get(cast<IntegerType>(NewIdx->getType()), 1);
} else if (ConstantInt *CI = dyn_cast<ConstantInt>(GEP->getOperand(1))) {
NewIdx = ConstantInt::get(CI->getType(), 1);
Scale = CI;
} else if (Instruction *Inst =dyn_cast<Instruction>(GEP->getOperand(1))){
if (Inst->getOpcode() == Instruction::Shl &&
isa<ConstantInt>(Inst->getOperand(1))) {
ConstantInt *ShAmt = cast<ConstantInt>(Inst->getOperand(1));
uint32_t ShAmtVal = ShAmt->getLimitedValue(64);
Scale = ConstantInt::get(cast<IntegerType>(Inst->getType()),
1ULL << ShAmtVal);
NewIdx = Inst->getOperand(0);
} else if (Inst->getOpcode() == Instruction::Mul &&
isa<ConstantInt>(Inst->getOperand(1))) {
Scale = cast<ConstantInt>(Inst->getOperand(1));
NewIdx = Inst->getOperand(0);
}
}
// If the index will be to exactly the right offset with the scale taken
// out, perform the transformation. Note, we don't know whether Scale is
// signed or not. We'll use unsigned version of division/modulo
// operation after making sure Scale doesn't have the sign bit set.
if (ArrayEltSize && Scale && Scale->getSExtValue() >= 0LL &&
Scale->getZExtValue() % ArrayEltSize == 0) {
Scale = ConstantInt::get(Scale->getType(),
Scale->getZExtValue() / ArrayEltSize);
if (Scale->getZExtValue() != 1) {
Constant *C = ConstantExpr::getIntegerCast(Scale, NewIdx->getType(),
false /*ZExt*/);
NewIdx = BinaryOperator::Create(BinaryOperator::Mul, NewIdx, C, "idxscale");
}
// Insert the new GEP instruction.
Value *Idx[2];
Idx[0] = Constant::getNullValue(Type::getInt32Ty(GEP->getContext()));
Idx[1] = NewIdx;
Value *NewGEP = GetElementPtrInst::Create(StrippedPtr, Idx,
GEP->getName(), GEP);
GEP->replaceAllUsesWith(new BitCastInst(NewGEP, GEP->getType(), GEP->getName(), GEP));
continue;
}
}
}
}
}
}
}
return true;
}
Value* LoopTripCount::insertTripCount(Loop* L, Instruction* InsertPos)
{
// inspired from Loop::getCanonicalInductionVariable
BasicBlock *H = L->getHeader();
BasicBlock* LoopPred = L->getLoopPredecessor();
BasicBlock* startBB = NULL;//which basicblock stores start value
int OneStep = 0;// the extra add or plus step for calc
Assert(LoopPred, "Require Loop has a Pred");
DEBUG(errs()<<"loop depth:"<<L->getLoopDepth()<<"\n");
/** whats difference on use of predecessor and preheader??*/
//RET_ON_FAIL(self->getLoopLatch()&&self->getLoopPreheader());
//assert(self->getLoopLatch() && self->getLoopPreheader() && "need loop simplify form" );
ret_null_fail(L->getLoopLatch(), "need loop simplify form");
BasicBlock* TE = NULL;//True Exit
SmallVector<BasicBlock*,4> Exits;
L->getExitingBlocks(Exits);
if(Exits.size()==1) TE = Exits.front();
else{
if(std::find(Exits.begin(),Exits.end(),L->getLoopLatch())!=Exits.end()) TE = L->getLoopLatch();
else{
SmallVector<llvm::Loop::Edge,4> ExitEdges;
L->getExitEdges(ExitEdges);
//stl 用法,先把所有满足条件的元素(出口的结束符是不可到达)移动到数组的末尾,再统一删除
ExitEdges.erase(std::remove_if(ExitEdges.begin(), ExitEdges.end(),
[](llvm::Loop::Edge& I){
return isa<UnreachableInst>(I.second->getTerminator());
}), ExitEdges.end());
if(ExitEdges.size()==1) TE = const_cast<BasicBlock*>(ExitEdges.front().first);
}
}
//process true exit
ret_null_fail(TE, "need have a true exit");
Instruction* IndOrNext = NULL;
Value* END = NULL;
//终止块的终止指令：分情况讨论branchinst,switchinst;
//跳转指令br bool a1,a2;condition<-->bool
if(isa<BranchInst>(TE->getTerminator())){
const BranchInst* EBR = cast<BranchInst>(TE->getTerminator());
Assert(EBR->isConditional(), "end branch is not conditional");
ICmpInst* EC = dyn_cast<ICmpInst>(EBR->getCondition());
if(EC->getPredicate() == EC->ICMP_SGT){
Assert(!L->contains(EBR->getSuccessor(0)), *EBR<<":abnormal exit with great than");//终止块的终止指令---->跳出执行循环外的指令
OneStep += 1;
} else if(EC->getPredicate() == EC->ICMP_EQ)
Assert(!L->contains(EBR->getSuccessor(0)), *EBR<<":abnormal exit with great than");
else if(EC->getPredicate() == EC->ICMP_SLT) {
ret_null_fail(!L->contains(EBR->getSuccessor(1)), *EBR<<":abnormal exit with less than");
} else {
ret_null_fail(0, *EC<<" unknow combination of end condition");
}
IndOrNext = dyn_cast<Instruction>(castoff(EC->getOperand(0)));//去掉类型转化
END = EC->getOperand(1);
DEBUG(errs()<<"end value:"<<*EC<<"\n");
}else if(isa<SwitchInst>(TE->getTerminator())){
SwitchInst* ESW = const_cast<SwitchInst*>(cast<SwitchInst>(TE->getTerminator()));
IndOrNext = dyn_cast<Instruction>(castoff(ESW->getCondition()));
for(auto I = ESW->case_begin(),E = ESW->case_end();I!=E;++I){
if(!L->contains(I.getCaseSuccessor())){
ret_null_fail(!END,"");
assert(!END && "shouldn't have two ends");
END = I.getCaseValue();
}
}
DEBUG(errs()<<"end value:"<<*ESW<<"\n");
}else{
assert(0 && "unknow terminator type");
}
ret_null_fail(L->isLoopInvariant(END), "end value should be loop invariant");//至此得END值
Value* start = NULL;
Value* ind = NULL;
Instruction* next = NULL;
bool addfirst = false;//add before icmp ed
DISABLE(errs()<<*IndOrNext<<"\n");
if(isa<LoadInst>(IndOrNext)){
//memory depend analysis
Value* PSi = IndOrNext->getOperand(0);//point type Step.i
int SICount[2] = {0};//store in predecessor count,store in loop body count
for( auto I = PSi->use_begin(),E = PSi->use_end();I!=E;++I){
DISABLE(errs()<<**I<<"\n");
StoreInst* SI = dyn_cast<StoreInst>(*I);
if(!SI || SI->getOperand(1) != PSi) continue;
if(!start&&L->isLoopInvariant(SI->getOperand(0))) {
if(SI->getParent() != LoopPred)
if(std::find(pred_begin(LoopPred),pred_end(LoopPred),SI->getParent()) == pred_end(LoopPred)) continue;
start = SI->getOperand(0);
startBB = SI->getParent();
++SICount[0];
}
Instruction* SI0 = dyn_cast<Instruction>(SI->getOperand(0));
if(L->contains(SI) && SI0 && SI0->getOpcode() == Instruction::Add){
next = SI0;
++SICount[1];
}
}
Assert(SICount[0]==1 && SICount[1]==1, "");
ind = IndOrNext;
}else{
if(isa<PHINode>(IndOrNext)){
PHINode* PHI = cast<PHINode>(IndOrNext);
ind = IndOrNext;
if(castoff(PHI->getIncomingValue(0)) == castoff(PHI->getIncomingValue(1)) && PHI->getParent() != H)
ind = castoff(PHI->getIncomingValue(0));
addfirst = false;
}else if(IndOrNext->getOpcode() == Instruction::Add){
next = IndOrNext;
addfirst = true;
}else{
Assert(0 ,"unknow how to analysis");
}
for(auto I = H->begin();isa<PHINode>(I);++I){
PHINode* P = cast<PHINode>(I);
if(ind && P == ind){
//start = P->getIncomingValueForBlock(L->getLoopPredecessor());
start = tryFindStart(P, L, startBB);
next = dyn_cast<Instruction>(P->getIncomingValueForBlock(L->getLoopLatch()));
}else if(next && P->getIncomingValueForBlock(L->getLoopLatch()) == next){
//start = P->getIncomingValueForBlock(L->getLoopPredecessor());
start = tryFindStart(P, L, startBB);
ind = P;
}
}
}
Assert(start ,"couldn't find a start value");
//process complex loops later
//DEBUG(if(L->getLoopDepth()>1 || !L->getSubLoops().empty()) return NULL);
DEBUG(errs()<<"start value:"<<*start<<"\n");
DEBUG(errs()<<"ind value:"<<*ind<<"\n");
DEBUG(errs()<<"next value:"<<*next<<"\n");
//process non add later
unsigned next_phi_idx = 0;
ConstantInt* Step = NULL,*PrevStep = NULL;/*only used if next is phi node*/
ret_null_fail(next, "");
PHINode* next_phi = dyn_cast<PHINode>(next);
do{
if(next_phi) {
next = dyn_cast<Instruction>(next_phi->getIncomingValue(next_phi_idx));
ret_null_fail(next, "");
DEBUG(errs()<<"next phi "<<next_phi_idx<<":"<<*next<<"\n");
if(Step&&PrevStep){
Assert(Step->getSExtValue() == PrevStep->getSExtValue(),"");
}
PrevStep = Step;
}
Assert(next->getOpcode() == Instruction::Add , "why induction increment is not Add");
Assert(next->getOperand(0) == ind ,"why induction increment is not add it self");
Step = dyn_cast<ConstantInt>(next->getOperand(1));
Assert(Step,"");
}while(next_phi && ++next_phi_idx<next_phi->getNumIncomingValues());
//RET_ON_FAIL(Step->equalsInt(1));
//assert(VERBOSE(Step->equalsInt(1),Step) && "why induction increment number is not 1");
Value* RES = NULL;
//if there are no predecessor, we can insert code into start value basicblock
IRBuilder<> Builder(InsertPos);
Assert(start->getType()->isIntegerTy() && END->getType()->isIntegerTy() , " why increment is not integer type");
if(start->getType() != END->getType()){
start = Builder.CreateCast(CastInst::getCastOpcode(start, false,
END->getType(), false),start,END->getType());
}
if(Step->getType() != END->getType()){
//Because Step is a Constant, so it casted is constant
Step = dyn_cast<ConstantInt>(Builder.CreateCast(CastInst::getCastOpcode(Step, false,
END->getType(), false),Step,END->getType()));
AssertRuntime(Step);
}
if(Step->isMinusOne())
RES = Builder.CreateSub(start,END);
else//Step Couldn't be zero
RES = Builder.CreateSub(END, start);
if(addfirst) OneStep -= 1;
if(Step->isMinusOne()) OneStep*=-1;
assert(OneStep<=1 && OneStep>=-1);
RES = (OneStep==1)?Builder.CreateAdd(RES,Step):(OneStep==-1)?Builder.CreateSub(RES, Step):RES;
if(!Step->isMinusOne()&&!Step->isOne())
RES = Builder.CreateSDiv(RES, Step);
RES->setName(H->getName()+".tc");
return RES;
}
bool InductionDescriptor::isInductionPHI(
PHINode *Phi, const Loop *TheLoop, ScalarEvolution *SE,
InductionDescriptor &D, const SCEV *Expr,
SmallVectorImpl<Instruction *> *CastsToIgnore) {
Type *PhiTy = Phi->getType();
// We only handle integer and pointer inductions variables.
if (!PhiTy->isIntegerTy() && !PhiTy->isPointerTy())
return false;
// Check that the PHI is consecutive.
const SCEV *PhiScev = Expr ? Expr : SE->getSCEV(Phi);
const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(PhiScev);
if (!AR) {
LLVM_DEBUG(dbgs() << "LV: PHI is not a poly recurrence.\n");
return false;
}
if (AR->getLoop() != TheLoop) {
// FIXME: We should treat this as a uniform. Unfortunately, we
// don't currently know how to handled uniform PHIs.
LLVM_DEBUG(
dbgs() << "LV: PHI is a recurrence with respect to an outer loop.\n");
return false;
}
Value *StartValue =
Phi->getIncomingValueForBlock(AR->getLoop()->getLoopPreheader());
BasicBlock *Latch = AR->getLoop()->getLoopLatch();
if (!Latch)
return false;
BinaryOperator *BOp =
dyn_cast<BinaryOperator>(Phi->getIncomingValueForBlock(Latch));
const SCEV *Step = AR->getStepRecurrence(*SE);
// Calculate the pointer stride and check if it is consecutive.
// The stride may be a constant or a loop invariant integer value.
const SCEVConstant *ConstStep = dyn_cast<SCEVConstant>(Step);
if (!ConstStep && !SE->isLoopInvariant(Step, TheLoop))
return false;
if (PhiTy->isIntegerTy()) {
D = InductionDescriptor(StartValue, IK_IntInduction, Step, BOp,
CastsToIgnore);
return true;
}
assert(PhiTy->isPointerTy() && "The PHI must be a pointer");
// Pointer induction should be a constant.
if (!ConstStep)
return false;
ConstantInt *CV = ConstStep->getValue();
Type *PointerElementType = PhiTy->getPointerElementType();
// The pointer stride cannot be determined if the pointer element type is not
// sized.
if (!PointerElementType->isSized())
return false;
const DataLayout &DL = Phi->getModule()->getDataLayout();
int64_t Size = static_cast<int64_t>(DL.getTypeAllocSize(PointerElementType));
if (!Size)
return false;
int64_t CVSize = CV->getSExtValue();
if (CVSize % Size)
return false;
auto *StepValue =
SE->getConstant(CV->getType(), CVSize / Size, true /* signed */);
D = InductionDescriptor(StartValue, IK_PtrInduction, StepValue, BOp);
return true;
}