ScalarEvolutionExpander.cpp [plain text]
#include "llvm/Analysis/ScalarEvolutionExpander.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/LLVMContext.h"
#include "llvm/Target/TargetData.h"
#include "llvm/ADT/STLExtras.h"
using namespace llvm;
Value *SCEVExpander::ReuseOrCreateCast(Value *V, const Type *Ty,
Instruction::CastOps Op,
BasicBlock::iterator IP) {
for (Value::use_iterator UI = V->use_begin(), E = V->use_end();
UI != E; ++UI) {
User *U = *UI;
if (U->getType() == Ty)
if (CastInst *CI = dyn_cast<CastInst>(U))
if (CI->getOpcode() == Op) {
if (BasicBlock::iterator(CI) != IP) {
Instruction *NewCI = CastInst::Create(Op, V, Ty, "", IP);
NewCI->takeName(CI);
CI->replaceAllUsesWith(NewCI);
CI->setOperand(0, UndefValue::get(V->getType()));
rememberInstruction(NewCI);
return NewCI;
}
rememberInstruction(CI);
return CI;
}
}
Instruction *I = CastInst::Create(Op, V, Ty, V->getName(), IP);
rememberInstruction(I);
return I;
}
Value *SCEVExpander::InsertNoopCastOfTo(Value *V, const Type *Ty) {
Instruction::CastOps Op = CastInst::getCastOpcode(V, false, Ty, false);
assert((Op == Instruction::BitCast ||
Op == Instruction::PtrToInt ||
Op == Instruction::IntToPtr) &&
"InsertNoopCastOfTo cannot perform non-noop casts!");
assert(SE.getTypeSizeInBits(V->getType()) == SE.getTypeSizeInBits(Ty) &&
"InsertNoopCastOfTo cannot change sizes!");
if (Op == Instruction::BitCast && V->getType() == Ty)
return V;
if ((Op == Instruction::PtrToInt || Op == Instruction::IntToPtr) &&
SE.getTypeSizeInBits(Ty) == SE.getTypeSizeInBits(V->getType())) {
if (CastInst *CI = dyn_cast<CastInst>(V))
if ((CI->getOpcode() == Instruction::PtrToInt ||
CI->getOpcode() == Instruction::IntToPtr) &&
SE.getTypeSizeInBits(CI->getType()) ==
SE.getTypeSizeInBits(CI->getOperand(0)->getType()))
return CI->getOperand(0);
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
if ((CE->getOpcode() == Instruction::PtrToInt ||
CE->getOpcode() == Instruction::IntToPtr) &&
SE.getTypeSizeInBits(CE->getType()) ==
SE.getTypeSizeInBits(CE->getOperand(0)->getType()))
return CE->getOperand(0);
}
if (Constant *C = dyn_cast<Constant>(V))
return ConstantExpr::getCast(Op, C, Ty);
if (Argument *A = dyn_cast<Argument>(V)) {
BasicBlock::iterator IP = A->getParent()->getEntryBlock().begin();
while ((isa<BitCastInst>(IP) &&
isa<Argument>(cast<BitCastInst>(IP)->getOperand(0)) &&
cast<BitCastInst>(IP)->getOperand(0) != A) ||
isa<DbgInfoIntrinsic>(IP))
++IP;
return ReuseOrCreateCast(A, Ty, Op, IP);
}
Instruction *I = cast<Instruction>(V);
BasicBlock::iterator IP = I; ++IP;
if (InvokeInst *II = dyn_cast<InvokeInst>(I))
IP = II->getNormalDest()->begin();
while (isa<PHINode>(IP) || isa<DbgInfoIntrinsic>(IP)) ++IP;
return ReuseOrCreateCast(I, Ty, Op, IP);
}
Value *SCEVExpander::InsertBinop(Instruction::BinaryOps Opcode,
Value *LHS, Value *RHS) {
if (Constant *CLHS = dyn_cast<Constant>(LHS))
if (Constant *CRHS = dyn_cast<Constant>(RHS))
return ConstantExpr::get(Opcode, CLHS, CRHS);
unsigned ScanLimit = 6;
BasicBlock::iterator BlockBegin = Builder.GetInsertBlock()->begin();
BasicBlock::iterator IP = Builder.GetInsertPoint();
if (IP != BlockBegin) {
--IP;
for (; ScanLimit; --IP, --ScanLimit) {
if (isa<DbgInfoIntrinsic>(IP))
ScanLimit++;
if (IP->getOpcode() == (unsigned)Opcode && IP->getOperand(0) == LHS &&
IP->getOperand(1) == RHS)
return IP;
if (IP == BlockBegin) break;
}
}
BasicBlock *SaveInsertBB = Builder.GetInsertBlock();
BasicBlock::iterator SaveInsertPt = Builder.GetInsertPoint();
while (const Loop *L = SE.LI->getLoopFor(Builder.GetInsertBlock())) {
if (!L->isLoopInvariant(LHS) || !L->isLoopInvariant(RHS)) break;
BasicBlock *Preheader = L->getLoopPreheader();
if (!Preheader) break;
Builder.SetInsertPoint(Preheader, Preheader->getTerminator());
}
Value *BO = Builder.CreateBinOp(Opcode, LHS, RHS, "tmp");
rememberInstruction(BO);
if (SaveInsertBB)
restoreInsertPoint(SaveInsertBB, SaveInsertPt);
return BO;
}
static bool FactorOutConstant(const SCEV *&S,
const SCEV *&Remainder,
const SCEV *Factor,
ScalarEvolution &SE,
const TargetData *TD) {
if (Factor->isOne())
return true;
if (S == Factor) {
S = SE.getConstant(S->getType(), 1);
return true;
}
if (const SCEVConstant *C = dyn_cast<SCEVConstant>(S)) {
if (C->isZero())
return true;
if (const SCEVConstant *FC = dyn_cast<SCEVConstant>(Factor)) {
ConstantInt *CI =
ConstantInt::get(SE.getContext(),
C->getValue()->getValue().sdiv(
FC->getValue()->getValue()));
if (!CI->isZero()) {
const SCEV *Div = SE.getConstant(CI);
S = Div;
Remainder =
SE.getAddExpr(Remainder,
SE.getConstant(C->getValue()->getValue().srem(
FC->getValue()->getValue())));
return true;
}
}
}
if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(S)) {
if (TD) {
const SCEVConstant *FC = cast<SCEVConstant>(Factor);
if (const SCEVConstant *C = dyn_cast<SCEVConstant>(M->getOperand(0)))
if (!C->getValue()->getValue().srem(FC->getValue()->getValue())) {
SmallVector<const SCEV *, 4> NewMulOps(M->op_begin(), M->op_end());
NewMulOps[0] =
SE.getConstant(C->getValue()->getValue().sdiv(
FC->getValue()->getValue()));
S = SE.getMulExpr(NewMulOps);
return true;
}
} else {
for (unsigned i = 0, e = M->getNumOperands(); i != e; ++i) {
const SCEV *SOp = M->getOperand(i);
const SCEV *Remainder = SE.getConstant(SOp->getType(), 0);
if (FactorOutConstant(SOp, Remainder, Factor, SE, TD) &&
Remainder->isZero()) {
SmallVector<const SCEV *, 4> NewMulOps(M->op_begin(), M->op_end());
NewMulOps[i] = SOp;
S = SE.getMulExpr(NewMulOps);
return true;
}
}
}
}
if (const SCEVAddRecExpr *A = dyn_cast<SCEVAddRecExpr>(S)) {
const SCEV *Step = A->getStepRecurrence(SE);
const SCEV *StepRem = SE.getConstant(Step->getType(), 0);
if (!FactorOutConstant(Step, StepRem, Factor, SE, TD))
return false;
if (!StepRem->isZero())
return false;
const SCEV *Start = A->getStart();
if (!FactorOutConstant(Start, Remainder, Factor, SE, TD))
return false;
S = SE.getAddRecExpr(Start, Step, A->getLoop());
return true;
}
return false;
}
static void SimplifyAddOperands(SmallVectorImpl<const SCEV *> &Ops,
const Type *Ty,
ScalarEvolution &SE) {
unsigned NumAddRecs = 0;
for (unsigned i = Ops.size(); i > 0 && isa<SCEVAddRecExpr>(Ops[i-1]); --i)
++NumAddRecs;
SmallVector<const SCEV *, 8> NoAddRecs(Ops.begin(), Ops.end() - NumAddRecs);
SmallVector<const SCEV *, 8> AddRecs(Ops.end() - NumAddRecs, Ops.end());
const SCEV *Sum = NoAddRecs.empty() ?
SE.getConstant(Ty, 0) :
SE.getAddExpr(NoAddRecs);
Ops.clear();
if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(Sum))
Ops.append(Add->op_begin(), Add->op_end());
else if (!Sum->isZero())
Ops.push_back(Sum);
Ops.append(AddRecs.begin(), AddRecs.end());
}
static void SplitAddRecs(SmallVectorImpl<const SCEV *> &Ops,
const Type *Ty,
ScalarEvolution &SE) {
SmallVector<const SCEV *, 8> AddRecs;
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
while (const SCEVAddRecExpr *A = dyn_cast<SCEVAddRecExpr>(Ops[i])) {
const SCEV *Start = A->getStart();
if (Start->isZero()) break;
const SCEV *Zero = SE.getConstant(Ty, 0);
AddRecs.push_back(SE.getAddRecExpr(Zero,
A->getStepRecurrence(SE),
A->getLoop()));
if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(Start)) {
Ops[i] = Zero;
Ops.append(Add->op_begin(), Add->op_end());
e += Add->getNumOperands();
} else {
Ops[i] = Start;
}
}
if (!AddRecs.empty()) {
Ops.append(AddRecs.begin(), AddRecs.end());
SimplifyAddOperands(Ops, Ty, SE);
}
}
Value *SCEVExpander::expandAddToGEP(const SCEV *const *op_begin,
const SCEV *const *op_end,
const PointerType *PTy,
const Type *Ty,
Value *V) {
const Type *ElTy = PTy->getElementType();
SmallVector<Value *, 4> GepIndices;
SmallVector<const SCEV *, 8> Ops(op_begin, op_end);
bool AnyNonZeroIndices = false;
SplitAddRecs(Ops, Ty, SE);
for (;;) {
SmallVector<const SCEV *, 8> ScaledOps;
if (ElTy->isSized()) {
const SCEV *ElSize = SE.getSizeOfExpr(ElTy);
if (!ElSize->isZero()) {
SmallVector<const SCEV *, 8> NewOps;
for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
const SCEV *Op = Ops[i];
const SCEV *Remainder = SE.getConstant(Ty, 0);
if (FactorOutConstant(Op, Remainder, ElSize, SE, SE.TD)) {
ScaledOps.push_back(Op);
if (!Remainder->isZero())
NewOps.push_back(Remainder);
AnyNonZeroIndices = true;
} else {
NewOps.push_back(Ops[i]);
}
}
if (!ScaledOps.empty()) {
Ops = NewOps;
SimplifyAddOperands(Ops, Ty, SE);
}
}
}
Value *Scaled = ScaledOps.empty() ?
Constant::getNullValue(Ty) :
expandCodeFor(SE.getAddExpr(ScaledOps), Ty);
GepIndices.push_back(Scaled);
while (const StructType *STy = dyn_cast<StructType>(ElTy)) {
bool FoundFieldNo = false;
if (STy->getNumElements() == 0) break;
if (SE.TD) {
if (Ops.empty()) break;
if (const SCEVConstant *C = dyn_cast<SCEVConstant>(Ops[0]))
if (SE.getTypeSizeInBits(C->getType()) <= 64) {
const StructLayout &SL = *SE.TD->getStructLayout(STy);
uint64_t FullOffset = C->getValue()->getZExtValue();
if (FullOffset < SL.getSizeInBytes()) {
unsigned ElIdx = SL.getElementContainingOffset(FullOffset);
GepIndices.push_back(
ConstantInt::get(Type::getInt32Ty(Ty->getContext()), ElIdx));
ElTy = STy->getTypeAtIndex(ElIdx);
Ops[0] =
SE.getConstant(Ty, FullOffset - SL.getElementOffset(ElIdx));
AnyNonZeroIndices = true;
FoundFieldNo = true;
}
}
} else {
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(Ops[i])) {
const Type *CTy;
Constant *FieldNo;
if (U->isOffsetOf(CTy, FieldNo) && CTy == STy) {
GepIndices.push_back(FieldNo);
ElTy =
STy->getTypeAtIndex(cast<ConstantInt>(FieldNo)->getZExtValue());
Ops[i] = SE.getConstant(Ty, 0);
AnyNonZeroIndices = true;
FoundFieldNo = true;
break;
}
}
}
if (!FoundFieldNo) {
ElTy = STy->getTypeAtIndex(0u);
GepIndices.push_back(
Constant::getNullValue(Type::getInt32Ty(Ty->getContext())));
}
}
if (const ArrayType *ATy = dyn_cast<ArrayType>(ElTy))
ElTy = ATy->getElementType();
else
break;
}
if (!AnyNonZeroIndices) {
V = InsertNoopCastOfTo(V,
Type::getInt8PtrTy(Ty->getContext(), PTy->getAddressSpace()));
Value *Idx = expandCodeFor(SE.getAddExpr(Ops), Ty);
if (Constant *CLHS = dyn_cast<Constant>(V))
if (Constant *CRHS = dyn_cast<Constant>(Idx))
return ConstantExpr::getGetElementPtr(CLHS, &CRHS, 1);
unsigned ScanLimit = 6;
BasicBlock::iterator BlockBegin = Builder.GetInsertBlock()->begin();
BasicBlock::iterator IP = Builder.GetInsertPoint();
if (IP != BlockBegin) {
--IP;
for (; ScanLimit; --IP, --ScanLimit) {
if (isa<DbgInfoIntrinsic>(IP))
ScanLimit++;
if (IP->getOpcode() == Instruction::GetElementPtr &&
IP->getOperand(0) == V && IP->getOperand(1) == Idx)
return IP;
if (IP == BlockBegin) break;
}
}
BasicBlock *SaveInsertBB = Builder.GetInsertBlock();
BasicBlock::iterator SaveInsertPt = Builder.GetInsertPoint();
while (const Loop *L = SE.LI->getLoopFor(Builder.GetInsertBlock())) {
if (!L->isLoopInvariant(V) || !L->isLoopInvariant(Idx)) break;
BasicBlock *Preheader = L->getLoopPreheader();
if (!Preheader) break;
Builder.SetInsertPoint(Preheader, Preheader->getTerminator());
}
Value *GEP = Builder.CreateGEP(V, Idx, "uglygep");
rememberInstruction(GEP);
if (SaveInsertBB)
restoreInsertPoint(SaveInsertBB, SaveInsertPt);
return GEP;
}
BasicBlock *SaveInsertBB = Builder.GetInsertBlock();
BasicBlock::iterator SaveInsertPt = Builder.GetInsertPoint();
while (const Loop *L = SE.LI->getLoopFor(Builder.GetInsertBlock())) {
if (!L->isLoopInvariant(V)) break;
bool AnyIndexNotLoopInvariant = false;
for (SmallVectorImpl<Value *>::const_iterator I = GepIndices.begin(),
E = GepIndices.end(); I != E; ++I)
if (!L->isLoopInvariant(*I)) {
AnyIndexNotLoopInvariant = true;
break;
}
if (AnyIndexNotLoopInvariant)
break;
BasicBlock *Preheader = L->getLoopPreheader();
if (!Preheader) break;
Builder.SetInsertPoint(Preheader, Preheader->getTerminator());
}
Value *Casted = V;
if (V->getType() != PTy)
Casted = InsertNoopCastOfTo(Casted, PTy);
Value *GEP = Builder.CreateGEP(Casted,
GepIndices.begin(),
GepIndices.end(),
"scevgep");
Ops.push_back(SE.getUnknown(GEP));
rememberInstruction(GEP);
if (SaveInsertBB)
restoreInsertPoint(SaveInsertBB, SaveInsertPt);
return expand(SE.getAddExpr(Ops));
}
static bool isNonConstantNegative(const SCEV *F) {
const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(F);
if (!Mul) return false;
const SCEVConstant *SC = dyn_cast<SCEVConstant>(Mul->getOperand(0));
if (!SC) return false;
return SC->getValue()->getValue().isNegative();
}
static const Loop *PickMostRelevantLoop(const Loop *A, const Loop *B,
DominatorTree &DT) {
if (!A) return B;
if (!B) return A;
if (A->contains(B)) return B;
if (B->contains(A)) return A;
if (DT.dominates(A->getHeader(), B->getHeader())) return B;
if (DT.dominates(B->getHeader(), A->getHeader())) return A;
return A; }
const Loop *SCEVExpander::getRelevantLoop(const SCEV *S) {
std::pair<DenseMap<const SCEV *, const Loop *>::iterator, bool> Pair =
RelevantLoops.insert(std::make_pair(S, static_cast<const Loop *>(0)));
if (!Pair.second)
return Pair.first->second;
if (isa<SCEVConstant>(S))
return 0;
if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(S)) {
if (const Instruction *I = dyn_cast<Instruction>(U->getValue()))
return Pair.first->second = SE.LI->getLoopFor(I->getParent());
return 0;
}
if (const SCEVNAryExpr *N = dyn_cast<SCEVNAryExpr>(S)) {
const Loop *L = 0;
if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S))
L = AR->getLoop();
for (SCEVNAryExpr::op_iterator I = N->op_begin(), E = N->op_end();
I != E; ++I)
L = PickMostRelevantLoop(L, getRelevantLoop(*I), *SE.DT);
return RelevantLoops[N] = L;
}
if (const SCEVCastExpr *C = dyn_cast<SCEVCastExpr>(S)) {
const Loop *Result = getRelevantLoop(C->getOperand());
return RelevantLoops[C] = Result;
}
if (const SCEVUDivExpr *D = dyn_cast<SCEVUDivExpr>(S)) {
const Loop *Result =
PickMostRelevantLoop(getRelevantLoop(D->getLHS()),
getRelevantLoop(D->getRHS()),
*SE.DT);
return RelevantLoops[D] = Result;
}
llvm_unreachable("Unexpected SCEV type!");
return 0;
}
namespace {
class LoopCompare {
DominatorTree &DT;
public:
explicit LoopCompare(DominatorTree &dt) : DT(dt) {}
bool operator()(std::pair<const Loop *, const SCEV *> LHS,
std::pair<const Loop *, const SCEV *> RHS) const {
if (LHS.second->getType()->isPointerTy() !=
RHS.second->getType()->isPointerTy())
return LHS.second->getType()->isPointerTy();
if (LHS.first != RHS.first)
return PickMostRelevantLoop(LHS.first, RHS.first, DT) != LHS.first;
if (isNonConstantNegative(LHS.second)) {
if (!isNonConstantNegative(RHS.second))
return false;
} else if (isNonConstantNegative(RHS.second))
return true;
return false;
}
};
}
Value *SCEVExpander::visitAddExpr(const SCEVAddExpr *S) {
const Type *Ty = SE.getEffectiveSCEVType(S->getType());
SmallVector<std::pair<const Loop *, const SCEV *>, 8> OpsAndLoops;
for (std::reverse_iterator<SCEVAddExpr::op_iterator> I(S->op_end()),
E(S->op_begin()); I != E; ++I)
OpsAndLoops.push_back(std::make_pair(getRelevantLoop(*I), *I));
std::stable_sort(OpsAndLoops.begin(), OpsAndLoops.end(), LoopCompare(*SE.DT));
Value *Sum = 0;
for (SmallVectorImpl<std::pair<const Loop *, const SCEV *> >::iterator
I = OpsAndLoops.begin(), E = OpsAndLoops.end(); I != E; ) {
const Loop *CurLoop = I->first;
const SCEV *Op = I->second;
if (!Sum) {
Sum = expand(Op);
++I;
} else if (const PointerType *PTy = dyn_cast<PointerType>(Sum->getType())) {
SmallVector<const SCEV *, 4> NewOps;
for (; I != E && I->first == CurLoop; ++I) {
const SCEV *X = I->second;
if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(X))
if (!isa<Instruction>(U->getValue()))
X = SE.getSCEV(U->getValue());
NewOps.push_back(X);
}
Sum = expandAddToGEP(NewOps.begin(), NewOps.end(), PTy, Ty, Sum);
} else if (const PointerType *PTy = dyn_cast<PointerType>(Op->getType())) {
SmallVector<const SCEV *, 4> NewOps;
NewOps.push_back(isa<Instruction>(Sum) ? SE.getUnknown(Sum) :
SE.getSCEV(Sum));
for (++I; I != E && I->first == CurLoop; ++I)
NewOps.push_back(I->second);
Sum = expandAddToGEP(NewOps.begin(), NewOps.end(), PTy, Ty, expand(Op));
} else if (isNonConstantNegative(Op)) {
Value *W = expandCodeFor(SE.getNegativeSCEV(Op), Ty);
Sum = InsertNoopCastOfTo(Sum, Ty);
Sum = InsertBinop(Instruction::Sub, Sum, W);
++I;
} else {
Value *W = expandCodeFor(Op, Ty);
Sum = InsertNoopCastOfTo(Sum, Ty);
if (isa<Constant>(Sum)) std::swap(Sum, W);
Sum = InsertBinop(Instruction::Add, Sum, W);
++I;
}
}
return Sum;
}
Value *SCEVExpander::visitMulExpr(const SCEVMulExpr *S) {
const Type *Ty = SE.getEffectiveSCEVType(S->getType());
SmallVector<std::pair<const Loop *, const SCEV *>, 8> OpsAndLoops;
for (std::reverse_iterator<SCEVMulExpr::op_iterator> I(S->op_end()),
E(S->op_begin()); I != E; ++I)
OpsAndLoops.push_back(std::make_pair(getRelevantLoop(*I), *I));
std::stable_sort(OpsAndLoops.begin(), OpsAndLoops.end(), LoopCompare(*SE.DT));
Value *Prod = 0;
for (SmallVectorImpl<std::pair<const Loop *, const SCEV *> >::iterator
I = OpsAndLoops.begin(), E = OpsAndLoops.end(); I != E; ) {
const SCEV *Op = I->second;
if (!Prod) {
Prod = expand(Op);
++I;
} else if (Op->isAllOnesValue()) {
Prod = InsertNoopCastOfTo(Prod, Ty);
Prod = InsertBinop(Instruction::Sub, Constant::getNullValue(Ty), Prod);
++I;
} else {
Value *W = expandCodeFor(Op, Ty);
Prod = InsertNoopCastOfTo(Prod, Ty);
if (isa<Constant>(Prod)) std::swap(Prod, W);
Prod = InsertBinop(Instruction::Mul, Prod, W);
++I;
}
}
return Prod;
}
Value *SCEVExpander::visitUDivExpr(const SCEVUDivExpr *S) {
const Type *Ty = SE.getEffectiveSCEVType(S->getType());
Value *LHS = expandCodeFor(S->getLHS(), Ty);
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(S->getRHS())) {
const APInt &RHS = SC->getValue()->getValue();
if (RHS.isPowerOf2())
return InsertBinop(Instruction::LShr, LHS,
ConstantInt::get(Ty, RHS.logBase2()));
}
Value *RHS = expandCodeFor(S->getRHS(), Ty);
return InsertBinop(Instruction::UDiv, LHS, RHS);
}
static void ExposePointerBase(const SCEV *&Base, const SCEV *&Rest,
ScalarEvolution &SE) {
while (const SCEVAddRecExpr *A = dyn_cast<SCEVAddRecExpr>(Base)) {
Base = A->getStart();
Rest = SE.getAddExpr(Rest,
SE.getAddRecExpr(SE.getConstant(A->getType(), 0),
A->getStepRecurrence(SE),
A->getLoop()));
}
if (const SCEVAddExpr *A = dyn_cast<SCEVAddExpr>(Base)) {
Base = A->getOperand(A->getNumOperands()-1);
SmallVector<const SCEV *, 8> NewAddOps(A->op_begin(), A->op_end());
NewAddOps.back() = Rest;
Rest = SE.getAddExpr(NewAddOps);
ExposePointerBase(Base, Rest, SE);
}
}
PHINode *
SCEVExpander::getAddRecExprPHILiterally(const SCEVAddRecExpr *Normalized,
const Loop *L,
const Type *ExpandTy,
const Type *IntTy) {
for (BasicBlock::iterator I = L->getHeader()->begin();
PHINode *PN = dyn_cast<PHINode>(I); ++I)
if (SE.isSCEVable(PN->getType()) &&
(SE.getEffectiveSCEVType(PN->getType()) ==
SE.getEffectiveSCEVType(Normalized->getType())) &&
SE.getSCEV(PN) == Normalized)
if (BasicBlock *LatchBlock = L->getLoopLatch()) {
Instruction *IncV =
cast<Instruction>(PN->getIncomingValueForBlock(LatchBlock));
do {
if (IncV->getNumOperands() == 0 || isa<PHINode>(IncV)) {
IncV = 0;
break;
}
for (User::op_iterator OI = IncV->op_begin()+1,
OE = IncV->op_end(); OI != OE; ++OI)
if (Instruction *OInst = dyn_cast<Instruction>(OI))
if (!SE.DT->dominates(OInst, IVIncInsertPos)) {
IncV = 0;
break;
}
if (!IncV)
break;
IncV = dyn_cast<Instruction>(IncV->getOperand(0));
if (!IncV)
break;
if (IncV->mayHaveSideEffects()) {
IncV = 0;
break;
}
} while (IncV != PN);
if (IncV) {
InsertedValues.insert(PN);
IncV = cast<Instruction>(PN->getIncomingValueForBlock(LatchBlock));
rememberInstruction(IncV);
if (L == IVIncInsertLoop)
do {
if (SE.DT->dominates(IncV, IVIncInsertPos))
break;
IncV->moveBefore(IVIncInsertPos);
IVIncInsertPos = IncV;
IncV = cast<Instruction>(IncV->getOperand(0));
} while (IncV != PN);
return PN;
}
}
BasicBlock *SaveInsertBB = Builder.GetInsertBlock();
BasicBlock::iterator SaveInsertPt = Builder.GetInsertPoint();
Value *StartV = expandCodeFor(Normalized->getStart(), ExpandTy,
L->getHeader()->begin());
const SCEV *Step = Normalized->getStepRecurrence(SE);
bool isPointer = ExpandTy->isPointerTy();
bool isNegative = !isPointer && isNonConstantNegative(Step);
if (isNegative)
Step = SE.getNegativeSCEV(Step);
Value *StepV = expandCodeFor(Step, IntTy, L->getHeader()->begin());
Builder.SetInsertPoint(L->getHeader(), L->getHeader()->begin());
PHINode *PN = Builder.CreatePHI(ExpandTy, "lsr.iv");
rememberInstruction(PN);
BasicBlock *Header = L->getHeader();
for (pred_iterator HPI = pred_begin(Header), HPE = pred_end(Header);
HPI != HPE; ++HPI) {
BasicBlock *Pred = *HPI;
if (!L->contains(Pred)) {
PN->addIncoming(StartV, Pred);
continue;
}
Instruction *InsertPos = L == IVIncInsertLoop ?
IVIncInsertPos : Pred->getTerminator();
Builder.SetInsertPoint(InsertPos->getParent(), InsertPos);
Value *IncV;
if (isPointer) {
const PointerType *GEPPtrTy = cast<PointerType>(ExpandTy);
if (!isa<ConstantInt>(StepV))
GEPPtrTy = PointerType::get(Type::getInt1Ty(SE.getContext()),
GEPPtrTy->getAddressSpace());
const SCEV *const StepArray[1] = { SE.getSCEV(StepV) };
IncV = expandAddToGEP(StepArray, StepArray+1, GEPPtrTy, IntTy, PN);
if (IncV->getType() != PN->getType()) {
IncV = Builder.CreateBitCast(IncV, PN->getType(), "tmp");
rememberInstruction(IncV);
}
} else {
IncV = isNegative ?
Builder.CreateSub(PN, StepV, "lsr.iv.next") :
Builder.CreateAdd(PN, StepV, "lsr.iv.next");
rememberInstruction(IncV);
}
PN->addIncoming(IncV, Pred);
}
if (SaveInsertBB)
restoreInsertPoint(SaveInsertBB, SaveInsertPt);
InsertedValues.insert(PN);
return PN;
}
Value *SCEVExpander::expandAddRecExprLiterally(const SCEVAddRecExpr *S) {
const Type *STy = S->getType();
const Type *IntTy = SE.getEffectiveSCEVType(STy);
const Loop *L = S->getLoop();
const SCEVAddRecExpr *Normalized = S;
if (PostIncLoops.count(L)) {
PostIncLoopSet Loops;
Loops.insert(L);
Normalized =
cast<SCEVAddRecExpr>(TransformForPostIncUse(Normalize, S, 0, 0,
Loops, SE, *SE.DT));
}
const SCEV *Start = Normalized->getStart();
const SCEV *PostLoopOffset = 0;
if (!SE.properlyDominates(Start, L->getHeader())) {
PostLoopOffset = Start;
Start = SE.getConstant(Normalized->getType(), 0);
Normalized =
cast<SCEVAddRecExpr>(SE.getAddRecExpr(Start,
Normalized->getStepRecurrence(SE),
Normalized->getLoop()));
}
const SCEV *Step = Normalized->getStepRecurrence(SE);
const SCEV *PostLoopScale = 0;
if (!SE.dominates(Step, L->getHeader())) {
PostLoopScale = Step;
Step = SE.getConstant(Normalized->getType(), 1);
Normalized =
cast<SCEVAddRecExpr>(SE.getAddRecExpr(Start, Step,
Normalized->getLoop()));
}
const Type *ExpandTy = PostLoopScale ? IntTy : STy;
PHINode *PN = getAddRecExprPHILiterally(Normalized, L, ExpandTy, IntTy);
Value *Result;
if (!PostIncLoops.count(L))
Result = PN;
else {
BasicBlock *LatchBlock = L->getLoopLatch();
assert(LatchBlock && "PostInc mode requires a unique loop latch!");
Result = PN->getIncomingValueForBlock(LatchBlock);
}
if (PostLoopScale) {
Result = InsertNoopCastOfTo(Result, IntTy);
Result = Builder.CreateMul(Result,
expandCodeFor(PostLoopScale, IntTy));
rememberInstruction(Result);
}
if (PostLoopOffset) {
if (const PointerType *PTy = dyn_cast<PointerType>(ExpandTy)) {
const SCEV *const OffsetArray[1] = { PostLoopOffset };
Result = expandAddToGEP(OffsetArray, OffsetArray+1, PTy, IntTy, Result);
} else {
Result = InsertNoopCastOfTo(Result, IntTy);
Result = Builder.CreateAdd(Result,
expandCodeFor(PostLoopOffset, IntTy));
rememberInstruction(Result);
}
}
return Result;
}
Value *SCEVExpander::visitAddRecExpr(const SCEVAddRecExpr *S) {
if (!CanonicalMode) return expandAddRecExprLiterally(S);
const Type *Ty = SE.getEffectiveSCEVType(S->getType());
const Loop *L = S->getLoop();
PHINode *CanonicalIV = 0;
if (PHINode *PN = L->getCanonicalInductionVariable())
if (SE.getTypeSizeInBits(PN->getType()) >= SE.getTypeSizeInBits(Ty))
CanonicalIV = PN;
if (CanonicalIV &&
SE.getTypeSizeInBits(CanonicalIV->getType()) >
SE.getTypeSizeInBits(Ty)) {
SmallVector<const SCEV *, 4> NewOps(S->getNumOperands());
for (unsigned i = 0, e = S->getNumOperands(); i != e; ++i)
NewOps[i] = SE.getAnyExtendExpr(S->op_begin()[i], CanonicalIV->getType());
Value *V = expand(SE.getAddRecExpr(NewOps, S->getLoop()));
BasicBlock *SaveInsertBB = Builder.GetInsertBlock();
BasicBlock::iterator SaveInsertPt = Builder.GetInsertPoint();
BasicBlock::iterator NewInsertPt =
llvm::next(BasicBlock::iterator(cast<Instruction>(V)));
while (isa<PHINode>(NewInsertPt) || isa<DbgInfoIntrinsic>(NewInsertPt))
++NewInsertPt;
V = expandCodeFor(SE.getTruncateExpr(SE.getUnknown(V), Ty), 0,
NewInsertPt);
restoreInsertPoint(SaveInsertBB, SaveInsertPt);
return V;
}
if (!S->getStart()->isZero()) {
SmallVector<const SCEV *, 4> NewOps(S->op_begin(), S->op_end());
NewOps[0] = SE.getConstant(Ty, 0);
const SCEV *Rest = SE.getAddRecExpr(NewOps, L);
const SCEV *Base = S->getStart();
const SCEV *RestArray[1] = { Rest };
ExposePointerBase(Base, RestArray[0], SE);
if (const PointerType *PTy = dyn_cast<PointerType>(Base->getType())) {
if (!isa<SCEVMulExpr>(Base) && !isa<SCEVUDivExpr>(Base)) {
Value *StartV = expand(Base);
assert(StartV->getType() == PTy && "Pointer type mismatch for GEP!");
return expandAddToGEP(RestArray, RestArray+1, PTy, Ty, StartV);
}
}
return expand(SE.getAddExpr(SE.getUnknown(expand(S->getStart())),
SE.getUnknown(expand(Rest))));
}
if (!CanonicalIV) {
BasicBlock *Header = L->getHeader();
CanonicalIV = PHINode::Create(Ty, "indvar", Header->begin());
rememberInstruction(CanonicalIV);
Constant *One = ConstantInt::get(Ty, 1);
for (pred_iterator HPI = pred_begin(Header), HPE = pred_end(Header);
HPI != HPE; ++HPI) {
BasicBlock *HP = *HPI;
if (L->contains(HP)) {
Instruction *Add = BinaryOperator::CreateAdd(CanonicalIV, One,
"indvar.next",
HP->getTerminator());
rememberInstruction(Add);
CanonicalIV->addIncoming(Add, HP);
} else {
CanonicalIV->addIncoming(Constant::getNullValue(Ty), HP);
}
}
}
if (S->isAffine() && S->getOperand(1)->isOne()) {
assert(Ty == SE.getEffectiveSCEVType(CanonicalIV->getType()) &&
"IVs with types different from the canonical IV should "
"already have been handled!");
return CanonicalIV;
}
if (S->isAffine()) return
expand(SE.getTruncateOrNoop(
SE.getMulExpr(SE.getUnknown(CanonicalIV),
SE.getNoopOrAnyExtend(S->getOperand(1),
CanonicalIV->getType())),
Ty));
const SCEV *IH = SE.getUnknown(CanonicalIV);
const SCEV *NewS = S;
const SCEV *Ext = SE.getNoopOrAnyExtend(S, CanonicalIV->getType());
if (isa<SCEVAddRecExpr>(Ext))
NewS = Ext;
const SCEV *V = cast<SCEVAddRecExpr>(NewS)->evaluateAtIteration(IH, SE);
const SCEV *T = SE.getTruncateOrNoop(V, Ty);
return expand(T);
}
Value *SCEVExpander::visitTruncateExpr(const SCEVTruncateExpr *S) {
const Type *Ty = SE.getEffectiveSCEVType(S->getType());
Value *V = expandCodeFor(S->getOperand(),
SE.getEffectiveSCEVType(S->getOperand()->getType()));
Value *I = Builder.CreateTrunc(V, Ty, "tmp");
rememberInstruction(I);
return I;
}
Value *SCEVExpander::visitZeroExtendExpr(const SCEVZeroExtendExpr *S) {
const Type *Ty = SE.getEffectiveSCEVType(S->getType());
Value *V = expandCodeFor(S->getOperand(),
SE.getEffectiveSCEVType(S->getOperand()->getType()));
Value *I = Builder.CreateZExt(V, Ty, "tmp");
rememberInstruction(I);
return I;
}
Value *SCEVExpander::visitSignExtendExpr(const SCEVSignExtendExpr *S) {
const Type *Ty = SE.getEffectiveSCEVType(S->getType());
Value *V = expandCodeFor(S->getOperand(),
SE.getEffectiveSCEVType(S->getOperand()->getType()));
Value *I = Builder.CreateSExt(V, Ty, "tmp");
rememberInstruction(I);
return I;
}
Value *SCEVExpander::visitSMaxExpr(const SCEVSMaxExpr *S) {
Value *LHS = expand(S->getOperand(S->getNumOperands()-1));
const Type *Ty = LHS->getType();
for (int i = S->getNumOperands()-2; i >= 0; --i) {
if (S->getOperand(i)->getType() != Ty) {
Ty = SE.getEffectiveSCEVType(Ty);
LHS = InsertNoopCastOfTo(LHS, Ty);
}
Value *RHS = expandCodeFor(S->getOperand(i), Ty);
Value *ICmp = Builder.CreateICmpSGT(LHS, RHS, "tmp");
rememberInstruction(ICmp);
Value *Sel = Builder.CreateSelect(ICmp, LHS, RHS, "smax");
rememberInstruction(Sel);
LHS = Sel;
}
if (LHS->getType() != S->getType())
LHS = InsertNoopCastOfTo(LHS, S->getType());
return LHS;
}
Value *SCEVExpander::visitUMaxExpr(const SCEVUMaxExpr *S) {
Value *LHS = expand(S->getOperand(S->getNumOperands()-1));
const Type *Ty = LHS->getType();
for (int i = S->getNumOperands()-2; i >= 0; --i) {
if (S->getOperand(i)->getType() != Ty) {
Ty = SE.getEffectiveSCEVType(Ty);
LHS = InsertNoopCastOfTo(LHS, Ty);
}
Value *RHS = expandCodeFor(S->getOperand(i), Ty);
Value *ICmp = Builder.CreateICmpUGT(LHS, RHS, "tmp");
rememberInstruction(ICmp);
Value *Sel = Builder.CreateSelect(ICmp, LHS, RHS, "umax");
rememberInstruction(Sel);
LHS = Sel;
}
if (LHS->getType() != S->getType())
LHS = InsertNoopCastOfTo(LHS, S->getType());
return LHS;
}
Value *SCEVExpander::expandCodeFor(const SCEV *SH, const Type *Ty,
Instruction *I) {
BasicBlock::iterator IP = I;
while (isInsertedInstruction(IP) || isa<DbgInfoIntrinsic>(IP))
++IP;
Builder.SetInsertPoint(IP->getParent(), IP);
return expandCodeFor(SH, Ty);
}
Value *SCEVExpander::expandCodeFor(const SCEV *SH, const Type *Ty) {
Value *V = expand(SH);
if (Ty) {
assert(SE.getTypeSizeInBits(Ty) == SE.getTypeSizeInBits(SH->getType()) &&
"non-trivial casts should be done with the SCEVs directly!");
V = InsertNoopCastOfTo(V, Ty);
}
return V;
}
Value *SCEVExpander::expand(const SCEV *S) {
Instruction *InsertPt = Builder.GetInsertPoint();
for (Loop *L = SE.LI->getLoopFor(Builder.GetInsertBlock()); ;
L = L->getParentLoop())
if (SE.isLoopInvariant(S, L)) {
if (!L) break;
if (BasicBlock *Preheader = L->getLoopPreheader())
InsertPt = Preheader->getTerminator();
} else {
if (L && SE.hasComputableLoopEvolution(S, L) && !PostIncLoops.count(L))
InsertPt = L->getHeader()->getFirstNonPHI();
while (isInsertedInstruction(InsertPt) || isa<DbgInfoIntrinsic>(InsertPt))
InsertPt = llvm::next(BasicBlock::iterator(InsertPt));
break;
}
std::map<std::pair<const SCEV *, Instruction *>,
AssertingVH<Value> >::iterator I =
InsertedExpressions.find(std::make_pair(S, InsertPt));
if (I != InsertedExpressions.end())
return I->second;
BasicBlock *SaveInsertBB = Builder.GetInsertBlock();
BasicBlock::iterator SaveInsertPt = Builder.GetInsertPoint();
Builder.SetInsertPoint(InsertPt->getParent(), InsertPt);
Value *V = visit(S);
if (PostIncLoops.empty())
InsertedExpressions[std::make_pair(S, InsertPt)] = V;
restoreInsertPoint(SaveInsertBB, SaveInsertPt);
return V;
}
void SCEVExpander::rememberInstruction(Value *I) {
if (!PostIncLoops.empty())
InsertedPostIncValues.insert(I);
else
InsertedValues.insert(I);
if (Builder.GetInsertPoint() == I) {
BasicBlock::iterator It = cast<Instruction>(I);
do { ++It; } while (isInsertedInstruction(It) ||
isa<DbgInfoIntrinsic>(It));
Builder.SetInsertPoint(Builder.GetInsertBlock(), It);
}
}
void SCEVExpander::restoreInsertPoint(BasicBlock *BB, BasicBlock::iterator I) {
while (isInsertedInstruction(I) || isa<DbgInfoIntrinsic>(I)) ++I;
Builder.SetInsertPoint(BB, I);
}
PHINode *
SCEVExpander::getOrInsertCanonicalInductionVariable(const Loop *L,
const Type *Ty) {
assert(Ty->isIntegerTy() && "Can only insert integer induction variables!");
const SCEV *H = SE.getAddRecExpr(SE.getConstant(Ty, 0),
SE.getConstant(Ty, 1), L);
BasicBlock *SaveInsertBB = Builder.GetInsertBlock();
BasicBlock::iterator SaveInsertPt = Builder.GetInsertPoint();
PHINode *V = cast<PHINode>(expandCodeFor(H, 0, L->getHeader()->begin()));
if (SaveInsertBB)
restoreInsertPoint(SaveInsertBB, SaveInsertPt);
return V;
}