}
bool SCEVNAryExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
- for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
- if (!getOperand(i)->dominates(BB, DT))
+ for (op_iterator I = op_begin(), E = op_end(); I != E; ++I)
+ if (!(*I)->dominates(BB, DT))
return false;
- }
return true;
}
bool SCEVNAryExpr::properlyDominates(BasicBlock *BB, DominatorTree *DT) const {
- for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
- if (!getOperand(i)->properlyDominates(BB, DT))
+ for (op_iterator I = op_begin(), E = op_end(); I != E; ++I)
+ if (!(*I)->properlyDominates(BB, DT))
+ return false;
+ return true;
+}
+
+bool SCEVNAryExpr::isLoopInvariant(const Loop *L) const {
+ for (op_iterator I = op_begin(), E = op_end(); I != E; ++I)
+ if (!(*I)->isLoopInvariant(L))
return false;
- }
return true;
}
+// hasComputableLoopEvolution - N-ary expressions have computable loop
+// evolutions iff they have at least one operand that varies with the loop,
+// but that all varying operands are computable.
+bool SCEVNAryExpr::hasComputableLoopEvolution(const Loop *L) const {
+ bool HasVarying = false;
+ for (op_iterator I = op_begin(), E = op_end(); I != E; ++I) {
+ const SCEV *S = *I;
+ if (!S->isLoopInvariant(L)) {
+ if (S->hasComputableLoopEvolution(L))
+ HasVarying = true;
+ else
+ return false;
+ }
+ }
+ return HasVarying;
+}
+
+bool SCEVNAryExpr::hasOperand(const SCEV *O) const {
+ for (op_iterator I = op_begin(), E = op_end(); I != E; ++I) {
+ const SCEV *S = *I;
+ if (O == S || S->hasOperand(O))
+ return true;
+ }
+ return false;
+}
+
bool SCEVUDivExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
return LHS->dominates(BB, DT) && RHS->dominates(BB, DT);
}
// This recurrence is variant w.r.t. QueryLoop if any of its operands
// are variant.
- for (unsigned i = 0, e = getNumOperands(); i != e; ++i)
- if (!getOperand(i)->isLoopInvariant(QueryLoop))
+ for (op_iterator I = op_begin(), E = op_end(); I != E; ++I)
+ if (!(*I)->isLoopInvariant(QueryLoop))
return false;
// Otherwise it's loop-invariant.
// SCEV Utilities
//===----------------------------------------------------------------------===//
-static bool CompareTypes(const Type *A, const Type *B) {
- if (A->getTypeID() != B->getTypeID())
- return A->getTypeID() < B->getTypeID();
- if (const IntegerType *AI = dyn_cast<IntegerType>(A)) {
- const IntegerType *BI = cast<IntegerType>(B);
- return AI->getBitWidth() < BI->getBitWidth();
- }
- if (const PointerType *AI = dyn_cast<PointerType>(A)) {
- const PointerType *BI = cast<PointerType>(B);
- return CompareTypes(AI->getElementType(), BI->getElementType());
- }
- if (const ArrayType *AI = dyn_cast<ArrayType>(A)) {
- const ArrayType *BI = cast<ArrayType>(B);
- if (AI->getNumElements() != BI->getNumElements())
- return AI->getNumElements() < BI->getNumElements();
- return CompareTypes(AI->getElementType(), BI->getElementType());
- }
- if (const VectorType *AI = dyn_cast<VectorType>(A)) {
- const VectorType *BI = cast<VectorType>(B);
- if (AI->getNumElements() != BI->getNumElements())
- return AI->getNumElements() < BI->getNumElements();
- return CompareTypes(AI->getElementType(), BI->getElementType());
- }
- if (const StructType *AI = dyn_cast<StructType>(A)) {
- const StructType *BI = cast<StructType>(B);
- if (AI->getNumElements() != BI->getNumElements())
- return AI->getNumElements() < BI->getNumElements();
- for (unsigned i = 0, e = AI->getNumElements(); i != e; ++i)
- if (CompareTypes(AI->getElementType(i), BI->getElementType(i)) ||
- CompareTypes(BI->getElementType(i), AI->getElementType(i)))
- return CompareTypes(AI->getElementType(i), BI->getElementType(i));
- }
- return false;
-}
-
namespace {
/// SCEVComplexityCompare - Return true if the complexity of the LHS is less
/// than the complexity of the RHS. This comparator is used to canonicalize
/// expressions.
class SCEVComplexityCompare {
- const LoopInfo *LI;
+ const LoopInfo *const LI;
public:
explicit SCEVComplexityCompare(const LoopInfo *li) : LI(li) {}
+ // Return true or false if LHS is less than, or at least RHS, respectively.
bool operator()(const SCEV *LHS, const SCEV *RHS) const {
+ return compare(LHS, RHS) < 0;
+ }
+
+ // Return negative, zero, or positive, if LHS is less than, equal to, or
+ // greater than RHS, respectively. A three-way result allows recursive
+ // comparisons to be more efficient.
+ int compare(const SCEV *LHS, const SCEV *RHS) const {
// Fast-path: SCEVs are uniqued so we can do a quick equality check.
if (LHS == RHS)
- return false;
+ return 0;
// Primarily, sort the SCEVs by their getSCEVType().
unsigned LType = LHS->getSCEVType(), RType = RHS->getSCEVType();
if (LType != RType)
- return LType < RType;
+ return (int)LType - (int)RType;
// Aside from the getSCEVType() ordering, the particular ordering
// isn't very important except that it's beneficial to be consistent,
// so that (a + b) and (b + a) don't end up as different expressions.
-
- // Sort SCEVUnknown values with some loose heuristics. TODO: This is
- // not as complete as it could be.
- if (const SCEVUnknown *LU = dyn_cast<SCEVUnknown>(LHS)) {
+ switch (LType) {
+ case scUnknown: {
+ const SCEVUnknown *LU = cast<SCEVUnknown>(LHS);
const SCEVUnknown *RU = cast<SCEVUnknown>(RHS);
+ // Sort SCEVUnknown values with some loose heuristics. TODO: This is
+ // not as complete as it could be.
+ const Value *LV = LU->getValue(), *RV = RU->getValue();
+
// Order pointer values after integer values. This helps SCEVExpander
// form GEPs.
- bool LIsPointer = LU->getType()->isPointerTy(),
- RIsPointer = RU->getType()->isPointerTy();
+ bool LIsPointer = LV->getType()->isPointerTy(),
+ RIsPointer = RV->getType()->isPointerTy();
if (LIsPointer != RIsPointer)
- return RIsPointer;
+ return (int)LIsPointer - (int)RIsPointer;
// Compare getValueID values.
- unsigned LID = LU->getValue()->getValueID(),
- RID = RU->getValue()->getValueID();
+ unsigned LID = LV->getValueID(),
+ RID = RV->getValueID();
if (LID != RID)
- return LID < RID;
+ return (int)LID - (int)RID;
// Sort arguments by their position.
- if (const Argument *LA = dyn_cast<Argument>(LU->getValue())) {
- const Argument *RA = cast<Argument>(RU->getValue());
- return LA->getArgNo() < RA->getArgNo();
+ if (const Argument *LA = dyn_cast<Argument>(LV)) {
+ const Argument *RA = cast<Argument>(RV);
+ unsigned LArgNo = LA->getArgNo(), RArgNo = RA->getArgNo();
+ return (int)LArgNo - (int)RArgNo;
}
- // For instructions, compare their loop depth, and their opcode.
- // This is pretty loose.
- if (const Instruction *LV = dyn_cast<Instruction>(LU->getValue())) {
- const Instruction *RV = cast<Instruction>(RU->getValue());
+ // For instructions, compare their loop depth, and their operand
+ // count. This is pretty loose.
+ if (const Instruction *LInst = dyn_cast<Instruction>(LV)) {
+ const Instruction *RInst = cast<Instruction>(RV);
// Compare loop depths.
- unsigned LDepth = LI->getLoopDepth(LV->getParent()),
- RDepth = LI->getLoopDepth(RV->getParent());
- if (LDepth != RDepth)
- return LDepth < RDepth;
+ const BasicBlock *LParent = LInst->getParent(),
+ *RParent = RInst->getParent();
+ if (LParent != RParent) {
+ unsigned LDepth = LI->getLoopDepth(LParent),
+ RDepth = LI->getLoopDepth(RParent);
+ if (LDepth != RDepth)
+ return (int)LDepth - (int)RDepth;
+ }
// Compare the number of operands.
- unsigned LNumOps = LV->getNumOperands(),
- RNumOps = RV->getNumOperands();
- if (LNumOps != RNumOps)
- return LNumOps < RNumOps;
+ unsigned LNumOps = LInst->getNumOperands(),
+ RNumOps = RInst->getNumOperands();
+ return (int)LNumOps - (int)RNumOps;
}
- return false;
+ return 0;
}
- // Compare constant values.
- if (const SCEVConstant *LC = dyn_cast<SCEVConstant>(LHS)) {
+ case scConstant: {
+ const SCEVConstant *LC = cast<SCEVConstant>(LHS);
const SCEVConstant *RC = cast<SCEVConstant>(RHS);
- const ConstantInt *LCC = LC->getValue();
- const ConstantInt *RCC = RC->getValue();
- unsigned LBitWidth = LCC->getBitWidth(), RBitWidth = RCC->getBitWidth();
+
+ // Compare constant values.
+ const APInt &LA = LC->getValue()->getValue();
+ const APInt &RA = RC->getValue()->getValue();
+ unsigned LBitWidth = LA.getBitWidth(), RBitWidth = RA.getBitWidth();
if (LBitWidth != RBitWidth)
- return LBitWidth < RBitWidth;
- return LCC->getValue().ult(RCC->getValue());
+ return (int)LBitWidth - (int)RBitWidth;
+ return LA.ult(RA) ? -1 : 1;
}
- // Compare addrec loop depths.
- if (const SCEVAddRecExpr *LA = dyn_cast<SCEVAddRecExpr>(LHS)) {
+ case scAddRecExpr: {
+ const SCEVAddRecExpr *LA = cast<SCEVAddRecExpr>(LHS);
const SCEVAddRecExpr *RA = cast<SCEVAddRecExpr>(RHS);
- unsigned LDepth = LA->getLoop()->getLoopDepth(),
- RDepth = RA->getLoop()->getLoopDepth();
- if (LDepth != RDepth)
- return LDepth < RDepth;
+
+ // Compare addrec loop depths.
+ const Loop *LLoop = LA->getLoop(), *RLoop = RA->getLoop();
+ if (LLoop != RLoop) {
+ unsigned LDepth = LLoop->getLoopDepth(),
+ RDepth = RLoop->getLoopDepth();
+ if (LDepth != RDepth)
+ return (int)LDepth - (int)RDepth;
+ }
+
+ // Addrec complexity grows with operand count.
+ unsigned LNumOps = LA->getNumOperands(), RNumOps = RA->getNumOperands();
+ if (LNumOps != RNumOps)
+ return (int)LNumOps - (int)RNumOps;
+
+ // Lexicographically compare.
+ for (unsigned i = 0; i != LNumOps; ++i) {
+ long X = compare(LA->getOperand(i), RA->getOperand(i));
+ if (X != 0)
+ return X;
+ }
+
+ return 0;
}
- // Lexicographically compare n-ary expressions.
- if (const SCEVNAryExpr *LC = dyn_cast<SCEVNAryExpr>(LHS)) {
+ case scAddExpr:
+ case scMulExpr:
+ case scSMaxExpr:
+ case scUMaxExpr: {
+ const SCEVNAryExpr *LC = cast<SCEVNAryExpr>(LHS);
const SCEVNAryExpr *RC = cast<SCEVNAryExpr>(RHS);
+
+ // Lexicographically compare n-ary expressions.
unsigned LNumOps = LC->getNumOperands(), RNumOps = RC->getNumOperands();
for (unsigned i = 0; i != LNumOps; ++i) {
if (i >= RNumOps)
- return false;
- const SCEV *LOp = LC->getOperand(i), *ROp = RC->getOperand(i);
- if (operator()(LOp, ROp))
- return true;
- if (operator()(ROp, LOp))
- return false;
+ return 1;
+ long X = compare(LC->getOperand(i), RC->getOperand(i));
+ if (X != 0)
+ return X;
}
- return LNumOps < RNumOps;
+ return (int)LNumOps - (int)RNumOps;
}
- // Lexicographically compare udiv expressions.
- if (const SCEVUDivExpr *LC = dyn_cast<SCEVUDivExpr>(LHS)) {
+ case scUDivExpr: {
+ const SCEVUDivExpr *LC = cast<SCEVUDivExpr>(LHS);
const SCEVUDivExpr *RC = cast<SCEVUDivExpr>(RHS);
- const SCEV *LL = LC->getLHS(), *LR = LC->getRHS(),
- *RL = RC->getLHS(), *RR = RC->getRHS();
- if (operator()(LL, RL))
- return true;
- if (operator()(RL, LL))
- return false;
- if (operator()(LR, RR))
- return true;
- if (operator()(RR, LR))
- return false;
- return false;
+
+ // Lexicographically compare udiv expressions.
+ long X = compare(LC->getLHS(), RC->getLHS());
+ if (X != 0)
+ return X;
+ return compare(LC->getRHS(), RC->getRHS());
}
- // Compare cast expressions by operand.
- if (const SCEVCastExpr *LC = dyn_cast<SCEVCastExpr>(LHS)) {
+ case scTruncate:
+ case scZeroExtend:
+ case scSignExtend: {
+ const SCEVCastExpr *LC = cast<SCEVCastExpr>(LHS);
const SCEVCastExpr *RC = cast<SCEVCastExpr>(RHS);
- return operator()(LC->getOperand(), RC->getOperand());
+
+ // Compare cast expressions by operand.
+ return compare(LC->getOperand(), RC->getOperand());
+ }
+
+ default:
+ break;
}
llvm_unreachable("Unknown SCEV kind!");
- return false;
+ return 0;
}
};
}
if (Ops.size() == 2) {
// This is the common case, which also happens to be trivially simple.
// Special case it.
- if (SCEVComplexityCompare(LI)(Ops[1], Ops[0]))
- std::swap(Ops[0], Ops[1]);
+ const SCEV *&LHS = Ops[0], *&RHS = Ops[1];
+ if (SCEVComplexityCompare(LI)(RHS, LHS))
+ std::swap(LHS, RHS);
return;
}
// If HasNSW is true and all the operands are non-negative, infer HasNUW.
if (!HasNUW && HasNSW) {
bool All = true;
- for (unsigned i = 0, e = Ops.size(); i != e; ++i)
- if (!isKnownNonNegative(Ops[i])) {
+ for (SmallVectorImpl<const SCEV *>::const_iterator I = Ops.begin(),
+ E = Ops.end(); I != E; ++I)
+ if (!isKnownNonNegative(*I)) {
All = false;
break;
}
if (Ops.size() == 1) return Ops[0];
}
- // Okay, check to see if the same value occurs in the operand list twice. If
- // so, merge them together into an multiply expression. Since we sorted the
- // list, these values are required to be adjacent.
+ // Okay, check to see if the same value occurs in the operand list more than
+ // once. If so, merge them together into an multiply expression. Since we
+ // sorted the list, these values are required to be adjacent.
const Type *Ty = Ops[0]->getType();
bool FoundMatch = false;
- for (unsigned i = 0, e = Ops.size()-1; i != e; ++i)
+ for (unsigned i = 0, e = Ops.size(); i != e-1; ++i)
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 = getConstant(Ty, 2);
- const SCEV *Mul = getMulExpr(Ops[i], Two);
- if (Ops.size() == 2)
+ // Scan ahead to count how many equal operands there are.
+ unsigned Count = 2;
+ while (i+Count != e && Ops[i+Count] == Ops[i])
+ ++Count;
+ // Merge the values into a multiply.
+ const SCEV *Scale = getConstant(Ty, Count);
+ const SCEV *Mul = getMulExpr(Scale, Ops[i]);
+ if (Ops.size() == Count)
return Mul;
Ops[i] = Mul;
- Ops.erase(Ops.begin()+i+1);
- --i; --e;
+ Ops.erase(Ops.begin()+i+1, Ops.begin()+i+Count);
+ --i; e -= Count - 1;
FoundMatch = true;
}
if (FoundMatch)
// 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(),
+ for (SmallVector<const SCEV *, 8>::const_iterator I = NewOps.begin(),
E = NewOps.end(); I != E; ++I)
MulOpLists[M.find(*I)->second].push_back(*I);
// Re-generate the operands list.
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());
- MulOps.erase(MulOps.begin()+MulOp);
+ SmallVector<const SCEV *, 4> MulOps(Mul->op_begin(),
+ Mul->op_begin()+MulOp);
+ MulOps.append(Mul->op_begin()+MulOp+1, Mul->op_end());
InnerMul = getMulExpr(MulOps);
}
const SCEV *One = getConstant(Ty, 1);
- const SCEV *AddOne = getAddExpr(InnerMul, One);
+ const SCEV *AddOne = getAddExpr(One, InnerMul);
const SCEV *OuterMul = getMulExpr(AddOne, MulOpSCEV);
if (Ops.size() == 2) return OuterMul;
if (AddOp < Idx) {
}
// Check this multiply against other multiplies being added together.
- bool AnyFold = false;
for (unsigned OtherMulIdx = Idx+1;
OtherMulIdx < Ops.size() && isa<SCEVMulExpr>(Ops[OtherMulIdx]);
++OtherMulIdx) {
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);
+ Mul->op_begin()+MulOp);
+ MulOps.append(Mul->op_begin()+MulOp+1, Mul->op_end());
InnerMul1 = getMulExpr(MulOps);
}
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);
+ OtherMul->op_begin()+OMulOp);
+ MulOps.append(OtherMul->op_begin()+OMulOp+1, OtherMul->op_end());
InnerMul2 = getMulExpr(MulOps);
}
const SCEV *InnerMulSum = getAddExpr(InnerMul1,InnerMul2);
const SCEV *OuterMul = getMulExpr(MulOpSCEV, InnerMulSum);
if (Ops.size() == 2) return OuterMul;
- Ops[Idx] = OuterMul;
- Ops.erase(Ops.begin()+OtherMulIdx);
- OtherMulIdx = Idx;
- AnyFold = true;
+ Ops.erase(Ops.begin()+Idx);
+ Ops.erase(Ops.begin()+OtherMulIdx-1);
+ Ops.push_back(OuterMul);
+ return getAddExpr(Ops);
}
}
- if (AnyFold)
- return getAddExpr(Ops);
}
}
// there are multiple AddRec's with the same loop induction variable being
// added together. If so, we can fold them.
for (unsigned OtherIdx = Idx+1;
- OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);++OtherIdx)
- if (OtherIdx != Idx) {
- const SCEVAddRecExpr *OtherAddRec = cast<SCEVAddRecExpr>(Ops[OtherIdx]);
- if (AddRecLoop == OtherAddRec->getLoop()) {
- // Other + {A,+,B} + {C,+,D} --> Other + {A+C,+,B+D}
- SmallVector<const SCEV *, 4> NewOps(AddRec->op_begin(),
- AddRec->op_end());
- for (unsigned i = 0, e = OtherAddRec->getNumOperands(); i != e; ++i) {
- if (i >= NewOps.size()) {
- NewOps.append(OtherAddRec->op_begin()+i,
- OtherAddRec->op_end());
- break;
+ OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);
+ ++OtherIdx)
+ if (AddRecLoop == cast<SCEVAddRecExpr>(Ops[OtherIdx])->getLoop()) {
+ // Other + {A,+,B}<L> + {C,+,D}<L> --> Other + {A+C,+,B+D}<L>
+ SmallVector<const SCEV *, 4> AddRecOps(AddRec->op_begin(),
+ AddRec->op_end());
+ for (; OtherIdx != Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);
+ ++OtherIdx)
+ if (const SCEVAddRecExpr *OtherAddRec =
+ dyn_cast<SCEVAddRecExpr>(Ops[OtherIdx]))
+ if (OtherAddRec->getLoop() == AddRecLoop) {
+ for (unsigned i = 0, e = OtherAddRec->getNumOperands();
+ i != e; ++i) {
+ if (i >= AddRecOps.size()) {
+ AddRecOps.append(OtherAddRec->op_begin()+i,
+ OtherAddRec->op_end());
+ break;
+ }
+ AddRecOps[i] = getAddExpr(AddRecOps[i],
+ OtherAddRec->getOperand(i));
+ }
+ Ops.erase(Ops.begin() + OtherIdx); --OtherIdx;
}
- NewOps[i] = getAddExpr(NewOps[i], OtherAddRec->getOperand(i));
- }
- const SCEV *NewAddRec = getAddRecExpr(NewOps, AddRecLoop);
-
- if (Ops.size() == 2) return NewAddRec;
-
- Ops.erase(Ops.begin()+Idx);
- Ops.erase(Ops.begin()+OtherIdx-1);
- Ops.push_back(NewAddRec);
- return getAddExpr(Ops);
- }
+ Ops[Idx] = getAddRecExpr(AddRecOps, AddRecLoop);
+ return getAddExpr(Ops);
}
// Otherwise couldn't fold anything into this recurrence. Move onto the
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scAddExpr);
- ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
assert(!Ops.empty() && "Cannot get empty mul!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
+ const Type *ETy = getEffectiveSCEVType(Ops[0]->getType());
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
- assert(getEffectiveSCEVType(Ops[i]->getType()) ==
- getEffectiveSCEVType(Ops[0]->getType()) &&
+ assert(getEffectiveSCEVType(Ops[i]->getType()) == ETy &&
"SCEVMulExpr operand types don't match!");
#endif
// If HasNSW is true and all the operands are non-negative, infer HasNUW.
if (!HasNUW && HasNSW) {
bool All = true;
- for (unsigned i = 0, e = Ops.size(); i != e; ++i)
- if (!isKnownNonNegative(Ops[i])) {
+ for (SmallVectorImpl<const SCEV *>::const_iterator I = Ops.begin(),
+ E = Ops.end(); I != E; ++I)
+ if (!isKnownNonNegative(*I)) {
All = false;
break;
}
// they are loop invariant w.r.t. the recurrence.
SmallVector<const SCEV *, 8> LIOps;
const SCEVAddRecExpr *AddRec = cast<SCEVAddRecExpr>(Ops[Idx]);
+ const Loop *AddRecLoop = AddRec->getLoop();
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
- if (Ops[i]->isLoopInvariant(AddRec->getLoop())) {
+ if (Ops[i]->isLoopInvariant(AddRecLoop)) {
LIOps.push_back(Ops[i]);
Ops.erase(Ops.begin()+i);
--i; --e;
// Build the new addrec. Propagate the NUW and NSW flags if both the
// outer mul and the inner addrec are guaranteed to have no overflow.
- const SCEV *NewRec = getAddRecExpr(NewOps, AddRec->getLoop(),
+ const SCEV *NewRec = getAddRecExpr(NewOps, AddRecLoop,
HasNUW && AddRec->hasNoUnsignedWrap(),
HasNSW && AddRec->hasNoSignedWrap());
// there are multiple AddRec's with the same loop induction variable being
// multiplied together. If so, we can fold them.
for (unsigned OtherIdx = Idx+1;
- OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);++OtherIdx)
- if (OtherIdx != Idx) {
- const SCEVAddRecExpr *OtherAddRec = cast<SCEVAddRecExpr>(Ops[OtherIdx]);
- 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(),
- G->getStart());
- 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,
- F->getLoop());
- if (Ops.size() == 2) return NewAddRec;
-
- Ops.erase(Ops.begin()+Idx);
- Ops.erase(Ops.begin()+OtherIdx-1);
- Ops.push_back(NewAddRec);
- return getMulExpr(Ops);
- }
+ OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);
+ ++OtherIdx)
+ if (AddRecLoop == cast<SCEVAddRecExpr>(Ops[OtherIdx])->getLoop()) {
+ // F * G, where F = {A,+,B}<L> and G = {C,+,D}<L> -->
+ // {A*C,+,F*D + G*B + B*D}<L>
+ for (; OtherIdx != Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);
+ ++OtherIdx)
+ if (const SCEVAddRecExpr *OtherAddRec =
+ dyn_cast<SCEVAddRecExpr>(Ops[OtherIdx]))
+ if (OtherAddRec->getLoop() == AddRecLoop) {
+ const SCEVAddRecExpr *F = AddRec, *G = OtherAddRec;
+ 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),
+ getMulExpr(G, B),
+ getMulExpr(B, D));
+ const SCEV *NewAddRec = getAddRecExpr(NewStart, NewStep,
+ F->getLoop());
+ if (Ops.size() == 2) return NewAddRec;
+ Ops[Idx] = AddRec = cast<SCEVAddRecExpr>(NewAddRec);
+ Ops.erase(Ops.begin() + OtherIdx); --OtherIdx;
+ }
+ return getMulExpr(Ops);
}
// Otherwise couldn't fold anything into this recurrence. Move onto the
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scMulExpr);
- ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
bool HasNUW, bool HasNSW) {
if (Operands.size() == 1) return Operands[0];
#ifndef NDEBUG
+ const Type *ETy = getEffectiveSCEVType(Operands[0]->getType());
for (unsigned i = 1, e = Operands.size(); i != e; ++i)
- assert(getEffectiveSCEVType(Operands[i]->getType()) ==
- getEffectiveSCEVType(Operands[0]->getType()) &&
+ assert(getEffectiveSCEVType(Operands[i]->getType()) == ETy &&
"SCEVAddRecExpr operand types don't match!");
#endif
// If HasNSW is true and all the operands are non-negative, infer HasNUW.
if (!HasNUW && HasNSW) {
bool All = true;
- for (unsigned i = 0, e = Operands.size(); i != e; ++i)
- if (!isKnownNonNegative(Operands[i])) {
+ for (SmallVectorImpl<const SCEV *>::const_iterator I = Operands.begin(),
+ E = Operands.end(); I != E; ++I)
+ if (!isKnownNonNegative(*I)) {
All = false;
break;
}
// Canonicalize nested AddRecs in by nesting them in order of loop depth.
if (const SCEVAddRecExpr *NestedAR = dyn_cast<SCEVAddRecExpr>(Operands[0])) {
const Loop *NestedLoop = NestedAR->getLoop();
- if (L->contains(NestedLoop->getHeader()) ?
+ if (L->contains(NestedLoop) ?
(L->getLoopDepth() < NestedLoop->getLoopDepth()) :
- (!NestedLoop->contains(L->getHeader()) &&
+ (!NestedLoop->contains(L) &&
DT->dominates(L->getHeader(), NestedLoop->getHeader()))) {
SmallVector<const SCEV *, 4> NestedOperands(NestedAR->op_begin(),
NestedAR->op_end());
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scAddRecExpr);
- ID.AddInteger(Operands.size());
for (unsigned i = 0, e = Operands.size(); i != e; ++i)
ID.AddPointer(Operands[i]);
ID.AddPointer(L);
assert(!Ops.empty() && "Cannot get empty smax!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
+ const Type *ETy = getEffectiveSCEVType(Ops[0]->getType());
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
- assert(getEffectiveSCEVType(Ops[i]->getType()) ==
- getEffectiveSCEVType(Ops[0]->getType()) &&
+ assert(getEffectiveSCEVType(Ops[i]->getType()) == ETy &&
"SCEVSMaxExpr operand types don't match!");
#endif
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scSMaxExpr);
- ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
assert(!Ops.empty() && "Cannot get empty umax!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
+ const Type *ETy = getEffectiveSCEVType(Ops[0]->getType());
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
- assert(getEffectiveSCEVType(Ops[i]->getType()) ==
- getEffectiveSCEVType(Ops[0]->getType()) &&
+ assert(getEffectiveSCEVType(Ops[i]->getType()) == ETy &&
"SCEVUMaxExpr operand types don't match!");
#endif
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scUMaxExpr);
- ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
const SCEV *ScalarEvolution::getSCEV(Value *V) {
assert(isSCEVable(V->getType()) && "Value is not SCEVable!");
- std::map<SCEVCallbackVH, const SCEV *>::iterator I = Scalars.find(V);
- if (I != Scalars.end()) return I->second;
+ ValueExprMapType::const_iterator I = ValueExprMap.find(V);
+ if (I != ValueExprMap.end()) return I->second;
const SCEV *S = createSCEV(V);
- Scalars.insert(std::make_pair(SCEVCallbackVH(V, this), S));
+
+ // The process of creating a SCEV for V may have caused other SCEVs
+ // to have been created, so it's necessary to insert the new entry
+ // from scratch, rather than trying to remember the insert position
+ // above.
+ ValueExprMap.insert(std::make_pair(SCEVCallbackVH(V, this), S));
return S;
}
/// ForgetSymbolicValue - This looks up computed SCEV values for all
/// instructions that depend on the given instruction and removes them from
-/// the Scalars map if they reference SymName. This is used during PHI
+/// the ValueExprMapType map if they reference SymName. This is used during PHI
/// resolution.
void
ScalarEvolution::ForgetSymbolicName(Instruction *PN, const SCEV *SymName) {
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()) {
+ ValueExprMapType::iterator It =
+ ValueExprMap.find(static_cast<Value *>(I));
+ if (It != ValueExprMap.end()) {
// Short-circuit the def-use traversal if the symbolic name
// ceases to appear in expressions.
if (It->second != SymName && !It->second->hasOperand(SymName))
!isa<SCEVUnknown>(It->second) ||
(I != PN && It->second == SymName)) {
ValuesAtScopes.erase(It->second);
- Scalars.erase(It);
+ ValueExprMap.erase(It);
}
}
if (BEValueV && StartValueV) {
// While we are analyzing this PHI node, handle its value symbolically.
const SCEV *SymbolicName = getUnknown(PN);
- assert(Scalars.find(PN) == Scalars.end() &&
+ assert(ValueExprMap.find(PN) == ValueExprMap.end() &&
"PHI node already processed?");
- Scalars.insert(std::make_pair(SCEVCallbackVH(PN, this), SymbolicName));
+ ValueExprMap.insert(std::make_pair(SCEVCallbackVH(PN, this), SymbolicName));
// Using this symbolic name for the PHI, analyze the value coming around
// the back-edge.
// to be symbolic. We now need to go back and purge all of the
// entries for the scalars that use the symbolic expression.
ForgetSymbolicName(PN, SymbolicName);
- Scalars[SCEVCallbackVH(PN, this)] = PHISCEV;
+ ValueExprMap[SCEVCallbackVH(PN, this)] = PHISCEV;
return PHISCEV;
}
}
// to be symbolic. We now need to go back and purge all of the
// entries for the scalars that use the symbolic expression.
ForgetSymbolicName(PN, SymbolicName);
- Scalars[SCEVCallbackVH(PN, this)] = PHISCEV;
+ ValueExprMap[SCEVCallbackVH(PN, this)] = PHISCEV;
return PHISCEV;
}
}
Operator *U = cast<Operator>(V);
switch (Opcode) {
- case Instruction::Add:
- return getAddExpr(getSCEV(U->getOperand(0)),
- getSCEV(U->getOperand(1)));
- case Instruction::Mul:
- return getMulExpr(getSCEV(U->getOperand(0)),
- getSCEV(U->getOperand(1)));
+ case Instruction::Add: {
+ // The simple thing to do would be to just call getSCEV on both operands
+ // and call getAddExpr with the result. However if we're looking at a
+ // bunch of things all added together, this can be quite inefficient,
+ // because it leads to N-1 getAddExpr calls for N ultimate operands.
+ // Instead, gather up all the operands and make a single getAddExpr call.
+ // LLVM IR canonical form means we need only traverse the left operands.
+ SmallVector<const SCEV *, 4> AddOps;
+ AddOps.push_back(getSCEV(U->getOperand(1)));
+ for (Value *Op = U->getOperand(0); ; Op = U->getOperand(0)) {
+ unsigned Opcode = Op->getValueID() - Value::InstructionVal;
+ if (Opcode != Instruction::Add && Opcode != Instruction::Sub)
+ break;
+ U = cast<Operator>(Op);
+ const SCEV *Op1 = getSCEV(U->getOperand(1));
+ if (Opcode == Instruction::Sub)
+ AddOps.push_back(getNegativeSCEV(Op1));
+ else
+ AddOps.push_back(Op1);
+ }
+ AddOps.push_back(getSCEV(U->getOperand(0)));
+ return getAddExpr(AddOps);
+ }
+ case Instruction::Mul: {
+ // See the Add code above.
+ SmallVector<const SCEV *, 4> MulOps;
+ MulOps.push_back(getSCEV(U->getOperand(1)));
+ for (Value *Op = U->getOperand(0);
+ Op->getValueID() == Instruction::Mul + Value::InstructionVal;
+ Op = U->getOperand(0)) {
+ U = cast<Operator>(Op);
+ MulOps.push_back(getSCEV(U->getOperand(1)));
+ }
+ MulOps.push_back(getSCEV(U->getOperand(0)));
+ return getMulExpr(MulOps);
+ }
case Instruction::UDiv:
return getUDivExpr(getSCEV(U->getOperand(0)),
getSCEV(U->getOperand(1)));
const SCEV *LDiff = getMinusSCEV(LA, LS);
const SCEV *RDiff = getMinusSCEV(RA, One);
if (LDiff == RDiff)
- return getAddExpr(getUMaxExpr(LS, One), LDiff);
+ return getAddExpr(getUMaxExpr(One, LS), LDiff);
}
break;
case ICmpInst::ICMP_EQ:
const SCEV *LDiff = getMinusSCEV(LA, One);
const SCEV *RDiff = getMinusSCEV(RA, LS);
if (LDiff == RDiff)
- return getAddExpr(getUMaxExpr(LS, One), LDiff);
+ return getAddExpr(getUMaxExpr(One, LS), LDiff);
}
break;
default:
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()) {
+ ValueExprMapType::iterator It =
+ ValueExprMap.find(static_cast<Value *>(I));
+ if (It != ValueExprMap.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
// own when it gets to that point.
if (!isa<PHINode>(I) || !isa<SCEVUnknown>(It->second)) {
ValuesAtScopes.erase(It->second);
- Scalars.erase(It);
+ ValueExprMap.erase(It);
}
if (PHINode *PN = dyn_cast<PHINode>(I))
ConstantEvolutionLoopExitValue.erase(PN);
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()) {
+ ValueExprMapType::iterator It = ValueExprMap.find(static_cast<Value *>(I));
+ if (It != ValueExprMap.end()) {
ValuesAtScopes.erase(It->second);
- Scalars.erase(It);
+ ValueExprMap.erase(It);
if (PHINode *PN = dyn_cast<PHINode>(I))
ConstantEvolutionLoopExitValue.erase(PN);
}
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()) {
+ ValueExprMapType::iterator It = ValueExprMap.find(static_cast<Value *>(I));
+ if (It != ValueExprMap.end()) {
ValuesAtScopes.erase(It->second);
- Scalars.erase(It);
+ ValueExprMap.erase(It);
if (PHINode *PN = dyn_cast<PHINode>(I))
ConstantEvolutionLoopExitValue.erase(PN);
}
ScalarEvolution::getConstantEvolutionLoopExitValue(PHINode *PN,
const APInt &BEs,
const Loop *L) {
- std::map<PHINode*, Constant*>::iterator I =
+ std::map<PHINode*, Constant*>::const_iterator I =
ConstantEvolutionLoopExitValue.find(PN);
if (I != ConstantEvolutionLoopExitValue.end())
return I->second;
assert(SE && "SCEVCallbackVH called with a null ScalarEvolution!");
if (PHINode *PN = dyn_cast<PHINode>(getValPtr()))
SE->ConstantEvolutionLoopExitValue.erase(PN);
- SE->Scalars.erase(getValPtr());
+ SE->ValueExprMap.erase(getValPtr());
// this now dangles!
}
continue;
if (PHINode *PN = dyn_cast<PHINode>(U))
SE->ConstantEvolutionLoopExitValue.erase(PN);
- SE->Scalars.erase(U);
+ SE->ValueExprMap.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 (PHINode *PN = dyn_cast<PHINode>(Old))
SE->ConstantEvolutionLoopExitValue.erase(PN);
- SE->Scalars.erase(Old);
+ SE->ValueExprMap.erase(Old);
// this now dangles!
}
U->~SCEVUnknown();
FirstUnknown = 0;
- Scalars.clear();
+ ValueExprMap.clear();
BackedgeTakenCounts.clear();
ConstantEvolutionLoopExitValue.clear();
ValuesAtScopes.clear();