/// AddToWorkList - Add the specified instruction to the worklist if it
/// isn't already in it.
void AddToWorkList(Instruction *I) {
- if (WorklistMap.insert(std::make_pair(I, Worklist.size())))
+ if (WorklistMap.insert(std::make_pair(I, Worklist.size())).second)
Worklist.push_back(I);
}
/// the work lists because they might get more simplified now.
///
void AddUsesToWorkList(Instruction &I) {
- for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i)
- if (Instruction *Op = dyn_cast<Instruction>(I.getOperand(i)))
+ for (User::op_iterator i = I.op_begin(), e = I.op_end(); i != e; ++i)
+ if (Instruction *Op = dyn_cast<Instruction>(*i))
AddToWorkList(Op);
}
Value *AddSoonDeadInstToWorklist(Instruction &I, unsigned op) {
Value *R = I.getOperand(op);
- for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i)
- if (Instruction *Op = dyn_cast<Instruction>(I.getOperand(i))) {
+ for (User::op_iterator i = I.op_begin(), e = I.op_end(); i != e; ++i)
+ if (Instruction *Op = dyn_cast<Instruction>(*i)) {
AddToWorkList(Op);
// Set the operand to undef to drop the use.
- I.setOperand(i, UndefValue::get(Op->getType()));
+ *i = UndefValue::get(Op->getType());
}
return R;
Instruction *visitInsertElementInst(InsertElementInst &IE);
Instruction *visitExtractElementInst(ExtractElementInst &EI);
Instruction *visitShuffleVectorInst(ShuffleVectorInst &SVI);
+ Instruction *visitExtractValueInst(ExtractValueInst &EV);
// visitInstruction - Specify what to return for unhandled instructions...
Instruction *visitInstruction(Instruction &I) { return 0; }
int &NumCastsRemoved);
unsigned GetOrEnforceKnownAlignment(Value *V,
unsigned PrefAlign = 0);
+
};
}
// If the input sign bit is known to be zero, or if none of the top bits
// are demanded, turn this into an unsigned shift right.
- if (RHSKnownZero[BitWidth-ShiftAmt-1] ||
+ if (BitWidth <= ShiftAmt || RHSKnownZero[BitWidth-ShiftAmt-1] ||
(HighBits & ~DemandedMask) == HighBits) {
// Perform the logical shift right.
Value *NewVal = BinaryOperator::CreateLShr(
KnownZero = APInt::getHighBitsSet(BitWidth, Leaders) & DemandedMask;
break;
}
+ case Instruction::Call:
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+ switch (II->getIntrinsicID()) {
+ default: break;
+ case Intrinsic::bswap: {
+ // If the only bits demanded come from one byte of the bswap result,
+ // just shift the input byte into position to eliminate the bswap.
+ unsigned NLZ = DemandedMask.countLeadingZeros();
+ unsigned NTZ = DemandedMask.countTrailingZeros();
+
+ // Round NTZ down to the next byte. If we have 11 trailing zeros, then
+ // we need all the bits down to bit 8. Likewise, round NLZ. If we
+ // have 14 leading zeros, round to 8.
+ NLZ &= ~7;
+ NTZ &= ~7;
+ // If we need exactly one byte, we can do this transformation.
+ if (BitWidth-NLZ-NTZ == 8) {
+ unsigned ResultBit = NTZ;
+ unsigned InputBit = BitWidth-NTZ-8;
+
+ // Replace this with either a left or right shift to get the byte into
+ // the right place.
+ Instruction *NewVal;
+ if (InputBit > ResultBit)
+ NewVal = BinaryOperator::CreateLShr(I->getOperand(1),
+ ConstantInt::get(I->getType(), InputBit-ResultBit));
+ else
+ NewVal = BinaryOperator::CreateShl(I->getOperand(1),
+ ConstantInt::get(I->getType(), ResultBit-InputBit));
+ NewVal->takeName(I);
+ InsertNewInstBefore(NewVal, *I);
+ return UpdateValueUsesWith(I, NewVal);
+ }
+
+ // TODO: Could compute known zero/one bits based on the input.
+ break;
+ }
+ }
+ }
+ ComputeMaskedBits(V, DemandedMask, RHSKnownZero, RHSKnownOne, Depth);
+ break;
}
// If the client is only demanding bits that we know, return the known
// If the functor wants to apply the optimization to the RHS of LHSI,
// reassociate the expression from ((? op A) op B) to (? op (A op B))
if (ShouldApply) {
- BasicBlock *BB = Root.getParent();
-
// Now all of the instructions are in the current basic block, go ahead
// and perform the reassociation.
Instruction *TmpLHSI = cast<Instruction>(Root.getOperand(0));
}
Root.replaceAllUsesWith(TmpLHSI); // Users now use TmpLHSI
TmpLHSI->setOperand(1, &Root); // TmpLHSI now uses the root
- TmpLHSI->getParent()->getInstList().remove(TmpLHSI);
BasicBlock::iterator ARI = &Root; ++ARI;
- BB->getInstList().insert(ARI, TmpLHSI); // Move TmpLHSI to after Root
+ TmpLHSI->moveBefore(ARI); // Move TmpLHSI to after Root
ARI = Root;
// Now propagate the ExtraOperand down the chain of instructions until we
Instruction *NextLHSI = cast<Instruction>(TmpLHSI->getOperand(0));
// Move the instruction to immediately before the chain we are
// constructing to avoid breaking dominance properties.
- NextLHSI->getParent()->getInstList().remove(NextLHSI);
- BB->getInstList().insert(ARI, NextLHSI);
+ NextLHSI->moveBefore(ARI);
ARI = NextLHSI;
Value *NextOp = NextLHSI->getOperand(1);
// ((A ^ N) - B) & AndRHS -> (A - B) & AndRHS iff N&AndRHS == 0
if (Value *V = FoldLogicalPlusAnd(Op0LHS, Op0RHS, AndRHS, true, I))
return BinaryOperator::CreateAnd(V, AndRHS);
+
+ // (A - N) & AndRHS -> -N & AndRHS where A & AndRHS == 0
+ if (Op0I->hasOneUse() && MaskedValueIsZero(Op0LHS, AndRHSMask)) {
+ ConstantInt *A = dyn_cast<ConstantInt>(Op0LHS);
+ if (!A || !A->isZero()) {
+ Instruction *NewNeg = BinaryOperator::CreateNeg(Op0RHS);
+ InsertNewInstBefore(NewNeg, I);
+ return BinaryOperator::CreateAnd(NewNeg, AndRHS);
+ }
+ }
+
break;
}
case ICmpInst::ICMP_UGT:
switch (RHSCC) {
default: assert(0 && "Unknown integer condition code!");
- case ICmpInst::ICMP_EQ: // (X u> 13 & X == 15) -> X > 13
- return ReplaceInstUsesWith(I, LHS);
+ case ICmpInst::ICMP_EQ: // (X u> 13 & X == 15) -> X == 15
case ICmpInst::ICMP_UGT: // (X u> 13 & X u> 15) -> X u> 15
return ReplaceInstUsesWith(I, RHS);
case ICmpInst::ICMP_SGT: // (X u> 13 & X s> 15) -> no change
}
}
}
-
+
return Changed ? &I : 0;
}
unsigned IntPtrWidth = TD.getPointerSizeInBits();
uint64_t PtrSizeMask = ~0ULL >> (64-IntPtrWidth);
- for (unsigned i = 1, e = GEP->getNumOperands(); i != e; ++i, ++GTI) {
- Value *Op = GEP->getOperand(i);
+ for (User::op_iterator i = GEP->op_begin() + 1, e = GEP->op_end(); i != e;
+ ++i, ++GTI) {
+ Value *Op = *i;
uint64_t Size = TD.getABITypeSize(GTI.getIndexedType()) & PtrSizeMask;
if (ConstantInt *OpC = dyn_cast<ConstantInt>(Op)) {
if (OpC->isZero()) continue;
if (Instruction *LHSI = dyn_cast<Instruction>(Op0))
switch (LHSI->getOpcode()) {
case Instruction::PHI:
- if (Instruction *NV = FoldOpIntoPhi(I))
- return NV;
+ // Only fold fcmp into the PHI if the phi and fcmp are in the same
+ // block. If in the same block, we're encouraging jump threading. If
+ // not, we are just pessimizing the code by making an i1 phi.
+ if (LHSI->getParent() == I.getParent())
+ if (Instruction *NV = FoldOpIntoPhi(I))
+ return NV;
break;
case Instruction::SIToFP:
case Instruction::UIToFP:
break;
case Instruction::PHI:
- if (Instruction *NV = FoldOpIntoPhi(I))
- return NV;
+ // Only fold icmp into the PHI if the phi and fcmp are in the same
+ // block. If in the same block, we're encouraging jump threading. If
+ // not, we are just pessimizing the code by making an i1 phi.
+ if (LHSI->getParent() == I.getParent())
+ if (Instruction *NV = FoldOpIntoPhi(I))
+ return NV;
break;
case Instruction::Select: {
// If either operand of the select is a constant, we can fold the
///
/// This is a truncation operation if Ty is smaller than V->getType(), or an
/// extension operation if Ty is larger.
+///
+/// If CastOpc is a truncation, then Ty will be a type smaller than V. We
+/// should return true if trunc(V) can be computed by computing V in the smaller
+/// type. If V is an instruction, then trunc(inst(x,y)) can be computed as
+/// inst(trunc(x),trunc(y)), which only makes sense if x and y can be
+/// efficiently truncated.
+///
+/// If CastOpc is a sext or zext, we are asking if the low bits of the value can
+/// bit computed in a larger type, which is then and'd or sext_in_reg'd to get
+/// the final result.
bool InstCombiner::CanEvaluateInDifferentType(Value *V, const IntegerType *Ty,
unsigned CastOpc,
int &NumCastsRemoved) {
// If the first operand is itself a cast, and is eliminable, do not count
// this as an eliminable cast. We would prefer to eliminate those two
// casts first.
- if (!isa<CastInst>(I->getOperand(0)))
+ if (!isa<CastInst>(I->getOperand(0)) && I->hasOneUse())
++NumCastsRemoved;
return true;
}
switch (I->getOpcode()) {
case Instruction::Add:
case Instruction::Sub:
+ case Instruction::Mul:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor:
CanEvaluateInDifferentType(I->getOperand(1), Ty, CastOpc,
NumCastsRemoved);
- case Instruction::Mul:
- // A multiply can be truncated by truncating its operands.
- return Ty->getBitWidth() < OrigTy->getBitWidth() &&
- CanEvaluateInDifferentType(I->getOperand(0), Ty, CastOpc,
- NumCastsRemoved) &&
- CanEvaluateInDifferentType(I->getOperand(1), Ty, CastOpc,
- NumCastsRemoved);
-
case Instruction::Shl:
// If we are truncating the result of this SHL, and if it's a shift of a
// constant amount, we can always perform a SHL in a smaller type.
// of casts in the input.
if (I->getOpcode() == CastOpc)
return true;
-
break;
+ case Instruction::Select: {
+ SelectInst *SI = cast<SelectInst>(I);
+ return CanEvaluateInDifferentType(SI->getTrueValue(), Ty, CastOpc,
+ NumCastsRemoved) &&
+ CanEvaluateInDifferentType(SI->getFalseValue(), Ty, CastOpc,
+ NumCastsRemoved);
+ }
+ case Instruction::PHI: {
+ // We can change a phi if we can change all operands.
+ PHINode *PN = cast<PHINode>(I);
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
+ if (!CanEvaluateInDifferentType(PN->getIncomingValue(i), Ty, CastOpc,
+ NumCastsRemoved))
+ return false;
+ return true;
+ }
default:
// TODO: Can handle more cases here.
break;
Value *LHS = EvaluateInDifferentType(I->getOperand(0), Ty, isSigned);
Value *RHS = EvaluateInDifferentType(I->getOperand(1), Ty, isSigned);
Res = BinaryOperator::Create((Instruction::BinaryOps)I->getOpcode(),
- LHS, RHS, I->getName());
+ LHS, RHS);
break;
}
case Instruction::Trunc:
if (I->getOperand(0)->getType() == Ty)
return I->getOperand(0);
- // Otherwise, must be the same type of case, so just reinsert a new one.
+ // Otherwise, must be the same type of cast, so just reinsert a new one.
Res = CastInst::Create(cast<CastInst>(I)->getOpcode(), I->getOperand(0),
- Ty, I->getName());
+ Ty);
break;
+ case Instruction::Select: {
+ Value *True = EvaluateInDifferentType(I->getOperand(1), Ty, isSigned);
+ Value *False = EvaluateInDifferentType(I->getOperand(2), Ty, isSigned);
+ Res = SelectInst::Create(I->getOperand(0), True, False);
+ break;
+ }
+ case Instruction::PHI: {
+ PHINode *OPN = cast<PHINode>(I);
+ PHINode *NPN = PHINode::Create(Ty);
+ for (unsigned i = 0, e = OPN->getNumIncomingValues(); i != e; ++i) {
+ Value *V =EvaluateInDifferentType(OPN->getIncomingValue(i), Ty, isSigned);
+ NPN->addIncoming(V, OPN->getIncomingBlock(i));
+ }
+ Res = NPN;
+ break;
+ }
default:
// TODO: Can handle more cases here.
assert(0 && "Unreachable!");
break;
}
+ Res->takeName(I);
return InsertNewInstBefore(Res, *I);
}
case Instruction::GetElementPtr: {
// If all indexes are zero, it is just the alignment of the base pointer.
bool AllZeroOperands = true;
- for (unsigned i = 1, e = U->getNumOperands(); i != e; ++i)
- if (!isa<Constant>(U->getOperand(i)) ||
- !cast<Constant>(U->getOperand(i))->isNullValue()) {
+ for (User::op_iterator i = U->op_begin() + 1, e = U->op_end(); i != e; ++i)
+ if (!isa<Constant>(*i) ||
+ !cast<Constant>(*i)->isNullValue()) {
AllZeroOperands = false;
break;
}
}
if (Changed) return II;
- } else {
- switch (II->getIntrinsicID()) {
- default: break;
- case Intrinsic::ppc_altivec_lvx:
- case Intrinsic::ppc_altivec_lvxl:
- case Intrinsic::x86_sse_loadu_ps:
- case Intrinsic::x86_sse2_loadu_pd:
- case Intrinsic::x86_sse2_loadu_dq:
- // Turn PPC lvx -> load if the pointer is known aligned.
- // Turn X86 loadups -> load if the pointer is known aligned.
- if (GetOrEnforceKnownAlignment(II->getOperand(1), 16) >= 16) {
- Value *Ptr = InsertBitCastBefore(II->getOperand(1),
- PointerType::getUnqual(II->getType()),
- CI);
- return new LoadInst(Ptr);
- }
- break;
- case Intrinsic::ppc_altivec_stvx:
- case Intrinsic::ppc_altivec_stvxl:
- // Turn stvx -> store if the pointer is known aligned.
- if (GetOrEnforceKnownAlignment(II->getOperand(2), 16) >= 16) {
- const Type *OpPtrTy =
- PointerType::getUnqual(II->getOperand(1)->getType());
- Value *Ptr = InsertBitCastBefore(II->getOperand(2), OpPtrTy, CI);
- return new StoreInst(II->getOperand(1), Ptr);
- }
- break;
- case Intrinsic::x86_sse_storeu_ps:
- case Intrinsic::x86_sse2_storeu_pd:
- case Intrinsic::x86_sse2_storeu_dq:
- case Intrinsic::x86_sse2_storel_dq:
- // Turn X86 storeu -> store if the pointer is known aligned.
- if (GetOrEnforceKnownAlignment(II->getOperand(1), 16) >= 16) {
- const Type *OpPtrTy =
- PointerType::getUnqual(II->getOperand(2)->getType());
- Value *Ptr = InsertBitCastBefore(II->getOperand(1), OpPtrTy, CI);
- return new StoreInst(II->getOperand(2), Ptr);
- }
- break;
+ }
+
+ switch (II->getIntrinsicID()) {
+ default: break;
+ case Intrinsic::bswap:
+ // bswap(bswap(x)) -> x
+ if (IntrinsicInst *Operand = dyn_cast<IntrinsicInst>(II->getOperand(1)))
+ if (Operand->getIntrinsicID() == Intrinsic::bswap)
+ return ReplaceInstUsesWith(CI, Operand->getOperand(1));
+ break;
+ case Intrinsic::ppc_altivec_lvx:
+ case Intrinsic::ppc_altivec_lvxl:
+ case Intrinsic::x86_sse_loadu_ps:
+ case Intrinsic::x86_sse2_loadu_pd:
+ case Intrinsic::x86_sse2_loadu_dq:
+ // Turn PPC lvx -> load if the pointer is known aligned.
+ // Turn X86 loadups -> load if the pointer is known aligned.
+ if (GetOrEnforceKnownAlignment(II->getOperand(1), 16) >= 16) {
+ Value *Ptr = InsertBitCastBefore(II->getOperand(1),
+ PointerType::getUnqual(II->getType()),
+ CI);
+ return new LoadInst(Ptr);
+ }
+ break;
+ case Intrinsic::ppc_altivec_stvx:
+ case Intrinsic::ppc_altivec_stvxl:
+ // Turn stvx -> store if the pointer is known aligned.
+ if (GetOrEnforceKnownAlignment(II->getOperand(2), 16) >= 16) {
+ const Type *OpPtrTy =
+ PointerType::getUnqual(II->getOperand(1)->getType());
+ Value *Ptr = InsertBitCastBefore(II->getOperand(2), OpPtrTy, CI);
+ return new StoreInst(II->getOperand(1), Ptr);
+ }
+ break;
+ case Intrinsic::x86_sse_storeu_ps:
+ case Intrinsic::x86_sse2_storeu_pd:
+ case Intrinsic::x86_sse2_storeu_dq:
+ case Intrinsic::x86_sse2_storel_dq:
+ // Turn X86 storeu -> store if the pointer is known aligned.
+ if (GetOrEnforceKnownAlignment(II->getOperand(1), 16) >= 16) {
+ const Type *OpPtrTy =
+ PointerType::getUnqual(II->getOperand(2)->getType());
+ Value *Ptr = InsertBitCastBefore(II->getOperand(1), OpPtrTy, CI);
+ return new StoreInst(II->getOperand(2), Ptr);
+ }
+ break;
+
+ case Intrinsic::x86_sse_cvttss2si: {
+ // These intrinsics only demands the 0th element of its input vector. If
+ // we can simplify the input based on that, do so now.
+ uint64_t UndefElts;
+ if (Value *V = SimplifyDemandedVectorElts(II->getOperand(1), 1,
+ UndefElts)) {
+ II->setOperand(1, V);
+ return II;
+ }
+ break;
+ }
+
+ case Intrinsic::ppc_altivec_vperm:
+ // Turn vperm(V1,V2,mask) -> shuffle(V1,V2,mask) if mask is a constant.
+ if (ConstantVector *Mask = dyn_cast<ConstantVector>(II->getOperand(3))) {
+ assert(Mask->getNumOperands() == 16 && "Bad type for intrinsic!");
- case Intrinsic::x86_sse_cvttss2si: {
- // These intrinsics only demands the 0th element of its input vector. If
- // we can simplify the input based on that, do so now.
- uint64_t UndefElts;
- if (Value *V = SimplifyDemandedVectorElts(II->getOperand(1), 1,
- UndefElts)) {
- II->setOperand(1, V);
- return II;
+ // Check that all of the elements are integer constants or undefs.
+ bool AllEltsOk = true;
+ for (unsigned i = 0; i != 16; ++i) {
+ if (!isa<ConstantInt>(Mask->getOperand(i)) &&
+ !isa<UndefValue>(Mask->getOperand(i))) {
+ AllEltsOk = false;
+ break;
+ }
}
- break;
- }
- case Intrinsic::ppc_altivec_vperm:
- // Turn vperm(V1,V2,mask) -> shuffle(V1,V2,mask) if mask is a constant.
- if (ConstantVector *Mask = dyn_cast<ConstantVector>(II->getOperand(3))) {
- assert(Mask->getNumOperands() == 16 && "Bad type for intrinsic!");
+ if (AllEltsOk) {
+ // Cast the input vectors to byte vectors.
+ Value *Op0 =InsertBitCastBefore(II->getOperand(1),Mask->getType(),CI);
+ Value *Op1 =InsertBitCastBefore(II->getOperand(2),Mask->getType(),CI);
+ Value *Result = UndefValue::get(Op0->getType());
- // Check that all of the elements are integer constants or undefs.
- bool AllEltsOk = true;
- for (unsigned i = 0; i != 16; ++i) {
- if (!isa<ConstantInt>(Mask->getOperand(i)) &&
- !isa<UndefValue>(Mask->getOperand(i))) {
- AllEltsOk = false;
- break;
- }
- }
+ // Only extract each element once.
+ Value *ExtractedElts[32];
+ memset(ExtractedElts, 0, sizeof(ExtractedElts));
- if (AllEltsOk) {
- // Cast the input vectors to byte vectors.
- Value *Op0 =InsertBitCastBefore(II->getOperand(1),Mask->getType(),CI);
- Value *Op1 =InsertBitCastBefore(II->getOperand(2),Mask->getType(),CI);
- Value *Result = UndefValue::get(Op0->getType());
-
- // Only extract each element once.
- Value *ExtractedElts[32];
- memset(ExtractedElts, 0, sizeof(ExtractedElts));
-
- for (unsigned i = 0; i != 16; ++i) {
- if (isa<UndefValue>(Mask->getOperand(i)))
- continue;
- unsigned Idx=cast<ConstantInt>(Mask->getOperand(i))->getZExtValue();
- Idx &= 31; // Match the hardware behavior.
-
- if (ExtractedElts[Idx] == 0) {
- Instruction *Elt =
- new ExtractElementInst(Idx < 16 ? Op0 : Op1, Idx&15, "tmp");
- InsertNewInstBefore(Elt, CI);
- ExtractedElts[Idx] = Elt;
- }
+ for (unsigned i = 0; i != 16; ++i) {
+ if (isa<UndefValue>(Mask->getOperand(i)))
+ continue;
+ unsigned Idx=cast<ConstantInt>(Mask->getOperand(i))->getZExtValue();
+ Idx &= 31; // Match the hardware behavior.
- // Insert this value into the result vector.
- Result = InsertElementInst::Create(Result, ExtractedElts[Idx],
- i, "tmp");
- InsertNewInstBefore(cast<Instruction>(Result), CI);
+ if (ExtractedElts[Idx] == 0) {
+ Instruction *Elt =
+ new ExtractElementInst(Idx < 16 ? Op0 : Op1, Idx&15, "tmp");
+ InsertNewInstBefore(Elt, CI);
+ ExtractedElts[Idx] = Elt;
}
- return CastInst::Create(Instruction::BitCast, Result, CI.getType());
+
+ // Insert this value into the result vector.
+ Result = InsertElementInst::Create(Result, ExtractedElts[Idx],
+ i, "tmp");
+ InsertNewInstBefore(cast<Instruction>(Result), CI);
}
+ return CastInst::Create(Instruction::BitCast, Result, CI.getType());
}
- break;
+ }
+ break;
- case Intrinsic::stackrestore: {
- // If the save is right next to the restore, remove the restore. This can
- // happen when variable allocas are DCE'd.
- if (IntrinsicInst *SS = dyn_cast<IntrinsicInst>(II->getOperand(1))) {
- if (SS->getIntrinsicID() == Intrinsic::stacksave) {
- BasicBlock::iterator BI = SS;
- if (&*++BI == II)
- return EraseInstFromFunction(CI);
- }
+ case Intrinsic::stackrestore: {
+ // If the save is right next to the restore, remove the restore. This can
+ // happen when variable allocas are DCE'd.
+ if (IntrinsicInst *SS = dyn_cast<IntrinsicInst>(II->getOperand(1))) {
+ if (SS->getIntrinsicID() == Intrinsic::stacksave) {
+ BasicBlock::iterator BI = SS;
+ if (&*++BI == II)
+ return EraseInstFromFunction(CI);
}
-
- // Scan down this block to see if there is another stack restore in the
- // same block without an intervening call/alloca.
- BasicBlock::iterator BI = II;
- TerminatorInst *TI = II->getParent()->getTerminator();
- bool CannotRemove = false;
- for (++BI; &*BI != TI; ++BI) {
- if (isa<AllocaInst>(BI)) {
+ }
+
+ // Scan down this block to see if there is another stack restore in the
+ // same block without an intervening call/alloca.
+ BasicBlock::iterator BI = II;
+ TerminatorInst *TI = II->getParent()->getTerminator();
+ bool CannotRemove = false;
+ for (++BI; &*BI != TI; ++BI) {
+ if (isa<AllocaInst>(BI)) {
+ CannotRemove = true;
+ break;
+ }
+ if (CallInst *BCI = dyn_cast<CallInst>(BI)) {
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(BCI)) {
+ // If there is a stackrestore below this one, remove this one.
+ if (II->getIntrinsicID() == Intrinsic::stackrestore)
+ return EraseInstFromFunction(CI);
+ // Otherwise, ignore the intrinsic.
+ } else {
+ // If we found a non-intrinsic call, we can't remove the stack
+ // restore.
CannotRemove = true;
break;
}
- if (isa<CallInst>(BI)) {
- if (!isa<IntrinsicInst>(BI)) {
- CannotRemove = true;
- break;
- }
- // If there is a stackrestore below this one, remove this one.
- return EraseInstFromFunction(CI);
- }
}
-
- // If the stack restore is in a return/unwind block and if there are no
- // allocas or calls between the restore and the return, nuke the restore.
- if (!CannotRemove && (isa<ReturnInst>(TI) || isa<UnwindInst>(TI)))
- return EraseInstFromFunction(CI);
- break;
- }
}
+
+ // If the stack restore is in a return/unwind block and if there are no
+ // allocas or calls between the restore and the return, nuke the restore.
+ if (!CannotRemove && (isa<ReturnInst>(TI) || isa<UnwindInst>(TI)))
+ return EraseInstFromFunction(CI);
+ break;
+ }
}
return visitCallSite(II);
// Conversion is ok if changing from one pointer type to another or from
// a pointer to an integer of the same size.
!((isa<PointerType>(OldRetTy) || OldRetTy == TD->getIntPtrType()) &&
- isa<PointerType>(NewRetTy) || NewRetTy == TD->getIntPtrType()))
+ (isa<PointerType>(NewRetTy) || NewRetTy == TD->getIntPtrType())))
return false; // Cannot transform this return value.
if (!Caller->use_empty() &&
if (LI->getParent() != PN.getIncomingBlock(0) ||
!isSafeToSinkLoad(LI))
return 0;
+
+ // If the PHI is of volatile loads and the load block has multiple
+ // successors, sinking it would remove a load of the volatile value from
+ // the path through the other successor.
+ if (isVolatile &&
+ LI->getParent()->getTerminator()->getNumSuccessors() != 1)
+ return 0;
+
} else if (isa<GetElementPtrInst>(FirstInst)) {
if (FirstInst->getNumOperands() == 2)
return FoldPHIArgBinOpIntoPHI(PN);
!isSafeToSinkLoad(LI))
return 0;
- // If the PHI is volatile and its block has multiple successors, sinking
- // it would remove a load of the volatile value from the path through the
- // other successor.
+ // If the PHI is of volatile loads and the load block has multiple
+ // successors, sinking it would remove a load of the volatile value from
+ // the path through the other successor.
if (isVolatile &&
LI->getParent()->getTerminator()->getNumSuccessors() != 1)
return 0;
bool MadeChange = false;
gep_type_iterator GTI = gep_type_begin(GEP);
- for (unsigned i = 1, e = GEP.getNumOperands(); i != e; ++i, ++GTI) {
+ for (User::op_iterator i = GEP.op_begin() + 1, e = GEP.op_end();
+ i != e; ++i, ++GTI) {
if (isa<SequentialType>(*GTI)) {
- if (CastInst *CI = dyn_cast<CastInst>(GEP.getOperand(i))) {
+ if (CastInst *CI = dyn_cast<CastInst>(*i)) {
if (CI->getOpcode() == Instruction::ZExt ||
CI->getOpcode() == Instruction::SExt) {
const Type *SrcTy = CI->getOperand(0)->getType();
// is a 32-bit pointer target.
if (SrcTy->getPrimitiveSizeInBits() >= TD->getPointerSizeInBits()) {
MadeChange = true;
- GEP.setOperand(i, CI->getOperand(0));
+ *i = CI->getOperand(0);
}
}
}
// to what we need. If the incoming value needs a cast instruction,
// insert it. This explicit cast can make subsequent optimizations more
// obvious.
- Value *Op = GEP.getOperand(i);
+ Value *Op = *i;
if (TD->getTypeSizeInBits(Op->getType()) > TD->getPointerSizeInBits()) {
if (Constant *C = dyn_cast<Constant>(Op)) {
- GEP.setOperand(i, ConstantExpr::getTrunc(C, TD->getIntPtrType()));
+ *i = ConstantExpr::getTrunc(C, TD->getIntPtrType());
MadeChange = true;
} else {
Op = InsertCastBefore(Instruction::Trunc, Op, TD->getIntPtrType(),
GEP);
- GEP.setOperand(i, Op);
+ *i = Op;
MadeChange = true;
}
}
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(CI)) {
// Instead of loading constant c string, use corresponding integer value
// directly if string length is small enough.
- const std::string &Str = CE->getOperand(0)->getStringValue();
- if (!Str.empty()) {
+ std::string Str;
+ if (GetConstantStringInfo(CE->getOperand(0), Str) && !Str.empty()) {
unsigned len = Str.length();
const Type *Ty = cast<PointerType>(CE->getType())->getElementType();
unsigned numBits = Ty->getPrimitiveSizeInBits();
while (BBI != E) {
--BBI;
+ // If we see a free or a call (which might do a free) the pointer could be
+ // marked invalid.
+ if (isa<FreeInst>(BBI) || isa<CallInst>(BBI))
+ return false;
+
if (LoadInst *LI = dyn_cast<LoadInst>(BBI)) {
if (LI->getOperand(0) == V) return true;
- } else if (StoreInst *SI = dyn_cast<StoreInst>(BBI))
+ } else if (StoreInst *SI = dyn_cast<StoreInst>(BBI)) {
if (SI->getOperand(1) == V) return true;
+ }
}
return false;
}
if (++PI != pred_end(DestBB))
return false;
-
-
+
+ // Bail out if all the relevant blocks aren't distinct (this can happen,
+ // for example, if SI is in an infinite loop)
+ if (StoreBB == DestBB || OtherBB == DestBB)
+ return false;
+
// Verify that the other block ends in a branch and is not otherwise empty.
BasicBlock::iterator BBI = OtherBB->getTerminator();
BranchInst *OtherBr = dyn_cast<BranchInst>(BBI);
return false;
break;
}
- // If we find something that may be using the stored value, or if we run
- // out of instructions, we can't do the xform.
- if (isa<LoadInst>(BBI) || BBI->mayWriteToMemory() ||
+ // If we find something that may be using or overwriting the stored
+ // value, or if we run out of instructions, we can't do the xform.
+ if (BBI->mayReadFromMemory() || BBI->mayWriteToMemory() ||
BBI == OtherBB->begin())
return false;
}
// In order to eliminate the store in OtherBr, we have to
- // make sure nothing reads the stored value in StoreBB.
+ // make sure nothing reads or overwrites the stored value in
+ // StoreBB.
for (BasicBlock::iterator I = StoreBB->begin(); &*I != &SI; ++I) {
// FIXME: This should really be AA driven.
- if (isa<LoadInst>(I) || I->mayWriteToMemory())
+ if (I->mayReadFromMemory() || I->mayWriteToMemory())
return false;
}
}
return 0;
}
+Instruction *InstCombiner::visitExtractValueInst(ExtractValueInst &EV) {
+ // See if we are trying to extract a known value. If so, use that instead.
+ if (Value *Elt = FindInsertedValue(EV.getOperand(0), EV.idx_begin(),
+ EV.idx_end(), &EV))
+ return ReplaceInstUsesWith(EV, Elt);
+
+ // No changes
+ return 0;
+}
+
/// CheapToScalarize - Return true if the value is cheaper to scalarize than it
/// is to leave as a vector operation.
static bool CheapToScalarize(Value *V, bool isConstant) {
std::vector<unsigned> Result;
const ConstantVector *CP = cast<ConstantVector>(SVI->getOperand(2));
- for (unsigned i = 0, e = CP->getNumOperands(); i != e; ++i)
- if (isa<UndefValue>(CP->getOperand(i)))
+ for (User::const_op_iterator i = CP->op_begin(), e = CP->op_end(); i!=e; ++i)
+ if (isa<UndefValue>(*i))
Result.push_back(NElts*2); // undef -> 8
else
- Result.push_back(cast<ConstantInt>(CP->getOperand(i))->getZExtValue());
+ Result.push_back(cast<ConstantInt>(*i)->getZExtValue());
return Result;
}
}
Instruction *InstCombiner::visitExtractElementInst(ExtractElementInst &EI) {
-
// If vector val is undef, replace extract with scalar undef.
if (isa<UndefValue>(EI.getOperand(0)))
return ReplaceInstUsesWith(EI, UndefValue::get(EI.getType()));
return ReplaceInstUsesWith(EI, Constant::getNullValue(EI.getType()));
if (ConstantVector *C = dyn_cast<ConstantVector>(EI.getOperand(0))) {
- // If vector val is constant with uniform operands, replace EI
- // with that operand
+ // If vector val is constant with all elements the same, replace EI with
+ // that element. When the elements are not identical, we cannot replace yet
+ // (we do that below, but only when the index is constant).
Constant *op0 = C->getOperand(0);
for (unsigned i = 1; i < C->getNumOperands(); ++i)
if (C->getOperand(i) != op0) {