// Implementation of r = rand (); a = b - r; a = a - c; a = a + r
void Substitution::subRand2(BinaryOperator *bo) {
BinaryOperator *op = NULL;
if (bo->getOpcode() == Instruction::Sub) {
Type *ty = bo->getType();
ConstantInt *co =
(ConstantInt *)ConstantInt::get(ty, llvm::cryptoutils->get_uint64_t());
op =
BinaryOperator::Create(Instruction::Sub, bo->getOperand(0), co, "", bo);
op =
BinaryOperator::Create(Instruction::Sub, op, bo->getOperand(1), "", bo);
op = BinaryOperator::Create(Instruction::Add, op, co, "", bo);
// Check signed wrap
op->setHasNoSignedWrap(bo->hasNoSignedWrap());
op->setHasNoUnsignedWrap(bo->hasNoUnsignedWrap());
bo->replaceAllUsesWith(op);
}
/* else {
Type *ty = bo->getType();
ConstantFP *co =
(ConstantFP*)ConstantFP::get(ty,(float)llvm::cryptoutils->get_uint64_t());
op = BinaryOperator::Create(Instruction::FSub,bo->getOperand(0),co,"",bo);
op = BinaryOperator::Create(Instruction::FSub,op,bo->getOperand(1),"",bo);
op = BinaryOperator::Create(Instruction::FAdd,op,co,"",bo);
} */
}
Value *ConstantOffsetExtractor::removeConstOffset(unsigned ChainIndex) {
if (ChainIndex == 0) {
assert(isa<ConstantInt>(UserChain[ChainIndex]));
return ConstantInt::getNullValue(UserChain[ChainIndex]->getType());
}
BinaryOperator *BO = cast<BinaryOperator>(UserChain[ChainIndex]);
unsigned OpNo = (BO->getOperand(0) == UserChain[ChainIndex - 1] ? 0 : 1);
assert(BO->getOperand(OpNo) == UserChain[ChainIndex - 1]);
Value *NextInChain = removeConstOffset(ChainIndex - 1);
Value *TheOther = BO->getOperand(1 - OpNo);
// If NextInChain is 0 and not the LHS of a sub, we can simplify the
// sub-expression to be just TheOther.
if (ConstantInt *CI = dyn_cast<ConstantInt>(NextInChain)) {
if (CI->isZero() && !(BO->getOpcode() == Instruction::Sub && OpNo == 0))
return TheOther;
}
if (BO->getOpcode() == Instruction::Or) {
// Rebuild "or" as "add", because "or" may be invalid for the new
// epxression.
//
// For instance, given
// a | (b + 5) where a and b + 5 have no common bits,
// we can extract 5 as the constant offset.
//
// However, reusing the "or" in the new index would give us
// (a | b) + 5
// which does not equal a | (b + 5).
//
// Replacing the "or" with "add" is fine, because
// a | (b + 5) = a + (b + 5) = (a + b) + 5
if (OpNo == 0) {
return BinaryOperator::CreateAdd(NextInChain, TheOther, BO->getName(),
IP);
} else {
return BinaryOperator::CreateAdd(TheOther, NextInChain, BO->getName(),
IP);
}
}
// We can reuse BO in this case, because the new expression shares the same
// instruction type and BO is used at most once.
assert(BO->getNumUses() <= 1 &&
"distributeExtsAndCloneChain clones each BinaryOperator in "
"UserChain, so no one should be used more than "
"once");
BO->setOperand(OpNo, NextInChain);
BO->setHasNoSignedWrap(false);
BO->setHasNoUnsignedWrap(false);
// Make sure it appears after all instructions we've inserted so far.
BO->moveBefore(IP);
return BO;
}
Value *AMDGPUCodeGenPrepare::copyFlags(
const BinaryOperator &I, Value *V) const {
assert(isa<BinaryOperator>(V) && "V must be binary operation");
BinaryOperator *BinOp = cast<BinaryOperator>(V);
if (isa<OverflowingBinaryOperator>(BinOp)) {
BinOp->setHasNoSignedWrap(I.hasNoSignedWrap());
BinOp->setHasNoUnsignedWrap(I.hasNoUnsignedWrap());
} else if (isa<PossiblyExactOperator>(BinOp))
BinOp->setIsExact(I.isExact());
return V;
}
// Implementation of a = b + (-c)
void Substitution::subNeg(BinaryOperator *bo) {
BinaryOperator *op = NULL;
if (bo->getOpcode() == Instruction::Sub) {
op = BinaryOperator::CreateNeg(bo->getOperand(1), "", bo);
op =
BinaryOperator::Create(Instruction::Add, bo->getOperand(0), op, "", bo);
// Check signed wrap
op->setHasNoSignedWrap(bo->hasNoSignedWrap());
op->setHasNoUnsignedWrap(bo->hasNoUnsignedWrap());
} else {
op = BinaryOperator::CreateFNeg(bo->getOperand(1), "", bo);
op = BinaryOperator::Create(Instruction::FAdd, bo->getOperand(0), op, "",
bo);
}
bo->replaceAllUsesWith(op);
}
TEST(CloneInstruction, OverflowBits) {
LLVMContext context;
Value *V = new Argument(Type::getInt32Ty(context));
BinaryOperator *Add = BinaryOperator::Create(Instruction::Add, V, V);
BinaryOperator *Sub = BinaryOperator::Create(Instruction::Sub, V, V);
BinaryOperator *Mul = BinaryOperator::Create(Instruction::Mul, V, V);
EXPECT_FALSE(cast<BinaryOperator>(Add->clone())->hasNoUnsignedWrap());
EXPECT_FALSE(cast<BinaryOperator>(Add->clone())->hasNoSignedWrap());
EXPECT_FALSE(cast<BinaryOperator>(Sub->clone())->hasNoUnsignedWrap());
EXPECT_FALSE(cast<BinaryOperator>(Sub->clone())->hasNoSignedWrap());
EXPECT_FALSE(cast<BinaryOperator>(Mul->clone())->hasNoUnsignedWrap());
EXPECT_FALSE(cast<BinaryOperator>(Mul->clone())->hasNoSignedWrap());
Add->setHasNoUnsignedWrap();
Sub->setHasNoUnsignedWrap();
Mul->setHasNoUnsignedWrap();
EXPECT_TRUE(cast<BinaryOperator>(Add->clone())->hasNoUnsignedWrap());
EXPECT_FALSE(cast<BinaryOperator>(Add->clone())->hasNoSignedWrap());
EXPECT_TRUE(cast<BinaryOperator>(Sub->clone())->hasNoUnsignedWrap());
EXPECT_FALSE(cast<BinaryOperator>(Sub->clone())->hasNoSignedWrap());
EXPECT_TRUE(cast<BinaryOperator>(Mul->clone())->hasNoUnsignedWrap());
EXPECT_FALSE(cast<BinaryOperator>(Mul->clone())->hasNoSignedWrap());
Add->setHasNoSignedWrap();
Sub->setHasNoSignedWrap();
Mul->setHasNoSignedWrap();
EXPECT_TRUE(cast<BinaryOperator>(Add->clone())->hasNoUnsignedWrap());
EXPECT_TRUE(cast<BinaryOperator>(Add->clone())->hasNoSignedWrap());
EXPECT_TRUE(cast<BinaryOperator>(Sub->clone())->hasNoUnsignedWrap());
EXPECT_TRUE(cast<BinaryOperator>(Sub->clone())->hasNoSignedWrap());
EXPECT_TRUE(cast<BinaryOperator>(Mul->clone())->hasNoUnsignedWrap());
EXPECT_TRUE(cast<BinaryOperator>(Mul->clone())->hasNoSignedWrap());
Add->setHasNoUnsignedWrap(false);
Sub->setHasNoUnsignedWrap(false);
Mul->setHasNoUnsignedWrap(false);
EXPECT_FALSE(cast<BinaryOperator>(Add->clone())->hasNoUnsignedWrap());
EXPECT_TRUE(cast<BinaryOperator>(Add->clone())->hasNoSignedWrap());
EXPECT_FALSE(cast<BinaryOperator>(Sub->clone())->hasNoUnsignedWrap());
EXPECT_TRUE(cast<BinaryOperator>(Sub->clone())->hasNoSignedWrap());
EXPECT_FALSE(cast<BinaryOperator>(Mul->clone())->hasNoUnsignedWrap());
EXPECT_TRUE(cast<BinaryOperator>(Mul->clone())->hasNoSignedWrap());
}
Instruction *InstCombiner::visitSub(BinaryOperator &I) {
Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
if (Value *V = SimplifySubInst(Op0, Op1, I.hasNoSignedWrap(),
I.hasNoUnsignedWrap(), TD))
return ReplaceInstUsesWith(I, V);
// (A*B)-(A*C) -> A*(B-C) etc
if (Value *V = SimplifyUsingDistributiveLaws(I))
return ReplaceInstUsesWith(I, V);
// If this is a 'B = x-(-A)', change to B = x+A. This preserves NSW/NUW.
if (Value *V = dyn_castNegVal(Op1)) {
BinaryOperator *Res = BinaryOperator::CreateAdd(Op0, V);
Res->setHasNoSignedWrap(I.hasNoSignedWrap());
Res->setHasNoUnsignedWrap(I.hasNoUnsignedWrap());
return Res;
}
if (I.getType()->isIntegerTy(1))
return BinaryOperator::CreateXor(Op0, Op1);
// Replace (-1 - A) with (~A).
if (match(Op0, m_AllOnes()))
return BinaryOperator::CreateNot(Op1);
if (ConstantInt *C = dyn_cast<ConstantInt>(Op0)) {
// C - ~X == X + (1+C)
Value *X = 0;
if (match(Op1, m_Not(m_Value(X))))
return BinaryOperator::CreateAdd(X, AddOne(C));
// -(X >>u 31) -> (X >>s 31)
// -(X >>s 31) -> (X >>u 31)
if (C->isZero()) {
Value *X; ConstantInt *CI;
if (match(Op1, m_LShr(m_Value(X), m_ConstantInt(CI))) &&
// Verify we are shifting out everything but the sign bit.
CI->getValue() == I.getType()->getPrimitiveSizeInBits()-1)
return BinaryOperator::CreateAShr(X, CI);
if (match(Op1, m_AShr(m_Value(X), m_ConstantInt(CI))) &&
// Verify we are shifting out everything but the sign bit.
CI->getValue() == I.getType()->getPrimitiveSizeInBits()-1)
return BinaryOperator::CreateLShr(X, CI);
}
// Try to fold constant sub into select arguments.
if (SelectInst *SI = dyn_cast<SelectInst>(Op1))
if (Instruction *R = FoldOpIntoSelect(I, SI))
return R;
// C - zext(bool) -> bool ? C - 1 : C
if (ZExtInst *ZI = dyn_cast<ZExtInst>(Op1))
if (ZI->getSrcTy()->isIntegerTy(1))
return SelectInst::Create(ZI->getOperand(0), SubOne(C), C);
// C-(X+C2) --> (C-C2)-X
ConstantInt *C2;
if (match(Op1, m_Add(m_Value(X), m_ConstantInt(C2))))
return BinaryOperator::CreateSub(ConstantExpr::getSub(C, C2), X);
}
{ Value *Y;
// X-(X+Y) == -Y X-(Y+X) == -Y
if (match(Op1, m_Add(m_Specific(Op0), m_Value(Y))) ||
match(Op1, m_Add(m_Value(Y), m_Specific(Op0))))
return BinaryOperator::CreateNeg(Y);
// (X-Y)-X == -Y
if (match(Op0, m_Sub(m_Specific(Op1), m_Value(Y))))
return BinaryOperator::CreateNeg(Y);
}
if (Op1->hasOneUse()) {
Value *X = 0, *Y = 0, *Z = 0;
Constant *C = 0;
ConstantInt *CI = 0;
// (X - (Y - Z)) --> (X + (Z - Y)).
if (match(Op1, m_Sub(m_Value(Y), m_Value(Z))))
return BinaryOperator::CreateAdd(Op0,
Builder->CreateSub(Z, Y, Op1->getName()));
// (X - (X & Y)) --> (X & ~Y)
//
if (match(Op1, m_And(m_Value(Y), m_Specific(Op0))) ||
match(Op1, m_And(m_Specific(Op0), m_Value(Y))))
return BinaryOperator::CreateAnd(Op0,
Builder->CreateNot(Y, Y->getName() + ".not"));
// 0 - (X sdiv C) -> (X sdiv -C)
if (match(Op1, m_SDiv(m_Value(X), m_Constant(C))) &&
match(Op0, m_Zero()))
return BinaryOperator::CreateSDiv(X, ConstantExpr::getNeg(C));
// 0 - (X << Y) -> (-X << Y) when X is freely negatable.
if (match(Op1, m_Shl(m_Value(X), m_Value(Y))) && match(Op0, m_Zero()))
if (Value *XNeg = dyn_castNegVal(X))
return BinaryOperator::CreateShl(XNeg, Y);
// X - X*C --> X * (1-C)
if (match(Op1, m_Mul(m_Specific(Op0), m_ConstantInt(CI)))) {
Constant *CP1 = ConstantExpr::getSub(ConstantInt::get(I.getType(),1), CI);
return BinaryOperator::CreateMul(Op0, CP1);
}
// X - X<<C --> X * (1-(1<<C))
if (match(Op1, m_Shl(m_Specific(Op0), m_ConstantInt(CI)))) {
Constant *One = ConstantInt::get(I.getType(), 1);
C = ConstantExpr::getSub(One, ConstantExpr::getShl(One, CI));
return BinaryOperator::CreateMul(Op0, C);
}
// X - A*-B -> X + A*B
// X - -A*B -> X + A*B
Value *A, *B;
if (match(Op1, m_Mul(m_Value(A), m_Neg(m_Value(B)))) ||
match(Op1, m_Mul(m_Neg(m_Value(A)), m_Value(B))))
return BinaryOperator::CreateAdd(Op0, Builder->CreateMul(A, B));
// X - A*CI -> X + A*-CI
// X - CI*A -> X + A*-CI
if (match(Op1, m_Mul(m_Value(A), m_ConstantInt(CI))) ||
match(Op1, m_Mul(m_ConstantInt(CI), m_Value(A)))) {
Value *NewMul = Builder->CreateMul(A, ConstantExpr::getNeg(CI));
return BinaryOperator::CreateAdd(Op0, NewMul);
}
}
ConstantInt *C1;
if (Value *X = dyn_castFoldableMul(Op0, C1)) {
if (X == Op1) // X*C - X --> X * (C-1)
return BinaryOperator::CreateMul(Op1, SubOne(C1));
ConstantInt *C2; // X*C1 - X*C2 -> X * (C1-C2)
if (X == dyn_castFoldableMul(Op1, C2))
return BinaryOperator::CreateMul(X, ConstantExpr::getSub(C1, C2));
}
// Optimize pointer differences into the same array into a size. Consider:
// &A[10] - &A[0]: we should compile this to "10".
if (TD) {
Value *LHSOp, *RHSOp;
if (match(Op0, m_PtrToInt(m_Value(LHSOp))) &&
match(Op1, m_PtrToInt(m_Value(RHSOp))))
if (Value *Res = OptimizePointerDifference(LHSOp, RHSOp, I.getType()))
return ReplaceInstUsesWith(I, Res);
// trunc(p)-trunc(q) -> trunc(p-q)
if (match(Op0, m_Trunc(m_PtrToInt(m_Value(LHSOp)))) &&
match(Op1, m_Trunc(m_PtrToInt(m_Value(RHSOp)))))
if (Value *Res = OptimizePointerDifference(LHSOp, RHSOp, I.getType()))
return ReplaceInstUsesWith(I, Res);
}
return 0;
}
bool SimplifyIndvar::eliminateOverflowIntrinsic(CallInst *CI) {
auto *F = CI->getCalledFunction();
if (!F)
return false;
typedef const SCEV *(ScalarEvolution::*OperationFunctionTy)(
const SCEV *, const SCEV *, SCEV::NoWrapFlags, unsigned);
typedef const SCEV *(ScalarEvolution::*ExtensionFunctionTy)(
const SCEV *, Type *, unsigned);
OperationFunctionTy Operation;
ExtensionFunctionTy Extension;
Instruction::BinaryOps RawOp;
// We always have exactly one of nsw or nuw. If NoSignedOverflow is false, we
// have nuw.
bool NoSignedOverflow;
switch (F->getIntrinsicID()) {
default:
return false;
case Intrinsic::sadd_with_overflow:
Operation = &ScalarEvolution::getAddExpr;
Extension = &ScalarEvolution::getSignExtendExpr;
RawOp = Instruction::Add;
NoSignedOverflow = true;
break;
case Intrinsic::uadd_with_overflow:
Operation = &ScalarEvolution::getAddExpr;
Extension = &ScalarEvolution::getZeroExtendExpr;
RawOp = Instruction::Add;
NoSignedOverflow = false;
break;
case Intrinsic::ssub_with_overflow:
Operation = &ScalarEvolution::getMinusSCEV;
Extension = &ScalarEvolution::getSignExtendExpr;
RawOp = Instruction::Sub;
NoSignedOverflow = true;
break;
case Intrinsic::usub_with_overflow:
Operation = &ScalarEvolution::getMinusSCEV;
Extension = &ScalarEvolution::getZeroExtendExpr;
RawOp = Instruction::Sub;
NoSignedOverflow = false;
break;
}
const SCEV *LHS = SE->getSCEV(CI->getArgOperand(0));
const SCEV *RHS = SE->getSCEV(CI->getArgOperand(1));
auto *NarrowTy = cast<IntegerType>(LHS->getType());
auto *WideTy =
IntegerType::get(NarrowTy->getContext(), NarrowTy->getBitWidth() * 2);
const SCEV *A =
(SE->*Extension)((SE->*Operation)(LHS, RHS, SCEV::FlagAnyWrap, 0),
WideTy, 0);
const SCEV *B =
(SE->*Operation)((SE->*Extension)(LHS, WideTy, 0),
(SE->*Extension)(RHS, WideTy, 0), SCEV::FlagAnyWrap, 0);
if (A != B)
return false;
// Proved no overflow, nuke the overflow check and, if possible, the overflow
// intrinsic as well.
BinaryOperator *NewResult = BinaryOperator::Create(
RawOp, CI->getArgOperand(0), CI->getArgOperand(1), "", CI);
if (NoSignedOverflow)
NewResult->setHasNoSignedWrap(true);
else
NewResult->setHasNoUnsignedWrap(true);
SmallVector<ExtractValueInst *, 4> ToDelete;
for (auto *U : CI->users()) {
if (auto *EVI = dyn_cast<ExtractValueInst>(U)) {
if (EVI->getIndices()[0] == 1)
EVI->replaceAllUsesWith(ConstantInt::getFalse(CI->getContext()));
else {
assert(EVI->getIndices()[0] == 0 && "Only two possibilities!");
EVI->replaceAllUsesWith(NewResult);
}
ToDelete.push_back(EVI);
}
}
for (auto *EVI : ToDelete)
EVI->eraseFromParent();
if (CI->use_empty())
CI->eraseFromParent();
return true;
}
/// FoldSelectIntoOp - Try fold the select into one of the operands to
/// facilitate further optimization.
Instruction *InstCombiner::FoldSelectIntoOp(SelectInst &SI, Value *TrueVal,
Value *FalseVal) {
// See the comment above GetSelectFoldableOperands for a description of the
// transformation we are doing here.
if (Instruction *TVI = dyn_cast<Instruction>(TrueVal)) {
if (TVI->hasOneUse() && TVI->getNumOperands() == 2 &&
!isa<Constant>(FalseVal)) {
if (unsigned SFO = GetSelectFoldableOperands(TVI)) {
unsigned OpToFold = 0;
if ((SFO & 1) && FalseVal == TVI->getOperand(0)) {
OpToFold = 1;
} else if ((SFO & 2) && FalseVal == TVI->getOperand(1)) {
OpToFold = 2;
}
if (OpToFold) {
Constant *C = GetSelectFoldableConstant(TVI);
Value *OOp = TVI->getOperand(2-OpToFold);
// Avoid creating select between 2 constants unless it's selecting
// between 0, 1 and -1.
if (!isa<Constant>(OOp) || isSelect01(C, cast<Constant>(OOp))) {
Value *NewSel = Builder->CreateSelect(SI.getCondition(), OOp, C);
NewSel->takeName(TVI);
BinaryOperator *TVI_BO = cast<BinaryOperator>(TVI);
BinaryOperator *BO = BinaryOperator::Create(TVI_BO->getOpcode(),
FalseVal, NewSel);
if (isa<PossiblyExactOperator>(BO))
BO->setIsExact(TVI_BO->isExact());
if (isa<OverflowingBinaryOperator>(BO)) {
BO->setHasNoUnsignedWrap(TVI_BO->hasNoUnsignedWrap());
BO->setHasNoSignedWrap(TVI_BO->hasNoSignedWrap());
}
return BO;
}
}
}
}
}
if (Instruction *FVI = dyn_cast<Instruction>(FalseVal)) {
if (FVI->hasOneUse() && FVI->getNumOperands() == 2 &&
!isa<Constant>(TrueVal)) {
if (unsigned SFO = GetSelectFoldableOperands(FVI)) {
unsigned OpToFold = 0;
if ((SFO & 1) && TrueVal == FVI->getOperand(0)) {
OpToFold = 1;
} else if ((SFO & 2) && TrueVal == FVI->getOperand(1)) {
OpToFold = 2;
}
if (OpToFold) {
Constant *C = GetSelectFoldableConstant(FVI);
Value *OOp = FVI->getOperand(2-OpToFold);
// Avoid creating select between 2 constants unless it's selecting
// between 0, 1 and -1.
if (!isa<Constant>(OOp) || isSelect01(C, cast<Constant>(OOp))) {
Value *NewSel = Builder->CreateSelect(SI.getCondition(), C, OOp);
NewSel->takeName(FVI);
BinaryOperator *FVI_BO = cast<BinaryOperator>(FVI);
BinaryOperator *BO = BinaryOperator::Create(FVI_BO->getOpcode(),
TrueVal, NewSel);
if (isa<PossiblyExactOperator>(BO))
BO->setIsExact(FVI_BO->isExact());
if (isa<OverflowingBinaryOperator>(BO)) {
BO->setHasNoUnsignedWrap(FVI_BO->hasNoUnsignedWrap());
BO->setHasNoSignedWrap(FVI_BO->hasNoSignedWrap());
}
return BO;
}
}
}
}
}
return nullptr;
}
/// Rebuild a new instruction just like 'I' but with the new operands given.
/// In the event of type mismatch, the type of the operands is correct.
static Value *BuildNew(Instruction *I, ArrayRef<Value*> NewOps) {
// We don't want to use the IRBuilder here because we want the replacement
// instructions to appear next to 'I', not the builder's insertion point.
switch (I->getOpcode()) {
case Instruction::Add:
case Instruction::FAdd:
case Instruction::Sub:
case Instruction::FSub:
case Instruction::Mul:
case Instruction::FMul:
case Instruction::UDiv:
case Instruction::SDiv:
case Instruction::FDiv:
case Instruction::URem:
case Instruction::SRem:
case Instruction::FRem:
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor: {
BinaryOperator *BO = cast<BinaryOperator>(I);
assert(NewOps.size() == 2 && "binary operator with #ops != 2");
BinaryOperator *New =
BinaryOperator::Create(cast<BinaryOperator>(I)->getOpcode(),
NewOps[0], NewOps[1], "", BO);
if (isa<OverflowingBinaryOperator>(BO)) {
New->setHasNoUnsignedWrap(BO->hasNoUnsignedWrap());
New->setHasNoSignedWrap(BO->hasNoSignedWrap());
}
if (isa<PossiblyExactOperator>(BO)) {
New->setIsExact(BO->isExact());
}
return New;
}
case Instruction::ICmp:
assert(NewOps.size() == 2 && "icmp with #ops != 2");
return new ICmpInst(I, cast<ICmpInst>(I)->getPredicate(),
NewOps[0], NewOps[1]);
case Instruction::FCmp:
assert(NewOps.size() == 2 && "fcmp with #ops != 2");
return new FCmpInst(I, cast<FCmpInst>(I)->getPredicate(),
NewOps[0], NewOps[1]);
case Instruction::Trunc:
case Instruction::ZExt:
case Instruction::SExt:
case Instruction::FPToUI:
case Instruction::FPToSI:
case Instruction::UIToFP:
case Instruction::SIToFP:
case Instruction::FPTrunc:
case Instruction::FPExt: {
// It's possible that the mask has a different number of elements from
// the original cast. We recompute the destination type to match the mask.
Type *DestTy =
VectorType::get(I->getType()->getScalarType(),
NewOps[0]->getType()->getVectorNumElements());
assert(NewOps.size() == 1 && "cast with #ops != 1");
return CastInst::Create(cast<CastInst>(I)->getOpcode(), NewOps[0], DestTy,
"", I);
}
case Instruction::GetElementPtr: {
Value *Ptr = NewOps[0];
ArrayRef<Value*> Idx = NewOps.slice(1);
GetElementPtrInst *GEP = GetElementPtrInst::Create(Ptr, Idx, "", I);
GEP->setIsInBounds(cast<GetElementPtrInst>(I)->isInBounds());
return GEP;
}
}
llvm_unreachable("failed to rebuild vector instructions");
}
Instruction *InstCombiner::visitMul(BinaryOperator &I) {
bool Changed = SimplifyAssociativeOrCommutative(I);
Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
if (Value *V = SimplifyMulInst(Op0, Op1, TD))
return ReplaceInstUsesWith(I, V);
if (Value *V = SimplifyUsingDistributiveLaws(I))
return ReplaceInstUsesWith(I, V);
if (match(Op1, m_AllOnes())) // X * -1 == 0 - X
return BinaryOperator::CreateNeg(Op0, I.getName());
if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
// ((X << C1)*C2) == (X * (C2 << C1))
if (BinaryOperator *SI = dyn_cast<BinaryOperator>(Op0))
if (SI->getOpcode() == Instruction::Shl)
if (Constant *ShOp = dyn_cast<Constant>(SI->getOperand(1)))
return BinaryOperator::CreateMul(SI->getOperand(0),
ConstantExpr::getShl(CI, ShOp));
const APInt &Val = CI->getValue();
if (Val.isPowerOf2()) { // Replace X*(2^C) with X << C
Constant *NewCst = ConstantInt::get(Op0->getType(), Val.logBase2());
BinaryOperator *Shl = BinaryOperator::CreateShl(Op0, NewCst);
if (I.hasNoSignedWrap()) Shl->setHasNoSignedWrap();
if (I.hasNoUnsignedWrap()) Shl->setHasNoUnsignedWrap();
return Shl;
}
// Canonicalize (X+C1)*CI -> X*CI+C1*CI.
{ Value *X; ConstantInt *C1;
if (Op0->hasOneUse() &&
match(Op0, m_Add(m_Value(X), m_ConstantInt(C1)))) {
Value *Add = Builder->CreateMul(X, CI);
return BinaryOperator::CreateAdd(Add, Builder->CreateMul(C1, CI));
}
}
// (Y - X) * (-(2**n)) -> (X - Y) * (2**n), for positive nonzero n
// (Y + const) * (-(2**n)) -> (-constY) * (2**n), for positive nonzero n
// The "* (2**n)" thus becomes a potential shifting opportunity.
{
const APInt & Val = CI->getValue();
const APInt &PosVal = Val.abs();
if (Val.isNegative() && PosVal.isPowerOf2()) {
Value *X = 0, *Y = 0;
if (Op0->hasOneUse()) {
ConstantInt *C1;
Value *Sub = 0;
if (match(Op0, m_Sub(m_Value(Y), m_Value(X))))
Sub = Builder->CreateSub(X, Y, "suba");
else if (match(Op0, m_Add(m_Value(Y), m_ConstantInt(C1))))
Sub = Builder->CreateSub(Builder->CreateNeg(C1), Y, "subc");
if (Sub)
return
BinaryOperator::CreateMul(Sub,
ConstantInt::get(Y->getType(), PosVal));
}
}
}
}
// Simplify mul instructions with a constant RHS.
if (isa<Constant>(Op1)) {
// Try to fold constant mul into select arguments.
if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
if (Instruction *R = FoldOpIntoSelect(I, SI))
return R;
if (isa<PHINode>(Op0))
if (Instruction *NV = FoldOpIntoPhi(I))
return NV;
}
if (Value *Op0v = dyn_castNegVal(Op0)) // -X * -Y = X*Y
if (Value *Op1v = dyn_castNegVal(Op1))
return BinaryOperator::CreateMul(Op0v, Op1v);
// (X / Y) * Y = X - (X % Y)
// (X / Y) * -Y = (X % Y) - X
{
Value *Op1C = Op1;
BinaryOperator *BO = dyn_cast<BinaryOperator>(Op0);
if (!BO ||
(BO->getOpcode() != Instruction::UDiv &&
BO->getOpcode() != Instruction::SDiv)) {
Op1C = Op0;
BO = dyn_cast<BinaryOperator>(Op1);
}
Value *Neg = dyn_castNegVal(Op1C);
if (BO && BO->hasOneUse() &&
(BO->getOperand(1) == Op1C || BO->getOperand(1) == Neg) &&
(BO->getOpcode() == Instruction::UDiv ||
BO->getOpcode() == Instruction::SDiv)) {
Value *Op0BO = BO->getOperand(0), *Op1BO = BO->getOperand(1);
// If the division is exact, X % Y is zero, so we end up with X or -X.
if (PossiblyExactOperator *SDiv = dyn_cast<PossiblyExactOperator>(BO))
if (SDiv->isExact()) {
if (Op1BO == Op1C)
return ReplaceInstUsesWith(I, Op0BO);
return BinaryOperator::CreateNeg(Op0BO);
}
Value *Rem;
if (BO->getOpcode() == Instruction::UDiv)
Rem = Builder->CreateURem(Op0BO, Op1BO);
else
Rem = Builder->CreateSRem(Op0BO, Op1BO);
Rem->takeName(BO);
if (Op1BO == Op1C)
return BinaryOperator::CreateSub(Op0BO, Rem);
return BinaryOperator::CreateSub(Rem, Op0BO);
}
}
/// i1 mul -> i1 and.
if (I.getType()->isIntegerTy(1))
return BinaryOperator::CreateAnd(Op0, Op1);
// X*(1 << Y) --> X << Y
// (1 << Y)*X --> X << Y
{
Value *Y;
if (match(Op0, m_Shl(m_One(), m_Value(Y))))
return BinaryOperator::CreateShl(Op1, Y);
if (match(Op1, m_Shl(m_One(), m_Value(Y))))
return BinaryOperator::CreateShl(Op0, Y);
}
// If one of the operands of the multiply is a cast from a boolean value, then
// we know the bool is either zero or one, so this is a 'masking' multiply.
// X * Y (where Y is 0 or 1) -> X & (0-Y)
if (!I.getType()->isVectorTy()) {
// -2 is "-1 << 1" so it is all bits set except the low one.
APInt Negative2(I.getType()->getPrimitiveSizeInBits(), (uint64_t)-2, true);
Value *BoolCast = 0, *OtherOp = 0;
if (MaskedValueIsZero(Op0, Negative2))
BoolCast = Op0, OtherOp = Op1;
else if (MaskedValueIsZero(Op1, Negative2))
BoolCast = Op1, OtherOp = Op0;
if (BoolCast) {
Value *V = Builder->CreateSub(Constant::getNullValue(I.getType()),
BoolCast);
return BinaryOperator::CreateAnd(V, OtherOp);
}
}
return Changed ? &I : 0;
}
static Value *genVal(int &Budget, unsigned Width, bool ConstOK, bool ArgOK) {
if (Budget > 0 && Choose(2)) {
if (Verbose)
errs() << "adding a phi, budget = " << Budget << "\n";
--Budget;
Value *V = Builder->CreatePHI(Type::getIntNTy(*C, Width), N);
Vals.push_back(V);
return V;
}
if (Budget > 0 && Budget != N && Choose(2)) {
if (Verbose)
errs() << "adding a branch, budget = " << Budget << "\n";
--Budget;
BranchInst *Br;
if (0 && Builder->GetInsertBlock()->size() > 0 && Choose(2)) {
Br = Builder->CreateBr(BBs[0]);
} else {
Value *C = genVal(Budget, 1, false, ArgOK);
Br = Builder->CreateCondBr(C, BBs[0], BBs[0]);
}
Branches.push_back(Br);
BasicBlock *BB = BasicBlock::Create(*C, "", F);
BBs.push_back(BB);
Builder->SetInsertPoint(BB);
return genVal(Budget, Width, ConstOK, ArgOK);
}
if (Budget > 0 && Width == W && Choose(2)) {
if (Verbose)
errs() << "adding a select with width = " << Width
<< " and budget = " << Budget << "\n";
--Budget;
Value *L, *R;
genLR(L, R, Budget, Width);
Value *C = genVal(Budget, 1, false);
Value *V = Builder->CreateSelect(C, L, R);
Vals.push_back(V);
return V;
}
if (Budget > 0 && Width == 1 && Choose(2)) {
if (Verbose)
errs() << "adding an icmp with width = " << Width
<< " and budget = " << Budget << "\n";
--Budget;
Value *L, *R;
genLR(L, R, Budget, Width);
CmpInst::Predicate P;
switch (OneICmp ? 0 : Choose(10)) {
case 0:
P = CmpInst::ICMP_EQ;
break;
case 1:
P = CmpInst::ICMP_NE;
break;
case 2:
P = CmpInst::ICMP_UGT;
break;
case 3:
P = CmpInst::ICMP_UGE;
break;
case 4:
P = CmpInst::ICMP_ULT;
break;
case 5:
P = CmpInst::ICMP_ULE;
break;
case 6:
P = CmpInst::ICMP_SGT;
break;
case 7:
P = CmpInst::ICMP_SGE;
break;
case 8:
P = CmpInst::ICMP_SLT;
break;
case 9:
P = CmpInst::ICMP_SLE;
break;
}
Value *V = Builder->CreateICmp(P, L, R);
Vals.push_back(V);
return V;
}
if (Budget > 0 && Width == W && Choose(2)) {
unsigned OldW = Width * 2;
if (Verbose)
errs() << "adding a trunc from " << OldW << " to " << Width
<< " and budget = " << Budget << "\n";
--Budget;
Value *V = Builder->CreateTrunc(genVal(Budget, OldW, false),
Type::getIntNTy(*C, Width));
Vals.push_back(V);
return V;
}
if (Budget > 0 && Width == W && Choose(2)) {
unsigned OldW = Width / 2;
if (OldW > 1 && Choose(2))
OldW = 1;
if (Verbose)
errs() << "adding a zext from " << OldW << " to " << Width
<< " and budget = " << Budget << "\n";
--Budget;
Value *V;
if (Choose(2))
V = Builder->CreateZExt(genVal(Budget, OldW, false),
Type::getIntNTy(*C, Width));
else
V = Builder->CreateSExt(genVal(Budget, OldW, false),
Type::getIntNTy(*C, Width));
Vals.push_back(V);
return V;
}
if (Budget > 0 && Width == W && Choose(2)) {
if (Verbose)
errs() << "adding a binop with width = " << Width
<< " and budget = " << Budget << "\n";
--Budget;
Instruction::BinaryOps Op;
switch (OneBinop ? 0 : Choose(10)) {
case 0:
Op = Instruction::Add;
break;
case 1:
Op = Instruction::Sub;
break;
case 2:
Op = Instruction::Mul;
break;
case 3:
Op = Instruction::SDiv;
break;
case 4:
Op = Instruction::UDiv;
break;
case 5:
Op = Instruction::SRem;
break;
case 6:
Op = Instruction::URem;
break;
case 7:
Op = Instruction::And;
break;
case 8:
Op = Instruction::Or;
break;
case 9:
Op = Instruction::Xor;
break;
}
Value *L, *R;
genLR(L, R, Budget, Width);
Value *V = Builder->CreateBinOp(Op, L, R);
if (!NoUB) {
if ((Op == Instruction::Add || Op == Instruction::Sub ||
Op == Instruction::Mul || Op == Instruction::Shl) &&
Choose(2)) {
BinaryOperator *B = cast<BinaryOperator>(V);
B->setHasNoSignedWrap(true);
}
if ((Op == Instruction::Add || Op == Instruction::Sub ||
Op == Instruction::Mul || Op == Instruction::Shl) &&
Choose(2)) {
BinaryOperator *B = cast<BinaryOperator>(V);
B->setHasNoUnsignedWrap(true);
}
if ((Op == Instruction::UDiv || Op == Instruction::SDiv ||
Op == Instruction::LShr || Op == Instruction::AShr) &&
Choose(2)) {
BinaryOperator *B = cast<BinaryOperator>(V);
B->setIsExact(true);
}
}
Vals.push_back(V);
return V;
}
if (ConstOK && Choose(2)) {
if (Verbose)
errs() << "adding a const with width = " << Width
<< " and budget = " << Budget << "\n";
if (OneConst) {
return ConstantInt::get(*C, APInt(Width, 1));
} else {
int n = Choose((1 << Width) + 1);
if (n == (1 << Width))
return UndefValue::get(Type::getIntNTy(*C, Width));
else
return ConstantInt::get(*C, APInt(Width, n));
}
}
if (Verbose)
errs() << "using existing val with width = " << Width
<< " and budget = " << Budget << " and ArgOK = " << ArgOK << "\n";
std::vector<Value *> Vs;
for (auto &it : Vals)
if (it->getType()->getPrimitiveSizeInBits() == Width)
Vs.push_back(it);
unsigned choices = Vs.size() + (ArgOK ? 1 : 0);
if (choices == 0)
exit(0);
unsigned which = Choose(choices);
if (which == Vs.size()) {
Value *V = 0;
for (auto it = F->arg_begin(); it != F->arg_end(); ++it) {
if (UsedArgs.find(it) == UsedArgs.end() &&
it->getType()->getPrimitiveSizeInBits() == Width) {
UsedArgs.insert(it);
V = it;
Vals.push_back(V);
break;
}
}
ensure(V);
return V;
} else {
return Vs.at(which);
}
}
/// InsertPHITranslatedPointer - Insert a computation of the PHI translated
/// version of 'V' for the edge PredBB->CurBB into the end of the PredBB
/// block. All newly created instructions are added to the NewInsts list.
/// This returns null on failure.
///
Value *PHITransAddr::
InsertPHITranslatedSubExpr(Value *InVal, BasicBlock *CurBB,
BasicBlock *PredBB, const DominatorTree &DT,
SmallVectorImpl<Instruction*> &NewInsts) {
// See if we have a version of this value already available and dominating
// PredBB. If so, there is no need to insert a new instance of it.
PHITransAddr Tmp(InVal, TD);
if (!Tmp.PHITranslateValue(CurBB, PredBB, &DT))
return Tmp.getAddr();
// If we don't have an available version of this value, it must be an
// instruction.
Instruction *Inst = cast<Instruction>(InVal);
// Handle bitcast of PHI translatable value.
if (BitCastInst *BC = dyn_cast<BitCastInst>(Inst)) {
Value *OpVal = InsertPHITranslatedSubExpr(BC->getOperand(0),
CurBB, PredBB, DT, NewInsts);
if (OpVal == 0) return 0;
// Otherwise insert a bitcast at the end of PredBB.
BitCastInst *New = new BitCastInst(OpVal, InVal->getType(),
InVal->getName()+".phi.trans.insert",
PredBB->getTerminator());
NewInsts.push_back(New);
return New;
}
// Handle getelementptr with at least one PHI operand.
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Inst)) {
SmallVector<Value*, 8> GEPOps;
BasicBlock *CurBB = GEP->getParent();
for (unsigned i = 0, e = GEP->getNumOperands(); i != e; ++i) {
Value *OpVal = InsertPHITranslatedSubExpr(GEP->getOperand(i),
CurBB, PredBB, DT, NewInsts);
if (OpVal == 0) return 0;
GEPOps.push_back(OpVal);
}
GetElementPtrInst *Result =
GetElementPtrInst::Create(GEPOps[0], GEPOps.begin()+1, GEPOps.end(),
InVal->getName()+".phi.trans.insert",
PredBB->getTerminator());
Result->setIsInBounds(GEP->isInBounds());
NewInsts.push_back(Result);
return Result;
}
#if 0
// FIXME: This code works, but it is unclear that we actually want to insert
// a big chain of computation in order to make a value available in a block.
// This needs to be evaluated carefully to consider its cost trade offs.
// Handle add with a constant RHS.
if (Inst->getOpcode() == Instruction::Add &&
isa<ConstantInt>(Inst->getOperand(1))) {
// PHI translate the LHS.
Value *OpVal = InsertPHITranslatedSubExpr(Inst->getOperand(0),
CurBB, PredBB, DT, NewInsts);
if (OpVal == 0) return 0;
BinaryOperator *Res = BinaryOperator::CreateAdd(OpVal, Inst->getOperand(1),
InVal->getName()+".phi.trans.insert",
PredBB->getTerminator());
Res->setHasNoSignedWrap(cast<BinaryOperator>(Inst)->hasNoSignedWrap());
Res->setHasNoUnsignedWrap(cast<BinaryOperator>(Inst)->hasNoUnsignedWrap());
NewInsts.push_back(Res);
return Res;
}
#endif
return 0;
}
Instruction *InstCombiner::FoldShiftByConstant(Value *Op0, Constant *Op1,
BinaryOperator &I) {
bool isLeftShift = I.getOpcode() == Instruction::Shl;
ConstantInt *COp1 = nullptr;
if (ConstantDataVector *CV = dyn_cast<ConstantDataVector>(Op1))
COp1 = dyn_cast_or_null<ConstantInt>(CV->getSplatValue());
else if (ConstantVector *CV = dyn_cast<ConstantVector>(Op1))
COp1 = dyn_cast_or_null<ConstantInt>(CV->getSplatValue());
else
COp1 = dyn_cast<ConstantInt>(Op1);
if (!COp1)
return nullptr;
// See if we can propagate this shift into the input, this covers the trivial
// cast of lshr(shl(x,c1),c2) as well as other more complex cases.
if (I.getOpcode() != Instruction::AShr &&
CanEvaluateShifted(Op0, COp1->getZExtValue(), isLeftShift, *this)) {
DEBUG(dbgs() << "ICE: GetShiftedValue propagating shift through expression"
" to eliminate shift:\n IN: " << *Op0 << "\n SH: " << I <<"\n");
return ReplaceInstUsesWith(I,
GetShiftedValue(Op0, COp1->getZExtValue(), isLeftShift, *this));
}
// See if we can simplify any instructions used by the instruction whose sole
// purpose is to compute bits we don't care about.
uint32_t TypeBits = Op0->getType()->getScalarSizeInBits();
assert(!COp1->uge(TypeBits) &&
"Shift over the type width should have been removed already");
// ((X*C1) << C2) == (X * (C1 << C2))
if (BinaryOperator *BO = dyn_cast<BinaryOperator>(Op0))
if (BO->getOpcode() == Instruction::Mul && isLeftShift)
if (Constant *BOOp = dyn_cast<Constant>(BO->getOperand(1)))
return BinaryOperator::CreateMul(BO->getOperand(0),
ConstantExpr::getShl(BOOp, Op1));
// Try to fold constant and into select arguments.
if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
if (Instruction *R = FoldOpIntoSelect(I, SI))
return R;
if (isa<PHINode>(Op0))
if (Instruction *NV = FoldOpIntoPhi(I))
return NV;
// Fold shift2(trunc(shift1(x,c1)), c2) -> trunc(shift2(shift1(x,c1),c2))
if (TruncInst *TI = dyn_cast<TruncInst>(Op0)) {
Instruction *TrOp = dyn_cast<Instruction>(TI->getOperand(0));
// If 'shift2' is an ashr, we would have to get the sign bit into a funny
// place. Don't try to do this transformation in this case. Also, we
// require that the input operand is a shift-by-constant so that we have
// confidence that the shifts will get folded together. We could do this
// xform in more cases, but it is unlikely to be profitable.
if (TrOp && I.isLogicalShift() && TrOp->isShift() &&
isa<ConstantInt>(TrOp->getOperand(1))) {
// Okay, we'll do this xform. Make the shift of shift.
Constant *ShAmt = ConstantExpr::getZExt(COp1, TrOp->getType());
// (shift2 (shift1 & 0x00FF), c2)
Value *NSh = Builder->CreateBinOp(I.getOpcode(), TrOp, ShAmt,I.getName());
// For logical shifts, the truncation has the effect of making the high
// part of the register be zeros. Emulate this by inserting an AND to
// clear the top bits as needed. This 'and' will usually be zapped by
// other xforms later if dead.
unsigned SrcSize = TrOp->getType()->getScalarSizeInBits();
unsigned DstSize = TI->getType()->getScalarSizeInBits();
APInt MaskV(APInt::getLowBitsSet(SrcSize, DstSize));
// The mask we constructed says what the trunc would do if occurring
// between the shifts. We want to know the effect *after* the second
// shift. We know that it is a logical shift by a constant, so adjust the
// mask as appropriate.
if (I.getOpcode() == Instruction::Shl)
MaskV <<= COp1->getZExtValue();
else {
assert(I.getOpcode() == Instruction::LShr && "Unknown logical shift");
MaskV = MaskV.lshr(COp1->getZExtValue());
}
// shift1 & 0x00FF
Value *And = Builder->CreateAnd(NSh,
ConstantInt::get(I.getContext(), MaskV),
TI->getName());
// Return the value truncated to the interesting size.
return new TruncInst(And, I.getType());
}
}
if (Op0->hasOneUse()) {
if (BinaryOperator *Op0BO = dyn_cast<BinaryOperator>(Op0)) {
// Turn ((X >> C) + Y) << C -> (X + (Y << C)) & (~0 << C)
Value *V1, *V2;
ConstantInt *CC;
switch (Op0BO->getOpcode()) {
default: break;
case Instruction::Add:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor: {
// These operators commute.
// Turn (Y + (X >> C)) << C -> (X + (Y << C)) & (~0 << C)
if (isLeftShift && Op0BO->getOperand(1)->hasOneUse() &&
match(Op0BO->getOperand(1), m_Shr(m_Value(V1),
m_Specific(Op1)))) {
Value *YS = // (Y << C)
Builder->CreateShl(Op0BO->getOperand(0), Op1, Op0BO->getName());
// (X + (Y << C))
Value *X = Builder->CreateBinOp(Op0BO->getOpcode(), YS, V1,
Op0BO->getOperand(1)->getName());
uint32_t Op1Val = COp1->getLimitedValue(TypeBits);
APInt Bits = APInt::getHighBitsSet(TypeBits, TypeBits - Op1Val);
Constant *Mask = ConstantInt::get(I.getContext(), Bits);
if (VectorType *VT = dyn_cast<VectorType>(X->getType()))
Mask = ConstantVector::getSplat(VT->getNumElements(), Mask);
return BinaryOperator::CreateAnd(X, Mask);
}
// Turn (Y + ((X >> C) & CC)) << C -> ((X & (CC << C)) + (Y << C))
Value *Op0BOOp1 = Op0BO->getOperand(1);
if (isLeftShift && Op0BOOp1->hasOneUse() &&
match(Op0BOOp1,
m_And(m_OneUse(m_Shr(m_Value(V1), m_Specific(Op1))),
m_ConstantInt(CC)))) {
Value *YS = // (Y << C)
Builder->CreateShl(Op0BO->getOperand(0), Op1,
Op0BO->getName());
// X & (CC << C)
Value *XM = Builder->CreateAnd(V1, ConstantExpr::getShl(CC, Op1),
V1->getName()+".mask");
return BinaryOperator::Create(Op0BO->getOpcode(), YS, XM);
}
}
// FALL THROUGH.
case Instruction::Sub: {
// Turn ((X >> C) + Y) << C -> (X + (Y << C)) & (~0 << C)
if (isLeftShift && Op0BO->getOperand(0)->hasOneUse() &&
match(Op0BO->getOperand(0), m_Shr(m_Value(V1),
m_Specific(Op1)))) {
Value *YS = // (Y << C)
Builder->CreateShl(Op0BO->getOperand(1), Op1, Op0BO->getName());
// (X + (Y << C))
Value *X = Builder->CreateBinOp(Op0BO->getOpcode(), V1, YS,
Op0BO->getOperand(0)->getName());
uint32_t Op1Val = COp1->getLimitedValue(TypeBits);
APInt Bits = APInt::getHighBitsSet(TypeBits, TypeBits - Op1Val);
Constant *Mask = ConstantInt::get(I.getContext(), Bits);
if (VectorType *VT = dyn_cast<VectorType>(X->getType()))
Mask = ConstantVector::getSplat(VT->getNumElements(), Mask);
return BinaryOperator::CreateAnd(X, Mask);
}
// Turn (((X >> C)&CC) + Y) << C -> (X + (Y << C)) & (CC << C)
if (isLeftShift && Op0BO->getOperand(0)->hasOneUse() &&
match(Op0BO->getOperand(0),
m_And(m_OneUse(m_Shr(m_Value(V1), m_Value(V2))),
m_ConstantInt(CC))) && V2 == Op1) {
Value *YS = // (Y << C)
Builder->CreateShl(Op0BO->getOperand(1), Op1, Op0BO->getName());
// X & (CC << C)
Value *XM = Builder->CreateAnd(V1, ConstantExpr::getShl(CC, Op1),
V1->getName()+".mask");
return BinaryOperator::Create(Op0BO->getOpcode(), XM, YS);
}
break;
}
}
// If the operand is an bitwise operator with a constant RHS, and the
// shift is the only use, we can pull it out of the shift.
if (ConstantInt *Op0C = dyn_cast<ConstantInt>(Op0BO->getOperand(1))) {
bool isValid = true; // Valid only for And, Or, Xor
bool highBitSet = false; // Transform if high bit of constant set?
switch (Op0BO->getOpcode()) {
default: isValid = false; break; // Do not perform transform!
case Instruction::Add:
isValid = isLeftShift;
break;
case Instruction::Or:
case Instruction::Xor:
highBitSet = false;
break;
case Instruction::And:
highBitSet = true;
break;
}
// If this is a signed shift right, and the high bit is modified
// by the logical operation, do not perform the transformation.
// The highBitSet boolean indicates the value of the high bit of
// the constant which would cause it to be modified for this
// operation.
//
if (isValid && I.getOpcode() == Instruction::AShr)
isValid = Op0C->getValue()[TypeBits-1] == highBitSet;
if (isValid) {
Constant *NewRHS = ConstantExpr::get(I.getOpcode(), Op0C, Op1);
Value *NewShift =
Builder->CreateBinOp(I.getOpcode(), Op0BO->getOperand(0), Op1);
NewShift->takeName(Op0BO);
return BinaryOperator::Create(Op0BO->getOpcode(), NewShift,
NewRHS);
}
}
}
}
// Find out if this is a shift of a shift by a constant.
BinaryOperator *ShiftOp = dyn_cast<BinaryOperator>(Op0);
if (ShiftOp && !ShiftOp->isShift())
ShiftOp = nullptr;
if (ShiftOp && isa<ConstantInt>(ShiftOp->getOperand(1))) {
// This is a constant shift of a constant shift. Be careful about hiding
// shl instructions behind bit masks. They are used to represent multiplies
// by a constant, and it is important that simple arithmetic expressions
// are still recognizable by scalar evolution.
//
// The transforms applied to shl are very similar to the transforms applied
// to mul by constant. We can be more aggressive about optimizing right
// shifts.
//
// Combinations of right and left shifts will still be optimized in
// DAGCombine where scalar evolution no longer applies.
ConstantInt *ShiftAmt1C = cast<ConstantInt>(ShiftOp->getOperand(1));
uint32_t ShiftAmt1 = ShiftAmt1C->getLimitedValue(TypeBits);
uint32_t ShiftAmt2 = COp1->getLimitedValue(TypeBits);
assert(ShiftAmt2 != 0 && "Should have been simplified earlier");
if (ShiftAmt1 == 0) return nullptr; // Will be simplified in the future.
Value *X = ShiftOp->getOperand(0);
IntegerType *Ty = cast<IntegerType>(I.getType());
// Check for (X << c1) << c2 and (X >> c1) >> c2
if (I.getOpcode() == ShiftOp->getOpcode()) {
uint32_t AmtSum = ShiftAmt1+ShiftAmt2; // Fold into one big shift.
// If this is oversized composite shift, then unsigned shifts get 0, ashr
// saturates.
if (AmtSum >= TypeBits) {
if (I.getOpcode() != Instruction::AShr)
return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
AmtSum = TypeBits-1; // Saturate to 31 for i32 ashr.
}
return BinaryOperator::Create(I.getOpcode(), X,
ConstantInt::get(Ty, AmtSum));
}
if (ShiftAmt1 == ShiftAmt2) {
// If we have ((X << C) >>u C), turn this into X & (-1 >>u C).
if (I.getOpcode() == Instruction::LShr &&
ShiftOp->getOpcode() == Instruction::Shl) {
APInt Mask(APInt::getLowBitsSet(TypeBits, TypeBits - ShiftAmt1));
return BinaryOperator::CreateAnd(X,
ConstantInt::get(I.getContext(), Mask));
}
} else if (ShiftAmt1 < ShiftAmt2) {
uint32_t ShiftDiff = ShiftAmt2-ShiftAmt1;
// (X >>?,exact C1) << C2 --> X << (C2-C1)
// The inexact version is deferred to DAGCombine so we don't hide shl
// behind a bit mask.
if (I.getOpcode() == Instruction::Shl &&
ShiftOp->getOpcode() != Instruction::Shl &&
ShiftOp->isExact()) {
assert(ShiftOp->getOpcode() == Instruction::LShr ||
ShiftOp->getOpcode() == Instruction::AShr);
ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
BinaryOperator *NewShl = BinaryOperator::Create(Instruction::Shl,
X, ShiftDiffCst);
NewShl->setHasNoUnsignedWrap(I.hasNoUnsignedWrap());
NewShl->setHasNoSignedWrap(I.hasNoSignedWrap());
return NewShl;
}
// (X << C1) >>u C2 --> X >>u (C2-C1) & (-1 >> C2)
if (I.getOpcode() == Instruction::LShr &&
ShiftOp->getOpcode() == Instruction::Shl) {
ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
// (X <<nuw C1) >>u C2 --> X >>u (C2-C1)
if (ShiftOp->hasNoUnsignedWrap()) {
BinaryOperator *NewLShr = BinaryOperator::Create(Instruction::LShr,
X, ShiftDiffCst);
NewLShr->setIsExact(I.isExact());
return NewLShr;
}
Value *Shift = Builder->CreateLShr(X, ShiftDiffCst);
APInt Mask(APInt::getLowBitsSet(TypeBits, TypeBits - ShiftAmt2));
return BinaryOperator::CreateAnd(Shift,
ConstantInt::get(I.getContext(),Mask));
}
// We can't handle (X << C1) >>s C2, it shifts arbitrary bits in. However,
// we can handle (X <<nsw C1) >>s C2 since it only shifts in sign bits.
if (I.getOpcode() == Instruction::AShr &&
ShiftOp->getOpcode() == Instruction::Shl) {
if (ShiftOp->hasNoSignedWrap()) {
// (X <<nsw C1) >>s C2 --> X >>s (C2-C1)
ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
BinaryOperator *NewAShr = BinaryOperator::Create(Instruction::AShr,
X, ShiftDiffCst);
NewAShr->setIsExact(I.isExact());
return NewAShr;
}
}
} else {
assert(ShiftAmt2 < ShiftAmt1);
uint32_t ShiftDiff = ShiftAmt1-ShiftAmt2;
// (X >>?exact C1) << C2 --> X >>?exact (C1-C2)
// The inexact version is deferred to DAGCombine so we don't hide shl
// behind a bit mask.
if (I.getOpcode() == Instruction::Shl &&
ShiftOp->getOpcode() != Instruction::Shl &&
ShiftOp->isExact()) {
ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
BinaryOperator *NewShr = BinaryOperator::Create(ShiftOp->getOpcode(),
X, ShiftDiffCst);
NewShr->setIsExact(true);
return NewShr;
}
// (X << C1) >>u C2 --> X << (C1-C2) & (-1 >> C2)
if (I.getOpcode() == Instruction::LShr &&
ShiftOp->getOpcode() == Instruction::Shl) {
ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
if (ShiftOp->hasNoUnsignedWrap()) {
// (X <<nuw C1) >>u C2 --> X <<nuw (C1-C2)
BinaryOperator *NewShl = BinaryOperator::Create(Instruction::Shl,
X, ShiftDiffCst);
NewShl->setHasNoUnsignedWrap(true);
return NewShl;
}
Value *Shift = Builder->CreateShl(X, ShiftDiffCst);
APInt Mask(APInt::getLowBitsSet(TypeBits, TypeBits - ShiftAmt2));
return BinaryOperator::CreateAnd(Shift,
ConstantInt::get(I.getContext(),Mask));
}
// We can't handle (X << C1) >>s C2, it shifts arbitrary bits in. However,
// we can handle (X <<nsw C1) >>s C2 since it only shifts in sign bits.
if (I.getOpcode() == Instruction::AShr &&
ShiftOp->getOpcode() == Instruction::Shl) {
if (ShiftOp->hasNoSignedWrap()) {
// (X <<nsw C1) >>s C2 --> X <<nsw (C1-C2)
ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
BinaryOperator *NewShl = BinaryOperator::Create(Instruction::Shl,
X, ShiftDiffCst);
NewShl->setHasNoSignedWrap(true);
return NewShl;
}
}
}
}
return nullptr;
}
/// GetShiftedValue - When CanEvaluateShifted returned true for an expression,
/// this value inserts the new computation that produces the shifted value.
static Value *GetShiftedValue(Value *V, unsigned NumBits, bool isLeftShift,
InstCombiner &IC) {
// We can always evaluate constants shifted.
if (Constant *C = dyn_cast<Constant>(V)) {
if (isLeftShift)
V = IC.Builder->CreateShl(C, NumBits);
else
V = IC.Builder->CreateLShr(C, NumBits);
// If we got a constantexpr back, try to simplify it with TD info.
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
V = ConstantFoldConstantExpression(CE, IC.getDataLayout(),
IC.getTargetLibraryInfo());
return V;
}
Instruction *I = cast<Instruction>(V);
IC.Worklist.Add(I);
switch (I->getOpcode()) {
default: llvm_unreachable("Inconsistency with CanEvaluateShifted");
case Instruction::And:
case Instruction::Or:
case Instruction::Xor:
// Bitwise operators can all arbitrarily be arbitrarily evaluated shifted.
I->setOperand(0, GetShiftedValue(I->getOperand(0), NumBits,isLeftShift,IC));
I->setOperand(1, GetShiftedValue(I->getOperand(1), NumBits,isLeftShift,IC));
return I;
case Instruction::Shl: {
BinaryOperator *BO = cast<BinaryOperator>(I);
unsigned TypeWidth = BO->getType()->getScalarSizeInBits();
// We only accept shifts-by-a-constant in CanEvaluateShifted.
ConstantInt *CI = cast<ConstantInt>(BO->getOperand(1));
// We can always fold shl(c1)+shl(c2) -> shl(c1+c2).
if (isLeftShift) {
// If this is oversized composite shift, then unsigned shifts get 0.
unsigned NewShAmt = NumBits+CI->getZExtValue();
if (NewShAmt >= TypeWidth)
return Constant::getNullValue(I->getType());
BO->setOperand(1, ConstantInt::get(BO->getType(), NewShAmt));
BO->setHasNoUnsignedWrap(false);
BO->setHasNoSignedWrap(false);
return I;
}
// We turn shl(c)+lshr(c) -> and(c2) if the input doesn't already have
// zeros.
if (CI->getValue() == NumBits) {
APInt Mask(APInt::getLowBitsSet(TypeWidth, TypeWidth - NumBits));
V = IC.Builder->CreateAnd(BO->getOperand(0),
ConstantInt::get(BO->getContext(), Mask));
if (Instruction *VI = dyn_cast<Instruction>(V)) {
VI->moveBefore(BO);
VI->takeName(BO);
}
return V;
}
// We turn shl(c1)+shr(c2) -> shl(c3)+and(c4), but only when we know that
// the and won't be needed.
assert(CI->getZExtValue() > NumBits);
BO->setOperand(1, ConstantInt::get(BO->getType(),
CI->getZExtValue() - NumBits));
BO->setHasNoUnsignedWrap(false);
BO->setHasNoSignedWrap(false);
return BO;
}
case Instruction::LShr: {
BinaryOperator *BO = cast<BinaryOperator>(I);
unsigned TypeWidth = BO->getType()->getScalarSizeInBits();
// We only accept shifts-by-a-constant in CanEvaluateShifted.
ConstantInt *CI = cast<ConstantInt>(BO->getOperand(1));
// We can always fold lshr(c1)+lshr(c2) -> lshr(c1+c2).
if (!isLeftShift) {
// If this is oversized composite shift, then unsigned shifts get 0.
unsigned NewShAmt = NumBits+CI->getZExtValue();
if (NewShAmt >= TypeWidth)
return Constant::getNullValue(BO->getType());
BO->setOperand(1, ConstantInt::get(BO->getType(), NewShAmt));
BO->setIsExact(false);
return I;
}
// We turn lshr(c)+shl(c) -> and(c2) if the input doesn't already have
// zeros.
if (CI->getValue() == NumBits) {
APInt Mask(APInt::getHighBitsSet(TypeWidth, TypeWidth - NumBits));
V = IC.Builder->CreateAnd(I->getOperand(0),
ConstantInt::get(BO->getContext(), Mask));
if (Instruction *VI = dyn_cast<Instruction>(V)) {
VI->moveBefore(I);
VI->takeName(I);
}
return V;
}
// We turn lshr(c1)+shl(c2) -> lshr(c3)+and(c4), but only when we know that
// the and won't be needed.
assert(CI->getZExtValue() > NumBits);
BO->setOperand(1, ConstantInt::get(BO->getType(),
CI->getZExtValue() - NumBits));
BO->setIsExact(false);
return BO;
}
case Instruction::Select:
I->setOperand(1, GetShiftedValue(I->getOperand(1), NumBits,isLeftShift,IC));
I->setOperand(2, GetShiftedValue(I->getOperand(2), NumBits,isLeftShift,IC));
return I;
case Instruction::PHI: {
// We can change a phi if we can change all operands. Note that we never
// get into trouble with cyclic PHIs here because we only consider
// instructions with a single use.
PHINode *PN = cast<PHINode>(I);
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
PN->setIncomingValue(i, GetShiftedValue(PN->getIncomingValue(i),
NumBits, isLeftShift, IC));
return PN;
}
}
}
Instruction *InstCombiner::visitAdd(BinaryOperator &I) {
bool Changed = SimplifyAssociativeOrCommutative(I);
Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);
if (Value *V = SimplifyAddInst(LHS, RHS, I.hasNoSignedWrap(),
I.hasNoUnsignedWrap(), TD))
return ReplaceInstUsesWith(I, V);
// (A*B)+(A*C) -> A*(B+C) etc
if (Value *V = SimplifyUsingDistributiveLaws(I))
return ReplaceInstUsesWith(I, V);
if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
// X + (signbit) --> X ^ signbit
const APInt &Val = CI->getValue();
if (Val.isSignBit())
return BinaryOperator::CreateXor(LHS, RHS);
// See if SimplifyDemandedBits can simplify this. This handles stuff like
// (X & 254)+1 -> (X&254)|1
if (SimplifyDemandedInstructionBits(I))
return &I;
// zext(bool) + C -> bool ? C + 1 : C
if (ZExtInst *ZI = dyn_cast<ZExtInst>(LHS))
if (ZI->getSrcTy()->isIntegerTy(1))
return SelectInst::Create(ZI->getOperand(0), AddOne(CI), CI);
Value *XorLHS = 0; ConstantInt *XorRHS = 0;
if (match(LHS, m_Xor(m_Value(XorLHS), m_ConstantInt(XorRHS)))) {
uint32_t TySizeBits = I.getType()->getScalarSizeInBits();
const APInt &RHSVal = CI->getValue();
unsigned ExtendAmt = 0;
// If we have ADD(XOR(AND(X, 0xFF), 0x80), 0xF..F80), it's a sext.
// If we have ADD(XOR(AND(X, 0xFF), 0xF..F80), 0x80), it's a sext.
if (XorRHS->getValue() == -RHSVal) {
if (RHSVal.isPowerOf2())
ExtendAmt = TySizeBits - RHSVal.logBase2() - 1;
else if (XorRHS->getValue().isPowerOf2())
ExtendAmt = TySizeBits - XorRHS->getValue().logBase2() - 1;
}
if (ExtendAmt) {
APInt Mask = APInt::getHighBitsSet(TySizeBits, ExtendAmt);
if (!MaskedValueIsZero(XorLHS, Mask))
ExtendAmt = 0;
}
if (ExtendAmt) {
Constant *ShAmt = ConstantInt::get(I.getType(), ExtendAmt);
Value *NewShl = Builder->CreateShl(XorLHS, ShAmt, "sext");
return BinaryOperator::CreateAShr(NewShl, ShAmt);
}
}
}
if (isa<Constant>(RHS) && isa<PHINode>(LHS))
if (Instruction *NV = FoldOpIntoPhi(I))
return NV;
if (I.getType()->isIntegerTy(1))
return BinaryOperator::CreateXor(LHS, RHS);
// X + X --> X << 1
if (LHS == RHS) {
BinaryOperator *New =
BinaryOperator::CreateShl(LHS, ConstantInt::get(I.getType(), 1));
New->setHasNoSignedWrap(I.hasNoSignedWrap());
New->setHasNoUnsignedWrap(I.hasNoUnsignedWrap());
return New;
}
// -A + B --> B - A
// -A + -B --> -(A + B)
if (Value *LHSV = dyn_castNegVal(LHS)) {
if (Value *RHSV = dyn_castNegVal(RHS)) {
Value *NewAdd = Builder->CreateAdd(LHSV, RHSV, "sum");
return BinaryOperator::CreateNeg(NewAdd);
}
return BinaryOperator::CreateSub(RHS, LHSV);
}
// A + -B --> A - B
if (!isa<Constant>(RHS))
if (Value *V = dyn_castNegVal(RHS))
return BinaryOperator::CreateSub(LHS, V);
ConstantInt *C2;
if (Value *X = dyn_castFoldableMul(LHS, C2)) {
if (X == RHS) // X*C + X --> X * (C+1)
return BinaryOperator::CreateMul(RHS, AddOne(C2));
// X*C1 + X*C2 --> X * (C1+C2)
ConstantInt *C1;
if (X == dyn_castFoldableMul(RHS, C1))
return BinaryOperator::CreateMul(X, ConstantExpr::getAdd(C1, C2));
}
// X + X*C --> X * (C+1)
if (dyn_castFoldableMul(RHS, C2) == LHS)
return BinaryOperator::CreateMul(LHS, AddOne(C2));
// A+B --> A|B iff A and B have no bits set in common.
if (const IntegerType *IT = dyn_cast<IntegerType>(I.getType())) {
APInt Mask = APInt::getAllOnesValue(IT->getBitWidth());
APInt LHSKnownOne(IT->getBitWidth(), 0);
APInt LHSKnownZero(IT->getBitWidth(), 0);
ComputeMaskedBits(LHS, Mask, LHSKnownZero, LHSKnownOne);
if (LHSKnownZero != 0) {
APInt RHSKnownOne(IT->getBitWidth(), 0);
APInt RHSKnownZero(IT->getBitWidth(), 0);
ComputeMaskedBits(RHS, Mask, RHSKnownZero, RHSKnownOne);
// No bits in common -> bitwise or.
if ((LHSKnownZero|RHSKnownZero).isAllOnesValue())
return BinaryOperator::CreateOr(LHS, RHS);
}
}
// W*X + Y*Z --> W * (X+Z) iff W == Y
{
Value *W, *X, *Y, *Z;
if (match(LHS, m_Mul(m_Value(W), m_Value(X))) &&
match(RHS, m_Mul(m_Value(Y), m_Value(Z)))) {
if (W != Y) {
if (W == Z) {
std::swap(Y, Z);
} else if (Y == X) {
std::swap(W, X);
} else if (X == Z) {
std::swap(Y, Z);
std::swap(W, X);
}
}
if (W == Y) {
Value *NewAdd = Builder->CreateAdd(X, Z, LHS->getName());
return BinaryOperator::CreateMul(W, NewAdd);
}
}
}
if (ConstantInt *CRHS = dyn_cast<ConstantInt>(RHS)) {
Value *X = 0;
if (match(LHS, m_Not(m_Value(X)))) // ~X + C --> (C-1) - X
return BinaryOperator::CreateSub(SubOne(CRHS), X);
// (X & FF00) + xx00 -> (X+xx00) & FF00
if (LHS->hasOneUse() &&
match(LHS, m_And(m_Value(X), m_ConstantInt(C2))) &&
CRHS->getValue() == (CRHS->getValue() & C2->getValue())) {
// See if all bits from the first bit set in the Add RHS up are included
// in the mask. First, get the rightmost bit.
const APInt &AddRHSV = CRHS->getValue();
// Form a mask of all bits from the lowest bit added through the top.
APInt AddRHSHighBits(~((AddRHSV & -AddRHSV)-1));
// See if the and mask includes all of these bits.
APInt AddRHSHighBitsAnd(AddRHSHighBits & C2->getValue());
if (AddRHSHighBits == AddRHSHighBitsAnd) {
// Okay, the xform is safe. Insert the new add pronto.
Value *NewAdd = Builder->CreateAdd(X, CRHS, LHS->getName());
return BinaryOperator::CreateAnd(NewAdd, C2);
}
}
// Try to fold constant add into select arguments.
if (SelectInst *SI = dyn_cast<SelectInst>(LHS))
if (Instruction *R = FoldOpIntoSelect(I, SI))
return R;
}
// add (select X 0 (sub n A)) A --> select X A n
{
SelectInst *SI = dyn_cast<SelectInst>(LHS);
Value *A = RHS;
if (!SI) {
SI = dyn_cast<SelectInst>(RHS);
A = LHS;
}
if (SI && SI->hasOneUse()) {
Value *TV = SI->getTrueValue();
Value *FV = SI->getFalseValue();
Value *N;
// Can we fold the add into the argument of the select?
// We check both true and false select arguments for a matching subtract.
if (match(FV, m_Zero()) && match(TV, m_Sub(m_Value(N), m_Specific(A))))
// Fold the add into the true select value.
return SelectInst::Create(SI->getCondition(), N, A);
if (match(TV, m_Zero()) && match(FV, m_Sub(m_Value(N), m_Specific(A))))
// Fold the add into the false select value.
return SelectInst::Create(SI->getCondition(), A, N);
}
}
// Check for (add (sext x), y), see if we can merge this into an
// integer add followed by a sext.
if (SExtInst *LHSConv = dyn_cast<SExtInst>(LHS)) {
// (add (sext x), cst) --> (sext (add x, cst'))
if (ConstantInt *RHSC = dyn_cast<ConstantInt>(RHS)) {
Constant *CI =
ConstantExpr::getTrunc(RHSC, LHSConv->getOperand(0)->getType());
if (LHSConv->hasOneUse() &&
ConstantExpr::getSExt(CI, I.getType()) == RHSC &&
WillNotOverflowSignedAdd(LHSConv->getOperand(0), CI)) {
// Insert the new, smaller add.
Value *NewAdd = Builder->CreateNSWAdd(LHSConv->getOperand(0),
CI, "addconv");
return new SExtInst(NewAdd, I.getType());
}
}
// (add (sext x), (sext y)) --> (sext (add int x, y))
if (SExtInst *RHSConv = dyn_cast<SExtInst>(RHS)) {
// Only do this if x/y have the same type, if at last one of them has a
// single use (so we don't increase the number of sexts), and if the
// integer add will not overflow.
if (LHSConv->getOperand(0)->getType()==RHSConv->getOperand(0)->getType()&&
(LHSConv->hasOneUse() || RHSConv->hasOneUse()) &&
WillNotOverflowSignedAdd(LHSConv->getOperand(0),
RHSConv->getOperand(0))) {
// Insert the new integer add.
Value *NewAdd = Builder->CreateNSWAdd(LHSConv->getOperand(0),
RHSConv->getOperand(0), "addconv");
return new SExtInst(NewAdd, I.getType());
}
}
}
return Changed ? &I : 0;
}
/// InsertPHITranslatedPointer - Insert a computation of the PHI translated
/// version of 'V' for the edge PredBB->CurBB into the end of the PredBB
/// block. All newly created instructions are added to the NewInsts list.
/// This returns null on failure.
///
Value *PHITransAddr::
InsertPHITranslatedSubExpr(Value *InVal, BasicBlock *CurBB,
BasicBlock *PredBB, const DominatorTree &DT,
SmallVectorImpl<Instruction*> &NewInsts) {
// See if we have a version of this value already available and dominating
// PredBB. If so, there is no need to insert a new instance of it.
PHITransAddr Tmp(InVal, DL, AC);
if (!Tmp.PHITranslateValue(CurBB, PredBB, &DT, /*MustDominate=*/true))
return Tmp.getAddr();
// We don't need to PHI translate values which aren't instructions.
auto *Inst = dyn_cast<Instruction>(InVal);
if (!Inst)
return nullptr;
// Handle cast of PHI translatable value.
if (CastInst *Cast = dyn_cast<CastInst>(Inst)) {
if (!isSafeToSpeculativelyExecute(Cast)) return nullptr;
Value *OpVal = InsertPHITranslatedSubExpr(Cast->getOperand(0),
CurBB, PredBB, DT, NewInsts);
if (!OpVal) return nullptr;
// Otherwise insert a cast at the end of PredBB.
CastInst *New = CastInst::Create(Cast->getOpcode(), OpVal, InVal->getType(),
InVal->getName() + ".phi.trans.insert",
PredBB->getTerminator());
New->setDebugLoc(Inst->getDebugLoc());
NewInsts.push_back(New);
return New;
}
// Handle getelementptr with at least one PHI operand.
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Inst)) {
SmallVector<Value*, 8> GEPOps;
BasicBlock *CurBB = GEP->getParent();
for (unsigned i = 0, e = GEP->getNumOperands(); i != e; ++i) {
Value *OpVal = InsertPHITranslatedSubExpr(GEP->getOperand(i),
CurBB, PredBB, DT, NewInsts);
if (!OpVal) return nullptr;
GEPOps.push_back(OpVal);
}
GetElementPtrInst *Result = GetElementPtrInst::Create(
GEP->getSourceElementType(), GEPOps[0], makeArrayRef(GEPOps).slice(1),
InVal->getName() + ".phi.trans.insert", PredBB->getTerminator());
Result->setDebugLoc(Inst->getDebugLoc());
Result->setIsInBounds(GEP->isInBounds());
NewInsts.push_back(Result);
return Result;
}
#if 0
// FIXME: This code works, but it is unclear that we actually want to insert
// a big chain of computation in order to make a value available in a block.
// This needs to be evaluated carefully to consider its cost trade offs.
// Handle add with a constant RHS.
if (Inst->getOpcode() == Instruction::Add &&
isa<ConstantInt>(Inst->getOperand(1))) {
// PHI translate the LHS.
Value *OpVal = InsertPHITranslatedSubExpr(Inst->getOperand(0),
CurBB, PredBB, DT, NewInsts);
if (OpVal == 0) return 0;
BinaryOperator *Res = BinaryOperator::CreateAdd(OpVal, Inst->getOperand(1),
InVal->getName()+".phi.trans.insert",
PredBB->getTerminator());
Res->setHasNoSignedWrap(cast<BinaryOperator>(Inst)->hasNoSignedWrap());
Res->setHasNoUnsignedWrap(cast<BinaryOperator>(Inst)->hasNoUnsignedWrap());
NewInsts.push_back(Res);
return Res;
}
#endif
return nullptr;
}
/// This function implements the transforms common to both integer division
/// instructions (udiv and sdiv). It is called by the visitors to those integer
/// division instructions.
/// @brief Common integer divide transforms
Instruction *InstCombiner::commonIDivTransforms(BinaryOperator &I) {
Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
// The RHS is known non-zero.
if (Value *V = simplifyValueKnownNonZero(I.getOperand(1), *this)) {
I.setOperand(1, V);
return &I;
}
// Handle cases involving: [su]div X, (select Cond, Y, Z)
// This does not apply for fdiv.
if (isa<SelectInst>(Op1) && SimplifyDivRemOfSelect(I))
return &I;
if (ConstantInt *RHS = dyn_cast<ConstantInt>(Op1)) {
if (Instruction *LHS = dyn_cast<Instruction>(Op0)) {
// (X / C1) / C2 -> X / (C1*C2)
if (Instruction::BinaryOps(LHS->getOpcode()) == I.getOpcode())
if (ConstantInt *LHSRHS = dyn_cast<ConstantInt>(LHS->getOperand(1))) {
if (MultiplyOverflows(RHS, LHSRHS,
I.getOpcode() == Instruction::SDiv))
return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
return BinaryOperator::Create(I.getOpcode(), LHS->getOperand(0),
ConstantExpr::getMul(RHS, LHSRHS));
}
Value *X;
const APInt *C1, *C2;
if (match(RHS, m_APInt(C2))) {
bool IsSigned = I.getOpcode() == Instruction::SDiv;
if ((IsSigned && match(LHS, m_NSWMul(m_Value(X), m_APInt(C1)))) ||
(!IsSigned && match(LHS, m_NUWMul(m_Value(X), m_APInt(C1))))) {
APInt Quotient(C1->getBitWidth(), /*Val=*/0ULL, IsSigned);
// (X * C1) / C2 -> X / (C2 / C1) if C2 is a multiple of C1.
if (IsMultiple(*C2, *C1, Quotient, IsSigned)) {
BinaryOperator *BO = BinaryOperator::Create(
I.getOpcode(), X, ConstantInt::get(X->getType(), Quotient));
BO->setIsExact(I.isExact());
return BO;
}
// (X * C1) / C2 -> X * (C1 / C2) if C1 is a multiple of C2.
if (IsMultiple(*C1, *C2, Quotient, IsSigned)) {
BinaryOperator *BO = BinaryOperator::Create(
Instruction::Mul, X, ConstantInt::get(X->getType(), Quotient));
BO->setHasNoUnsignedWrap(
!IsSigned &&
cast<OverflowingBinaryOperator>(LHS)->hasNoUnsignedWrap());
BO->setHasNoSignedWrap(
cast<OverflowingBinaryOperator>(LHS)->hasNoSignedWrap());
return BO;
}
}
if ((IsSigned && match(LHS, m_NSWShl(m_Value(X), m_APInt(C1)))) ||
(!IsSigned && match(LHS, m_NUWShl(m_Value(X), m_APInt(C1))))) {
APInt Quotient(C1->getBitWidth(), /*Val=*/0ULL, IsSigned);
APInt C1Shifted = APInt::getOneBitSet(
C1->getBitWidth(), static_cast<unsigned>(C1->getLimitedValue()));
// (X << C1) / C2 -> X / (C2 >> C1) if C2 is a multiple of C1.
if (IsMultiple(*C2, C1Shifted, Quotient, IsSigned)) {
BinaryOperator *BO = BinaryOperator::Create(
I.getOpcode(), X, ConstantInt::get(X->getType(), Quotient));
BO->setIsExact(I.isExact());
return BO;
}
// (X << C1) / C2 -> X * (C2 >> C1) if C1 is a multiple of C2.
if (IsMultiple(C1Shifted, *C2, Quotient, IsSigned)) {
BinaryOperator *BO = BinaryOperator::Create(
Instruction::Mul, X, ConstantInt::get(X->getType(), Quotient));
BO->setHasNoUnsignedWrap(
!IsSigned &&
cast<OverflowingBinaryOperator>(LHS)->hasNoUnsignedWrap());
BO->setHasNoSignedWrap(
cast<OverflowingBinaryOperator>(LHS)->hasNoSignedWrap());
return BO;
}
}
}
}
if (!RHS->isZero()) { // avoid X udiv 0
if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
if (Instruction *R = FoldOpIntoSelect(I, SI))
return R;
if (isa<PHINode>(Op0))
if (Instruction *NV = FoldOpIntoPhi(I))
return NV;
}
}
if (ConstantInt *One = dyn_cast<ConstantInt>(Op0)) {
if (One->isOne() && !I.getType()->isIntegerTy(1)) {
bool isSigned = I.getOpcode() == Instruction::SDiv;
if (isSigned) {
// If Op1 is 0 then it's undefined behaviour, if Op1 is 1 then the
// result is one, if Op1 is -1 then the result is minus one, otherwise
// it's zero.
Value *Inc = Builder->CreateAdd(Op1, One);
Value *Cmp = Builder->CreateICmpULT(
Inc, ConstantInt::get(I.getType(), 3));
return SelectInst::Create(Cmp, Op1, ConstantInt::get(I.getType(), 0));
} else {
// If Op1 is 0 then it's undefined behaviour. If Op1 is 1 then the
// result is one, otherwise it's zero.
return new ZExtInst(Builder->CreateICmpEQ(Op1, One), I.getType());
}
}
}
// See if we can fold away this div instruction.
if (SimplifyDemandedInstructionBits(I))
return &I;
// (X - (X rem Y)) / Y -> X / Y; usually originates as ((X / Y) * Y) / Y
Value *X = nullptr, *Z = nullptr;
if (match(Op0, m_Sub(m_Value(X), m_Value(Z)))) { // (X - Z) / Y; Y = Op1
bool isSigned = I.getOpcode() == Instruction::SDiv;
if ((isSigned && match(Z, m_SRem(m_Specific(X), m_Specific(Op1)))) ||
(!isSigned && match(Z, m_URem(m_Specific(X), m_Specific(Op1)))))
return BinaryOperator::Create(I.getOpcode(), X, Op1);
}
return nullptr;
}
