///
/// If we can't emit an optimized form for this expression, this returns null.
///
-static Value *EvaluateGEPOffsetExpression(User *GEP, Instruction &I,
- InstCombiner &IC) {
+static Value *EvaluateGEPOffsetExpression(User *GEP, InstCombiner &IC) {
TargetData &TD = *IC.getTargetData();
gep_type_iterator GTI = gep_type_begin(GEP);
// Cast to intptrty in case a truncation occurs. If an extension is needed,
// we don't need to bother extending: the extension won't affect where the
// computation crosses zero.
- if (VariableIdx->getType()->getPrimitiveSizeInBits() > IntPtrWidth)
- VariableIdx = new TruncInst(VariableIdx,
- TD.getIntPtrType(VariableIdx->getContext()),
- VariableIdx->getName(), &I);
+ if (VariableIdx->getType()->getPrimitiveSizeInBits() > IntPtrWidth) {
+ const Type *IntPtrTy = TD.getIntPtrType(VariableIdx->getContext());
+ VariableIdx = IC.Builder->CreateTrunc(VariableIdx, IntPtrTy);
+ }
return VariableIdx;
}
// Okay, we can do this evaluation. Start by converting the index to intptr.
const Type *IntPtrTy = TD.getIntPtrType(VariableIdx->getContext());
if (VariableIdx->getType() != IntPtrTy)
- VariableIdx = CastInst::CreateIntegerCast(VariableIdx, IntPtrTy,
- true /*SExt*/,
- VariableIdx->getName(), &I);
+ VariableIdx = IC.Builder->CreateIntCast(VariableIdx, IntPtrTy,
+ true /*Signed*/);
Constant *OffsetVal = ConstantInt::get(IntPtrTy, NewOffs);
- return BinaryOperator::CreateAdd(VariableIdx, OffsetVal, "offset", &I);
+ return IC.Builder->CreateAdd(VariableIdx, OffsetVal, "offset");
}
/// FoldGEPICmp - Fold comparisons between a GEP instruction and something
// This transformation (ignoring the base and scales) is valid because we
// know pointers can't overflow since the gep is inbounds. See if we can
// output an optimized form.
- Value *Offset = EvaluateGEPOffsetExpression(GEPLHS, I, *this);
+ Value *Offset = EvaluateGEPOffsetExpression(GEPLHS, *this);
// If not, synthesize the offset the hard way.
if (Offset == 0)
if (AllZeros)
return FoldGEPICmp(GEPLHS, GEPRHS->getOperand(0), Cond, I);
+ bool GEPsInBounds = GEPLHS->isInBounds() && GEPRHS->isInBounds();
if (GEPLHS->getNumOperands() == GEPRHS->getNumOperands()) {
// If the GEPs only differ by one index, compare it.
unsigned NumDifferences = 0; // Keep track of # differences.
ConstantInt::get(Type::getInt1Ty(I.getContext()),
ICmpInst::isTrueWhenEqual(Cond)));
- else if (NumDifferences == 1) {
+ else if (NumDifferences == 1 && GEPsInBounds) {
Value *LHSV = GEPLHS->getOperand(DiffOperand);
Value *RHSV = GEPRHS->getOperand(DiffOperand);
// Make sure we do a signed comparison here.
// Only lower this if the icmp is the only user of the GEP or if we expect
// the result to fold to a constant!
if (TD &&
+ GEPsInBounds &&
(isa<ConstantExpr>(GEPLHS) || GEPLHS->hasOneUse()) &&
(isa<ConstantExpr>(GEPRHS) || GEPRHS->hasOneUse())) {
// ((gep Ptr, OFFSET1) cmp (gep Ptr, OFFSET2) ---> (OFFSET1 cmp OFFSET2)
if (ICI.isSigned() != (Shr->getOpcode() == Instruction::AShr))
return 0;
- // Otherwise, all lshr and all exact ashr's are equivalent to a udiv/sdiv by
- // a power of 2. Since we already have logic to simplify these, transform
- // to div and then simplify the resultant comparison.
+ // Otherwise, all lshr and most exact ashr's are equivalent to a udiv/sdiv
+ // by a power of 2. Since we already have logic to simplify these,
+ // transform to div and then simplify the resultant comparison.
if (Shr->getOpcode() == Instruction::AShr &&
- !Shr->isExact())
+ (!Shr->isExact() || ShAmtVal == TypeBits - 1))
return 0;
// Revisit the shift (to delete it).
// have its sign bit set or if it is an equality comparison.
// Extending a relational comparison when we're checking the sign
// bit would not work.
- if (Cast->hasOneUse() &&
- (ICI.isEquality() ||
- (AndCST->getValue().isNonNegative() && RHSV.isNonNegative()))) {
- uint32_t BitWidth =
- cast<IntegerType>(Cast->getOperand(0)->getType())->getBitWidth();
- APInt NewCST = AndCST->getValue().zext(BitWidth);
- APInt NewCI = RHSV.zext(BitWidth);
- Value *NewAnd =
+ if (ICI.isEquality() ||
+ (AndCST->getValue().isNonNegative() && RHSV.isNonNegative())) {
+ Value *NewAnd =
Builder->CreateAnd(Cast->getOperand(0),
- ConstantInt::get(ICI.getContext(), NewCST),
- LHSI->getName());
+ ConstantExpr::getZExt(AndCST, Cast->getSrcTy()));
+ NewAnd->takeName(LHSI);
return new ICmpInst(ICI.getPredicate(), NewAnd,
- ConstantInt::get(ICI.getContext(), NewCI));
+ ConstantExpr::getZExt(RHS, Cast->getSrcTy()));
}
}
-
+
+ // If the LHS is an AND of a zext, and we have an equality compare, we can
+ // shrink the and/compare to the smaller type, eliminating the cast.
+ if (ZExtInst *Cast = dyn_cast<ZExtInst>(LHSI->getOperand(0))) {
+ const IntegerType *Ty = cast<IntegerType>(Cast->getSrcTy());
+ // Make sure we don't compare the upper bits, SimplifyDemandedBits
+ // should fold the icmp to true/false in that case.
+ if (ICI.isEquality() && RHSV.getActiveBits() <= Ty->getBitWidth()) {
+ Value *NewAnd =
+ Builder->CreateAnd(Cast->getOperand(0),
+ ConstantExpr::getTrunc(AndCST, Ty));
+ NewAnd->takeName(LHSI);
+ return new ICmpInst(ICI.getPredicate(), NewAnd,
+ ConstantExpr::getTrunc(RHS, Ty));
+ }
+ }
+
// If this is: (X >> C1) & C2 != C3 (where any shift and any compare
// could exist), turn it into (X & (C2 << C1)) != (C3 << C1). This
// happens a LOT in code produced by the C front-end, for bitfield
case Instruction::Xor:
// For the xor case, we can xor two constants together, eliminating
// the explicit xor.
- if (Constant *BOC = dyn_cast<Constant>(BO->getOperand(1)))
- return new ICmpInst(ICI.getPredicate(), BO->getOperand(0),
+ if (Constant *BOC = dyn_cast<Constant>(BO->getOperand(1))) {
+ return new ICmpInst(ICI.getPredicate(), BO->getOperand(0),
ConstantExpr::getXor(RHS, BOC));
-
- // FALLTHROUGH
+ } else if (RHSV == 0) {
+ // Replace ((xor A, B) != 0) with (A != B)
+ return new ICmpInst(ICI.getPredicate(), BO->getOperand(0),
+ BO->getOperand(1));
+ }
+ break;
case Instruction::Sub:
- // Replace (([sub|xor] A, B) != 0) with (A != B)
- if (RHSV == 0)
+ // Replace ((sub A, B) != C) with (B != A-C) if A & C are constants.
+ if (ConstantInt *BOp0C = dyn_cast<ConstantInt>(BO->getOperand(0))) {
+ if (BO->hasOneUse())
+ return new ICmpInst(ICI.getPredicate(), BO->getOperand(1),
+ ConstantExpr::getSub(BOp0C, RHS));
+ } else if (RHSV == 0) {
+ // Replace ((sub A, B) != 0) with (A != B)
return new ICmpInst(ICI.getPredicate(), BO->getOperand(0),
BO->getOperand(1));
+ }
break;
-
case Instruction::Or:
// If bits are being or'd in that are not present in the constant we
// are comparing against, then the comparison could never succeed!
return new ICmpInst(isICMP_NE ? ICmpInst::ICMP_EQ :
ICmpInst::ICMP_NE, LHSI,
Constant::getNullValue(RHS->getType()));
-
+
+ // Don't perform the following transforms if the AND has multiple uses
+ if (!BO->hasOneUse())
+ break;
+
// Replace (and X, (1 << size(X)-1) != 0) with x s< 0
if (BOC->getValue().isSignBit()) {
Value *X = BO->getOperand(0);
// result and the overflow bit.
Module *M = I.getParent()->getParent()->getParent();
- const Type *NewType = IntegerType::get(OrigAdd->getContext(), NewWidth);
+ Type *NewType = IntegerType::get(OrigAdd->getContext(), NewWidth);
Value *F = Intrinsic::getDeclaration(M, Intrinsic::sadd_with_overflow,
&NewType, 1);
Builder->SetInsertPoint(OrigAdd);
Module *M = I.getParent()->getParent()->getParent();
- const Type *Ty = LHS->getType();
+ Type *Ty = LHS->getType();
Value *F = Intrinsic::getDeclaration(M, Intrinsic::uadd_with_overflow, &Ty,1);
CallInst *Call = Builder->CreateCall2(F, LHS, RHS, "uadd");
Value *Add = Builder->CreateExtractValue(Call, 0);
// fall-through
case Instruction::SDiv:
case Instruction::AShr:
- if (!BO0->isExact() && !BO1->isExact())
+ if (!BO0->isExact() || !BO1->isExact())
break;
return new ICmpInst(I.getPredicate(), BO0->getOperand(0),
BO1->getOperand(0));