// There are several aspects to this library. First is the representation of
// scalar expressions, which are represented as subclasses of the SCEV class.
// These classes are used to represent certain types of subexpressions that we
-// can handle. These classes are reference counted, managed by the const SCEV*
+// can handle. These classes are reference counted, managed by the const SCEV *
// class. We only create one SCEV of a particular shape, so pointer-comparisons
// for equality are legal.
//
#include "llvm/DerivedTypes.h"
#include "llvm/GlobalVariable.h"
#include "llvm/Instructions.h"
+#include "llvm/LLVMContext.h"
+#include "llvm/Operator.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/ConstantRange.h"
+#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/InstIterator.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/SmallPtrSet.h"
#include <algorithm>
using namespace llvm;
}
SCEVCouldNotCompute::SCEVCouldNotCompute() :
- SCEV(scCouldNotCompute) {}
-
-void SCEVCouldNotCompute::Profile(FoldingSetNodeID &ID) const {
- assert(0 && "Attempt to use a SCEVCouldNotCompute object!");
-}
+ SCEV(FoldingSetNodeID(), scCouldNotCompute) {}
bool SCEVCouldNotCompute::isLoopInvariant(const Loop *L) const {
- assert(0 && "Attempt to use a SCEVCouldNotCompute object!");
+ llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
return false;
}
const Type *SCEVCouldNotCompute::getType() const {
- assert(0 && "Attempt to use a SCEVCouldNotCompute object!");
+ llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
return 0;
}
bool SCEVCouldNotCompute::hasComputableLoopEvolution(const Loop *L) const {
- assert(0 && "Attempt to use a SCEVCouldNotCompute object!");
+ llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
return false;
}
return S->getSCEVType() == scCouldNotCompute;
}
-const SCEV* ScalarEvolution::getConstant(ConstantInt *V) {
+const SCEV *ScalarEvolution::getConstant(ConstantInt *V) {
FoldingSetNodeID ID;
ID.AddInteger(scConstant);
ID.AddPointer(V);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVConstant>();
- new (S) SCEVConstant(V);
+ new (S) SCEVConstant(ID, V);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
-const SCEV* ScalarEvolution::getConstant(const APInt& Val) {
- return getConstant(ConstantInt::get(Val));
+const SCEV *ScalarEvolution::getConstant(const APInt& Val) {
+ return getConstant(Context->getConstantInt(Val));
}
-const SCEV*
+const SCEV *
ScalarEvolution::getConstant(const Type *Ty, uint64_t V, bool isSigned) {
- return getConstant(ConstantInt::get(cast<IntegerType>(Ty), V, isSigned));
-}
-
-void SCEVConstant::Profile(FoldingSetNodeID &ID) const {
- ID.AddInteger(scConstant);
- ID.AddPointer(V);
+ return getConstant(
+ Context->getConstantInt(cast<IntegerType>(Ty), V, isSigned));
}
const Type *SCEVConstant::getType() const { return V->getType(); }
WriteAsOperand(OS, V, false);
}
-SCEVCastExpr::SCEVCastExpr(unsigned SCEVTy,
- const SCEV* op, const Type *ty)
- : SCEV(SCEVTy), Op(op), Ty(ty) {}
-
-void SCEVCastExpr::Profile(FoldingSetNodeID &ID) const {
- ID.AddInteger(getSCEVType());
- ID.AddPointer(Op);
- ID.AddPointer(Ty);
-}
+SCEVCastExpr::SCEVCastExpr(const FoldingSetNodeID &ID,
+ unsigned SCEVTy, const SCEV *op, const Type *ty)
+ : SCEV(ID, SCEVTy), Op(op), Ty(ty) {}
bool SCEVCastExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
return Op->dominates(BB, DT);
}
-SCEVTruncateExpr::SCEVTruncateExpr(const SCEV* op, const Type *ty)
- : SCEVCastExpr(scTruncate, op, ty) {
+SCEVTruncateExpr::SCEVTruncateExpr(const FoldingSetNodeID &ID,
+ const SCEV *op, const Type *ty)
+ : SCEVCastExpr(ID, scTruncate, op, ty) {
assert((Op->getType()->isInteger() || isa<PointerType>(Op->getType())) &&
(Ty->isInteger() || isa<PointerType>(Ty)) &&
"Cannot truncate non-integer value!");
OS << "(trunc " << *Op->getType() << " " << *Op << " to " << *Ty << ")";
}
-SCEVZeroExtendExpr::SCEVZeroExtendExpr(const SCEV* op, const Type *ty)
- : SCEVCastExpr(scZeroExtend, op, ty) {
+SCEVZeroExtendExpr::SCEVZeroExtendExpr(const FoldingSetNodeID &ID,
+ const SCEV *op, const Type *ty)
+ : SCEVCastExpr(ID, scZeroExtend, op, ty) {
assert((Op->getType()->isInteger() || isa<PointerType>(Op->getType())) &&
(Ty->isInteger() || isa<PointerType>(Ty)) &&
"Cannot zero extend non-integer value!");
OS << "(zext " << *Op->getType() << " " << *Op << " to " << *Ty << ")";
}
-SCEVSignExtendExpr::SCEVSignExtendExpr(const SCEV* op, const Type *ty)
- : SCEVCastExpr(scSignExtend, op, ty) {
+SCEVSignExtendExpr::SCEVSignExtendExpr(const FoldingSetNodeID &ID,
+ const SCEV *op, const Type *ty)
+ : SCEVCastExpr(ID, scSignExtend, op, ty) {
assert((Op->getType()->isInteger() || isa<PointerType>(Op->getType())) &&
(Ty->isInteger() || isa<PointerType>(Ty)) &&
"Cannot sign extend non-integer value!");
const SCEV *Conc,
ScalarEvolution &SE) const {
for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
- const SCEV* H =
+ const SCEV *H =
getOperand(i)->replaceSymbolicValuesWithConcrete(Sym, Conc, SE);
if (H != getOperand(i)) {
- SmallVector<const SCEV*, 8> NewOps;
+ SmallVector<const SCEV *, 8> NewOps;
NewOps.reserve(getNumOperands());
for (unsigned j = 0; j != i; ++j)
NewOps.push_back(getOperand(j));
else if (isa<SCEVUMaxExpr>(this))
return SE.getUMaxExpr(NewOps);
else
- assert(0 && "Unknown commutative expr!");
+ llvm_unreachable("Unknown commutative expr!");
}
}
return this;
}
-void SCEVNAryExpr::Profile(FoldingSetNodeID &ID) const {
- ID.AddInteger(getSCEVType());
- ID.AddInteger(Operands.size());
- for (unsigned i = 0, e = Operands.size(); i != e; ++i)
- ID.AddPointer(Operands[i]);
-}
-
bool SCEVNAryExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
if (!getOperand(i)->dominates(BB, DT))
return true;
}
-void SCEVUDivExpr::Profile(FoldingSetNodeID &ID) const {
- ID.AddInteger(scUDivExpr);
- ID.AddPointer(LHS);
- ID.AddPointer(RHS);
-}
-
bool SCEVUDivExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
return LHS->dominates(BB, DT) && RHS->dominates(BB, DT);
}
return RHS->getType();
}
-void SCEVAddRecExpr::Profile(FoldingSetNodeID &ID) const {
- ID.AddInteger(scAddRecExpr);
- ID.AddInteger(Operands.size());
- for (unsigned i = 0, e = Operands.size(); i != e; ++i)
- ID.AddPointer(Operands[i]);
- ID.AddPointer(L);
-}
-
const SCEV *
SCEVAddRecExpr::replaceSymbolicValuesWithConcrete(const SCEV *Sym,
const SCEV *Conc,
ScalarEvolution &SE) const {
for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
- const SCEV* H =
+ const SCEV *H =
getOperand(i)->replaceSymbolicValuesWithConcrete(Sym, Conc, SE);
if (H != getOperand(i)) {
- SmallVector<const SCEV*, 8> NewOps;
+ SmallVector<const SCEV *, 8> NewOps;
NewOps.reserve(getNumOperands());
for (unsigned j = 0; j != i; ++j)
NewOps.push_back(getOperand(j));
OS << "}<" << L->getHeader()->getName() + ">";
}
-void SCEVUnknown::Profile(FoldingSetNodeID &ID) const {
- ID.AddInteger(scUnknown);
- ID.AddPointer(V);
-}
-
bool SCEVUnknown::isLoopInvariant(const Loop *L) const {
// All non-instruction values are loop invariant. All instructions are loop
// invariant if they are not contained in the specified loop.
return operator()(LC->getOperand(), RC->getOperand());
}
- assert(0 && "Unknown SCEV kind!");
+ llvm_unreachable("Unknown SCEV kind!");
return false;
}
};
/// this to depend on where the addresses of various SCEV objects happened to
/// land in memory.
///
-static void GroupByComplexity(SmallVectorImpl<const SCEV*> &Ops,
+static void GroupByComplexity(SmallVectorImpl<const SCEV *> &Ops,
LoopInfo *LI) {
if (Ops.size() < 2) return; // Noop
if (Ops.size() == 2) {
/// BinomialCoefficient - Compute BC(It, K). The result has width W.
/// Assume, K > 0.
-static const SCEV* BinomialCoefficient(const SCEV* It, unsigned K,
+static const SCEV *BinomialCoefficient(const SCEV *It, unsigned K,
ScalarEvolution &SE,
const Type* ResultTy) {
// Handle the simplest case efficiently.
// Calculate the product, at width T+W
const IntegerType *CalculationTy = IntegerType::get(CalculationBits);
- const SCEV* Dividend = SE.getTruncateOrZeroExtend(It, CalculationTy);
+ const SCEV *Dividend = SE.getTruncateOrZeroExtend(It, CalculationTy);
for (unsigned i = 1; i != K; ++i) {
- const SCEV* S = SE.getMinusSCEV(It, SE.getIntegerSCEV(i, It->getType()));
+ const SCEV *S = SE.getMinusSCEV(It, SE.getIntegerSCEV(i, It->getType()));
Dividend = SE.getMulExpr(Dividend,
SE.getTruncateOrZeroExtend(S, CalculationTy));
}
// Divide by 2^T
- const SCEV* DivResult = SE.getUDivExpr(Dividend, SE.getConstant(DivFactor));
+ const SCEV *DivResult = SE.getUDivExpr(Dividend, SE.getConstant(DivFactor));
// Truncate the result, and divide by K! / 2^T.
///
/// where BC(It, k) stands for binomial coefficient.
///
-const SCEV* SCEVAddRecExpr::evaluateAtIteration(const SCEV* It,
+const SCEV *SCEVAddRecExpr::evaluateAtIteration(const SCEV *It,
ScalarEvolution &SE) const {
- const SCEV* Result = getStart();
+ const SCEV *Result = getStart();
for (unsigned i = 1, e = getNumOperands(); i != e; ++i) {
// The computation is correct in the face of overflow provided that the
// multiplication is performed _after_ the evaluation of the binomial
// coefficient.
- const SCEV* Coeff = BinomialCoefficient(It, i, SE, getType());
+ const SCEV *Coeff = BinomialCoefficient(It, i, SE, getType());
if (isa<SCEVCouldNotCompute>(Coeff))
return Coeff;
// SCEV Expression folder implementations
//===----------------------------------------------------------------------===//
-const SCEV* ScalarEvolution::getTruncateExpr(const SCEV* Op,
- const Type *Ty) {
+const SCEV *ScalarEvolution::getTruncateExpr(const SCEV *Op,
+ const Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) > getTypeSizeInBits(Ty) &&
"This is not a truncating conversion!");
assert(isSCEVable(Ty) &&
"This is not a conversion to a SCEVable type!");
Ty = getEffectiveSCEVType(Ty);
+ FoldingSetNodeID ID;
+ ID.AddInteger(scTruncate);
+ ID.AddPointer(Op);
+ ID.AddPointer(Ty);
+ void *IP = 0;
+ if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
+
// Fold if the operand is constant.
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(Op))
return getConstant(
// If the input value is a chrec scev, truncate the chrec's operands.
if (const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(Op)) {
- SmallVector<const SCEV*, 4> Operands;
+ SmallVector<const SCEV *, 4> Operands;
for (unsigned i = 0, e = AddRec->getNumOperands(); i != e; ++i)
Operands.push_back(getTruncateExpr(AddRec->getOperand(i), Ty));
return getAddRecExpr(Operands, AddRec->getLoop());
}
- FoldingSetNodeID ID;
- ID.AddInteger(scTruncate);
- ID.AddPointer(Op);
- ID.AddPointer(Ty);
- void *IP = 0;
+ // The cast wasn't folded; create an explicit cast node.
+ // Recompute the insert position, as it may have been invalidated.
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVTruncateExpr>();
- new (S) SCEVTruncateExpr(Op, Ty);
+ new (S) SCEVTruncateExpr(ID, Op, Ty);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
-const SCEV* ScalarEvolution::getZeroExtendExpr(const SCEV* Op,
- const Type *Ty) {
+const SCEV *ScalarEvolution::getZeroExtendExpr(const SCEV *Op,
+ const Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) < getTypeSizeInBits(Ty) &&
"This is not an extending conversion!");
assert(isSCEVable(Ty) &&
if (const SCEVZeroExtendExpr *SZ = dyn_cast<SCEVZeroExtendExpr>(Op))
return getZeroExtendExpr(SZ->getOperand(), Ty);
+ // Before doing any expensive analysis, check to see if we've already
+ // computed a SCEV for this Op and Ty.
+ FoldingSetNodeID ID;
+ ID.AddInteger(scZeroExtend);
+ ID.AddPointer(Op);
+ ID.AddPointer(Ty);
+ void *IP = 0;
+ if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
+
// If the input value is a chrec scev, and we can prove that the value
// did not overflow the old, smaller, value, we can zero extend all of the
// operands (often constants). This allows analysis of something like
// this: for (unsigned char X = 0; X < 100; ++X) { int Y = X; }
if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Op))
if (AR->isAffine()) {
+ const SCEV *Start = AR->getStart();
+ const SCEV *Step = AR->getStepRecurrence(*this);
+ unsigned BitWidth = getTypeSizeInBits(AR->getType());
+ const Loop *L = AR->getLoop();
+
// Check whether the backedge-taken count is SCEVCouldNotCompute.
// Note that this serves two purposes: It filters out loops that are
// simply not analyzable, and it covers the case where this code is
// in infinite recursion. In the later case, the analysis code will
// cope with a conservative value, and it will take care to purge
// that value once it has finished.
- const SCEV* MaxBECount = getMaxBackedgeTakenCount(AR->getLoop());
+ const SCEV *MaxBECount = getMaxBackedgeTakenCount(L);
if (!isa<SCEVCouldNotCompute>(MaxBECount)) {
// Manually compute the final value for AR, checking for
// overflow.
- const SCEV* Start = AR->getStart();
- const SCEV* Step = AR->getStepRecurrence(*this);
// Check whether the backedge-taken count can be losslessly casted to
// the addrec's type. The count is always unsigned.
- const SCEV* CastedMaxBECount =
+ const SCEV *CastedMaxBECount =
getTruncateOrZeroExtend(MaxBECount, Start->getType());
- const SCEV* RecastedMaxBECount =
+ const SCEV *RecastedMaxBECount =
getTruncateOrZeroExtend(CastedMaxBECount, MaxBECount->getType());
if (MaxBECount == RecastedMaxBECount) {
- const Type *WideTy =
- IntegerType::get(getTypeSizeInBits(Start->getType()) * 2);
+ const Type *WideTy = IntegerType::get(BitWidth * 2);
// Check whether Start+Step*MaxBECount has no unsigned overflow.
- const SCEV* ZMul =
+ const SCEV *ZMul =
getMulExpr(CastedMaxBECount,
getTruncateOrZeroExtend(Step, Start->getType()));
- const SCEV* Add = getAddExpr(Start, ZMul);
- const SCEV* OperandExtendedAdd =
+ const SCEV *Add = getAddExpr(Start, ZMul);
+ const SCEV *OperandExtendedAdd =
getAddExpr(getZeroExtendExpr(Start, WideTy),
getMulExpr(getZeroExtendExpr(CastedMaxBECount, WideTy),
getZeroExtendExpr(Step, WideTy)));
// Return the expression with the addrec on the outside.
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getZeroExtendExpr(Step, Ty),
- AR->getLoop());
+ L);
// Similar to above, only this time treat the step value as signed.
// This covers loops that count down.
- const SCEV* SMul =
+ const SCEV *SMul =
getMulExpr(CastedMaxBECount,
getTruncateOrSignExtend(Step, Start->getType()));
Add = getAddExpr(Start, SMul);
// Return the expression with the addrec on the outside.
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getSignExtendExpr(Step, Ty),
- AR->getLoop());
+ L);
+ }
+
+ // If the backedge is guarded by a comparison with the pre-inc value
+ // the addrec is safe. Also, if the entry is guarded by a comparison
+ // with the start value and the backedge is guarded by a comparison
+ // with the post-inc value, the addrec is safe.
+ if (isKnownPositive(Step)) {
+ const SCEV *N = getConstant(APInt::getMinValue(BitWidth) -
+ getUnsignedRange(Step).getUnsignedMax());
+ if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_ULT, AR, N) ||
+ (isLoopGuardedByCond(L, ICmpInst::ICMP_ULT, Start, N) &&
+ isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_ULT,
+ AR->getPostIncExpr(*this), N)))
+ // Return the expression with the addrec on the outside.
+ return getAddRecExpr(getZeroExtendExpr(Start, Ty),
+ getZeroExtendExpr(Step, Ty),
+ L);
+ } else if (isKnownNegative(Step)) {
+ const SCEV *N = getConstant(APInt::getMaxValue(BitWidth) -
+ getSignedRange(Step).getSignedMin());
+ if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_UGT, AR, N) &&
+ (isLoopGuardedByCond(L, ICmpInst::ICMP_UGT, Start, N) ||
+ isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_UGT,
+ AR->getPostIncExpr(*this), N)))
+ // Return the expression with the addrec on the outside.
+ return getAddRecExpr(getZeroExtendExpr(Start, Ty),
+ getSignExtendExpr(Step, Ty),
+ L);
}
}
}
- FoldingSetNodeID ID;
- ID.AddInteger(scZeroExtend);
- ID.AddPointer(Op);
- ID.AddPointer(Ty);
- void *IP = 0;
+ // The cast wasn't folded; create an explicit cast node.
+ // Recompute the insert position, as it may have been invalidated.
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVZeroExtendExpr>();
- new (S) SCEVZeroExtendExpr(Op, Ty);
+ new (S) SCEVZeroExtendExpr(ID, Op, Ty);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
-const SCEV* ScalarEvolution::getSignExtendExpr(const SCEV* Op,
- const Type *Ty) {
+const SCEV *ScalarEvolution::getSignExtendExpr(const SCEV *Op,
+ const Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) < getTypeSizeInBits(Ty) &&
"This is not an extending conversion!");
assert(isSCEVable(Ty) &&
if (const SCEVSignExtendExpr *SS = dyn_cast<SCEVSignExtendExpr>(Op))
return getSignExtendExpr(SS->getOperand(), Ty);
+ // Before doing any expensive analysis, check to see if we've already
+ // computed a SCEV for this Op and Ty.
+ FoldingSetNodeID ID;
+ ID.AddInteger(scSignExtend);
+ ID.AddPointer(Op);
+ ID.AddPointer(Ty);
+ void *IP = 0;
+ if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
+
// If the input value is a chrec scev, and we can prove that the value
// did not overflow the old, smaller, value, we can sign extend all of the
// operands (often constants). This allows analysis of something like
// this: for (signed char X = 0; X < 100; ++X) { int Y = X; }
if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Op))
if (AR->isAffine()) {
+ const SCEV *Start = AR->getStart();
+ const SCEV *Step = AR->getStepRecurrence(*this);
+ unsigned BitWidth = getTypeSizeInBits(AR->getType());
+ const Loop *L = AR->getLoop();
+
// Check whether the backedge-taken count is SCEVCouldNotCompute.
// Note that this serves two purposes: It filters out loops that are
// simply not analyzable, and it covers the case where this code is
// in infinite recursion. In the later case, the analysis code will
// cope with a conservative value, and it will take care to purge
// that value once it has finished.
- const SCEV* MaxBECount = getMaxBackedgeTakenCount(AR->getLoop());
+ const SCEV *MaxBECount = getMaxBackedgeTakenCount(L);
if (!isa<SCEVCouldNotCompute>(MaxBECount)) {
// Manually compute the final value for AR, checking for
// overflow.
- const SCEV* Start = AR->getStart();
- const SCEV* Step = AR->getStepRecurrence(*this);
// Check whether the backedge-taken count can be losslessly casted to
// the addrec's type. The count is always unsigned.
- const SCEV* CastedMaxBECount =
+ const SCEV *CastedMaxBECount =
getTruncateOrZeroExtend(MaxBECount, Start->getType());
- const SCEV* RecastedMaxBECount =
+ const SCEV *RecastedMaxBECount =
getTruncateOrZeroExtend(CastedMaxBECount, MaxBECount->getType());
if (MaxBECount == RecastedMaxBECount) {
- const Type *WideTy =
- IntegerType::get(getTypeSizeInBits(Start->getType()) * 2);
+ const Type *WideTy = IntegerType::get(BitWidth * 2);
// Check whether Start+Step*MaxBECount has no signed overflow.
- const SCEV* SMul =
+ const SCEV *SMul =
getMulExpr(CastedMaxBECount,
getTruncateOrSignExtend(Step, Start->getType()));
- const SCEV* Add = getAddExpr(Start, SMul);
- const SCEV* OperandExtendedAdd =
+ const SCEV *Add = getAddExpr(Start, SMul);
+ const SCEV *OperandExtendedAdd =
getAddExpr(getSignExtendExpr(Start, WideTy),
getMulExpr(getZeroExtendExpr(CastedMaxBECount, WideTy),
getSignExtendExpr(Step, WideTy)));
// Return the expression with the addrec on the outside.
return getAddRecExpr(getSignExtendExpr(Start, Ty),
getSignExtendExpr(Step, Ty),
- AR->getLoop());
+ L);
+
+ // Similar to above, only this time treat the step value as unsigned.
+ // This covers loops that count up with an unsigned step.
+ const SCEV *UMul =
+ getMulExpr(CastedMaxBECount,
+ getTruncateOrZeroExtend(Step, Start->getType()));
+ Add = getAddExpr(Start, UMul);
+ OperandExtendedAdd =
+ getAddExpr(getZeroExtendExpr(Start, WideTy),
+ getMulExpr(getZeroExtendExpr(CastedMaxBECount, WideTy),
+ getZeroExtendExpr(Step, WideTy)));
+ if (getZeroExtendExpr(Add, WideTy) == OperandExtendedAdd)
+ // Return the expression with the addrec on the outside.
+ return getAddRecExpr(getSignExtendExpr(Start, Ty),
+ getZeroExtendExpr(Step, Ty),
+ L);
+ }
+
+ // If the backedge is guarded by a comparison with the pre-inc value
+ // the addrec is safe. Also, if the entry is guarded by a comparison
+ // with the start value and the backedge is guarded by a comparison
+ // with the post-inc value, the addrec is safe.
+ if (isKnownPositive(Step)) {
+ const SCEV *N = getConstant(APInt::getSignedMinValue(BitWidth) -
+ getSignedRange(Step).getSignedMax());
+ if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SLT, AR, N) ||
+ (isLoopGuardedByCond(L, ICmpInst::ICMP_SLT, Start, N) &&
+ isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SLT,
+ AR->getPostIncExpr(*this), N)))
+ // Return the expression with the addrec on the outside.
+ return getAddRecExpr(getSignExtendExpr(Start, Ty),
+ getSignExtendExpr(Step, Ty),
+ L);
+ } else if (isKnownNegative(Step)) {
+ const SCEV *N = getConstant(APInt::getSignedMaxValue(BitWidth) -
+ getSignedRange(Step).getSignedMin());
+ if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SGT, AR, N) ||
+ (isLoopGuardedByCond(L, ICmpInst::ICMP_SGT, Start, N) &&
+ isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SGT,
+ AR->getPostIncExpr(*this), N)))
+ // Return the expression with the addrec on the outside.
+ return getAddRecExpr(getSignExtendExpr(Start, Ty),
+ getSignExtendExpr(Step, Ty),
+ L);
}
}
}
- FoldingSetNodeID ID;
- ID.AddInteger(scSignExtend);
- ID.AddPointer(Op);
- ID.AddPointer(Ty);
- void *IP = 0;
+ // The cast wasn't folded; create an explicit cast node.
+ // Recompute the insert position, as it may have been invalidated.
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVSignExtendExpr>();
- new (S) SCEVSignExtendExpr(Op, Ty);
+ new (S) SCEVSignExtendExpr(ID, Op, Ty);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
/// getAnyExtendExpr - Return a SCEV for the given operand extended with
/// unspecified bits out to the given type.
///
-const SCEV* ScalarEvolution::getAnyExtendExpr(const SCEV* Op,
+const SCEV *ScalarEvolution::getAnyExtendExpr(const SCEV *Op,
const Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) < getTypeSizeInBits(Ty) &&
"This is not an extending conversion!");
// Peel off a truncate cast.
if (const SCEVTruncateExpr *T = dyn_cast<SCEVTruncateExpr>(Op)) {
- const SCEV* NewOp = T->getOperand();
+ const SCEV *NewOp = T->getOperand();
if (getTypeSizeInBits(NewOp->getType()) < getTypeSizeInBits(Ty))
return getAnyExtendExpr(NewOp, Ty);
return getTruncateOrNoop(NewOp, Ty);
}
// Next try a zext cast. If the cast is folded, use it.
- const SCEV* ZExt = getZeroExtendExpr(Op, Ty);
+ const SCEV *ZExt = getZeroExtendExpr(Op, Ty);
if (!isa<SCEVZeroExtendExpr>(ZExt))
return ZExt;
// Next try a sext cast. If the cast is folded, use it.
- const SCEV* SExt = getSignExtendExpr(Op, Ty);
+ const SCEV *SExt = getSignExtendExpr(Op, Ty);
if (!isa<SCEVSignExtendExpr>(SExt))
return SExt;
/// is also used as a check to avoid infinite recursion.
///
static bool
-CollectAddOperandsWithScales(DenseMap<const SCEV*, APInt> &M,
- SmallVector<const SCEV*, 8> &NewOps,
+CollectAddOperandsWithScales(DenseMap<const SCEV *, APInt> &M,
+ SmallVector<const SCEV *, 8> &NewOps,
APInt &AccumulatedConstant,
- const SmallVectorImpl<const SCEV*> &Ops,
+ const SmallVectorImpl<const SCEV *> &Ops,
const APInt &Scale,
ScalarEvolution &SE) {
bool Interesting = false;
} else {
// A multiplication of a constant with some other value. Update
// the map.
- SmallVector<const SCEV*, 4> MulOps(Mul->op_begin()+1, Mul->op_end());
- const SCEV* Key = SE.getMulExpr(MulOps);
- std::pair<DenseMap<const SCEV*, APInt>::iterator, bool> Pair =
+ SmallVector<const SCEV *, 4> MulOps(Mul->op_begin()+1, Mul->op_end());
+ const SCEV *Key = SE.getMulExpr(MulOps);
+ std::pair<DenseMap<const SCEV *, APInt>::iterator, bool> Pair =
M.insert(std::make_pair(Key, NewScale));
if (Pair.second) {
NewOps.push_back(Pair.first->first);
AccumulatedConstant += Scale * C->getValue()->getValue();
} else {
// An ordinary operand. Update the map.
- std::pair<DenseMap<const SCEV*, APInt>::iterator, bool> Pair =
+ std::pair<DenseMap<const SCEV *, APInt>::iterator, bool> Pair =
M.insert(std::make_pair(Ops[i], Scale));
if (Pair.second) {
NewOps.push_back(Pair.first->first);
/// getAddExpr - Get a canonical add expression, or something simpler if
/// possible.
-const SCEV* ScalarEvolution::getAddExpr(SmallVectorImpl<const SCEV*> &Ops) {
+const SCEV *ScalarEvolution::getAddExpr(SmallVectorImpl<const SCEV *> &Ops) {
assert(!Ops.empty() && "Cannot get empty add!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
if (Ops[i] == Ops[i+1]) { // X + Y + Y --> X + Y*2
// Found a match, merge the two values into a multiply, and add any
// remaining values to the result.
- const SCEV* Two = getIntegerSCEV(2, Ty);
- const SCEV* Mul = getMulExpr(Ops[i], Two);
+ const SCEV *Two = getIntegerSCEV(2, Ty);
+ const SCEV *Mul = getMulExpr(Ops[i], Two);
if (Ops.size() == 2)
return Mul;
Ops.erase(Ops.begin()+i, Ops.begin()+i+2);
const SCEVTruncateExpr *Trunc = cast<SCEVTruncateExpr>(Ops[Idx]);
const Type *DstType = Trunc->getType();
const Type *SrcType = Trunc->getOperand()->getType();
- SmallVector<const SCEV*, 8> LargeOps;
+ SmallVector<const SCEV *, 8> LargeOps;
bool Ok = true;
// Check all the operands to see if they can be represented in the
// source type of the truncate.
// is much more likely to be foldable here.
LargeOps.push_back(getSignExtendExpr(C, SrcType));
} else if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(Ops[i])) {
- SmallVector<const SCEV*, 8> LargeMulOps;
+ SmallVector<const SCEV *, 8> LargeMulOps;
for (unsigned j = 0, f = M->getNumOperands(); j != f && Ok; ++j) {
if (const SCEVTruncateExpr *T =
dyn_cast<SCEVTruncateExpr>(M->getOperand(j))) {
}
if (Ok) {
// Evaluate the expression in the larger type.
- const SCEV* Fold = getAddExpr(LargeOps);
+ const SCEV *Fold = getAddExpr(LargeOps);
// If it folds to something simple, use it. Otherwise, don't.
if (isa<SCEVConstant>(Fold) || isa<SCEVUnknown>(Fold))
return getTruncateExpr(Fold, DstType);
// operands multiplied by constant values.
if (Idx < Ops.size() && isa<SCEVMulExpr>(Ops[Idx])) {
uint64_t BitWidth = getTypeSizeInBits(Ty);
- DenseMap<const SCEV*, APInt> M;
- SmallVector<const SCEV*, 8> NewOps;
+ DenseMap<const SCEV *, APInt> M;
+ SmallVector<const SCEV *, 8> NewOps;
APInt AccumulatedConstant(BitWidth, 0);
if (CollectAddOperandsWithScales(M, NewOps, AccumulatedConstant,
Ops, APInt(BitWidth, 1), *this)) {
// Some interesting folding opportunity is present, so its worthwhile to
// re-generate the operands list. Group the operands by constant scale,
// to avoid multiplying by the same constant scale multiple times.
- std::map<APInt, SmallVector<const SCEV*, 4>, APIntCompare> MulOpLists;
- for (SmallVector<const SCEV*, 8>::iterator I = NewOps.begin(),
+ std::map<APInt, SmallVector<const SCEV *, 4>, APIntCompare> MulOpLists;
+ for (SmallVector<const SCEV *, 8>::iterator I = NewOps.begin(),
E = NewOps.end(); I != E; ++I)
MulOpLists[M.find(*I)->second].push_back(*I);
// Re-generate the operands list.
for (unsigned AddOp = 0, e = Ops.size(); AddOp != e; ++AddOp)
if (MulOpSCEV == Ops[AddOp] && !isa<SCEVConstant>(Ops[AddOp])) {
// Fold W + X + (X * Y * Z) --> W + (X * ((Y*Z)+1))
- const SCEV* InnerMul = Mul->getOperand(MulOp == 0);
+ const SCEV *InnerMul = Mul->getOperand(MulOp == 0);
if (Mul->getNumOperands() != 2) {
// If the multiply has more than two operands, we must get the
// Y*Z term.
- SmallVector<const SCEV*, 4> MulOps(Mul->op_begin(), Mul->op_end());
+ SmallVector<const SCEV *, 4> MulOps(Mul->op_begin(), Mul->op_end());
MulOps.erase(MulOps.begin()+MulOp);
InnerMul = getMulExpr(MulOps);
}
- const SCEV* One = getIntegerSCEV(1, Ty);
- const SCEV* AddOne = getAddExpr(InnerMul, One);
- const SCEV* OuterMul = getMulExpr(AddOne, Ops[AddOp]);
+ const SCEV *One = getIntegerSCEV(1, Ty);
+ const SCEV *AddOne = getAddExpr(InnerMul, One);
+ const SCEV *OuterMul = getMulExpr(AddOne, Ops[AddOp]);
if (Ops.size() == 2) return OuterMul;
if (AddOp < Idx) {
Ops.erase(Ops.begin()+AddOp);
OMulOp != e; ++OMulOp)
if (OtherMul->getOperand(OMulOp) == MulOpSCEV) {
// Fold X + (A*B*C) + (A*D*E) --> X + (A*(B*C+D*E))
- const SCEV* InnerMul1 = Mul->getOperand(MulOp == 0);
+ const SCEV *InnerMul1 = Mul->getOperand(MulOp == 0);
if (Mul->getNumOperands() != 2) {
SmallVector<const SCEV *, 4> MulOps(Mul->op_begin(),
Mul->op_end());
MulOps.erase(MulOps.begin()+MulOp);
InnerMul1 = getMulExpr(MulOps);
}
- const SCEV* InnerMul2 = OtherMul->getOperand(OMulOp == 0);
+ const SCEV *InnerMul2 = OtherMul->getOperand(OMulOp == 0);
if (OtherMul->getNumOperands() != 2) {
SmallVector<const SCEV *, 4> MulOps(OtherMul->op_begin(),
OtherMul->op_end());
MulOps.erase(MulOps.begin()+OMulOp);
InnerMul2 = getMulExpr(MulOps);
}
- const SCEV* InnerMulSum = getAddExpr(InnerMul1,InnerMul2);
- const SCEV* OuterMul = getMulExpr(MulOpSCEV, InnerMulSum);
+ const SCEV *InnerMulSum = getAddExpr(InnerMul1,InnerMul2);
+ const SCEV *OuterMul = getMulExpr(MulOpSCEV, InnerMulSum);
if (Ops.size() == 2) return OuterMul;
Ops.erase(Ops.begin()+Idx);
Ops.erase(Ops.begin()+OtherMulIdx-1);
for (; Idx < Ops.size() && isa<SCEVAddRecExpr>(Ops[Idx]); ++Idx) {
// Scan all of the other operands to this add and add them to the vector if
// they are loop invariant w.r.t. the recurrence.
- SmallVector<const SCEV*, 8> LIOps;
+ SmallVector<const SCEV *, 8> LIOps;
const SCEVAddRecExpr *AddRec = cast<SCEVAddRecExpr>(Ops[Idx]);
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
if (Ops[i]->isLoopInvariant(AddRec->getLoop())) {
// NLI + LI + {Start,+,Step} --> NLI + {LI+Start,+,Step}
LIOps.push_back(AddRec->getStart());
- SmallVector<const SCEV*, 4> AddRecOps(AddRec->op_begin(),
+ SmallVector<const SCEV *, 4> AddRecOps(AddRec->op_begin(),
AddRec->op_end());
AddRecOps[0] = getAddExpr(LIOps);
- const SCEV* NewRec = getAddRecExpr(AddRecOps, AddRec->getLoop());
+ const SCEV *NewRec = getAddRecExpr(AddRecOps, AddRec->getLoop());
// If all of the other operands were loop invariant, we are done.
if (Ops.size() == 1) return NewRec;
}
NewOps[i] = getAddExpr(NewOps[i], OtherAddRec->getOperand(i));
}
- const SCEV* NewAddRec = getAddRecExpr(NewOps, AddRec->getLoop());
+ const SCEV *NewAddRec = getAddRecExpr(NewOps, AddRec->getLoop());
if (Ops.size() == 2) return NewAddRec;
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVAddExpr>();
- new (S) SCEVAddExpr(Ops);
+ new (S) SCEVAddExpr(ID, Ops);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
/// getMulExpr - Get a canonical multiply expression, or something simpler if
/// possible.
-const SCEV* ScalarEvolution::getMulExpr(SmallVectorImpl<const SCEV*> &Ops) {
+const SCEV *ScalarEvolution::getMulExpr(SmallVectorImpl<const SCEV *> &Ops) {
assert(!Ops.empty() && "Cannot get empty mul!");
#ifndef NDEBUG
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
++Idx;
while (const SCEVConstant *RHSC = dyn_cast<SCEVConstant>(Ops[Idx])) {
// We found two constants, fold them together!
- ConstantInt *Fold = ConstantInt::get(LHSC->getValue()->getValue() *
+ ConstantInt *Fold = Context->getConstantInt(LHSC->getValue()->getValue() *
RHSC->getValue()->getValue());
Ops[0] = getConstant(Fold);
Ops.erase(Ops.begin()+1); // Erase the folded element
for (; Idx < Ops.size() && isa<SCEVAddRecExpr>(Ops[Idx]); ++Idx) {
// Scan all of the other operands to this mul and add them to the vector if
// they are loop invariant w.r.t. the recurrence.
- SmallVector<const SCEV*, 8> LIOps;
+ SmallVector<const SCEV *, 8> LIOps;
const SCEVAddRecExpr *AddRec = cast<SCEVAddRecExpr>(Ops[Idx]);
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
if (Ops[i]->isLoopInvariant(AddRec->getLoop())) {
// If we found some loop invariants, fold them into the recurrence.
if (!LIOps.empty()) {
// NLI * LI * {Start,+,Step} --> NLI * {LI*Start,+,LI*Step}
- SmallVector<const SCEV*, 4> NewOps;
+ SmallVector<const SCEV *, 4> NewOps;
NewOps.reserve(AddRec->getNumOperands());
if (LIOps.size() == 1) {
const SCEV *Scale = LIOps[0];
NewOps.push_back(getMulExpr(Scale, AddRec->getOperand(i)));
} else {
for (unsigned i = 0, e = AddRec->getNumOperands(); i != e; ++i) {
- SmallVector<const SCEV*, 4> MulOps(LIOps.begin(), LIOps.end());
+ SmallVector<const SCEV *, 4> MulOps(LIOps.begin(), LIOps.end());
MulOps.push_back(AddRec->getOperand(i));
NewOps.push_back(getMulExpr(MulOps));
}
}
- const SCEV* NewRec = getAddRecExpr(NewOps, AddRec->getLoop());
+ const SCEV *NewRec = getAddRecExpr(NewOps, AddRec->getLoop());
// If all of the other operands were loop invariant, we are done.
if (Ops.size() == 1) return NewRec;
if (AddRec->getLoop() == OtherAddRec->getLoop()) {
// F * G --> {A,+,B} * {C,+,D} --> {A*C,+,F*D + G*B + B*D}
const SCEVAddRecExpr *F = AddRec, *G = OtherAddRec;
- const SCEV* NewStart = getMulExpr(F->getStart(),
+ const SCEV *NewStart = getMulExpr(F->getStart(),
G->getStart());
- const SCEV* B = F->getStepRecurrence(*this);
- const SCEV* D = G->getStepRecurrence(*this);
- const SCEV* NewStep = getAddExpr(getMulExpr(F, D),
+ const SCEV *B = F->getStepRecurrence(*this);
+ const SCEV *D = G->getStepRecurrence(*this);
+ const SCEV *NewStep = getAddExpr(getMulExpr(F, D),
getMulExpr(G, B),
getMulExpr(B, D));
- const SCEV* NewAddRec = getAddRecExpr(NewStart, NewStep,
+ const SCEV *NewAddRec = getAddRecExpr(NewStart, NewStep,
F->getLoop());
if (Ops.size() == 2) return NewAddRec;
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVMulExpr>();
- new (S) SCEVMulExpr(Ops);
+ new (S) SCEVMulExpr(ID, Ops);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
getAddRecExpr(getZeroExtendExpr(AR->getStart(), ExtTy),
getZeroExtendExpr(Step, ExtTy),
AR->getLoop())) {
- SmallVector<const SCEV*, 4> Operands;
+ SmallVector<const SCEV *, 4> Operands;
for (unsigned i = 0, e = AR->getNumOperands(); i != e; ++i)
Operands.push_back(getUDivExpr(AR->getOperand(i), RHS));
return getAddRecExpr(Operands, AR->getLoop());
}
// (A*B)/C --> A*(B/C) if safe and B/C can be folded.
if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(LHS)) {
- SmallVector<const SCEV*, 4> Operands;
+ SmallVector<const SCEV *, 4> Operands;
for (unsigned i = 0, e = M->getNumOperands(); i != e; ++i)
Operands.push_back(getZeroExtendExpr(M->getOperand(i), ExtTy));
if (getZeroExtendExpr(M, ExtTy) == getMulExpr(Operands))
// Find an operand that's safely divisible.
for (unsigned i = 0, e = M->getNumOperands(); i != e; ++i) {
- const SCEV* Op = M->getOperand(i);
- const SCEV* Div = getUDivExpr(Op, RHSC);
+ const SCEV *Op = M->getOperand(i);
+ const SCEV *Div = getUDivExpr(Op, RHSC);
if (!isa<SCEVUDivExpr>(Div) && getMulExpr(Div, RHSC) == Op) {
- const SmallVectorImpl<const SCEV*> &MOperands = M->getOperands();
- Operands = SmallVector<const SCEV*, 4>(MOperands.begin(),
+ const SmallVectorImpl<const SCEV *> &MOperands = M->getOperands();
+ Operands = SmallVector<const SCEV *, 4>(MOperands.begin(),
MOperands.end());
Operands[i] = Div;
return getMulExpr(Operands);
}
// (A+B)/C --> (A/C + B/C) if safe and A/C and B/C can be folded.
if (const SCEVAddRecExpr *A = dyn_cast<SCEVAddRecExpr>(LHS)) {
- SmallVector<const SCEV*, 4> Operands;
+ SmallVector<const SCEV *, 4> Operands;
for (unsigned i = 0, e = A->getNumOperands(); i != e; ++i)
Operands.push_back(getZeroExtendExpr(A->getOperand(i), ExtTy));
if (getZeroExtendExpr(A, ExtTy) == getAddExpr(Operands)) {
Operands.clear();
for (unsigned i = 0, e = A->getNumOperands(); i != e; ++i) {
- const SCEV* Op = getUDivExpr(A->getOperand(i), RHS);
+ const SCEV *Op = getUDivExpr(A->getOperand(i), RHS);
if (isa<SCEVUDivExpr>(Op) || getMulExpr(Op, RHS) != A->getOperand(i))
break;
Operands.push_back(Op);
if (const SCEVConstant *LHSC = dyn_cast<SCEVConstant>(LHS)) {
Constant *LHSCV = LHSC->getValue();
Constant *RHSCV = RHSC->getValue();
- return getConstant(cast<ConstantInt>(ConstantExpr::getUDiv(LHSCV,
+ return getConstant(cast<ConstantInt>(Context->getConstantExprUDiv(LHSCV,
RHSCV)));
}
}
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVUDivExpr>();
- new (S) SCEVUDivExpr(LHS, RHS);
+ new (S) SCEVUDivExpr(ID, LHS, RHS);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
/// getAddRecExpr - Get an add recurrence expression for the specified loop.
/// Simplify the expression as much as possible.
-const SCEV* ScalarEvolution::getAddRecExpr(const SCEV* Start,
- const SCEV* Step, const Loop *L) {
- SmallVector<const SCEV*, 4> Operands;
+const SCEV *ScalarEvolution::getAddRecExpr(const SCEV *Start,
+ const SCEV *Step, const Loop *L) {
+ SmallVector<const SCEV *, 4> Operands;
Operands.push_back(Start);
if (const SCEVAddRecExpr *StepChrec = dyn_cast<SCEVAddRecExpr>(Step))
if (StepChrec->getLoop() == L) {
/// getAddRecExpr - Get an add recurrence expression for the specified loop.
/// Simplify the expression as much as possible.
const SCEV *
-ScalarEvolution::getAddRecExpr(SmallVectorImpl<const SCEV*> &Operands,
+ScalarEvolution::getAddRecExpr(SmallVectorImpl<const SCEV *> &Operands,
const Loop *L) {
if (Operands.size() == 1) return Operands[0];
#ifndef NDEBUG
if (const SCEVAddRecExpr *NestedAR = dyn_cast<SCEVAddRecExpr>(Operands[0])) {
const Loop* NestedLoop = NestedAR->getLoop();
if (L->getLoopDepth() < NestedLoop->getLoopDepth()) {
- SmallVector<const SCEV*, 4> NestedOperands(NestedAR->op_begin(),
+ SmallVector<const SCEV *, 4> NestedOperands(NestedAR->op_begin(),
NestedAR->op_end());
Operands[0] = NestedAR->getStart();
// AddRecs require their operands be loop-invariant with respect to their
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVAddRecExpr>();
- new (S) SCEVAddRecExpr(Operands, L);
+ new (S) SCEVAddRecExpr(ID, Operands, L);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
const SCEV *ScalarEvolution::getSMaxExpr(const SCEV *LHS,
const SCEV *RHS) {
- SmallVector<const SCEV*, 2> Ops;
+ SmallVector<const SCEV *, 2> Ops;
Ops.push_back(LHS);
Ops.push_back(RHS);
return getSMaxExpr(Ops);
}
-const SCEV*
-ScalarEvolution::getSMaxExpr(SmallVectorImpl<const SCEV*> &Ops) {
+const SCEV *
+ScalarEvolution::getSMaxExpr(SmallVectorImpl<const SCEV *> &Ops) {
assert(!Ops.empty() && "Cannot get empty smax!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
assert(Idx < Ops.size());
while (const SCEVConstant *RHSC = dyn_cast<SCEVConstant>(Ops[Idx])) {
// We found two constants, fold them together!
- ConstantInt *Fold = ConstantInt::get(
+ ConstantInt *Fold = Context->getConstantInt(
APIntOps::smax(LHSC->getValue()->getValue(),
RHSC->getValue()->getValue()));
Ops[0] = getConstant(Fold);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVSMaxExpr>();
- new (S) SCEVSMaxExpr(Ops);
+ new (S) SCEVSMaxExpr(ID, Ops);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
const SCEV *ScalarEvolution::getUMaxExpr(const SCEV *LHS,
const SCEV *RHS) {
- SmallVector<const SCEV*, 2> Ops;
+ SmallVector<const SCEV *, 2> Ops;
Ops.push_back(LHS);
Ops.push_back(RHS);
return getUMaxExpr(Ops);
}
-const SCEV*
-ScalarEvolution::getUMaxExpr(SmallVectorImpl<const SCEV*> &Ops) {
+const SCEV *
+ScalarEvolution::getUMaxExpr(SmallVectorImpl<const SCEV *> &Ops) {
assert(!Ops.empty() && "Cannot get empty umax!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
assert(Idx < Ops.size());
while (const SCEVConstant *RHSC = dyn_cast<SCEVConstant>(Ops[Idx])) {
// We found two constants, fold them together!
- ConstantInt *Fold = ConstantInt::get(
+ ConstantInt *Fold = Context->getConstantInt(
APIntOps::umax(LHSC->getValue()->getValue(),
RHSC->getValue()->getValue()));
Ops[0] = getConstant(Fold);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVUMaxExpr>();
- new (S) SCEVUMaxExpr(Ops);
+ new (S) SCEVUMaxExpr(ID, Ops);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
return getNotSCEV(getUMaxExpr(getNotSCEV(LHS), getNotSCEV(RHS)));
}
-const SCEV* ScalarEvolution::getUnknown(Value *V) {
+const SCEV *ScalarEvolution::getUnknown(Value *V) {
// Don't attempt to do anything other than create a SCEVUnknown object
// here. createSCEV only calls getUnknown after checking for all other
// interesting possibilities, and any other code that calls getUnknown
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVUnknown>();
- new (S) SCEVUnknown(V);
+ new (S) SCEVUnknown(ID, V);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
return TD->getIntPtrType();
}
-const SCEV* ScalarEvolution::getCouldNotCompute() {
+const SCEV *ScalarEvolution::getCouldNotCompute() {
return &CouldNotCompute;
}
-/// hasSCEV - Return true if the SCEV for this value has already been
-/// computed.
-bool ScalarEvolution::hasSCEV(Value *V) const {
- return Scalars.count(V);
-}
-
/// getSCEV - Return an existing SCEV if it exists, otherwise analyze the
/// expression and create a new one.
-const SCEV* ScalarEvolution::getSCEV(Value *V) {
+const SCEV *ScalarEvolution::getSCEV(Value *V) {
assert(isSCEVable(V->getType()) && "Value is not SCEVable!");
- std::map<SCEVCallbackVH, const SCEV*>::iterator I = Scalars.find(V);
+ std::map<SCEVCallbackVH, const SCEV *>::iterator I = Scalars.find(V);
if (I != Scalars.end()) return I->second;
- const SCEV* S = createSCEV(V);
+ const SCEV *S = createSCEV(V);
Scalars.insert(std::make_pair(SCEVCallbackVH(V, this), S));
return S;
}
/// getIntegerSCEV - Given a SCEVable type, create a constant for the
/// specified signed integer value and return a SCEV for the constant.
-const SCEV* ScalarEvolution::getIntegerSCEV(int Val, const Type *Ty) {
+const SCEV *ScalarEvolution::getIntegerSCEV(int Val, const Type *Ty) {
const IntegerType *ITy = cast<IntegerType>(getEffectiveSCEVType(Ty));
- return getConstant(ConstantInt::get(ITy, Val));
+ return getConstant(Context->getConstantInt(ITy, Val));
}
/// getNegativeSCEV - Return a SCEV corresponding to -V = -1*V
///
-const SCEV* ScalarEvolution::getNegativeSCEV(const SCEV* V) {
+const SCEV *ScalarEvolution::getNegativeSCEV(const SCEV *V) {
if (const SCEVConstant *VC = dyn_cast<SCEVConstant>(V))
- return getConstant(cast<ConstantInt>(ConstantExpr::getNeg(VC->getValue())));
+ return getConstant(
+ cast<ConstantInt>(Context->getConstantExprNeg(VC->getValue())));
const Type *Ty = V->getType();
Ty = getEffectiveSCEVType(Ty);
- return getMulExpr(V, getConstant(ConstantInt::getAllOnesValue(Ty)));
+ return getMulExpr(V,
+ getConstant(cast<ConstantInt>(Context->getAllOnesValue(Ty))));
}
/// getNotSCEV - Return a SCEV corresponding to ~V = -1-V
-const SCEV* ScalarEvolution::getNotSCEV(const SCEV* V) {
+const SCEV *ScalarEvolution::getNotSCEV(const SCEV *V) {
if (const SCEVConstant *VC = dyn_cast<SCEVConstant>(V))
- return getConstant(cast<ConstantInt>(ConstantExpr::getNot(VC->getValue())));
+ return getConstant(
+ cast<ConstantInt>(Context->getConstantExprNot(VC->getValue())));
const Type *Ty = V->getType();
Ty = getEffectiveSCEVType(Ty);
- const SCEV* AllOnes = getConstant(ConstantInt::getAllOnesValue(Ty));
+ const SCEV *AllOnes =
+ getConstant(cast<ConstantInt>(Context->getAllOnesValue(Ty)));
return getMinusSCEV(AllOnes, V);
}
/// getTruncateOrZeroExtend - Return a SCEV corresponding to a conversion of the
/// input value to the specified type. If the type must be extended, it is zero
/// extended.
-const SCEV*
-ScalarEvolution::getTruncateOrZeroExtend(const SCEV* V,
+const SCEV *
+ScalarEvolution::getTruncateOrZeroExtend(const SCEV *V,
const Type *Ty) {
const Type *SrcTy = V->getType();
assert((SrcTy->isInteger() || (TD && isa<PointerType>(SrcTy))) &&
/// getTruncateOrSignExtend - Return a SCEV corresponding to a conversion of the
/// input value to the specified type. If the type must be extended, it is sign
/// extended.
-const SCEV*
-ScalarEvolution::getTruncateOrSignExtend(const SCEV* V,
+const SCEV *
+ScalarEvolution::getTruncateOrSignExtend(const SCEV *V,
const Type *Ty) {
const Type *SrcTy = V->getType();
assert((SrcTy->isInteger() || (TD && isa<PointerType>(SrcTy))) &&
/// getNoopOrZeroExtend - Return a SCEV corresponding to a conversion of the
/// input value to the specified type. If the type must be extended, it is zero
/// extended. The conversion must not be narrowing.
-const SCEV*
-ScalarEvolution::getNoopOrZeroExtend(const SCEV* V, const Type *Ty) {
+const SCEV *
+ScalarEvolution::getNoopOrZeroExtend(const SCEV *V, const Type *Ty) {
const Type *SrcTy = V->getType();
assert((SrcTy->isInteger() || (TD && isa<PointerType>(SrcTy))) &&
(Ty->isInteger() || (TD && isa<PointerType>(Ty))) &&
/// getNoopOrSignExtend - Return a SCEV corresponding to a conversion of the
/// input value to the specified type. If the type must be extended, it is sign
/// extended. The conversion must not be narrowing.
-const SCEV*
-ScalarEvolution::getNoopOrSignExtend(const SCEV* V, const Type *Ty) {
+const SCEV *
+ScalarEvolution::getNoopOrSignExtend(const SCEV *V, const Type *Ty) {
const Type *SrcTy = V->getType();
assert((SrcTy->isInteger() || (TD && isa<PointerType>(SrcTy))) &&
(Ty->isInteger() || (TD && isa<PointerType>(Ty))) &&
/// the input value to the specified type. If the type must be extended,
/// it is extended with unspecified bits. The conversion must not be
/// narrowing.
-const SCEV*
-ScalarEvolution::getNoopOrAnyExtend(const SCEV* V, const Type *Ty) {
+const SCEV *
+ScalarEvolution::getNoopOrAnyExtend(const SCEV *V, const Type *Ty) {
const Type *SrcTy = V->getType();
assert((SrcTy->isInteger() || (TD && isa<PointerType>(SrcTy))) &&
(Ty->isInteger() || (TD && isa<PointerType>(Ty))) &&
/// getTruncateOrNoop - Return a SCEV corresponding to a conversion of the
/// input value to the specified type. The conversion must not be widening.
-const SCEV*
-ScalarEvolution::getTruncateOrNoop(const SCEV* V, const Type *Ty) {
+const SCEV *
+ScalarEvolution::getTruncateOrNoop(const SCEV *V, const Type *Ty) {
const Type *SrcTy = V->getType();
assert((SrcTy->isInteger() || (TD && isa<PointerType>(SrcTy))) &&
(Ty->isInteger() || (TD && isa<PointerType>(Ty))) &&
/// with them.
const SCEV *ScalarEvolution::getUMaxFromMismatchedTypes(const SCEV *LHS,
const SCEV *RHS) {
- const SCEV* PromotedLHS = LHS;
- const SCEV* PromotedRHS = RHS;
+ const SCEV *PromotedLHS = LHS;
+ const SCEV *PromotedRHS = RHS;
if (getTypeSizeInBits(LHS->getType()) > getTypeSizeInBits(RHS->getType()))
PromotedRHS = getZeroExtendExpr(RHS, LHS->getType());
/// with them.
const SCEV *ScalarEvolution::getUMinFromMismatchedTypes(const SCEV *LHS,
const SCEV *RHS) {
- const SCEV* PromotedLHS = LHS;
- const SCEV* PromotedRHS = RHS;
+ const SCEV *PromotedLHS = LHS;
+ const SCEV *PromotedRHS = RHS;
if (getTypeSizeInBits(LHS->getType()) > getTypeSizeInBits(RHS->getType()))
PromotedRHS = getZeroExtendExpr(RHS, LHS->getType());
ScalarEvolution::ReplaceSymbolicValueWithConcrete(Instruction *I,
const SCEV *SymName,
const SCEV *NewVal) {
- std::map<SCEVCallbackVH, const SCEV*>::iterator SI =
+ std::map<SCEVCallbackVH, const SCEV *>::iterator SI =
Scalars.find(SCEVCallbackVH(I, this));
if (SI == Scalars.end()) return;
- const SCEV* NV =
+ const SCEV *NV =
SI->second->replaceSymbolicValuesWithConcrete(SymName, NewVal, *this);
if (NV == SI->second) return; // No change.
/// createNodeForPHI - PHI nodes have two cases. Either the PHI node exists in
/// a loop header, making it a potential recurrence, or it doesn't.
///
-const SCEV* ScalarEvolution::createNodeForPHI(PHINode *PN) {
+const SCEV *ScalarEvolution::createNodeForPHI(PHINode *PN) {
if (PN->getNumIncomingValues() == 2) // The loops have been canonicalized.
if (const Loop *L = LI->getLoopFor(PN->getParent()))
if (L->getHeader() == PN->getParent()) {
unsigned BackEdge = IncomingEdge^1;
// While we are analyzing this PHI node, handle its value symbolically.
- const SCEV* SymbolicName = getUnknown(PN);
+ const SCEV *SymbolicName = getUnknown(PN);
assert(Scalars.find(PN) == Scalars.end() &&
"PHI node already processed?");
Scalars.insert(std::make_pair(SCEVCallbackVH(PN, this), SymbolicName));
// Using this symbolic name for the PHI, analyze the value coming around
// the back-edge.
- const SCEV* BEValue = getSCEV(PN->getIncomingValue(BackEdge));
+ const SCEV *BEValue = getSCEV(PN->getIncomingValue(BackEdge));
// NOTE: If BEValue is loop invariant, we know that the PHI node just
// has a special value for the first iteration of the loop.
if (FoundIndex != Add->getNumOperands()) {
// Create an add with everything but the specified operand.
- SmallVector<const SCEV*, 8> Ops;
+ SmallVector<const SCEV *, 8> Ops;
for (unsigned i = 0, e = Add->getNumOperands(); i != e; ++i)
if (i != FoundIndex)
Ops.push_back(Add->getOperand(i));
- const SCEV* Accum = getAddExpr(Ops);
+ const SCEV *Accum = getAddExpr(Ops);
// This is not a valid addrec if the step amount is varying each
// loop iteration, but is not itself an addrec in this loop.
// Because the other in-value of i (0) fits the evolution of BEValue
// i really is an addrec evolution.
if (AddRec->getLoop() == L && AddRec->isAffine()) {
- const SCEV* StartVal = getSCEV(PN->getIncomingValue(IncomingEdge));
+ const SCEV *StartVal = getSCEV(PN->getIncomingValue(IncomingEdge));
// If StartVal = j.start - j.stride, we can use StartVal as the
// initial step of the addrec evolution.
if (StartVal == getMinusSCEV(AddRec->getOperand(0),
AddRec->getOperand(1))) {
- const SCEV* PHISCEV =
+ const SCEV *PHISCEV =
getAddRecExpr(StartVal, AddRec->getOperand(1), L);
// Okay, for the entire analysis of this edge we assumed the PHI
return SymbolicName;
}
+ // It's tempting to recognize PHIs with a unique incoming value, however
+ // this leads passes like indvars to break LCSSA form. Fortunately, such
+ // PHIs are rare, as instcombine zaps them.
+
// If it's not a loop phi, we can't handle it yet.
return getUnknown(PN);
}
/// createNodeForGEP - Expand GEP instructions into add and multiply
/// operations. This allows them to be analyzed by regular SCEV code.
///
-const SCEV* ScalarEvolution::createNodeForGEP(User *GEP) {
+const SCEV *ScalarEvolution::createNodeForGEP(Operator *GEP) {
const Type *IntPtrTy = TD->getIntPtrType();
Value *Base = GEP->getOperand(0);
// Don't attempt to analyze GEPs over unsized objects.
if (!cast<PointerType>(Base->getType())->getElementType()->isSized())
return getUnknown(GEP);
- const SCEV* TotalOffset = getIntegerSCEV(0, IntPtrTy);
+ const SCEV *TotalOffset = getIntegerSCEV(0, IntPtrTy);
gep_type_iterator GTI = gep_type_begin(GEP);
for (GetElementPtrInst::op_iterator I = next(GEP->op_begin()),
E = GEP->op_end();
const StructLayout &SL = *TD->getStructLayout(STy);
unsigned FieldNo = cast<ConstantInt>(Index)->getZExtValue();
uint64_t Offset = SL.getElementOffset(FieldNo);
- TotalOffset = getAddExpr(TotalOffset,
- getIntegerSCEV(Offset, IntPtrTy));
+ TotalOffset = getAddExpr(TotalOffset, getIntegerSCEV(Offset, IntPtrTy));
} else {
// For an array, add the element offset, explicitly scaled.
- const SCEV* LocalOffset = getSCEV(Index);
+ const SCEV *LocalOffset = getSCEV(Index);
if (!isa<PointerType>(LocalOffset->getType()))
// Getelementptr indicies are signed.
- LocalOffset = getTruncateOrSignExtend(LocalOffset,
- IntPtrTy);
+ LocalOffset = getTruncateOrSignExtend(LocalOffset, IntPtrTy);
LocalOffset =
getMulExpr(LocalOffset,
- getIntegerSCEV(TD->getTypeAllocSize(*GTI),
- IntPtrTy));
+ getIntegerSCEV(TD->getTypeAllocSize(*GTI), IntPtrTy));
TotalOffset = getAddExpr(TotalOffset, LocalOffset);
}
}
/// the minimum number of times S is divisible by 2. For example, given {4,+,8}
/// it returns 2. If S is guaranteed to be 0, it returns the bitwidth of S.
uint32_t
-ScalarEvolution::GetMinTrailingZeros(const SCEV* S) {
+ScalarEvolution::GetMinTrailingZeros(const SCEV *S) {
if (const SCEVConstant *C = dyn_cast<SCEVConstant>(S))
return C->getValue()->getValue().countTrailingZeros();
return 0;
}
-uint32_t
-ScalarEvolution::GetMinLeadingZeros(const SCEV* S) {
- // TODO: Handle other SCEV expression types here.
+/// getUnsignedRange - Determine the unsigned range for a particular SCEV.
+///
+ConstantRange
+ScalarEvolution::getUnsignedRange(const SCEV *S) {
if (const SCEVConstant *C = dyn_cast<SCEVConstant>(S))
- return C->getValue()->getValue().countLeadingZeros();
+ return ConstantRange(C->getValue()->getValue());
+
+ if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
+ ConstantRange X = getUnsignedRange(Add->getOperand(0));
+ for (unsigned i = 1, e = Add->getNumOperands(); i != e; ++i)
+ X = X.add(getUnsignedRange(Add->getOperand(i)));
+ return X;
+ }
+
+ if (const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(S)) {
+ ConstantRange X = getUnsignedRange(Mul->getOperand(0));
+ for (unsigned i = 1, e = Mul->getNumOperands(); i != e; ++i)
+ X = X.multiply(getUnsignedRange(Mul->getOperand(i)));
+ return X;
+ }
+
+ if (const SCEVSMaxExpr *SMax = dyn_cast<SCEVSMaxExpr>(S)) {
+ ConstantRange X = getUnsignedRange(SMax->getOperand(0));
+ for (unsigned i = 1, e = SMax->getNumOperands(); i != e; ++i)
+ X = X.smax(getUnsignedRange(SMax->getOperand(i)));
+ return X;
+ }
+
+ if (const SCEVUMaxExpr *UMax = dyn_cast<SCEVUMaxExpr>(S)) {
+ ConstantRange X = getUnsignedRange(UMax->getOperand(0));
+ for (unsigned i = 1, e = UMax->getNumOperands(); i != e; ++i)
+ X = X.umax(getUnsignedRange(UMax->getOperand(i)));
+ return X;
+ }
+
+ if (const SCEVUDivExpr *UDiv = dyn_cast<SCEVUDivExpr>(S)) {
+ ConstantRange X = getUnsignedRange(UDiv->getLHS());
+ ConstantRange Y = getUnsignedRange(UDiv->getRHS());
+ return X.udiv(Y);
+ }
+
+ if (const SCEVZeroExtendExpr *ZExt = dyn_cast<SCEVZeroExtendExpr>(S)) {
+ ConstantRange X = getUnsignedRange(ZExt->getOperand());
+ return X.zeroExtend(cast<IntegerType>(ZExt->getType())->getBitWidth());
+ }
+
+ if (const SCEVSignExtendExpr *SExt = dyn_cast<SCEVSignExtendExpr>(S)) {
+ ConstantRange X = getUnsignedRange(SExt->getOperand());
+ return X.signExtend(cast<IntegerType>(SExt->getType())->getBitWidth());
+ }
- if (const SCEVZeroExtendExpr *C = dyn_cast<SCEVZeroExtendExpr>(S)) {
- // A zero-extension cast adds zero bits.
- return GetMinLeadingZeros(C->getOperand()) +
- (getTypeSizeInBits(C->getType()) -
- getTypeSizeInBits(C->getOperand()->getType()));
+ if (const SCEVTruncateExpr *Trunc = dyn_cast<SCEVTruncateExpr>(S)) {
+ ConstantRange X = getUnsignedRange(Trunc->getOperand());
+ return X.truncate(cast<IntegerType>(Trunc->getType())->getBitWidth());
+ }
+
+ ConstantRange FullSet(getTypeSizeInBits(S->getType()), true);
+
+ if (const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(S)) {
+ const SCEV *T = getBackedgeTakenCount(AddRec->getLoop());
+ const SCEVConstant *Trip = dyn_cast<SCEVConstant>(T);
+ if (!Trip) return FullSet;
+
+ // TODO: non-affine addrec
+ if (AddRec->isAffine()) {
+ const Type *Ty = AddRec->getType();
+ const SCEV *MaxBECount = getMaxBackedgeTakenCount(AddRec->getLoop());
+ if (getTypeSizeInBits(MaxBECount->getType()) <= getTypeSizeInBits(Ty)) {
+ MaxBECount = getNoopOrZeroExtend(MaxBECount, Ty);
+
+ const SCEV *Start = AddRec->getStart();
+ const SCEV *End = AddRec->evaluateAtIteration(MaxBECount, *this);
+
+ // Check for overflow.
+ if (!isKnownPredicate(ICmpInst::ICMP_ULE, Start, End))
+ return FullSet;
+
+ ConstantRange StartRange = getUnsignedRange(Start);
+ ConstantRange EndRange = getUnsignedRange(End);
+ APInt Min = APIntOps::umin(StartRange.getUnsignedMin(),
+ EndRange.getUnsignedMin());
+ APInt Max = APIntOps::umax(StartRange.getUnsignedMax(),
+ EndRange.getUnsignedMax());
+ if (Min.isMinValue() && Max.isMaxValue())
+ return FullSet;
+ return ConstantRange(Min, Max+1);
+ }
+ }
}
if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(S)) {
APInt Mask = APInt::getAllOnesValue(BitWidth);
APInt Zeros(BitWidth, 0), Ones(BitWidth, 0);
ComputeMaskedBits(U->getValue(), Mask, Zeros, Ones, TD);
- return Zeros.countLeadingOnes();
+ if (Ones == ~Zeros + 1)
+ return FullSet;
+ return ConstantRange(Ones, ~Zeros + 1);
}
- return 1;
+ return FullSet;
}
-uint32_t
-ScalarEvolution::GetMinSignBits(const SCEV* S) {
- // TODO: Handle other SCEV expression types here.
+/// getSignedRange - Determine the signed range for a particular SCEV.
+///
+ConstantRange
+ScalarEvolution::getSignedRange(const SCEV *S) {
+
+ if (const SCEVConstant *C = dyn_cast<SCEVConstant>(S))
+ return ConstantRange(C->getValue()->getValue());
- if (const SCEVConstant *C = dyn_cast<SCEVConstant>(S)) {
- const APInt &A = C->getValue()->getValue();
- return A.isNegative() ? A.countLeadingOnes() :
- A.countLeadingZeros();
+ if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
+ ConstantRange X = getSignedRange(Add->getOperand(0));
+ for (unsigned i = 1, e = Add->getNumOperands(); i != e; ++i)
+ X = X.add(getSignedRange(Add->getOperand(i)));
+ return X;
}
- if (const SCEVSignExtendExpr *C = dyn_cast<SCEVSignExtendExpr>(S)) {
- // A sign-extension cast adds sign bits.
- return GetMinSignBits(C->getOperand()) +
- (getTypeSizeInBits(C->getType()) -
- getTypeSizeInBits(C->getOperand()->getType()));
+ if (const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(S)) {
+ ConstantRange X = getSignedRange(Mul->getOperand(0));
+ for (unsigned i = 1, e = Mul->getNumOperands(); i != e; ++i)
+ X = X.multiply(getSignedRange(Mul->getOperand(i)));
+ return X;
}
- if (const SCEVAddExpr *A = dyn_cast<SCEVAddExpr>(S)) {
- unsigned BitWidth = getTypeSizeInBits(A->getType());
-
- // Special case decrementing a value (ADD X, -1):
- if (const SCEVConstant *CRHS = dyn_cast<SCEVConstant>(A->getOperand(0)))
- if (CRHS->isAllOnesValue()) {
- SmallVector<const SCEV *, 4> OtherOps(A->op_begin() + 1, A->op_end());
- const SCEV *OtherOpsAdd = getAddExpr(OtherOps);
- unsigned LZ = GetMinLeadingZeros(OtherOpsAdd);
-
- // If the input is known to be 0 or 1, the output is 0/-1, which is all
- // sign bits set.
- if (LZ == BitWidth - 1)
- return BitWidth;
-
- // If we are subtracting one from a positive number, there is no carry
- // out of the result.
- if (LZ > 0)
- return GetMinSignBits(OtherOpsAdd);
- }
+ if (const SCEVSMaxExpr *SMax = dyn_cast<SCEVSMaxExpr>(S)) {
+ ConstantRange X = getSignedRange(SMax->getOperand(0));
+ for (unsigned i = 1, e = SMax->getNumOperands(); i != e; ++i)
+ X = X.smax(getSignedRange(SMax->getOperand(i)));
+ return X;
+ }
+
+ if (const SCEVUMaxExpr *UMax = dyn_cast<SCEVUMaxExpr>(S)) {
+ ConstantRange X = getSignedRange(UMax->getOperand(0));
+ for (unsigned i = 1, e = UMax->getNumOperands(); i != e; ++i)
+ X = X.umax(getSignedRange(UMax->getOperand(i)));
+ return X;
+ }
+
+ if (const SCEVUDivExpr *UDiv = dyn_cast<SCEVUDivExpr>(S)) {
+ ConstantRange X = getSignedRange(UDiv->getLHS());
+ ConstantRange Y = getSignedRange(UDiv->getRHS());
+ return X.udiv(Y);
+ }
+
+ if (const SCEVZeroExtendExpr *ZExt = dyn_cast<SCEVZeroExtendExpr>(S)) {
+ ConstantRange X = getSignedRange(ZExt->getOperand());
+ return X.zeroExtend(cast<IntegerType>(ZExt->getType())->getBitWidth());
+ }
- // Add can have at most one carry bit. Thus we know that the output
- // is, at worst, one more bit than the inputs.
- unsigned Min = BitWidth;
- for (unsigned i = 0, e = A->getNumOperands(); i != e; ++i) {
- unsigned N = GetMinSignBits(A->getOperand(i));
- Min = std::min(Min, N) - 1;
- if (Min == 0) return 1;
+ if (const SCEVSignExtendExpr *SExt = dyn_cast<SCEVSignExtendExpr>(S)) {
+ ConstantRange X = getSignedRange(SExt->getOperand());
+ return X.signExtend(cast<IntegerType>(SExt->getType())->getBitWidth());
+ }
+
+ if (const SCEVTruncateExpr *Trunc = dyn_cast<SCEVTruncateExpr>(S)) {
+ ConstantRange X = getSignedRange(Trunc->getOperand());
+ return X.truncate(cast<IntegerType>(Trunc->getType())->getBitWidth());
+ }
+
+ ConstantRange FullSet(getTypeSizeInBits(S->getType()), true);
+
+ if (const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(S)) {
+ const SCEV *T = getBackedgeTakenCount(AddRec->getLoop());
+ const SCEVConstant *Trip = dyn_cast<SCEVConstant>(T);
+ if (!Trip) return FullSet;
+
+ // TODO: non-affine addrec
+ if (AddRec->isAffine()) {
+ const Type *Ty = AddRec->getType();
+ const SCEV *MaxBECount = getMaxBackedgeTakenCount(AddRec->getLoop());
+ if (getTypeSizeInBits(MaxBECount->getType()) <= getTypeSizeInBits(Ty)) {
+ MaxBECount = getNoopOrZeroExtend(MaxBECount, Ty);
+
+ const SCEV *Start = AddRec->getStart();
+ const SCEV *Step = AddRec->getStepRecurrence(*this);
+ const SCEV *End = AddRec->evaluateAtIteration(MaxBECount, *this);
+
+ // Check for overflow.
+ if (!(isKnownPositive(Step) &&
+ isKnownPredicate(ICmpInst::ICMP_SLT, Start, End)) &&
+ !(isKnownNegative(Step) &&
+ isKnownPredicate(ICmpInst::ICMP_SGT, Start, End)))
+ return FullSet;
+
+ ConstantRange StartRange = getSignedRange(Start);
+ ConstantRange EndRange = getSignedRange(End);
+ APInt Min = APIntOps::smin(StartRange.getSignedMin(),
+ EndRange.getSignedMin());
+ APInt Max = APIntOps::smax(StartRange.getSignedMax(),
+ EndRange.getSignedMax());
+ if (Min.isMinSignedValue() && Max.isMaxSignedValue())
+ return ConstantRange(Min.getBitWidth(), /*isFullSet=*/true);
+ return ConstantRange(Min, Max+1);
+ }
}
- return 1;
}
if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(S)) {
// For a SCEVUnknown, ask ValueTracking.
- return ComputeNumSignBits(U->getValue(), TD);
+ unsigned BitWidth = getTypeSizeInBits(U->getType());
+ unsigned NS = ComputeNumSignBits(U->getValue(), TD);
+ if (NS == 1)
+ return FullSet;
+ return
+ ConstantRange(APInt::getSignedMinValue(BitWidth).ashr(NS - 1),
+ APInt::getSignedMaxValue(BitWidth).ashr(NS - 1)+1);
}
- return 1;
+ return FullSet;
}
/// createSCEV - We know that there is no SCEV for the specified value.
/// Analyze the expression.
///
-const SCEV* ScalarEvolution::createSCEV(Value *V) {
+const SCEV *ScalarEvolution::createSCEV(Value *V) {
if (!isSCEVable(V->getType()))
return getUnknown(V);
else
return getUnknown(V);
- User *U = cast<User>(V);
+ Operator *U = cast<Operator>(V);
switch (Opcode) {
case Instruction::Add:
return getAddExpr(getSCEV(U->getOperand(0)),
// In order for this transformation to be safe, the LHS must be of the
// form X*(2^n) and the Or constant must be less than 2^n.
if (ConstantInt *CI = dyn_cast<ConstantInt>(U->getOperand(1))) {
- const SCEV* LHS = getSCEV(U->getOperand(0));
+ const SCEV *LHS = getSCEV(U->getOperand(0));
const APInt &CIVal = CI->getValue();
if (GetMinTrailingZeros(LHS) >=
(CIVal.getBitWidth() - CIVal.countLeadingZeros()))
if (const SCEVZeroExtendExpr *Z =
dyn_cast<SCEVZeroExtendExpr>(getSCEV(U->getOperand(0)))) {
const Type *UTy = U->getType();
- const SCEV* Z0 = Z->getOperand();
+ const SCEV *Z0 = Z->getOperand();
const Type *Z0Ty = Z0->getType();
unsigned Z0TySize = getTypeSizeInBits(Z0Ty);
// Turn shift left of a constant amount into a multiply.
if (ConstantInt *SA = dyn_cast<ConstantInt>(U->getOperand(1))) {
uint32_t BitWidth = cast<IntegerType>(V->getType())->getBitWidth();
- Constant *X = ConstantInt::get(
+ Constant *X = Context->getConstantInt(
APInt(BitWidth, 1).shl(SA->getLimitedValue(BitWidth)));
return getMulExpr(getSCEV(U->getOperand(0)), getSCEV(X));
}
// Turn logical shift right of a constant into a unsigned divide.
if (ConstantInt *SA = dyn_cast<ConstantInt>(U->getOperand(1))) {
uint32_t BitWidth = cast<IntegerType>(V->getType())->getBitWidth();
- Constant *X = ConstantInt::get(
+ Constant *X = Context->getConstantInt(
APInt(BitWidth, 1).shl(SA->getLimitedValue(BitWidth)));
return getUDivExpr(getSCEV(U->getOperand(0)), getSCEV(X));
}
return getSCEV(U->getOperand(0));
break;
- case Instruction::IntToPtr:
- if (!TD) break; // Without TD we can't analyze pointers.
- return getTruncateOrZeroExtend(getSCEV(U->getOperand(0)),
- TD->getIntPtrType());
-
- case Instruction::PtrToInt:
- if (!TD) break; // Without TD we can't analyze pointers.
- return getTruncateOrZeroExtend(getSCEV(U->getOperand(0)),
- U->getType());
+ // It's tempting to handle inttoptr and ptrtoint, however this can
+ // lead to pointer expressions which cannot be expanded to GEPs
+ // (because they may overflow). For now, the only pointer-typed
+ // expressions we handle are GEPs and address literals.
case Instruction::GetElementPtr:
if (!TD) break; // Without TD we can't analyze pointers.
/// loop-invariant backedge-taken count (see
/// hasLoopInvariantBackedgeTakenCount).
///
-const SCEV* ScalarEvolution::getBackedgeTakenCount(const Loop *L) {
+const SCEV *ScalarEvolution::getBackedgeTakenCount(const Loop *L) {
return getBackedgeTakenInfo(L).Exact;
}
/// getMaxBackedgeTakenCount - Similar to getBackedgeTakenCount, except
/// return the least SCEV value that is known never to be less than the
/// actual backedge taken count.
-const SCEV* ScalarEvolution::getMaxBackedgeTakenCount(const Loop *L) {
+const SCEV *ScalarEvolution::getMaxBackedgeTakenCount(const Loop *L) {
return getBackedgeTakenInfo(L).Max;
}
+/// PushLoopPHIs - Push PHI nodes in the header of the given loop
+/// onto the given Worklist.
+static void
+PushLoopPHIs(const Loop *L, SmallVectorImpl<Instruction *> &Worklist) {
+ BasicBlock *Header = L->getHeader();
+
+ // Push all Loop-header PHIs onto the Worklist stack.
+ for (BasicBlock::iterator I = Header->begin();
+ PHINode *PN = dyn_cast<PHINode>(I); ++I)
+ Worklist.push_back(PN);
+}
+
+/// PushDefUseChildren - Push users of the given Instruction
+/// onto the given Worklist.
+static void
+PushDefUseChildren(Instruction *I,
+ SmallVectorImpl<Instruction *> &Worklist) {
+ // Push the def-use children onto the Worklist stack.
+ for (Value::use_iterator UI = I->use_begin(), UE = I->use_end();
+ UI != UE; ++UI)
+ Worklist.push_back(cast<Instruction>(UI));
+}
+
const ScalarEvolution::BackedgeTakenInfo &
ScalarEvolution::getBackedgeTakenInfo(const Loop *L) {
// Initially insert a CouldNotCompute for this loop. If the insertion
// Now that we know more about the trip count for this loop, forget any
// existing SCEV values for PHI nodes in this loop since they are only
- // conservative estimates made without the benefit
- // of trip count information.
- if (ItCount.hasAnyInfo())
- forgetLoopPHIs(L);
+ // conservative estimates made without the benefit of trip count
+ // information. This is similar to the code in
+ // forgetLoopBackedgeTakenCount, except that it handles SCEVUnknown PHI
+ // nodes specially.
+ if (ItCount.hasAnyInfo()) {
+ SmallVector<Instruction *, 16> Worklist;
+ PushLoopPHIs(L, Worklist);
+
+ SmallPtrSet<Instruction *, 8> Visited;
+ while (!Worklist.empty()) {
+ Instruction *I = Worklist.pop_back_val();
+ if (!Visited.insert(I)) continue;
+
+ std::map<SCEVCallbackVH, const SCEV*>::iterator It =
+ Scalars.find(static_cast<Value *>(I));
+ if (It != Scalars.end()) {
+ // SCEVUnknown for a PHI either means that it has an unrecognized
+ // structure, or it's a PHI that's in the progress of being computed
+ // by createNodeForPHI. In the former case, additional loop trip
+ // count information isn't going to change anything. In the later
+ // case, createNodeForPHI will perform the necessary updates on its
+ // own when it gets to that point.
+ if (!isa<PHINode>(I) || !isa<SCEVUnknown>(It->second))
+ Scalars.erase(It);
+ ValuesAtScopes.erase(I);
+ if (PHINode *PN = dyn_cast<PHINode>(I))
+ ConstantEvolutionLoopExitValue.erase(PN);
+ }
+
+ PushDefUseChildren(I, Worklist);
+ }
+ }
}
return Pair.first->second;
}
/// is deleted.
void ScalarEvolution::forgetLoopBackedgeTakenCount(const Loop *L) {
BackedgeTakenCounts.erase(L);
- forgetLoopPHIs(L);
-}
-
-/// forgetLoopPHIs - Delete the memoized SCEVs associated with the
-/// PHI nodes in the given loop. This is used when the trip count of
-/// the loop may have changed.
-void ScalarEvolution::forgetLoopPHIs(const Loop *L) {
- BasicBlock *Header = L->getHeader();
- // Push all Loop-header PHIs onto the Worklist stack, except those
- // that are presently represented via a SCEVUnknown. SCEVUnknown for
- // a PHI either means that it has an unrecognized structure, or it's
- // a PHI that's in the progress of being computed by createNodeForPHI.
- // In the former case, additional loop trip count information isn't
- // going to change anything. In the later case, createNodeForPHI will
- // perform the necessary updates on its own when it gets to that point.
SmallVector<Instruction *, 16> Worklist;
- for (BasicBlock::iterator I = Header->begin();
- PHINode *PN = dyn_cast<PHINode>(I); ++I) {
- std::map<SCEVCallbackVH, const SCEV*>::iterator It =
- Scalars.find((Value*)I);
- if (It != Scalars.end() && !isa<SCEVUnknown>(It->second))
- Worklist.push_back(PN);
- }
+ PushLoopPHIs(L, Worklist);
+ SmallPtrSet<Instruction *, 8> Visited;
while (!Worklist.empty()) {
Instruction *I = Worklist.pop_back_val();
- if (Scalars.erase(I))
- for (Value::use_iterator UI = I->use_begin(), UE = I->use_end();
- UI != UE; ++UI)
- Worklist.push_back(cast<Instruction>(UI));
+ if (!Visited.insert(I)) continue;
+
+ std::map<SCEVCallbackVH, const SCEV*>::iterator It =
+ Scalars.find(static_cast<Value *>(I));
+ if (It != Scalars.end()) {
+ Scalars.erase(It);
+ ValuesAtScopes.erase(I);
+ if (PHINode *PN = dyn_cast<PHINode>(I))
+ ConstantEvolutionLoopExitValue.erase(PN);
+ }
+
+ PushDefUseChildren(I, Worklist);
}
}
L->getExitingBlocks(ExitingBlocks);
// Examine all exits and pick the most conservative values.
- const SCEV* BECount = getCouldNotCompute();
- const SCEV* MaxBECount = getCouldNotCompute();
+ const SCEV *BECount = getCouldNotCompute();
+ const SCEV *MaxBECount = getCouldNotCompute();
bool CouldNotComputeBECount = false;
for (unsigned i = 0, e = ExitingBlocks.size(); i != e; ++i) {
BackedgeTakenInfo NewBTI =
ComputeBackedgeTakenCountFromExitCond(L, BO->getOperand(0), TBB, FBB);
BackedgeTakenInfo BTI1 =
ComputeBackedgeTakenCountFromExitCond(L, BO->getOperand(1), TBB, FBB);
- const SCEV* BECount = getCouldNotCompute();
- const SCEV* MaxBECount = getCouldNotCompute();
+ const SCEV *BECount = getCouldNotCompute();
+ const SCEV *MaxBECount = getCouldNotCompute();
if (L->contains(TBB)) {
// Both conditions must be true for the loop to continue executing.
// Choose the less conservative count.
ComputeBackedgeTakenCountFromExitCond(L, BO->getOperand(0), TBB, FBB);
BackedgeTakenInfo BTI1 =
ComputeBackedgeTakenCountFromExitCond(L, BO->getOperand(1), TBB, FBB);
- const SCEV* BECount = getCouldNotCompute();
- const SCEV* MaxBECount = getCouldNotCompute();
+ const SCEV *BECount = getCouldNotCompute();
+ const SCEV *MaxBECount = getCouldNotCompute();
if (L->contains(FBB)) {
// Both conditions must be false for the loop to continue executing.
// Choose the less conservative count.
// Handle common loops like: for (X = "string"; *X; ++X)
if (LoadInst *LI = dyn_cast<LoadInst>(ExitCond->getOperand(0)))
if (Constant *RHS = dyn_cast<Constant>(ExitCond->getOperand(1))) {
- const SCEV* ItCnt =
+ const SCEV *ItCnt =
ComputeLoadConstantCompareBackedgeTakenCount(LI, RHS, L, Cond);
if (!isa<SCEVCouldNotCompute>(ItCnt)) {
unsigned BitWidth = getTypeSizeInBits(ItCnt->getType());
}
}
- const SCEV* LHS = getSCEV(ExitCond->getOperand(0));
- const SCEV* RHS = getSCEV(ExitCond->getOperand(1));
+ const SCEV *LHS = getSCEV(ExitCond->getOperand(0));
+ const SCEV *RHS = getSCEV(ExitCond->getOperand(1));
// Try to evaluate any dependencies out of the loop.
LHS = getSCEVAtScope(LHS, L);
ConstantRange CompRange(
ICmpInst::makeConstantRange(Cond, RHSC->getValue()->getValue()));
- const SCEV* Ret = AddRec->getNumIterationsInRange(CompRange, *this);
+ const SCEV *Ret = AddRec->getNumIterationsInRange(CompRange, *this);
if (!isa<SCEVCouldNotCompute>(Ret)) return Ret;
}
switch (Cond) {
case ICmpInst::ICMP_NE: { // while (X != Y)
// Convert to: while (X-Y != 0)
- const SCEV* TC = HowFarToZero(getMinusSCEV(LHS, RHS), L);
+ const SCEV *TC = HowFarToZero(getMinusSCEV(LHS, RHS), L);
if (!isa<SCEVCouldNotCompute>(TC)) return TC;
break;
}
case ICmpInst::ICMP_EQ: {
// Convert to: while (X-Y == 0) // while (X == Y)
- const SCEV* TC = HowFarToNonZero(getMinusSCEV(LHS, RHS), L);
+ const SCEV *TC = HowFarToNonZero(getMinusSCEV(LHS, RHS), L);
if (!isa<SCEVCouldNotCompute>(TC)) return TC;
break;
}
static ConstantInt *
EvaluateConstantChrecAtConstant(const SCEVAddRecExpr *AddRec, ConstantInt *C,
ScalarEvolution &SE) {
- const SCEV* InVal = SE.getConstant(C);
- const SCEV* Val = AddRec->evaluateAtIteration(InVal, SE);
+ const SCEV *InVal = SE.getConstant(C);
+ const SCEV *Val = AddRec->evaluateAtIteration(InVal, SE);
assert(isa<SCEVConstant>(Val) &&
"Evaluation of SCEV at constant didn't fold correctly?");
return cast<SCEVConstant>(Val)->getValue();
/// the addressed element of the initializer or null if the index expression is
/// invalid.
static Constant *
-GetAddressedElementFromGlobal(GlobalVariable *GV,
+GetAddressedElementFromGlobal(LLVMContext *Context, GlobalVariable *GV,
const std::vector<ConstantInt*> &Indices) {
Constant *Init = GV->getInitializer();
for (unsigned i = 0, e = Indices.size(); i != e; ++i) {
} else if (isa<ConstantAggregateZero>(Init)) {
if (const StructType *STy = dyn_cast<StructType>(Init->getType())) {
assert(Idx < STy->getNumElements() && "Bad struct index!");
- Init = Constant::getNullValue(STy->getElementType(Idx));
+ Init = Context->getNullValue(STy->getElementType(Idx));
} else if (const ArrayType *ATy = dyn_cast<ArrayType>(Init->getType())) {
if (Idx >= ATy->getNumElements()) return 0; // Bogus program
- Init = Constant::getNullValue(ATy->getElementType());
+ Init = Context->getNullValue(ATy->getElementType());
} else {
- assert(0 && "Unknown constant aggregate type!");
+ llvm_unreachable("Unknown constant aggregate type!");
}
return 0;
} else {
// Okay, we know we have a (load (gep GV, 0, X)) comparison with a constant.
// Check to see if X is a loop variant variable value now.
- const SCEV* Idx = getSCEV(VarIdx);
+ const SCEV *Idx = getSCEV(VarIdx);
Idx = getSCEVAtScope(Idx, L);
// We can only recognize very limited forms of loop index expressions, in
unsigned MaxSteps = MaxBruteForceIterations;
for (unsigned IterationNum = 0; IterationNum != MaxSteps; ++IterationNum) {
- ConstantInt *ItCst =
- ConstantInt::get(cast<IntegerType>(IdxExpr->getType()), IterationNum);
+ ConstantInt *ItCst = Context->getConstantInt(
+ cast<IntegerType>(IdxExpr->getType()), IterationNum);
ConstantInt *Val = EvaluateConstantChrecAtConstant(IdxExpr, ItCst, *this);
// Form the GEP offset.
Indexes[VarIdxNum] = Val;
- Constant *Result = GetAddressedElementFromGlobal(GV, Indexes);
+ Constant *Result = GetAddressedElementFromGlobal(Context, GV, Indexes);
if (Result == 0) break; // Cannot compute!
// Evaluate the condition for this iteration.
if (Constant *C = dyn_cast<Constant>(V)) return C;
if (GlobalValue *GV = dyn_cast<GlobalValue>(V)) return GV;
Instruction *I = cast<Instruction>(V);
+ LLVMContext *Context = I->getParent()->getContext();
std::vector<Constant*> Operands;
Operands.resize(I->getNumOperands());
if (const CmpInst *CI = dyn_cast<CmpInst>(I))
return ConstantFoldCompareInstOperands(CI->getPredicate(),
- &Operands[0], Operands.size());
+ &Operands[0], Operands.size(),
+ Context);
else
return ConstantFoldInstOperands(I->getOpcode(), I->getType(),
- &Operands[0], Operands.size());
+ &Operands[0], Operands.size(),
+ Context);
}
/// getConstantEvolutionLoopExitValue - If we know that the specified Phi is
///
/// In the case that a relevant loop exit value cannot be computed, the
/// original value V is returned.
-const SCEV* ScalarEvolution::getSCEVAtScope(const SCEV *V, const Loop *L) {
+const SCEV *ScalarEvolution::getSCEVAtScope(const SCEV *V, const Loop *L) {
// FIXME: this should be turned into a virtual method on SCEV!
if (isa<SCEVConstant>(V)) return V;
// to see if the loop that contains it has a known backedge-taken
// count. If so, we may be able to force computation of the exit
// value.
- const SCEV* BackedgeTakenCount = getBackedgeTakenCount(LI);
+ const SCEV *BackedgeTakenCount = getBackedgeTakenCount(LI);
if (const SCEVConstant *BTCC =
dyn_cast<SCEVConstant>(BackedgeTakenCount)) {
// Okay, we know how many times the containing loop executes. If
if (!isSCEVable(Op->getType()))
return V;
- const SCEV* OpV = getSCEVAtScope(getSCEV(Op), L);
+ const SCEV* OpV = getSCEVAtScope(Op, L);
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(OpV)) {
Constant *C = SC->getValue();
if (C->getType() != Op->getType())
Constant *C;
if (const CmpInst *CI = dyn_cast<CmpInst>(I))
C = ConstantFoldCompareInstOperands(CI->getPredicate(),
- &Operands[0], Operands.size());
+ &Operands[0], Operands.size(),
+ Context);
else
C = ConstantFoldInstOperands(I->getOpcode(), I->getType(),
- &Operands[0], Operands.size());
+ &Operands[0], Operands.size(), Context);
Pair.first->second = C;
return getSCEV(C);
}
// Avoid performing the look-up in the common case where the specified
// expression has no loop-variant portions.
for (unsigned i = 0, e = Comm->getNumOperands(); i != e; ++i) {
- const SCEV* OpAtScope = getSCEVAtScope(Comm->getOperand(i), L);
+ const SCEV *OpAtScope = getSCEVAtScope(Comm->getOperand(i), L);
if (OpAtScope != Comm->getOperand(i)) {
// Okay, at least one of these operands is loop variant but might be
// foldable. Build a new instance of the folded commutative expression.
return getSMaxExpr(NewOps);
if (isa<SCEVUMaxExpr>(Comm))
return getUMaxExpr(NewOps);
- assert(0 && "Unknown commutative SCEV type!");
+ llvm_unreachable("Unknown commutative SCEV type!");
}
}
// If we got here, all operands are loop invariant.
}
if (const SCEVUDivExpr *Div = dyn_cast<SCEVUDivExpr>(V)) {
- const SCEV* LHS = getSCEVAtScope(Div->getLHS(), L);
- const SCEV* RHS = getSCEVAtScope(Div->getRHS(), L);
+ const SCEV *LHS = getSCEVAtScope(Div->getLHS(), L);
+ const SCEV *RHS = getSCEVAtScope(Div->getRHS(), L);
if (LHS == Div->getLHS() && RHS == Div->getRHS())
return Div; // must be loop invariant
return getUDivExpr(LHS, RHS);
if (!L || !AddRec->getLoop()->contains(L->getHeader())) {
// To evaluate this recurrence, we need to know how many times the AddRec
// loop iterates. Compute this now.
- const SCEV* BackedgeTakenCount = getBackedgeTakenCount(AddRec->getLoop());
+ const SCEV *BackedgeTakenCount = getBackedgeTakenCount(AddRec->getLoop());
if (BackedgeTakenCount == getCouldNotCompute()) return AddRec;
// Then, evaluate the AddRec.
}
if (const SCEVZeroExtendExpr *Cast = dyn_cast<SCEVZeroExtendExpr>(V)) {
- const SCEV* Op = getSCEVAtScope(Cast->getOperand(), L);
+ const SCEV *Op = getSCEVAtScope(Cast->getOperand(), L);
if (Op == Cast->getOperand())
return Cast; // must be loop invariant
return getZeroExtendExpr(Op, Cast->getType());
}
if (const SCEVSignExtendExpr *Cast = dyn_cast<SCEVSignExtendExpr>(V)) {
- const SCEV* Op = getSCEVAtScope(Cast->getOperand(), L);
+ const SCEV *Op = getSCEVAtScope(Cast->getOperand(), L);
if (Op == Cast->getOperand())
return Cast; // must be loop invariant
return getSignExtendExpr(Op, Cast->getType());
}
if (const SCEVTruncateExpr *Cast = dyn_cast<SCEVTruncateExpr>(V)) {
- const SCEV* Op = getSCEVAtScope(Cast->getOperand(), L);
+ const SCEV *Op = getSCEVAtScope(Cast->getOperand(), L);
if (Op == Cast->getOperand())
return Cast; // must be loop invariant
return getTruncateExpr(Op, Cast->getType());
}
- assert(0 && "Unknown SCEV type!");
+ llvm_unreachable("Unknown SCEV type!");
return 0;
}
/// getSCEVAtScope - This is a convenience function which does
/// getSCEVAtScope(getSCEV(V), L).
-const SCEV* ScalarEvolution::getSCEVAtScope(Value *V, const Loop *L) {
+const SCEV *ScalarEvolution::getSCEVAtScope(Value *V, const Loop *L) {
return getSCEVAtScope(getSCEV(V), L);
}
/// A and B isn't important.
///
/// If the equation does not have a solution, SCEVCouldNotCompute is returned.
-static const SCEV* SolveLinEquationWithOverflow(const APInt &A, const APInt &B,
+static const SCEV *SolveLinEquationWithOverflow(const APInt &A, const APInt &B,
ScalarEvolution &SE) {
uint32_t BW = A.getBitWidth();
assert(BW == B.getBitWidth() && "Bit widths must be the same.");
/// given quadratic chrec {L,+,M,+,N}. This returns either the two roots (which
/// might be the same) or two SCEVCouldNotCompute objects.
///
-static std::pair<const SCEV*,const SCEV*>
+static std::pair<const SCEV *,const SCEV *>
SolveQuadraticEquation(const SCEVAddRecExpr *AddRec, ScalarEvolution &SE) {
assert(AddRec->getNumOperands() == 3 && "This is not a quadratic chrec!");
const SCEVConstant *LC = dyn_cast<SCEVConstant>(AddRec->getOperand(0));
return std::make_pair(CNC, CNC);
}
- ConstantInt *Solution1 = ConstantInt::get((NegB + SqrtVal).sdiv(TwoA));
- ConstantInt *Solution2 = ConstantInt::get((NegB - SqrtVal).sdiv(TwoA));
+ LLVMContext *Context = SE.getContext();
+
+ ConstantInt *Solution1 =
+ Context->getConstantInt((NegB + SqrtVal).sdiv(TwoA));
+ ConstantInt *Solution2 =
+ Context->getConstantInt((NegB - SqrtVal).sdiv(TwoA));
return std::make_pair(SE.getConstant(Solution1),
SE.getConstant(Solution2));
/// HowFarToZero - Return the number of times a backedge comparing the specified
/// value to zero will execute. If not computable, return CouldNotCompute.
-const SCEV* ScalarEvolution::HowFarToZero(const SCEV *V, const Loop *L) {
+const SCEV *ScalarEvolution::HowFarToZero(const SCEV *V, const Loop *L) {
// If the value is a constant
if (const SCEVConstant *C = dyn_cast<SCEVConstant>(V)) {
// If the value is already zero, the branch will execute zero times.
} else if (AddRec->isQuadratic() && AddRec->getType()->isInteger()) {
// If this is a quadratic (3-term) AddRec {L,+,M,+,N}, find the roots of
// the quadratic equation to solve it.
- std::pair<const SCEV*,const SCEV*> Roots = SolveQuadraticEquation(AddRec,
+ std::pair<const SCEV *,const SCEV *> Roots = SolveQuadraticEquation(AddRec,
*this);
const SCEVConstant *R1 = dyn_cast<SCEVConstant>(Roots.first);
const SCEVConstant *R2 = dyn_cast<SCEVConstant>(Roots.second);
#endif
// Pick the smallest positive root value.
if (ConstantInt *CB =
- dyn_cast<ConstantInt>(ConstantExpr::getICmp(ICmpInst::ICMP_ULT,
+ dyn_cast<ConstantInt>(Context->getConstantExprICmp(ICmpInst::ICMP_ULT,
R1->getValue(), R2->getValue()))) {
if (CB->getZExtValue() == false)
std::swap(R1, R2); // R1 is the minimum root now.
// We can only use this value if the chrec ends up with an exact zero
// value at this index. When solving for "X*X != 5", for example, we
// should not accept a root of 2.
- const SCEV* Val = AddRec->evaluateAtIteration(R1, *this);
+ const SCEV *Val = AddRec->evaluateAtIteration(R1, *this);
if (Val->isZero())
return R1; // We found a quadratic root!
}
/// HowFarToNonZero - Return the number of times a backedge checking the
/// specified value for nonzero will execute. If not computable, return
/// CouldNotCompute
-const SCEV* ScalarEvolution::HowFarToNonZero(const SCEV *V, const Loop *L) {
+const SCEV *ScalarEvolution::HowFarToNonZero(const SCEV *V, const Loop *L) {
// Loops that look like: while (X == 0) are very strange indeed. We don't
// handle them yet except for the trivial case. This could be expanded in the
// future as needed.
/// more general, since a front-end may have replicated the controlling
/// expression.
///
-static bool HasSameValue(const SCEV* A, const SCEV* B) {
+static bool HasSameValue(const SCEV *A, const SCEV *B) {
// Quick check to see if they are the same SCEV.
if (A == B) return true;
return false;
}
-/// isLoopGuardedByCond - Test whether entry to the loop is protected by
-/// a conditional between LHS and RHS. This is used to help avoid max
-/// expressions in loop trip counts.
-bool ScalarEvolution::isLoopGuardedByCond(const Loop *L,
- ICmpInst::Predicate Pred,
- const SCEV *LHS, const SCEV *RHS) {
+bool ScalarEvolution::isKnownNegative(const SCEV *S) {
+ return getSignedRange(S).getSignedMax().isNegative();
+}
+
+bool ScalarEvolution::isKnownPositive(const SCEV *S) {
+ return getSignedRange(S).getSignedMin().isStrictlyPositive();
+}
+
+bool ScalarEvolution::isKnownNonNegative(const SCEV *S) {
+ return !getSignedRange(S).getSignedMin().isNegative();
+}
+
+bool ScalarEvolution::isKnownNonPositive(const SCEV *S) {
+ return !getSignedRange(S).getSignedMax().isStrictlyPositive();
+}
+
+bool ScalarEvolution::isKnownNonZero(const SCEV *S) {
+ return isKnownNegative(S) || isKnownPositive(S);
+}
+
+bool ScalarEvolution::isKnownPredicate(ICmpInst::Predicate Pred,
+ const SCEV *LHS, const SCEV *RHS) {
+
+ if (HasSameValue(LHS, RHS))
+ return ICmpInst::isTrueWhenEqual(Pred);
+
+ switch (Pred) {
+ default:
+ llvm_unreachable("Unexpected ICmpInst::Predicate value!");
+ break;
+ case ICmpInst::ICMP_SGT:
+ Pred = ICmpInst::ICMP_SLT;
+ std::swap(LHS, RHS);
+ case ICmpInst::ICMP_SLT: {
+ ConstantRange LHSRange = getSignedRange(LHS);
+ ConstantRange RHSRange = getSignedRange(RHS);
+ if (LHSRange.getSignedMax().slt(RHSRange.getSignedMin()))
+ return true;
+ if (LHSRange.getSignedMin().sge(RHSRange.getSignedMax()))
+ return false;
+
+ const SCEV *Diff = getMinusSCEV(LHS, RHS);
+ ConstantRange DiffRange = getUnsignedRange(Diff);
+ if (isKnownNegative(Diff)) {
+ if (DiffRange.getUnsignedMax().ult(LHSRange.getUnsignedMin()))
+ return true;
+ if (DiffRange.getUnsignedMin().uge(LHSRange.getUnsignedMax()))
+ return false;
+ } else if (isKnownPositive(Diff)) {
+ if (LHSRange.getUnsignedMax().ult(DiffRange.getUnsignedMin()))
+ return true;
+ if (LHSRange.getUnsignedMin().uge(DiffRange.getUnsignedMax()))
+ return false;
+ }
+ break;
+ }
+ case ICmpInst::ICMP_SGE:
+ Pred = ICmpInst::ICMP_SLE;
+ std::swap(LHS, RHS);
+ case ICmpInst::ICMP_SLE: {
+ ConstantRange LHSRange = getSignedRange(LHS);
+ ConstantRange RHSRange = getSignedRange(RHS);
+ if (LHSRange.getSignedMax().sle(RHSRange.getSignedMin()))
+ return true;
+ if (LHSRange.getSignedMin().sgt(RHSRange.getSignedMax()))
+ return false;
+
+ const SCEV *Diff = getMinusSCEV(LHS, RHS);
+ ConstantRange DiffRange = getUnsignedRange(Diff);
+ if (isKnownNonPositive(Diff)) {
+ if (DiffRange.getUnsignedMax().ule(LHSRange.getUnsignedMin()))
+ return true;
+ if (DiffRange.getUnsignedMin().ugt(LHSRange.getUnsignedMax()))
+ return false;
+ } else if (isKnownNonNegative(Diff)) {
+ if (LHSRange.getUnsignedMax().ule(DiffRange.getUnsignedMin()))
+ return true;
+ if (LHSRange.getUnsignedMin().ugt(DiffRange.getUnsignedMax()))
+ return false;
+ }
+ break;
+ }
+ case ICmpInst::ICMP_UGT:
+ Pred = ICmpInst::ICMP_ULT;
+ std::swap(LHS, RHS);
+ case ICmpInst::ICMP_ULT: {
+ ConstantRange LHSRange = getUnsignedRange(LHS);
+ ConstantRange RHSRange = getUnsignedRange(RHS);
+ if (LHSRange.getUnsignedMax().ult(RHSRange.getUnsignedMin()))
+ return true;
+ if (LHSRange.getUnsignedMin().uge(RHSRange.getUnsignedMax()))
+ return false;
+
+ const SCEV *Diff = getMinusSCEV(LHS, RHS);
+ ConstantRange DiffRange = getUnsignedRange(Diff);
+ if (LHSRange.getUnsignedMax().ult(DiffRange.getUnsignedMin()))
+ return true;
+ if (LHSRange.getUnsignedMin().uge(DiffRange.getUnsignedMax()))
+ return false;
+ break;
+ }
+ case ICmpInst::ICMP_UGE:
+ Pred = ICmpInst::ICMP_ULE;
+ std::swap(LHS, RHS);
+ case ICmpInst::ICMP_ULE: {
+ ConstantRange LHSRange = getUnsignedRange(LHS);
+ ConstantRange RHSRange = getUnsignedRange(RHS);
+ if (LHSRange.getUnsignedMax().ule(RHSRange.getUnsignedMin()))
+ return true;
+ if (LHSRange.getUnsignedMin().ugt(RHSRange.getUnsignedMax()))
+ return false;
+
+ const SCEV *Diff = getMinusSCEV(LHS, RHS);
+ ConstantRange DiffRange = getUnsignedRange(Diff);
+ if (LHSRange.getUnsignedMax().ule(DiffRange.getUnsignedMin()))
+ return true;
+ if (LHSRange.getUnsignedMin().ugt(DiffRange.getUnsignedMax()))
+ return false;
+ break;
+ }
+ case ICmpInst::ICMP_NE: {
+ if (getUnsignedRange(LHS).intersectWith(getUnsignedRange(RHS)).isEmptySet())
+ return true;
+ if (getSignedRange(LHS).intersectWith(getSignedRange(RHS)).isEmptySet())
+ return true;
+
+ const SCEV *Diff = getMinusSCEV(LHS, RHS);
+ if (isKnownNonZero(Diff))
+ return true;
+ break;
+ }
+ case ICmpInst::ICMP_EQ:
+ break;
+ }
+ return false;
+}
+
+/// isLoopBackedgeGuardedByCond - Test whether the backedge of the loop is
+/// protected by a conditional between LHS and RHS. This is used to
+/// to eliminate casts.
+bool
+ScalarEvolution::isLoopBackedgeGuardedByCond(const Loop *L,
+ ICmpInst::Predicate Pred,
+ const SCEV *LHS, const SCEV *RHS) {
+ // Interpret a null as meaning no loop, where there is obviously no guard
+ // (interprocedural conditions notwithstanding).
+ if (!L) return true;
+
+ BasicBlock *Latch = L->getLoopLatch();
+ if (!Latch)
+ return false;
+
+ BranchInst *LoopContinuePredicate =
+ dyn_cast<BranchInst>(Latch->getTerminator());
+ if (!LoopContinuePredicate ||
+ LoopContinuePredicate->isUnconditional())
+ return false;
+
+ return
+ isNecessaryCond(LoopContinuePredicate->getCondition(), Pred, LHS, RHS,
+ LoopContinuePredicate->getSuccessor(0) != L->getHeader());
+}
+
+/// isLoopGuardedByCond - Test whether entry to the loop is protected
+/// by a conditional between LHS and RHS. This is used to help avoid max
+/// expressions in loop trip counts, and to eliminate casts.
+bool
+ScalarEvolution::isLoopGuardedByCond(const Loop *L,
+ ICmpInst::Predicate Pred,
+ const SCEV *LHS, const SCEV *RHS) {
// Interpret a null as meaning no loop, where there is obviously no guard
// (interprocedural conditions notwithstanding).
if (!L) return false;
return false;
}
-/// isNecessaryCond - Test whether the given CondValue value is a condition
-/// which is at least as strict as the one described by Pred, LHS, and RHS.
+/// isNecessaryCond - Test whether the condition described by Pred, LHS,
+/// and RHS is a necessary condition for the given Cond value to evaluate
+/// to true.
bool ScalarEvolution::isNecessaryCond(Value *CondValue,
ICmpInst::Predicate Pred,
const SCEV *LHS, const SCEV *RHS,
// see if it is the comparison we are looking for.
Value *PreCondLHS = ICI->getOperand(0);
Value *PreCondRHS = ICI->getOperand(1);
- ICmpInst::Predicate Cond;
+ ICmpInst::Predicate FoundPred;
if (Inverse)
- Cond = ICI->getInversePredicate();
+ FoundPred = ICI->getInversePredicate();
else
- Cond = ICI->getPredicate();
+ FoundPred = ICI->getPredicate();
- if (Cond == Pred)
+ if (FoundPred == Pred)
; // An exact match.
- else if (!ICmpInst::isTrueWhenEqual(Cond) && Pred == ICmpInst::ICMP_NE)
- ; // The actual condition is beyond sufficient.
- else
+ else if (!ICmpInst::isTrueWhenEqual(FoundPred) && Pred == ICmpInst::ICMP_NE) {
+ // The actual condition is beyond sufficient.
+ FoundPred = ICmpInst::ICMP_NE;
+ // NE is symmetric but the original comparison may not be. Swap
+ // the operands if necessary so that they match below.
+ if (isa<SCEVConstant>(LHS))
+ std::swap(PreCondLHS, PreCondRHS);
+ } else
// Check a few special cases.
- switch (Cond) {
+ switch (FoundPred) {
case ICmpInst::ICMP_UGT:
if (Pred == ICmpInst::ICMP_ULT) {
std::swap(PreCondLHS, PreCondRHS);
- Cond = ICmpInst::ICMP_ULT;
+ FoundPred = ICmpInst::ICMP_ULT;
break;
}
return false;
case ICmpInst::ICMP_SGT:
if (Pred == ICmpInst::ICMP_SLT) {
std::swap(PreCondLHS, PreCondRHS);
- Cond = ICmpInst::ICMP_SLT;
+ FoundPred = ICmpInst::ICMP_SLT;
break;
}
return false;
// so check for this case by checking if the NE is comparing against
// a minimum or maximum constant.
if (!ICmpInst::isTrueWhenEqual(Pred))
- if (ConstantInt *CI = dyn_cast<ConstantInt>(PreCondRHS)) {
- const APInt &A = CI->getValue();
+ if (const SCEVConstant *C = dyn_cast<SCEVConstant>(RHS)) {
+ const APInt &A = C->getValue()->getValue();
switch (Pred) {
case ICmpInst::ICMP_SLT:
if (A.isMaxSignedValue()) break;
default:
return false;
}
- Cond = ICmpInst::ICMP_NE;
+ FoundPred = Pred;
// NE is symmetric but the original comparison may not be. Swap
// the operands if necessary so that they match below.
if (isa<SCEVConstant>(LHS))
return false;
}
- if (!PreCondLHS->getType()->isInteger()) return false;
+ assert(Pred == FoundPred && "Conditions were not reconciled!");
+
+ // Bail if the ICmp's operands' types are wider than the needed type
+ // before attempting to call getSCEV on them. This avoids infinite
+ // recursion, since the analysis of widening casts can require loop
+ // exit condition information for overflow checking, which would
+ // lead back here.
+ if (getTypeSizeInBits(LHS->getType()) <
+ getTypeSizeInBits(PreCondLHS->getType()))
+ return false;
+
+ const SCEV *FoundLHS = getSCEV(PreCondLHS);
+ const SCEV *FoundRHS = getSCEV(PreCondRHS);
+
+ // Balance the types. The case where FoundLHS' type is wider than
+ // LHS' type is checked for above.
+ if (getTypeSizeInBits(LHS->getType()) >
+ getTypeSizeInBits(FoundLHS->getType())) {
+ if (CmpInst::isSigned(Pred)) {
+ FoundLHS = getSignExtendExpr(FoundLHS, LHS->getType());
+ FoundRHS = getSignExtendExpr(FoundRHS, LHS->getType());
+ } else {
+ FoundLHS = getZeroExtendExpr(FoundLHS, LHS->getType());
+ FoundRHS = getZeroExtendExpr(FoundRHS, LHS->getType());
+ }
+ }
+
+ return isNecessaryCondOperands(Pred, LHS, RHS,
+ FoundLHS, FoundRHS) ||
+ // ~x < ~y --> x > y
+ isNecessaryCondOperands(Pred, LHS, RHS,
+ getNotSCEV(FoundRHS), getNotSCEV(FoundLHS));
+}
+
+/// isNecessaryCondOperands - Test whether the condition described by Pred,
+/// LHS, and RHS is a necessary condition for the condition described by
+/// Pred, FoundLHS, and FoundRHS to evaluate to true.
+bool
+ScalarEvolution::isNecessaryCondOperands(ICmpInst::Predicate Pred,
+ const SCEV *LHS, const SCEV *RHS,
+ const SCEV *FoundLHS,
+ const SCEV *FoundRHS) {
+ switch (Pred) {
+ default: llvm_unreachable("Unexpected ICmpInst::Predicate value!");
+ case ICmpInst::ICMP_EQ:
+ case ICmpInst::ICMP_NE:
+ if (HasSameValue(LHS, FoundLHS) && HasSameValue(RHS, FoundRHS))
+ return true;
+ break;
+ case ICmpInst::ICMP_SLT:
+ case ICmpInst::ICMP_SLE:
+ if (isKnownPredicate(ICmpInst::ICMP_SLE, LHS, FoundLHS) &&
+ isKnownPredicate(ICmpInst::ICMP_SGE, RHS, FoundRHS))
+ return true;
+ break;
+ case ICmpInst::ICMP_SGT:
+ case ICmpInst::ICMP_SGE:
+ if (isKnownPredicate(ICmpInst::ICMP_SGE, LHS, FoundLHS) &&
+ isKnownPredicate(ICmpInst::ICMP_SLE, RHS, FoundRHS))
+ return true;
+ break;
+ case ICmpInst::ICMP_ULT:
+ case ICmpInst::ICMP_ULE:
+ if (isKnownPredicate(ICmpInst::ICMP_ULE, LHS, FoundLHS) &&
+ isKnownPredicate(ICmpInst::ICMP_UGE, RHS, FoundRHS))
+ return true;
+ break;
+ case ICmpInst::ICMP_UGT:
+ case ICmpInst::ICMP_UGE:
+ if (isKnownPredicate(ICmpInst::ICMP_UGE, LHS, FoundLHS) &&
+ isKnownPredicate(ICmpInst::ICMP_ULE, RHS, FoundRHS))
+ return true;
+ break;
+ }
- const SCEV *PreCondLHSSCEV = getSCEV(PreCondLHS);
- const SCEV *PreCondRHSSCEV = getSCEV(PreCondRHS);
- return (HasSameValue(LHS, PreCondLHSSCEV) &&
- HasSameValue(RHS, PreCondRHSSCEV)) ||
- (HasSameValue(LHS, getNotSCEV(PreCondRHSSCEV)) &&
- HasSameValue(RHS, getNotSCEV(PreCondLHSSCEV)));
+ return false;
}
/// getBECount - Subtract the end and start values and divide by the step,
/// rounding up, to get the number of times the backedge is executed. Return
/// CouldNotCompute if an intermediate computation overflows.
-const SCEV* ScalarEvolution::getBECount(const SCEV* Start,
- const SCEV* End,
- const SCEV* Step) {
+const SCEV *ScalarEvolution::getBECount(const SCEV *Start,
+ const SCEV *End,
+ const SCEV *Step) {
const Type *Ty = Start->getType();
- const SCEV* NegOne = getIntegerSCEV(-1, Ty);
- const SCEV* Diff = getMinusSCEV(End, Start);
- const SCEV* RoundUp = getAddExpr(Step, NegOne);
+ const SCEV *NegOne = getIntegerSCEV(-1, Ty);
+ const SCEV *Diff = getMinusSCEV(End, Start);
+ const SCEV *RoundUp = getAddExpr(Step, NegOne);
// Add an adjustment to the difference between End and Start so that
// the division will effectively round up.
- const SCEV* Add = getAddExpr(Diff, RoundUp);
+ const SCEV *Add = getAddExpr(Diff, RoundUp);
// Check Add for unsigned overflow.
// TODO: More sophisticated things could be done here.
- const Type *WideTy = IntegerType::get(getTypeSizeInBits(Ty) + 1);
- const SCEV* OperandExtendedAdd =
- getAddExpr(getZeroExtendExpr(Diff, WideTy),
- getZeroExtendExpr(RoundUp, WideTy));
+ const Type *WideTy = Context->getIntegerType(getTypeSizeInBits(Ty) + 1);
+ const SCEV *EDiff = getZeroExtendExpr(Diff, WideTy);
+ const SCEV *ERoundUp = getZeroExtendExpr(RoundUp, WideTy);
+ const SCEV *OperandExtendedAdd = getAddExpr(EDiff, ERoundUp);
if (getZeroExtendExpr(Add, WideTy) != OperandExtendedAdd)
return getCouldNotCompute();
if (AddRec->isAffine()) {
// FORNOW: We only support unit strides.
unsigned BitWidth = getTypeSizeInBits(AddRec->getType());
- const SCEV* Step = AddRec->getStepRecurrence(*this);
+ const SCEV *Step = AddRec->getStepRecurrence(*this);
// TODO: handle non-constant strides.
const SCEVConstant *CStep = dyn_cast<SCEVConstant>(Step);
// treat m-n as signed nor unsigned due to overflow possibility.
// First, we get the value of the LHS in the first iteration: n
- const SCEV* Start = AddRec->getOperand(0);
+ const SCEV *Start = AddRec->getOperand(0);
// Determine the minimum constant start value.
- const SCEV *MinStart = isa<SCEVConstant>(Start) ? Start :
- getConstant(isSigned ? APInt::getSignedMinValue(BitWidth) :
- APInt::getMinValue(BitWidth));
+ const SCEV *MinStart = getConstant(isSigned ?
+ getSignedRange(Start).getSignedMin() :
+ getUnsignedRange(Start).getUnsignedMin());
// If we know that the condition is true in order to enter the loop,
// then we know that it will run exactly (m-n)/s times. Otherwise, we
// only know that it will execute (max(m,n)-n)/s times. In both cases,
// the division must round up.
- const SCEV* End = RHS;
+ const SCEV *End = RHS;
if (!isLoopGuardedByCond(L,
- isSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT,
+ isSigned ? ICmpInst::ICMP_SLT :
+ ICmpInst::ICMP_ULT,
getMinusSCEV(Start, Step), RHS))
End = isSigned ? getSMaxExpr(RHS, Start)
: getUMaxExpr(RHS, Start);
// Determine the maximum constant end value.
- const SCEV* MaxEnd =
- isa<SCEVConstant>(End) ? End :
- getConstant(isSigned ? APInt::getSignedMaxValue(BitWidth)
- .ashr(GetMinSignBits(End) - 1) :
- APInt::getMaxValue(BitWidth)
- .lshr(GetMinLeadingZeros(End)));
+ const SCEV *MaxEnd = getConstant(isSigned ?
+ getSignedRange(End).getSignedMax() :
+ getUnsignedRange(End).getUnsignedMax());
// Finally, we subtract these two values and divide, rounding up, to get
// the number of times the backedge is executed.
- const SCEV* BECount = getBECount(Start, End, Step);
+ const SCEV *BECount = getBECount(Start, End, Step);
// The maximum backedge count is similar, except using the minimum start
// value and the maximum end value.
- const SCEV* MaxBECount = getBECount(MinStart, MaxEnd, Step);
+ const SCEV *MaxBECount = getBECount(MinStart, MaxEnd, Step);
return BackedgeTakenInfo(BECount, MaxBECount);
}
/// this is that it returns the first iteration number where the value is not in
/// the condition, thus computing the exit count. If the iteration count can't
/// be computed, an instance of SCEVCouldNotCompute is returned.
-const SCEV* SCEVAddRecExpr::getNumIterationsInRange(ConstantRange Range,
+const SCEV *SCEVAddRecExpr::getNumIterationsInRange(ConstantRange Range,
ScalarEvolution &SE) const {
if (Range.isFullSet()) // Infinite loop.
return SE.getCouldNotCompute();
// If the start is a non-zero constant, shift the range to simplify things.
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(getStart()))
if (!SC->getValue()->isZero()) {
- SmallVector<const SCEV*, 4> Operands(op_begin(), op_end());
+ SmallVector<const SCEV *, 4> Operands(op_begin(), op_end());
Operands[0] = SE.getIntegerSCEV(0, SC->getType());
- const SCEV* Shifted = SE.getAddRecExpr(Operands, getLoop());
+ const SCEV *Shifted = SE.getAddRecExpr(Operands, getLoop());
if (const SCEVAddRecExpr *ShiftedAddRec =
dyn_cast<SCEVAddRecExpr>(Shifted))
return ShiftedAddRec->getNumIterationsInRange(
// The exit value should be (End+A)/A.
APInt ExitVal = (End + A).udiv(A);
- ConstantInt *ExitValue = ConstantInt::get(ExitVal);
+ ConstantInt *ExitValue = SE.getContext()->getConstantInt(ExitVal);
// Evaluate at the exit value. If we really did fall out of the valid
// range, then we computed our trip count, otherwise wrap around or other
// Ensure that the previous value is in the range. This is a sanity check.
assert(Range.contains(
EvaluateConstantChrecAtConstant(this,
- ConstantInt::get(ExitVal - One), SE)->getValue()) &&
+ SE.getContext()->getConstantInt(ExitVal - One), SE)->getValue()) &&
"Linear scev computation is off in a bad way!");
return SE.getConstant(ExitValue);
} else if (isQuadratic()) {
// quadratic equation to solve it. To do this, we must frame our problem in
// terms of figuring out when zero is crossed, instead of when
// Range.getUpper() is crossed.
- SmallVector<const SCEV*, 4> NewOps(op_begin(), op_end());
+ SmallVector<const SCEV *, 4> NewOps(op_begin(), op_end());
NewOps[0] = SE.getNegativeSCEV(SE.getConstant(Range.getUpper()));
- const SCEV* NewAddRec = SE.getAddRecExpr(NewOps, getLoop());
+ const SCEV *NewAddRec = SE.getAddRecExpr(NewOps, getLoop());
// Next, solve the constructed addrec
- std::pair<const SCEV*,const SCEV*> Roots =
+ std::pair<const SCEV *,const SCEV *> Roots =
SolveQuadraticEquation(cast<SCEVAddRecExpr>(NewAddRec), SE);
const SCEVConstant *R1 = dyn_cast<SCEVConstant>(Roots.first);
const SCEVConstant *R2 = dyn_cast<SCEVConstant>(Roots.second);
if (R1) {
// Pick the smallest positive root value.
if (ConstantInt *CB =
- dyn_cast<ConstantInt>(ConstantExpr::getICmp(ICmpInst::ICMP_ULT,
- R1->getValue(), R2->getValue()))) {
+ dyn_cast<ConstantInt>(
+ SE.getContext()->getConstantExprICmp(ICmpInst::ICMP_ULT,
+ R1->getValue(), R2->getValue()))) {
if (CB->getZExtValue() == false)
std::swap(R1, R2); // R1 is the minimum root now.
SE);
if (Range.contains(R1Val->getValue())) {
// The next iteration must be out of the range...
- ConstantInt *NextVal = ConstantInt::get(R1->getValue()->getValue()+1);
+ ConstantInt *NextVal =
+ SE.getContext()->getConstantInt(R1->getValue()->getValue()+1);
R1Val = EvaluateConstantChrecAtConstant(this, NextVal, SE);
if (!Range.contains(R1Val->getValue()))
// If R1 was not in the range, then it is a good return value. Make
// sure that R1-1 WAS in the range though, just in case.
- ConstantInt *NextVal = ConstantInt::get(R1->getValue()->getValue()-1);
+ ConstantInt *NextVal =
+ SE.getContext()->getConstantInt(R1->getValue()->getValue()-1);
R1Val = EvaluateConstantChrecAtConstant(this, NextVal, SE);
if (Range.contains(R1Val->getValue()))
return R1;
//===----------------------------------------------------------------------===//
void ScalarEvolution::SCEVCallbackVH::deleted() {
- assert(SE && "SCEVCallbackVH called with a non-null ScalarEvolution!");
+ assert(SE && "SCEVCallbackVH called with a null ScalarEvolution!");
if (PHINode *PN = dyn_cast<PHINode>(getValPtr()))
SE->ConstantEvolutionLoopExitValue.erase(PN);
if (Instruction *I = dyn_cast<Instruction>(getValPtr()))
}
void ScalarEvolution::SCEVCallbackVH::allUsesReplacedWith(Value *) {
- assert(SE && "SCEVCallbackVH called with a non-null ScalarEvolution!");
+ assert(SE && "SCEVCallbackVH called with a null ScalarEvolution!");
// Forget all the expressions associated with users of the old value,
// so that future queries will recompute the expressions using the new
// value.
SmallVector<User *, 16> Worklist;
+ SmallPtrSet<User *, 8> Visited;
Value *Old = getValPtr();
bool DeleteOld = false;
for (Value::use_iterator UI = Old->use_begin(), UE = Old->use_end();
DeleteOld = true;
continue;
}
+ if (!Visited.insert(U))
+ continue;
if (PHINode *PN = dyn_cast<PHINode>(U))
SE->ConstantEvolutionLoopExitValue.erase(PN);
if (Instruction *I = dyn_cast<Instruction>(U))
SE->ValuesAtScopes.erase(I);
- if (SE->Scalars.erase(U))
- for (Value::use_iterator UI = U->use_begin(), UE = U->use_end();
- UI != UE; ++UI)
- Worklist.push_back(*UI);
+ SE->Scalars.erase(U);
+ for (Value::use_iterator UI = U->use_begin(), UE = U->use_end();
+ UI != UE; ++UI)
+ Worklist.push_back(*UI);
}
+ // Delete the Old value if it (indirectly) references itself.
if (DeleteOld) {
if (PHINode *PN = dyn_cast<PHINode>(Old))
SE->ConstantEvolutionLoopExitValue.erase(PN);
// out SCEV values of all instructions that are interesting. Doing
// this potentially causes it to create new SCEV objects though,
// which technically conflicts with the const qualifier. This isn't
- // observable from outside the class though (the hasSCEV function
- // notwithstanding), so casting away the const isn't dangerous.
+ // observable from outside the class though, so casting away the
+ // const isn't dangerous.
ScalarEvolution &SE = *const_cast<ScalarEvolution*>(this);
OS << "Classifying expressions for: " << F->getName() << "\n";
for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
if (isSCEVable(I->getType())) {
- OS << *I;
+ OS << *I << '\n';
OS << " --> ";
- const SCEV* SV = SE.getSCEV(&*I);
+ const SCEV *SV = SE.getSCEV(&*I);
SV->print(OS);
const Loop *L = LI->getLoopFor((*I).getParent());
- const SCEV* AtUse = SE.getSCEVAtScope(SV, L);
+ const SCEV *AtUse = SE.getSCEVAtScope(SV, L);
if (AtUse != SV) {
OS << " --> ";
AtUse->print(OS);
if (L) {
OS << "\t\t" "Exits: ";
- const SCEV* ExitValue = SE.getSCEVAtScope(SV, L->getParentLoop());
+ const SCEV *ExitValue = SE.getSCEVAtScope(SV, L->getParentLoop());
if (!ExitValue->isLoopInvariant(L)) {
OS << "<<Unknown>>";
} else {
projects / oota-llvm.git / blobdiff