//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "instsimplify"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/ConstantFolding.h"
-#include "llvm/Analysis/Dominators.h"
+#include "llvm/Analysis/MemoryBuiltins.h"
#include "llvm/Analysis/ValueTracking.h"
-#include "llvm/DataLayout.h"
-#include "llvm/GlobalAlias.h"
-#include "llvm/Operator.h"
-#include "llvm/Support/ConstantRange.h"
-#include "llvm/Support/GetElementPtrTypeIterator.h"
-#include "llvm/Support/PatternMatch.h"
-#include "llvm/Support/ValueHandle.h"
+#include "llvm/IR/ConstantRange.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/GetElementPtrTypeIterator.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/Operator.h"
+#include "llvm/IR/PatternMatch.h"
+#include "llvm/IR/ValueHandle.h"
+#include <algorithm>
using namespace llvm;
using namespace llvm::PatternMatch;
+#define DEBUG_TYPE "instsimplify"
+
enum { RecursionLimit = 3 };
STATISTIC(NumExpand, "Number of expansions");
-STATISTIC(NumFactor , "Number of factorizations");
STATISTIC(NumReassoc, "Number of reassociations");
+namespace {
struct Query {
- const DataLayout *TD;
+ const DataLayout *DL;
const TargetLibraryInfo *TLI;
const DominatorTree *DT;
+ AssumptionTracker *AT;
+ const Instruction *CxtI;
- Query(const DataLayout *td, const TargetLibraryInfo *tli,
- const DominatorTree *dt) : TD(td), TLI(tli), DT(dt) {}
+ Query(const DataLayout *DL, const TargetLibraryInfo *tli,
+ const DominatorTree *dt, AssumptionTracker *at = nullptr,
+ const Instruction *cxti = nullptr)
+ : DL(DL), TLI(tli), DT(dt), AT(at), CxtI(cxti) {}
};
+} // end anonymous namespace
static Value *SimplifyAndInst(Value *, Value *, const Query &, unsigned);
static Value *SimplifyBinOp(unsigned, Value *, Value *, const Query &,
Instruction::BinaryOps OpcodeToExpand = (Instruction::BinaryOps)OpcToExpand;
// Recursion is always used, so bail out at once if we already hit the limit.
if (!MaxRecurse--)
- return 0;
+ return nullptr;
// Check whether the expression has the form "(A op' B) op C".
if (BinaryOperator *Op0 = dyn_cast<BinaryOperator>(LHS))
}
}
- return 0;
-}
-
-/// FactorizeBinOp - Simplify "LHS Opcode RHS" by factorizing out a common term
-/// using the operation OpCodeToExtract. For example, when Opcode is Add and
-/// OpCodeToExtract is Mul then this tries to turn "(A*B)+(A*C)" into "A*(B+C)".
-/// Returns the simplified value, or null if no simplification was performed.
-static Value *FactorizeBinOp(unsigned Opcode, Value *LHS, Value *RHS,
- unsigned OpcToExtract, const Query &Q,
- unsigned MaxRecurse) {
- Instruction::BinaryOps OpcodeToExtract = (Instruction::BinaryOps)OpcToExtract;
- // Recursion is always used, so bail out at once if we already hit the limit.
- if (!MaxRecurse--)
- return 0;
-
- BinaryOperator *Op0 = dyn_cast<BinaryOperator>(LHS);
- BinaryOperator *Op1 = dyn_cast<BinaryOperator>(RHS);
-
- if (!Op0 || Op0->getOpcode() != OpcodeToExtract ||
- !Op1 || Op1->getOpcode() != OpcodeToExtract)
- return 0;
-
- // The expression has the form "(A op' B) op (C op' D)".
- Value *A = Op0->getOperand(0), *B = Op0->getOperand(1);
- Value *C = Op1->getOperand(0), *D = Op1->getOperand(1);
-
- // Use left distributivity, i.e. "X op' (Y op Z) = (X op' Y) op (X op' Z)".
- // Does the instruction have the form "(A op' B) op (A op' D)" or, in the
- // commutative case, "(A op' B) op (C op' A)"?
- if (A == C || (Instruction::isCommutative(OpcodeToExtract) && A == D)) {
- Value *DD = A == C ? D : C;
- // Form "A op' (B op DD)" if it simplifies completely.
- // Does "B op DD" simplify?
- if (Value *V = SimplifyBinOp(Opcode, B, DD, Q, MaxRecurse)) {
- // It does! Return "A op' V" if it simplifies or is already available.
- // If V equals B then "A op' V" is just the LHS. If V equals DD then
- // "A op' V" is just the RHS.
- if (V == B || V == DD) {
- ++NumFactor;
- return V == B ? LHS : RHS;
- }
- // Otherwise return "A op' V" if it simplifies.
- if (Value *W = SimplifyBinOp(OpcodeToExtract, A, V, Q, MaxRecurse)) {
- ++NumFactor;
- return W;
- }
- }
- }
-
- // Use right distributivity, i.e. "(X op Y) op' Z = (X op' Z) op (Y op' Z)".
- // Does the instruction have the form "(A op' B) op (C op' B)" or, in the
- // commutative case, "(A op' B) op (B op' D)"?
- if (B == D || (Instruction::isCommutative(OpcodeToExtract) && B == C)) {
- Value *CC = B == D ? C : D;
- // Form "(A op CC) op' B" if it simplifies completely..
- // Does "A op CC" simplify?
- if (Value *V = SimplifyBinOp(Opcode, A, CC, Q, MaxRecurse)) {
- // It does! Return "V op' B" if it simplifies or is already available.
- // If V equals A then "V op' B" is just the LHS. If V equals CC then
- // "V op' B" is just the RHS.
- if (V == A || V == CC) {
- ++NumFactor;
- return V == A ? LHS : RHS;
- }
- // Otherwise return "V op' B" if it simplifies.
- if (Value *W = SimplifyBinOp(OpcodeToExtract, V, B, Q, MaxRecurse)) {
- ++NumFactor;
- return W;
- }
- }
- }
-
- return 0;
+ return nullptr;
}
/// SimplifyAssociativeBinOp - Generic simplifications for associative binary
// Recursion is always used, so bail out at once if we already hit the limit.
if (!MaxRecurse--)
- return 0;
+ return nullptr;
BinaryOperator *Op0 = dyn_cast<BinaryOperator>(LHS);
BinaryOperator *Op1 = dyn_cast<BinaryOperator>(RHS);
// The remaining transforms require commutativity as well as associativity.
if (!Instruction::isCommutative(Opcode))
- return 0;
+ return nullptr;
// Transform: "(A op B) op C" ==> "(C op A) op B" if it simplifies completely.
if (Op0 && Op0->getOpcode() == Opcode) {
}
}
- return 0;
+ return nullptr;
}
/// ThreadBinOpOverSelect - In the case of a binary operation with a select
const Query &Q, unsigned MaxRecurse) {
// Recursion is always used, so bail out at once if we already hit the limit.
if (!MaxRecurse--)
- return 0;
+ return nullptr;
SelectInst *SI;
if (isa<SelectInst>(LHS)) {
}
}
- return 0;
+ return nullptr;
}
/// ThreadCmpOverSelect - In the case of a comparison with a select instruction,
unsigned MaxRecurse) {
// Recursion is always used, so bail out at once if we already hit the limit.
if (!MaxRecurse--)
- return 0;
+ return nullptr;
// Make sure the select is on the LHS.
if (!isa<SelectInst>(LHS)) {
// It didn't simplify. However if "cmp TV, RHS" is equal to the select
// condition then we can replace it with 'true'. Otherwise give up.
if (!isSameCompare(Cond, Pred, TV, RHS))
- return 0;
+ return nullptr;
TCmp = getTrue(Cond->getType());
}
// It didn't simplify. However if "cmp FV, RHS" is equal to the select
// condition then we can replace it with 'false'. Otherwise give up.
if (!isSameCompare(Cond, Pred, FV, RHS))
- return 0;
+ return nullptr;
FCmp = getFalse(Cond->getType());
}
// The remaining cases only make sense if the select condition has the same
// type as the result of the comparison, so bail out if this is not so.
if (Cond->getType()->isVectorTy() != RHS->getType()->isVectorTy())
- return 0;
+ return nullptr;
// If the false value simplified to false, then the result of the compare
// is equal to "Cond && TCmp". This also catches the case when the false
// value simplified to false and the true value to true, returning "Cond".
Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
/// ThreadBinOpOverPHI - In the case of a binary operation with an operand that
const Query &Q, unsigned MaxRecurse) {
// Recursion is always used, so bail out at once if we already hit the limit.
if (!MaxRecurse--)
- return 0;
+ return nullptr;
PHINode *PI;
if (isa<PHINode>(LHS)) {
PI = cast<PHINode>(LHS);
// Bail out if RHS and the phi may be mutually interdependent due to a loop.
if (!ValueDominatesPHI(RHS, PI, Q.DT))
- return 0;
+ return nullptr;
} else {
assert(isa<PHINode>(RHS) && "No PHI instruction operand!");
PI = cast<PHINode>(RHS);
// Bail out if LHS and the phi may be mutually interdependent due to a loop.
if (!ValueDominatesPHI(LHS, PI, Q.DT))
- return 0;
+ return nullptr;
}
// Evaluate the BinOp on the incoming phi values.
- Value *CommonValue = 0;
+ Value *CommonValue = nullptr;
for (unsigned i = 0, e = PI->getNumIncomingValues(); i != e; ++i) {
Value *Incoming = PI->getIncomingValue(i);
// If the incoming value is the phi node itself, it can safely be skipped.
// If the operation failed to simplify, or simplified to a different value
// to previously, then give up.
if (!V || (CommonValue && V != CommonValue))
- return 0;
+ return nullptr;
CommonValue = V;
}
const Query &Q, unsigned MaxRecurse) {
// Recursion is always used, so bail out at once if we already hit the limit.
if (!MaxRecurse--)
- return 0;
+ return nullptr;
// Make sure the phi is on the LHS.
if (!isa<PHINode>(LHS)) {
// Bail out if RHS and the phi may be mutually interdependent due to a loop.
if (!ValueDominatesPHI(RHS, PI, Q.DT))
- return 0;
+ return nullptr;
// Evaluate the BinOp on the incoming phi values.
- Value *CommonValue = 0;
+ Value *CommonValue = nullptr;
for (unsigned i = 0, e = PI->getNumIncomingValues(); i != e; ++i) {
Value *Incoming = PI->getIncomingValue(i);
// If the incoming value is the phi node itself, it can safely be skipped.
// If the operation failed to simplify, or simplified to a different value
// to previously, then give up.
if (!V || (CommonValue && V != CommonValue))
- return 0;
+ return nullptr;
CommonValue = V;
}
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::Add, CLHS->getType(), Ops,
- Q.TD, Q.TLI);
+ Q.DL, Q.TLI);
}
// Canonicalize the constant to the RHS.
// X + (Y - X) -> Y
// (Y - X) + X -> Y
// Eg: X + -X -> 0
- Value *Y = 0;
+ Value *Y = nullptr;
if (match(Op1, m_Sub(m_Value(Y), m_Specific(Op0))) ||
match(Op0, m_Sub(m_Value(Y), m_Specific(Op1))))
return Y;
MaxRecurse))
return V;
- // Mul distributes over Add. Try some generic simplifications based on this.
- if (Value *V = FactorizeBinOp(Instruction::Add, Op0, Op1, Instruction::Mul,
- Q, MaxRecurse))
- return V;
-
// Threading Add over selects and phi nodes is pointless, so don't bother.
// Threading over the select in "A + select(cond, B, C)" means evaluating
// "A+B" and "A+C" and seeing if they are equal; but they are equal if and
// "A+B" and "A+C" thus gains nothing, but costs compile time. Similarly
// for threading over phi nodes.
- return 0;
+ return nullptr;
}
Value *llvm::SimplifyAddInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW,
- const DataLayout *TD, const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyAddInst(Op0, Op1, isNSW, isNUW, Query (TD, TLI, DT),
- RecursionLimit);
+ const DataLayout *DL, const TargetLibraryInfo *TLI,
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyAddInst(Op0, Op1, isNSW, isNUW,
+ Query (DL, TLI, DT, AT, CxtI), RecursionLimit);
}
/// \brief Compute the base pointer and cumulative constant offsets for V.
/// accumulates the total constant offset applied in the returned constant. It
/// returns 0 if V is not a pointer, and returns the constant '0' if there are
/// no constant offsets applied.
-static Constant *stripAndComputeConstantOffsets(const DataLayout &TD,
- Value *&V) {
- if (!V->getType()->isPointerTy())
- return 0;
-
- unsigned IntPtrWidth = TD.getPointerSizeInBits();
- APInt Offset = APInt::getNullValue(IntPtrWidth);
+///
+/// This is very similar to GetPointerBaseWithConstantOffset except it doesn't
+/// follow non-inbounds geps. This allows it to remain usable for icmp ult/etc.
+/// folding.
+static Constant *stripAndComputeConstantOffsets(const DataLayout *DL,
+ Value *&V,
+ bool AllowNonInbounds = false) {
+ assert(V->getType()->getScalarType()->isPointerTy());
+
+ // Without DataLayout, just be conservative for now. Theoretically, more could
+ // be done in this case.
+ if (!DL)
+ return ConstantInt::get(IntegerType::get(V->getContext(), 64), 0);
+
+ Type *IntPtrTy = DL->getIntPtrType(V->getType())->getScalarType();
+ APInt Offset = APInt::getNullValue(IntPtrTy->getIntegerBitWidth());
// Even though we don't look through PHI nodes, we could be called on an
// instruction in an unreachable block, which may be on a cycle.
Visited.insert(V);
do {
if (GEPOperator *GEP = dyn_cast<GEPOperator>(V)) {
- if (!GEP->isInBounds() || !GEP->accumulateConstantOffset(TD, Offset))
+ if ((!AllowNonInbounds && !GEP->isInBounds()) ||
+ !GEP->accumulateConstantOffset(*DL, Offset))
break;
V = GEP->getPointerOperand();
} else if (Operator::getOpcode(V) == Instruction::BitCast) {
} else {
break;
}
- assert(V->getType()->isPointerTy() && "Unexpected operand type!");
- } while (Visited.insert(V));
+ assert(V->getType()->getScalarType()->isPointerTy() &&
+ "Unexpected operand type!");
+ } while (Visited.insert(V).second);
- Type *IntPtrTy = TD.getIntPtrType(V->getContext());
- return ConstantInt::get(IntPtrTy, Offset);
+ Constant *OffsetIntPtr = ConstantInt::get(IntPtrTy, Offset);
+ if (V->getType()->isVectorTy())
+ return ConstantVector::getSplat(V->getType()->getVectorNumElements(),
+ OffsetIntPtr);
+ return OffsetIntPtr;
}
/// \brief Compute the constant difference between two pointer values.
/// If the difference is not a constant, returns zero.
-static Constant *computePointerDifference(const DataLayout &TD,
+static Constant *computePointerDifference(const DataLayout *DL,
Value *LHS, Value *RHS) {
- Constant *LHSOffset = stripAndComputeConstantOffsets(TD, LHS);
- if (!LHSOffset)
- return 0;
- Constant *RHSOffset = stripAndComputeConstantOffsets(TD, RHS);
- if (!RHSOffset)
- return 0;
+ Constant *LHSOffset = stripAndComputeConstantOffsets(DL, LHS);
+ Constant *RHSOffset = stripAndComputeConstantOffsets(DL, RHS);
// If LHS and RHS are not related via constant offsets to the same base
// value, there is nothing we can do here.
if (LHS != RHS)
- return 0;
+ return nullptr;
// Otherwise, the difference of LHS - RHS can be computed as:
// LHS - RHS
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::Sub, CLHS->getType(),
- Ops, Q.TD, Q.TLI);
+ Ops, Q.DL, Q.TLI);
}
// X - undef -> undef
if (Op0 == Op1)
return Constant::getNullValue(Op0->getType());
- // (X*2) - X -> X
- // (X<<1) - X -> X
- Value *X = 0;
- if (match(Op0, m_Mul(m_Specific(Op1), m_ConstantInt<2>())) ||
- match(Op0, m_Shl(m_Specific(Op1), m_One())))
- return Op1;
+ // 0 - X -> 0 if the sub is NUW.
+ if (isNUW && match(Op0, m_Zero()))
+ return Op0;
// (X + Y) - Z -> X + (Y - Z) or Y + (X - Z) if everything simplifies.
// For example, (X + Y) - Y -> X; (Y + X) - Y -> X
- Value *Y = 0, *Z = Op1;
+ Value *X = nullptr, *Y = nullptr, *Z = Op1;
if (MaxRecurse && match(Op0, m_Add(m_Value(X), m_Value(Y)))) { // (X + Y) - Z
// See if "V === Y - Z" simplifies.
if (Value *V = SimplifyBinOp(Instruction::Sub, Y, Z, Q, MaxRecurse-1))
return W;
// Variations on GEP(base, I, ...) - GEP(base, i, ...) -> GEP(null, I-i, ...).
- if (Q.TD && match(Op0, m_PtrToInt(m_Value(X))) &&
+ if (match(Op0, m_PtrToInt(m_Value(X))) &&
match(Op1, m_PtrToInt(m_Value(Y))))
- if (Constant *Result = computePointerDifference(*Q.TD, X, Y))
+ if (Constant *Result = computePointerDifference(Q.DL, X, Y))
return ConstantExpr::getIntegerCast(Result, Op0->getType(), true);
- // Mul distributes over Sub. Try some generic simplifications based on this.
- if (Value *V = FactorizeBinOp(Instruction::Sub, Op0, Op1, Instruction::Mul,
- Q, MaxRecurse))
- return V;
-
// i1 sub -> xor.
if (MaxRecurse && Op0->getType()->isIntegerTy(1))
if (Value *V = SimplifyXorInst(Op0, Op1, Q, MaxRecurse-1))
// "A-B" and "A-C" thus gains nothing, but costs compile time. Similarly
// for threading over phi nodes.
- return 0;
+ return nullptr;
}
Value *llvm::SimplifySubInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW,
- const DataLayout *TD, const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifySubInst(Op0, Op1, isNSW, isNUW, Query (TD, TLI, DT),
- RecursionLimit);
+ const DataLayout *DL, const TargetLibraryInfo *TLI,
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifySubInst(Op0, Op1, isNSW, isNUW,
+ Query (DL, TLI, DT, AT, CxtI), RecursionLimit);
}
/// Given operands for an FAdd, see if we can fold the result. If not, this
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::FAdd, CLHS->getType(),
- Ops, Q.TD, Q.TLI);
+ Ops, Q.DL, Q.TLI);
}
// Canonicalize the constant to the RHS.
// fadd [nnan ninf] X, (fsub [nnan ninf] 0, X) ==> 0
// where nnan and ninf have to occur at least once somewhere in this
// expression
- Value *SubOp = 0;
+ Value *SubOp = nullptr;
if (match(Op1, m_FSub(m_AnyZero(), m_Specific(Op0))))
SubOp = Op1;
else if (match(Op0, m_FSub(m_AnyZero(), m_Specific(Op1))))
return Constant::getNullValue(Op0->getType());
}
- return 0;
+ return nullptr;
}
/// Given operands for an FSub, see if we can fold the result. If not, this
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::FSub, CLHS->getType(),
- Ops, Q.TD, Q.TLI);
+ Ops, Q.DL, Q.TLI);
}
}
if (FMF.noNaNs() && FMF.noInfs() && Op0 == Op1)
return Constant::getNullValue(Op0->getType());
- return 0;
+ return nullptr;
}
/// Given the operands for an FMul, see if we can fold the result
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::FMul, CLHS->getType(),
- Ops, Q.TD, Q.TLI);
+ Ops, Q.DL, Q.TLI);
}
// Canonicalize the constant to the RHS.
if (FMF.noNaNs() && FMF.noSignedZeros() && match(Op1, m_AnyZero()))
return Op1;
- return 0;
+ return nullptr;
}
/// SimplifyMulInst - Given operands for a Mul, see if we can
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::Mul, CLHS->getType(),
- Ops, Q.TD, Q.TLI);
+ Ops, Q.DL, Q.TLI);
}
// Canonicalize the constant to the RHS.
return Op0;
// (X / Y) * Y -> X if the division is exact.
- Value *X = 0;
+ Value *X = nullptr;
if (match(Op0, m_Exact(m_IDiv(m_Value(X), m_Specific(Op1)))) || // (X / Y) * Y
match(Op1, m_Exact(m_IDiv(m_Value(X), m_Specific(Op0))))) // Y * (X / Y)
return X;
MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
Value *llvm::SimplifyFAddInst(Value *Op0, Value *Op1, FastMathFlags FMF,
- const DataLayout *TD, const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyFAddInst(Op0, Op1, FMF, Query (TD, TLI, DT), RecursionLimit);
+ const DataLayout *DL, const TargetLibraryInfo *TLI,
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyFAddInst(Op0, Op1, FMF, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
Value *llvm::SimplifyFSubInst(Value *Op0, Value *Op1, FastMathFlags FMF,
- const DataLayout *TD, const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyFSubInst(Op0, Op1, FMF, Query (TD, TLI, DT), RecursionLimit);
+ const DataLayout *DL, const TargetLibraryInfo *TLI,
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyFSubInst(Op0, Op1, FMF, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
Value *llvm::SimplifyFMulInst(Value *Op0, Value *Op1,
FastMathFlags FMF,
- const DataLayout *TD,
+ const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyFMulInst(Op0, Op1, FMF, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyFMulInst(Op0, Op1, FMF, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
-Value *llvm::SimplifyMulInst(Value *Op0, Value *Op1, const DataLayout *TD,
+Value *llvm::SimplifyMulInst(Value *Op0, Value *Op1, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyMulInst(Op0, Op1, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyMulInst(Op0, Op1, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
/// SimplifyDiv - Given operands for an SDiv or UDiv, see if we can
if (Constant *C0 = dyn_cast<Constant>(Op0)) {
if (Constant *C1 = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { C0, C1 };
- return ConstantFoldInstOperands(Opcode, C0->getType(), Ops, Q.TD, Q.TLI);
+ return ConstantFoldInstOperands(Opcode, C0->getType(), Ops, Q.DL, Q.TLI);
}
}
if (match(Op1, m_Undef()))
return Op1;
+ // X / 0 -> undef, we don't need to preserve faults!
+ if (match(Op1, m_Zero()))
+ return UndefValue::get(Op1->getType());
+
// undef / X -> 0
if (match(Op0, m_Undef()))
return Constant::getNullValue(Op0->getType());
return ConstantInt::get(Op0->getType(), 1);
// (X * Y) / Y -> X if the multiplication does not overflow.
- Value *X = 0, *Y = 0;
+ Value *X = nullptr, *Y = nullptr;
if (match(Op0, m_Mul(m_Value(X), m_Value(Y))) && (X == Op1 || Y == Op1)) {
if (Y != Op1) std::swap(X, Y); // Ensure expression is (X * Y) / Y, Y = Op1
OverflowingBinaryOperator *Mul = cast<OverflowingBinaryOperator>(Op0);
(!isSigned && match(Op0, m_URem(m_Value(), m_Specific(Op1)))))
return Constant::getNullValue(Op0->getType());
+ // (X /u C1) /u C2 -> 0 if C1 * C2 overflow
+ ConstantInt *C1, *C2;
+ if (!isSigned && match(Op0, m_UDiv(m_Value(X), m_ConstantInt(C1))) &&
+ match(Op1, m_ConstantInt(C2))) {
+ bool Overflow;
+ C1->getValue().umul_ov(C2->getValue(), Overflow);
+ if (Overflow)
+ return Constant::getNullValue(Op0->getType());
+ }
+
// If the operation is with the result of a select instruction, check whether
// operating on either branch of the select always yields the same value.
if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1))
if (Value *V = ThreadBinOpOverPHI(Opcode, Op0, Op1, Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
/// SimplifySDivInst - Given operands for an SDiv, see if we can
if (Value *V = SimplifyDiv(Instruction::SDiv, Op0, Op1, Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
-Value *llvm::SimplifySDivInst(Value *Op0, Value *Op1, const DataLayout *TD,
+Value *llvm::SimplifySDivInst(Value *Op0, Value *Op1, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifySDivInst(Op0, Op1, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifySDivInst(Op0, Op1, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
/// SimplifyUDivInst - Given operands for a UDiv, see if we can
if (Value *V = SimplifyDiv(Instruction::UDiv, Op0, Op1, Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
-Value *llvm::SimplifyUDivInst(Value *Op0, Value *Op1, const DataLayout *TD,
+Value *llvm::SimplifyUDivInst(Value *Op0, Value *Op1, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyUDivInst(Op0, Op1, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyUDivInst(Op0, Op1, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
static Value *SimplifyFDivInst(Value *Op0, Value *Op1, const Query &Q,
if (match(Op1, m_Undef()))
return Op1;
- return 0;
+ return nullptr;
}
-Value *llvm::SimplifyFDivInst(Value *Op0, Value *Op1, const DataLayout *TD,
+Value *llvm::SimplifyFDivInst(Value *Op0, Value *Op1, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyFDivInst(Op0, Op1, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyFDivInst(Op0, Op1, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
/// SimplifyRem - Given operands for an SRem or URem, see if we can
if (Constant *C0 = dyn_cast<Constant>(Op0)) {
if (Constant *C1 = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { C0, C1 };
- return ConstantFoldInstOperands(Opcode, C0->getType(), Ops, Q.TD, Q.TLI);
+ return ConstantFoldInstOperands(Opcode, C0->getType(), Ops, Q.DL, Q.TLI);
}
}
if (Op0 == Op1)
return Constant::getNullValue(Op0->getType());
+ // (X % Y) % Y -> X % Y
+ if ((Opcode == Instruction::SRem &&
+ match(Op0, m_SRem(m_Value(), m_Specific(Op1)))) ||
+ (Opcode == Instruction::URem &&
+ match(Op0, m_URem(m_Value(), m_Specific(Op1)))))
+ return Op0;
+
// If the operation is with the result of a select instruction, check whether
// operating on either branch of the select always yields the same value.
if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1))
if (Value *V = ThreadBinOpOverPHI(Opcode, Op0, Op1, Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
/// SimplifySRemInst - Given operands for an SRem, see if we can
if (Value *V = SimplifyRem(Instruction::SRem, Op0, Op1, Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
-Value *llvm::SimplifySRemInst(Value *Op0, Value *Op1, const DataLayout *TD,
+Value *llvm::SimplifySRemInst(Value *Op0, Value *Op1, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifySRemInst(Op0, Op1, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifySRemInst(Op0, Op1, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
/// SimplifyURemInst - Given operands for a URem, see if we can
if (Value *V = SimplifyRem(Instruction::URem, Op0, Op1, Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
-Value *llvm::SimplifyURemInst(Value *Op0, Value *Op1, const DataLayout *TD,
+Value *llvm::SimplifyURemInst(Value *Op0, Value *Op1, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyURemInst(Op0, Op1, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyURemInst(Op0, Op1, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
static Value *SimplifyFRemInst(Value *Op0, Value *Op1, const Query &,
if (match(Op1, m_Undef()))
return Op1;
- return 0;
+ return nullptr;
}
-Value *llvm::SimplifyFRemInst(Value *Op0, Value *Op1, const DataLayout *TD,
+Value *llvm::SimplifyFRemInst(Value *Op0, Value *Op1, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyFRemInst(Op0, Op1, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyFRemInst(Op0, Op1, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
+}
+
+/// isUndefShift - Returns true if a shift by \c Amount always yields undef.
+static bool isUndefShift(Value *Amount) {
+ Constant *C = dyn_cast<Constant>(Amount);
+ if (!C)
+ return false;
+
+ // X shift by undef -> undef because it may shift by the bitwidth.
+ if (isa<UndefValue>(C))
+ return true;
+
+ // Shifting by the bitwidth or more is undefined.
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(C))
+ if (CI->getValue().getLimitedValue() >=
+ CI->getType()->getScalarSizeInBits())
+ return true;
+
+ // If all lanes of a vector shift are undefined the whole shift is.
+ if (isa<ConstantVector>(C) || isa<ConstantDataVector>(C)) {
+ for (unsigned I = 0, E = C->getType()->getVectorNumElements(); I != E; ++I)
+ if (!isUndefShift(C->getAggregateElement(I)))
+ return false;
+ return true;
+ }
+
+ return false;
}
/// SimplifyShift - Given operands for an Shl, LShr or AShr, see if we can
if (Constant *C0 = dyn_cast<Constant>(Op0)) {
if (Constant *C1 = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { C0, C1 };
- return ConstantFoldInstOperands(Opcode, C0->getType(), Ops, Q.TD, Q.TLI);
+ return ConstantFoldInstOperands(Opcode, C0->getType(), Ops, Q.DL, Q.TLI);
}
}
if (match(Op1, m_Zero()))
return Op0;
- // X shift by undef -> undef because it may shift by the bitwidth.
- if (match(Op1, m_Undef()))
- return Op1;
-
- // Shifting by the bitwidth or more is undefined.
- if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1))
- if (CI->getValue().getLimitedValue() >=
- Op0->getType()->getScalarSizeInBits())
- return UndefValue::get(Op0->getType());
+ // Fold undefined shifts.
+ if (isUndefShift(Op1))
+ return UndefValue::get(Op0->getType());
// If the operation is with the result of a select instruction, check whether
// operating on either branch of the select always yields the same value.
if (Value *V = ThreadBinOpOverPHI(Opcode, Op0, Op1, Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
+}
+
+/// \brief Given operands for an Shl, LShr or AShr, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifyRightShift(unsigned Opcode, Value *Op0, Value *Op1,
+ bool isExact, const Query &Q,
+ unsigned MaxRecurse) {
+ if (Value *V = SimplifyShift(Opcode, Op0, Op1, Q, MaxRecurse))
+ return V;
+
+ // X >> X -> 0
+ if (Op0 == Op1)
+ return Constant::getNullValue(Op0->getType());
+
+ // undef >> X -> 0
+ // undef >> X -> undef (if it's exact)
+ if (match(Op0, m_Undef()))
+ return isExact ? Op0 : Constant::getNullValue(Op0->getType());
+
+ // The low bit cannot be shifted out of an exact shift if it is set.
+ if (isExact) {
+ unsigned BitWidth = Op0->getType()->getScalarSizeInBits();
+ APInt Op0KnownZero(BitWidth, 0);
+ APInt Op0KnownOne(BitWidth, 0);
+ computeKnownBits(Op0, Op0KnownZero, Op0KnownOne, Q.DL, /*Depth=*/0, Q.AT, Q.CxtI,
+ Q.DT);
+ if (Op0KnownOne[0])
+ return Op0;
+ }
+
+ return nullptr;
}
/// SimplifyShlInst - Given operands for an Shl, see if we can
return V;
// undef << X -> 0
+ // undef << X -> undef if (if it's NSW/NUW)
if (match(Op0, m_Undef()))
- return Constant::getNullValue(Op0->getType());
+ return isNSW || isNUW ? Op0 : Constant::getNullValue(Op0->getType());
// (X >> A) << A -> X
Value *X;
if (match(Op0, m_Exact(m_Shr(m_Value(X), m_Specific(Op1)))))
return X;
- return 0;
+ return nullptr;
}
Value *llvm::SimplifyShlInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW,
- const DataLayout *TD, const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyShlInst(Op0, Op1, isNSW, isNUW, Query (TD, TLI, DT),
+ const DataLayout *DL, const TargetLibraryInfo *TLI,
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyShlInst(Op0, Op1, isNSW, isNUW, Query (DL, TLI, DT, AT, CxtI),
RecursionLimit);
}
/// fold the result. If not, this returns null.
static Value *SimplifyLShrInst(Value *Op0, Value *Op1, bool isExact,
const Query &Q, unsigned MaxRecurse) {
- if (Value *V = SimplifyShift(Instruction::LShr, Op0, Op1, Q, MaxRecurse))
- return V;
-
- // undef >>l X -> 0
- if (match(Op0, m_Undef()))
- return Constant::getNullValue(Op0->getType());
+ if (Value *V = SimplifyRightShift(Instruction::LShr, Op0, Op1, isExact, Q,
+ MaxRecurse))
+ return V;
// (X << A) >> A -> X
Value *X;
- if (match(Op0, m_Shl(m_Value(X), m_Specific(Op1))) &&
- cast<OverflowingBinaryOperator>(Op0)->hasNoUnsignedWrap())
+ if (match(Op0, m_NUWShl(m_Value(X), m_Specific(Op1))))
return X;
- return 0;
+ return nullptr;
}
Value *llvm::SimplifyLShrInst(Value *Op0, Value *Op1, bool isExact,
- const DataLayout *TD,
+ const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyLShrInst(Op0, Op1, isExact, Query (TD, TLI, DT),
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyLShrInst(Op0, Op1, isExact, Query (DL, TLI, DT, AT, CxtI),
RecursionLimit);
}
/// fold the result. If not, this returns null.
static Value *SimplifyAShrInst(Value *Op0, Value *Op1, bool isExact,
const Query &Q, unsigned MaxRecurse) {
- if (Value *V = SimplifyShift(Instruction::AShr, Op0, Op1, Q, MaxRecurse))
+ if (Value *V = SimplifyRightShift(Instruction::AShr, Op0, Op1, isExact, Q,
+ MaxRecurse))
return V;
// all ones >>a X -> all ones
if (match(Op0, m_AllOnes()))
return Op0;
- // undef >>a X -> all ones
- if (match(Op0, m_Undef()))
- return Constant::getAllOnesValue(Op0->getType());
-
// (X << A) >> A -> X
Value *X;
- if (match(Op0, m_Shl(m_Value(X), m_Specific(Op1))) &&
- cast<OverflowingBinaryOperator>(Op0)->hasNoSignedWrap())
+ if (match(Op0, m_NSWShl(m_Value(X), m_Specific(Op1))))
return X;
- return 0;
+ // Arithmetic shifting an all-sign-bit value is a no-op.
+ unsigned NumSignBits = ComputeNumSignBits(Op0, Q.DL, 0, Q.AT, Q.CxtI, Q.DT);
+ if (NumSignBits == Op0->getType()->getScalarSizeInBits())
+ return Op0;
+
+ return nullptr;
}
Value *llvm::SimplifyAShrInst(Value *Op0, Value *Op1, bool isExact,
- const DataLayout *TD,
+ const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyAShrInst(Op0, Op1, isExact, Query (TD, TLI, DT),
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyAShrInst(Op0, Op1, isExact, Query (DL, TLI, DT, AT, CxtI),
RecursionLimit);
}
+static Value *simplifyUnsignedRangeCheck(ICmpInst *ZeroICmp,
+ ICmpInst *UnsignedICmp, bool IsAnd) {
+ Value *X, *Y;
+
+ ICmpInst::Predicate EqPred;
+ if (!match(ZeroICmp, m_ICmp(EqPred, m_Value(Y), m_Zero())) ||
+ !ICmpInst::isEquality(EqPred))
+ return nullptr;
+
+ ICmpInst::Predicate UnsignedPred;
+ if (match(UnsignedICmp, m_ICmp(UnsignedPred, m_Value(X), m_Specific(Y))) &&
+ ICmpInst::isUnsigned(UnsignedPred))
+ ;
+ else if (match(UnsignedICmp,
+ m_ICmp(UnsignedPred, m_Value(Y), m_Specific(X))) &&
+ ICmpInst::isUnsigned(UnsignedPred))
+ UnsignedPred = ICmpInst::getSwappedPredicate(UnsignedPred);
+ else
+ return nullptr;
+
+ // X < Y && Y != 0 --> X < Y
+ // X < Y || Y != 0 --> Y != 0
+ if (UnsignedPred == ICmpInst::ICMP_ULT && EqPred == ICmpInst::ICMP_NE)
+ return IsAnd ? UnsignedICmp : ZeroICmp;
+
+ // X >= Y || Y != 0 --> true
+ // X >= Y || Y == 0 --> X >= Y
+ if (UnsignedPred == ICmpInst::ICMP_UGE && !IsAnd) {
+ if (EqPred == ICmpInst::ICMP_NE)
+ return getTrue(UnsignedICmp->getType());
+ return UnsignedICmp;
+ }
+
+ // X < Y && Y == 0 --> false
+ if (UnsignedPred == ICmpInst::ICMP_ULT && EqPred == ICmpInst::ICMP_EQ &&
+ IsAnd)
+ return getFalse(UnsignedICmp->getType());
+
+ return nullptr;
+}
+
+// Simplify (and (icmp ...) (icmp ...)) to true when we can tell that the range
+// of possible values cannot be satisfied.
+static Value *SimplifyAndOfICmps(ICmpInst *Op0, ICmpInst *Op1) {
+ ICmpInst::Predicate Pred0, Pred1;
+ ConstantInt *CI1, *CI2;
+ Value *V;
+
+ if (Value *X = simplifyUnsignedRangeCheck(Op0, Op1, /*IsAnd=*/true))
+ return X;
+
+ if (!match(Op0, m_ICmp(Pred0, m_Add(m_Value(V), m_ConstantInt(CI1)),
+ m_ConstantInt(CI2))))
+ return nullptr;
+
+ if (!match(Op1, m_ICmp(Pred1, m_Specific(V), m_Specific(CI1))))
+ return nullptr;
+
+ Type *ITy = Op0->getType();
+
+ auto *AddInst = cast<BinaryOperator>(Op0->getOperand(0));
+ bool isNSW = AddInst->hasNoSignedWrap();
+ bool isNUW = AddInst->hasNoUnsignedWrap();
+
+ const APInt &CI1V = CI1->getValue();
+ const APInt &CI2V = CI2->getValue();
+ const APInt Delta = CI2V - CI1V;
+ if (CI1V.isStrictlyPositive()) {
+ if (Delta == 2) {
+ if (Pred0 == ICmpInst::ICMP_ULT && Pred1 == ICmpInst::ICMP_SGT)
+ return getFalse(ITy);
+ if (Pred0 == ICmpInst::ICMP_SLT && Pred1 == ICmpInst::ICMP_SGT && isNSW)
+ return getFalse(ITy);
+ }
+ if (Delta == 1) {
+ if (Pred0 == ICmpInst::ICMP_ULE && Pred1 == ICmpInst::ICMP_SGT)
+ return getFalse(ITy);
+ if (Pred0 == ICmpInst::ICMP_SLE && Pred1 == ICmpInst::ICMP_SGT && isNSW)
+ return getFalse(ITy);
+ }
+ }
+ if (CI1V.getBoolValue() && isNUW) {
+ if (Delta == 2)
+ if (Pred0 == ICmpInst::ICMP_ULT && Pred1 == ICmpInst::ICMP_UGT)
+ return getFalse(ITy);
+ if (Delta == 1)
+ if (Pred0 == ICmpInst::ICMP_ULE && Pred1 == ICmpInst::ICMP_UGT)
+ return getFalse(ITy);
+ }
+
+ return nullptr;
+}
+
/// SimplifyAndInst - Given operands for an And, see if we can
/// fold the result. If not, this returns null.
static Value *SimplifyAndInst(Value *Op0, Value *Op1, const Query &Q,
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::And, CLHS->getType(),
- Ops, Q.TD, Q.TLI);
+ Ops, Q.DL, Q.TLI);
}
// Canonicalize the constant to the RHS.
return Constant::getNullValue(Op0->getType());
// (A | ?) & A = A
- Value *A = 0, *B = 0;
+ Value *A = nullptr, *B = nullptr;
if (match(Op0, m_Or(m_Value(A), m_Value(B))) &&
(A == Op1 || B == Op1))
return Op1;
// A & (-A) = A if A is a power of two or zero.
if (match(Op0, m_Neg(m_Specific(Op1))) ||
match(Op1, m_Neg(m_Specific(Op0)))) {
- if (isKnownToBeAPowerOfTwo(Op0, /*OrZero*/true))
+ if (isKnownToBeAPowerOfTwo(Op0, /*OrZero*/true, 0, Q.AT, Q.CxtI, Q.DT))
return Op0;
- if (isKnownToBeAPowerOfTwo(Op1, /*OrZero*/true))
+ if (isKnownToBeAPowerOfTwo(Op1, /*OrZero*/true, 0, Q.AT, Q.CxtI, Q.DT))
return Op1;
}
+ if (auto *ICILHS = dyn_cast<ICmpInst>(Op0)) {
+ if (auto *ICIRHS = dyn_cast<ICmpInst>(Op1)) {
+ if (Value *V = SimplifyAndOfICmps(ICILHS, ICIRHS))
+ return V;
+ if (Value *V = SimplifyAndOfICmps(ICIRHS, ICILHS))
+ return V;
+ }
+ }
+
// Try some generic simplifications for associative operations.
if (Value *V = SimplifyAssociativeBinOp(Instruction::And, Op0, Op1, Q,
MaxRecurse))
Q, MaxRecurse))
return V;
- // Or distributes over And. Try some generic simplifications based on this.
- if (Value *V = FactorizeBinOp(Instruction::And, Op0, Op1, Instruction::Or,
- Q, MaxRecurse))
- return V;
-
// If the operation is with the result of a select instruction, check whether
// operating on either branch of the select always yields the same value.
if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1))
MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
-Value *llvm::SimplifyAndInst(Value *Op0, Value *Op1, const DataLayout *TD,
+Value *llvm::SimplifyAndInst(Value *Op0, Value *Op1, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyAndInst(Op0, Op1, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyAndInst(Op0, Op1, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
+}
+
+// Simplify (or (icmp ...) (icmp ...)) to true when we can tell that the union
+// contains all possible values.
+static Value *SimplifyOrOfICmps(ICmpInst *Op0, ICmpInst *Op1) {
+ ICmpInst::Predicate Pred0, Pred1;
+ ConstantInt *CI1, *CI2;
+ Value *V;
+
+ if (Value *X = simplifyUnsignedRangeCheck(Op0, Op1, /*IsAnd=*/false))
+ return X;
+
+ if (!match(Op0, m_ICmp(Pred0, m_Add(m_Value(V), m_ConstantInt(CI1)),
+ m_ConstantInt(CI2))))
+ return nullptr;
+
+ if (!match(Op1, m_ICmp(Pred1, m_Specific(V), m_Specific(CI1))))
+ return nullptr;
+
+ Type *ITy = Op0->getType();
+
+ auto *AddInst = cast<BinaryOperator>(Op0->getOperand(0));
+ bool isNSW = AddInst->hasNoSignedWrap();
+ bool isNUW = AddInst->hasNoUnsignedWrap();
+
+ const APInt &CI1V = CI1->getValue();
+ const APInt &CI2V = CI2->getValue();
+ const APInt Delta = CI2V - CI1V;
+ if (CI1V.isStrictlyPositive()) {
+ if (Delta == 2) {
+ if (Pred0 == ICmpInst::ICMP_UGE && Pred1 == ICmpInst::ICMP_SLE)
+ return getTrue(ITy);
+ if (Pred0 == ICmpInst::ICMP_SGE && Pred1 == ICmpInst::ICMP_SLE && isNSW)
+ return getTrue(ITy);
+ }
+ if (Delta == 1) {
+ if (Pred0 == ICmpInst::ICMP_UGT && Pred1 == ICmpInst::ICMP_SLE)
+ return getTrue(ITy);
+ if (Pred0 == ICmpInst::ICMP_SGT && Pred1 == ICmpInst::ICMP_SLE && isNSW)
+ return getTrue(ITy);
+ }
+ }
+ if (CI1V.getBoolValue() && isNUW) {
+ if (Delta == 2)
+ if (Pred0 == ICmpInst::ICMP_UGE && Pred1 == ICmpInst::ICMP_ULE)
+ return getTrue(ITy);
+ if (Delta == 1)
+ if (Pred0 == ICmpInst::ICMP_UGT && Pred1 == ICmpInst::ICMP_ULE)
+ return getTrue(ITy);
+ }
+
+ return nullptr;
}
/// SimplifyOrInst - Given operands for an Or, see if we can
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::Or, CLHS->getType(),
- Ops, Q.TD, Q.TLI);
+ Ops, Q.DL, Q.TLI);
}
// Canonicalize the constant to the RHS.
return Constant::getAllOnesValue(Op0->getType());
// (A & ?) | A = A
- Value *A = 0, *B = 0;
+ Value *A = nullptr, *B = nullptr;
if (match(Op0, m_And(m_Value(A), m_Value(B))) &&
(A == Op1 || B == Op1))
return Op1;
(A == Op0 || B == Op0))
return Constant::getAllOnesValue(Op0->getType());
+ if (auto *ICILHS = dyn_cast<ICmpInst>(Op0)) {
+ if (auto *ICIRHS = dyn_cast<ICmpInst>(Op1)) {
+ if (Value *V = SimplifyOrOfICmps(ICILHS, ICIRHS))
+ return V;
+ if (Value *V = SimplifyOrOfICmps(ICIRHS, ICILHS))
+ return V;
+ }
+ }
+
// Try some generic simplifications for associative operations.
if (Value *V = SimplifyAssociativeBinOp(Instruction::Or, Op0, Op1, Q,
MaxRecurse))
MaxRecurse))
return V;
- // And distributes over Or. Try some generic simplifications based on this.
- if (Value *V = FactorizeBinOp(Instruction::Or, Op0, Op1, Instruction::And,
- Q, MaxRecurse))
- return V;
-
// If the operation is with the result of a select instruction, check whether
// operating on either branch of the select always yields the same value.
if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1))
MaxRecurse))
return V;
+ // (A & C)|(B & D)
+ Value *C = nullptr, *D = nullptr;
+ if (match(Op0, m_And(m_Value(A), m_Value(C))) &&
+ match(Op1, m_And(m_Value(B), m_Value(D)))) {
+ ConstantInt *C1 = dyn_cast<ConstantInt>(C);
+ ConstantInt *C2 = dyn_cast<ConstantInt>(D);
+ if (C1 && C2 && (C1->getValue() == ~C2->getValue())) {
+ // (A & C1)|(B & C2)
+ // If we have: ((V + N) & C1) | (V & C2)
+ // .. and C2 = ~C1 and C2 is 0+1+ and (N & C2) == 0
+ // replace with V+N.
+ Value *V1, *V2;
+ if ((C2->getValue() & (C2->getValue() + 1)) == 0 && // C2 == 0+1+
+ match(A, m_Add(m_Value(V1), m_Value(V2)))) {
+ // Add commutes, try both ways.
+ if (V1 == B && MaskedValueIsZero(V2, C2->getValue(), Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT))
+ return A;
+ if (V2 == B && MaskedValueIsZero(V1, C2->getValue(), Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT))
+ return A;
+ }
+ // Or commutes, try both ways.
+ if ((C1->getValue() & (C1->getValue() + 1)) == 0 &&
+ match(B, m_Add(m_Value(V1), m_Value(V2)))) {
+ // Add commutes, try both ways.
+ if (V1 == A && MaskedValueIsZero(V2, C1->getValue(), Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT))
+ return B;
+ if (V2 == A && MaskedValueIsZero(V1, C1->getValue(), Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT))
+ return B;
+ }
+ }
+ }
+
// If the operation is with the result of a phi instruction, check whether
// operating on all incoming values of the phi always yields the same value.
if (isa<PHINode>(Op0) || isa<PHINode>(Op1))
if (Value *V = ThreadBinOpOverPHI(Instruction::Or, Op0, Op1, Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
-Value *llvm::SimplifyOrInst(Value *Op0, Value *Op1, const DataLayout *TD,
+Value *llvm::SimplifyOrInst(Value *Op0, Value *Op1, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyOrInst(Op0, Op1, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyOrInst(Op0, Op1, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
/// SimplifyXorInst - Given operands for a Xor, see if we can
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::Xor, CLHS->getType(),
- Ops, Q.TD, Q.TLI);
+ Ops, Q.DL, Q.TLI);
}
// Canonicalize the constant to the RHS.
MaxRecurse))
return V;
- // And distributes over Xor. Try some generic simplifications based on this.
- if (Value *V = FactorizeBinOp(Instruction::Xor, Op0, Op1, Instruction::And,
- Q, MaxRecurse))
- return V;
-
// Threading Xor over selects and phi nodes is pointless, so don't bother.
// Threading over the select in "A ^ select(cond, B, C)" means evaluating
// "A^B" and "A^C" and seeing if they are equal; but they are equal if and
// "A^B" and "A^C" thus gains nothing, but costs compile time. Similarly
// for threading over phi nodes.
- return 0;
+ return nullptr;
}
-Value *llvm::SimplifyXorInst(Value *Op0, Value *Op1, const DataLayout *TD,
+Value *llvm::SimplifyXorInst(Value *Op0, Value *Op1, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyXorInst(Op0, Op1, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyXorInst(Op0, Op1, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
static Type *GetCompareTy(Value *Op) {
Value *LHS, Value *RHS) {
SelectInst *SI = dyn_cast<SelectInst>(V);
if (!SI)
- return 0;
+ return nullptr;
CmpInst *Cmp = dyn_cast<CmpInst>(SI->getCondition());
if (!Cmp)
- return 0;
+ return nullptr;
Value *CmpLHS = Cmp->getOperand(0), *CmpRHS = Cmp->getOperand(1);
if (Pred == Cmp->getPredicate() && LHS == CmpLHS && RHS == CmpRHS)
return Cmp;
if (Pred == CmpInst::getSwappedPredicate(Cmp->getPredicate()) &&
LHS == CmpRHS && RHS == CmpLHS)
return Cmp;
- return 0;
+ return nullptr;
}
-static Constant *computePointerICmp(const DataLayout &TD,
+// A significant optimization not implemented here is assuming that alloca
+// addresses are not equal to incoming argument values. They don't *alias*,
+// as we say, but that doesn't mean they aren't equal, so we take a
+// conservative approach.
+//
+// This is inspired in part by C++11 5.10p1:
+// "Two pointers of the same type compare equal if and only if they are both
+// null, both point to the same function, or both represent the same
+// address."
+//
+// This is pretty permissive.
+//
+// It's also partly due to C11 6.5.9p6:
+// "Two pointers compare equal if and only if both are null pointers, both are
+// pointers to the same object (including a pointer to an object and a
+// subobject at its beginning) or function, both are pointers to one past the
+// last element of the same array object, or one is a pointer to one past the
+// end of one array object and the other is a pointer to the start of a
+// different array object that happens to immediately follow the first array
+// object in the address space.)
+//
+// C11's version is more restrictive, however there's no reason why an argument
+// couldn't be a one-past-the-end value for a stack object in the caller and be
+// equal to the beginning of a stack object in the callee.
+//
+// If the C and C++ standards are ever made sufficiently restrictive in this
+// area, it may be possible to update LLVM's semantics accordingly and reinstate
+// this optimization.
+static Constant *computePointerICmp(const DataLayout *DL,
+ const TargetLibraryInfo *TLI,
CmpInst::Predicate Pred,
Value *LHS, Value *RHS) {
+ // First, skip past any trivial no-ops.
+ LHS = LHS->stripPointerCasts();
+ RHS = RHS->stripPointerCasts();
+
+ // A non-null pointer is not equal to a null pointer.
+ if (llvm::isKnownNonNull(LHS, TLI) && isa<ConstantPointerNull>(RHS) &&
+ (Pred == CmpInst::ICMP_EQ || Pred == CmpInst::ICMP_NE))
+ return ConstantInt::get(GetCompareTy(LHS),
+ !CmpInst::isTrueWhenEqual(Pred));
+
// We can only fold certain predicates on pointer comparisons.
switch (Pred) {
default:
- return 0;
+ return nullptr;
// Equality comaprisons are easy to fold.
case CmpInst::ICMP_EQ:
break;
}
- Constant *LHSOffset = stripAndComputeConstantOffsets(TD, LHS);
- if (!LHSOffset)
- return 0;
- Constant *RHSOffset = stripAndComputeConstantOffsets(TD, RHS);
- if (!RHSOffset)
- return 0;
+ // Strip off any constant offsets so that we can reason about them.
+ // It's tempting to use getUnderlyingObject or even just stripInBoundsOffsets
+ // here and compare base addresses like AliasAnalysis does, however there are
+ // numerous hazards. AliasAnalysis and its utilities rely on special rules
+ // governing loads and stores which don't apply to icmps. Also, AliasAnalysis
+ // doesn't need to guarantee pointer inequality when it says NoAlias.
+ Constant *LHSOffset = stripAndComputeConstantOffsets(DL, LHS);
+ Constant *RHSOffset = stripAndComputeConstantOffsets(DL, RHS);
+
+ // If LHS and RHS are related via constant offsets to the same base
+ // value, we can replace it with an icmp which just compares the offsets.
+ if (LHS == RHS)
+ return ConstantExpr::getICmp(Pred, LHSOffset, RHSOffset);
+
+ // Various optimizations for (in)equality comparisons.
+ if (Pred == CmpInst::ICMP_EQ || Pred == CmpInst::ICMP_NE) {
+ // Different non-empty allocations that exist at the same time have
+ // different addresses (if the program can tell). Global variables always
+ // exist, so they always exist during the lifetime of each other and all
+ // allocas. Two different allocas usually have different addresses...
+ //
+ // However, if there's an @llvm.stackrestore dynamically in between two
+ // allocas, they may have the same address. It's tempting to reduce the
+ // scope of the problem by only looking at *static* allocas here. That would
+ // cover the majority of allocas while significantly reducing the likelihood
+ // of having an @llvm.stackrestore pop up in the middle. However, it's not
+ // actually impossible for an @llvm.stackrestore to pop up in the middle of
+ // an entry block. Also, if we have a block that's not attached to a
+ // function, we can't tell if it's "static" under the current definition.
+ // Theoretically, this problem could be fixed by creating a new kind of
+ // instruction kind specifically for static allocas. Such a new instruction
+ // could be required to be at the top of the entry block, thus preventing it
+ // from being subject to a @llvm.stackrestore. Instcombine could even
+ // convert regular allocas into these special allocas. It'd be nifty.
+ // However, until then, this problem remains open.
+ //
+ // So, we'll assume that two non-empty allocas have different addresses
+ // for now.
+ //
+ // With all that, if the offsets are within the bounds of their allocations
+ // (and not one-past-the-end! so we can't use inbounds!), and their
+ // allocations aren't the same, the pointers are not equal.
+ //
+ // Note that it's not necessary to check for LHS being a global variable
+ // address, due to canonicalization and constant folding.
+ if (isa<AllocaInst>(LHS) &&
+ (isa<AllocaInst>(RHS) || isa<GlobalVariable>(RHS))) {
+ ConstantInt *LHSOffsetCI = dyn_cast<ConstantInt>(LHSOffset);
+ ConstantInt *RHSOffsetCI = dyn_cast<ConstantInt>(RHSOffset);
+ uint64_t LHSSize, RHSSize;
+ if (LHSOffsetCI && RHSOffsetCI &&
+ getObjectSize(LHS, LHSSize, DL, TLI) &&
+ getObjectSize(RHS, RHSSize, DL, TLI)) {
+ const APInt &LHSOffsetValue = LHSOffsetCI->getValue();
+ const APInt &RHSOffsetValue = RHSOffsetCI->getValue();
+ if (!LHSOffsetValue.isNegative() &&
+ !RHSOffsetValue.isNegative() &&
+ LHSOffsetValue.ult(LHSSize) &&
+ RHSOffsetValue.ult(RHSSize)) {
+ return ConstantInt::get(GetCompareTy(LHS),
+ !CmpInst::isTrueWhenEqual(Pred));
+ }
+ }
- // If LHS and RHS are not related via constant offsets to the same base
- // value, there is nothing we can do here.
- if (LHS != RHS)
- return 0;
+ // Repeat the above check but this time without depending on DataLayout
+ // or being able to compute a precise size.
+ if (!cast<PointerType>(LHS->getType())->isEmptyTy() &&
+ !cast<PointerType>(RHS->getType())->isEmptyTy() &&
+ LHSOffset->isNullValue() &&
+ RHSOffset->isNullValue())
+ return ConstantInt::get(GetCompareTy(LHS),
+ !CmpInst::isTrueWhenEqual(Pred));
+ }
- return ConstantExpr::getICmp(Pred, LHSOffset, RHSOffset);
+ // Even if an non-inbounds GEP occurs along the path we can still optimize
+ // equality comparisons concerning the result. We avoid walking the whole
+ // chain again by starting where the last calls to
+ // stripAndComputeConstantOffsets left off and accumulate the offsets.
+ Constant *LHSNoBound = stripAndComputeConstantOffsets(DL, LHS, true);
+ Constant *RHSNoBound = stripAndComputeConstantOffsets(DL, RHS, true);
+ if (LHS == RHS)
+ return ConstantExpr::getICmp(Pred,
+ ConstantExpr::getAdd(LHSOffset, LHSNoBound),
+ ConstantExpr::getAdd(RHSOffset, RHSNoBound));
+
+ // If one side of the equality comparison must come from a noalias call
+ // (meaning a system memory allocation function), and the other side must
+ // come from a pointer that cannot overlap with dynamically-allocated
+ // memory within the lifetime of the current function (allocas, byval
+ // arguments, globals), then determine the comparison result here.
+ SmallVector<Value *, 8> LHSUObjs, RHSUObjs;
+ GetUnderlyingObjects(LHS, LHSUObjs, DL);
+ GetUnderlyingObjects(RHS, RHSUObjs, DL);
+
+ // Is the set of underlying objects all noalias calls?
+ auto IsNAC = [](SmallVectorImpl<Value *> &Objects) {
+ return std::all_of(Objects.begin(), Objects.end(),
+ [](Value *V){ return isNoAliasCall(V); });
+ };
+
+ // Is the set of underlying objects all things which must be disjoint from
+ // noalias calls. For allocas, we consider only static ones (dynamic
+ // allocas might be transformed into calls to malloc not simultaneously
+ // live with the compared-to allocation). For globals, we exclude symbols
+ // that might be resolve lazily to symbols in another dynamically-loaded
+ // library (and, thus, could be malloc'ed by the implementation).
+ auto IsAllocDisjoint = [](SmallVectorImpl<Value *> &Objects) {
+ return std::all_of(Objects.begin(), Objects.end(),
+ [](Value *V){
+ if (const AllocaInst *AI = dyn_cast<AllocaInst>(V))
+ return AI->getParent() && AI->getParent()->getParent() &&
+ AI->isStaticAlloca();
+ if (const GlobalValue *GV = dyn_cast<GlobalValue>(V))
+ return (GV->hasLocalLinkage() ||
+ GV->hasHiddenVisibility() ||
+ GV->hasProtectedVisibility() ||
+ GV->hasUnnamedAddr()) &&
+ !GV->isThreadLocal();
+ if (const Argument *A = dyn_cast<Argument>(V))
+ return A->hasByValAttr();
+ return false;
+ });
+ };
+
+ if ((IsNAC(LHSUObjs) && IsAllocDisjoint(RHSUObjs)) ||
+ (IsNAC(RHSUObjs) && IsAllocDisjoint(LHSUObjs)))
+ return ConstantInt::get(GetCompareTy(LHS),
+ !CmpInst::isTrueWhenEqual(Pred));
+ }
+
+ // Otherwise, fail.
+ return nullptr;
}
/// SimplifyICmpInst - Given operands for an ICmpInst, see if we can
if (Constant *CLHS = dyn_cast<Constant>(LHS)) {
if (Constant *CRHS = dyn_cast<Constant>(RHS))
- return ConstantFoldCompareInstOperands(Pred, CLHS, CRHS, Q.TD, Q.TLI);
+ return ConstantFoldCompareInstOperands(Pred, CLHS, CRHS, Q.DL, Q.TLI);
// If we have a constant, make sure it is on the RHS.
std::swap(LHS, RHS);
}
}
- // icmp <object*>, <object*/null> - Different identified objects have
- // different addresses (unless null), and what's more the address of an
- // identified local is never equal to another argument (again, barring null).
- // Note that generalizing to the case where LHS is a global variable address
- // or null is pointless, since if both LHS and RHS are constants then we
- // already constant folded the compare, and if only one of them is then we
- // moved it to RHS already.
- Value *LHSPtr = LHS->stripPointerCasts();
- Value *RHSPtr = RHS->stripPointerCasts();
- if (LHSPtr == RHSPtr)
- return ConstantInt::get(ITy, CmpInst::isTrueWhenEqual(Pred));
-
- // Be more aggressive about stripping pointer adjustments when checking a
- // comparison of an alloca address to another object. We can rip off all
- // inbounds GEP operations, even if they are variable.
- LHSPtr = LHSPtr->stripInBoundsOffsets();
- if (llvm::isIdentifiedObject(LHSPtr)) {
- RHSPtr = RHSPtr->stripInBoundsOffsets();
- if (llvm::isKnownNonNull(LHSPtr) || llvm::isKnownNonNull(RHSPtr)) {
- // If both sides are different identified objects, they aren't equal
- // unless they're null.
- if (LHSPtr != RHSPtr && llvm::isIdentifiedObject(RHSPtr) &&
- Pred == CmpInst::ICMP_EQ)
- return ConstantInt::get(ITy, false);
-
- // A local identified object (alloca or noalias call) can't equal any
- // incoming argument, unless they're both null or they belong to
- // different functions. The latter happens during inlining.
- if (Instruction *LHSInst = dyn_cast<Instruction>(LHSPtr))
- if (Argument *RHSArg = dyn_cast<Argument>(RHSPtr))
- if (LHSInst->getParent()->getParent() == RHSArg->getParent() &&
- Pred == CmpInst::ICMP_EQ)
- return ConstantInt::get(ITy, false);
- }
-
- // Assume that the constant null is on the right.
- if (llvm::isKnownNonNull(LHSPtr) && isa<ConstantPointerNull>(RHSPtr)) {
- if (Pred == CmpInst::ICMP_EQ)
- return ConstantInt::get(ITy, false);
- else if (Pred == CmpInst::ICMP_NE)
- return ConstantInt::get(ITy, true);
- }
- } else if (Argument *LHSArg = dyn_cast<Argument>(LHSPtr)) {
- RHSPtr = RHSPtr->stripInBoundsOffsets();
- // An alloca can't be equal to an argument unless they come from separate
- // functions via inlining.
- if (AllocaInst *RHSInst = dyn_cast<AllocaInst>(RHSPtr)) {
- if (LHSArg->getParent() == RHSInst->getParent()->getParent()) {
- if (Pred == CmpInst::ICMP_EQ)
- return ConstantInt::get(ITy, false);
- else if (Pred == CmpInst::ICMP_NE)
- return ConstantInt::get(ITy, true);
- }
- }
- }
-
// If we are comparing with zero then try hard since this is a common case.
if (match(RHS, m_Zero())) {
bool LHSKnownNonNegative, LHSKnownNegative;
return getTrue(ITy);
case ICmpInst::ICMP_EQ:
case ICmpInst::ICMP_ULE:
- if (isKnownNonZero(LHS, Q.TD))
+ if (isKnownNonZero(LHS, Q.DL, 0, Q.AT, Q.CxtI, Q.DT))
return getFalse(ITy);
break;
case ICmpInst::ICMP_NE:
case ICmpInst::ICMP_UGT:
- if (isKnownNonZero(LHS, Q.TD))
+ if (isKnownNonZero(LHS, Q.DL, 0, Q.AT, Q.CxtI, Q.DT))
return getTrue(ITy);
break;
case ICmpInst::ICMP_SLT:
- ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.TD);
+ ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT);
if (LHSKnownNegative)
return getTrue(ITy);
if (LHSKnownNonNegative)
return getFalse(ITy);
break;
case ICmpInst::ICMP_SLE:
- ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.TD);
+ ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT);
if (LHSKnownNegative)
return getTrue(ITy);
- if (LHSKnownNonNegative && isKnownNonZero(LHS, Q.TD))
+ if (LHSKnownNonNegative && isKnownNonZero(LHS, Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT))
return getFalse(ITy);
break;
case ICmpInst::ICMP_SGE:
- ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.TD);
+ ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT);
if (LHSKnownNegative)
return getFalse(ITy);
if (LHSKnownNonNegative)
return getTrue(ITy);
break;
case ICmpInst::ICMP_SGT:
- ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.TD);
+ ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT);
if (LHSKnownNegative)
return getFalse(ITy);
- if (LHSKnownNonNegative && isKnownNonZero(LHS, Q.TD))
+ if (LHSKnownNonNegative && isKnownNonZero(LHS, Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT))
return getTrue(ITy);
break;
}
// Many binary operators with constant RHS have easy to compute constant
// range. Use them to check whether the comparison is a tautology.
- uint32_t Width = CI->getBitWidth();
+ unsigned Width = CI->getBitWidth();
APInt Lower = APInt(Width, 0);
APInt Upper = APInt(Width, 0);
ConstantInt *CI2;
APInt NegOne = APInt::getAllOnesValue(Width);
if (!CI2->isZero())
Upper = NegOne.udiv(CI2->getValue()) + 1;
+ } else if (match(LHS, m_SDiv(m_ConstantInt(CI2), m_Value()))) {
+ if (CI2->isMinSignedValue()) {
+ // 'sdiv INT_MIN, x' produces [INT_MIN, INT_MIN / -2].
+ Lower = CI2->getValue();
+ Upper = Lower.lshr(1) + 1;
+ } else {
+ // 'sdiv CI2, x' produces [-|CI2|, |CI2|].
+ Upper = CI2->getValue().abs() + 1;
+ Lower = (-Upper) + 1;
+ }
} else if (match(LHS, m_SDiv(m_Value(), m_ConstantInt(CI2)))) {
- // 'sdiv x, CI2' produces [INT_MIN / CI2, INT_MAX / CI2].
APInt IntMin = APInt::getSignedMinValue(Width);
APInt IntMax = APInt::getSignedMaxValue(Width);
- APInt Val = CI2->getValue().abs();
- if (!Val.isMinValue()) {
+ APInt Val = CI2->getValue();
+ if (Val.isAllOnesValue()) {
+ // 'sdiv x, -1' produces [INT_MIN + 1, INT_MAX]
+ // where CI2 != -1 and CI2 != 0 and CI2 != 1
+ Lower = IntMin + 1;
+ Upper = IntMax + 1;
+ } else if (Val.countLeadingZeros() < Width - 1) {
+ // 'sdiv x, CI2' produces [INT_MIN / CI2, INT_MAX / CI2]
+ // where CI2 != -1 and CI2 != 0 and CI2 != 1
Lower = IntMin.sdiv(Val);
- Upper = IntMax.sdiv(Val) + 1;
+ Upper = IntMax.sdiv(Val);
+ if (Lower.sgt(Upper))
+ std::swap(Lower, Upper);
+ Upper = Upper + 1;
+ assert(Upper != Lower && "Upper part of range has wrapped!");
+ }
+ } else if (match(LHS, m_NUWShl(m_ConstantInt(CI2), m_Value()))) {
+ // 'shl nuw CI2, x' produces [CI2, CI2 << CLZ(CI2)]
+ Lower = CI2->getValue();
+ Upper = Lower.shl(Lower.countLeadingZeros()) + 1;
+ } else if (match(LHS, m_NSWShl(m_ConstantInt(CI2), m_Value()))) {
+ if (CI2->isNegative()) {
+ // 'shl nsw CI2, x' produces [CI2 << CLO(CI2)-1, CI2]
+ unsigned ShiftAmount = CI2->getValue().countLeadingOnes() - 1;
+ Lower = CI2->getValue().shl(ShiftAmount);
+ Upper = CI2->getValue() + 1;
+ } else {
+ // 'shl nsw CI2, x' produces [CI2, CI2 << CLZ(CI2)-1]
+ unsigned ShiftAmount = CI2->getValue().countLeadingZeros() - 1;
+ Lower = CI2->getValue();
+ Upper = CI2->getValue().shl(ShiftAmount) + 1;
}
} else if (match(LHS, m_LShr(m_Value(), m_ConstantInt(CI2)))) {
// 'lshr x, CI2' produces [0, UINT_MAX >> CI2].
APInt NegOne = APInt::getAllOnesValue(Width);
if (CI2->getValue().ult(Width))
Upper = NegOne.lshr(CI2->getValue()) + 1;
+ } else if (match(LHS, m_LShr(m_ConstantInt(CI2), m_Value()))) {
+ // 'lshr CI2, x' produces [CI2 >> (Width-1), CI2].
+ unsigned ShiftAmount = Width - 1;
+ if (!CI2->isZero() && cast<BinaryOperator>(LHS)->isExact())
+ ShiftAmount = CI2->getValue().countTrailingZeros();
+ Lower = CI2->getValue().lshr(ShiftAmount);
+ Upper = CI2->getValue() + 1;
} else if (match(LHS, m_AShr(m_Value(), m_ConstantInt(CI2)))) {
// 'ashr x, CI2' produces [INT_MIN >> CI2, INT_MAX >> CI2].
APInt IntMin = APInt::getSignedMinValue(Width);
Lower = IntMin.ashr(CI2->getValue());
Upper = IntMax.ashr(CI2->getValue()) + 1;
}
+ } else if (match(LHS, m_AShr(m_ConstantInt(CI2), m_Value()))) {
+ unsigned ShiftAmount = Width - 1;
+ if (!CI2->isZero() && cast<BinaryOperator>(LHS)->isExact())
+ ShiftAmount = CI2->getValue().countTrailingZeros();
+ if (CI2->isNegative()) {
+ // 'ashr CI2, x' produces [CI2, CI2 >> (Width-1)]
+ Lower = CI2->getValue();
+ Upper = CI2->getValue().ashr(ShiftAmount) + 1;
+ } else {
+ // 'ashr CI2, x' produces [CI2 >> (Width-1), CI2]
+ Lower = CI2->getValue().ashr(ShiftAmount);
+ Upper = CI2->getValue() + 1;
+ }
} else if (match(LHS, m_Or(m_Value(), m_ConstantInt(CI2)))) {
// 'or x, CI2' produces [CI2, UINT_MAX].
Lower = CI2->getValue();
// Turn icmp (ptrtoint x), (ptrtoint/constant) into a compare of the input
// if the integer type is the same size as the pointer type.
- if (MaxRecurse && Q.
TD && isa<PtrToIntInst>(LI) &&
- Q.TD->getPointerSizeInBits() == DstTy->getPrimitiveSizeInBits()) {
+ if (MaxRecurse && Q.
DL && isa<PtrToIntInst>(LI) &&
+ Q.DL->getTypeSizeInBits(SrcTy) == DstTy->getPrimitiveSizeInBits()) {
if (Constant *RHSC = dyn_cast<Constant>(RHS)) {
// Transfer the cast to the constant.
if (Value *V = SimplifyICmpInst(Pred, SrcOp,
BinaryOperator *RBO = dyn_cast<BinaryOperator>(RHS);
if (MaxRecurse && (LBO || RBO)) {
// Analyze the case when either LHS or RHS is an add instruction.
- Value *A = 0, *B = 0, *C = 0, *D = 0;
+ Value *A = nullptr, *B = nullptr, *C = nullptr, *D = nullptr;
// LHS = A + B (or A and B are null); RHS = C + D (or C and D are null).
bool NoLHSWrapProblem = false, NoRHSWrapProblem = false;
if (LBO && LBO->getOpcode() == Instruction::Add) {
}
}
+ // icmp pred (or X, Y), X
+ if (LBO && match(LBO, m_CombineOr(m_Or(m_Value(), m_Specific(RHS)),
+ m_Or(m_Specific(RHS), m_Value())))) {
+ if (Pred == ICmpInst::ICMP_ULT)
+ return getFalse(ITy);
+ if (Pred == ICmpInst::ICMP_UGE)
+ return getTrue(ITy);
+ }
+ // icmp pred X, (or X, Y)
+ if (RBO && match(RBO, m_CombineOr(m_Or(m_Value(), m_Specific(LHS)),
+ m_Or(m_Specific(LHS), m_Value())))) {
+ if (Pred == ICmpInst::ICMP_ULE)
+ return getTrue(ITy);
+ if (Pred == ICmpInst::ICMP_UGT)
+ return getFalse(ITy);
+ }
+
+ // icmp pred (and X, Y), X
+ if (LBO && match(LBO, m_CombineOr(m_And(m_Value(), m_Specific(RHS)),
+ m_And(m_Specific(RHS), m_Value())))) {
+ if (Pred == ICmpInst::ICMP_UGT)
+ return getFalse(ITy);
+ if (Pred == ICmpInst::ICMP_ULE)
+ return getTrue(ITy);
+ }
+ // icmp pred X, (and X, Y)
+ if (RBO && match(RBO, m_CombineOr(m_And(m_Value(), m_Specific(LHS)),
+ m_And(m_Specific(LHS), m_Value())))) {
+ if (Pred == ICmpInst::ICMP_UGE)
+ return getTrue(ITy);
+ if (Pred == ICmpInst::ICMP_ULT)
+ return getFalse(ITy);
+ }
+
+ // 0 - (zext X) pred C
+ if (!CmpInst::isUnsigned(Pred) && match(LHS, m_Neg(m_ZExt(m_Value())))) {
+ if (ConstantInt *RHSC = dyn_cast<ConstantInt>(RHS)) {
+ if (RHSC->getValue().isStrictlyPositive()) {
+ if (Pred == ICmpInst::ICMP_SLT)
+ return ConstantInt::getTrue(RHSC->getContext());
+ if (Pred == ICmpInst::ICMP_SGE)
+ return ConstantInt::getFalse(RHSC->getContext());
+ if (Pred == ICmpInst::ICMP_EQ)
+ return ConstantInt::getFalse(RHSC->getContext());
+ if (Pred == ICmpInst::ICMP_NE)
+ return ConstantInt::getTrue(RHSC->getContext());
+ }
+ if (RHSC->getValue().isNonNegative()) {
+ if (Pred == ICmpInst::ICMP_SLE)
+ return ConstantInt::getTrue(RHSC->getContext());
+ if (Pred == ICmpInst::ICMP_SGT)
+ return ConstantInt::getFalse(RHSC->getContext());
+ }
+ }
+ }
+
+ // icmp pred (urem X, Y), Y
if (LBO && match(LBO, m_URem(m_Value(), m_Specific(RHS)))) {
bool KnownNonNegative, KnownNegative;
switch (Pred) {
break;
case ICmpInst::ICMP_SGT:
case ICmpInst::ICMP_SGE:
- ComputeSignBit(LHS, KnownNonNegative, KnownNegative, Q.TD);
+ ComputeSignBit(RHS, KnownNonNegative, KnownNegative, Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT);
if (!KnownNonNegative)
break;
// fall-through
return getFalse(ITy);
case ICmpInst::ICMP_SLT:
case ICmpInst::ICMP_SLE:
- ComputeSignBit(LHS, KnownNonNegative, KnownNegative, Q.TD);
+ ComputeSignBit(RHS, KnownNonNegative, KnownNegative, Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT);
if (!KnownNonNegative)
break;
// fall-through
return getTrue(ITy);
}
}
+
+ // icmp pred X, (urem Y, X)
if (RBO && match(RBO, m_URem(m_Value(), m_Specific(LHS)))) {
bool KnownNonNegative, KnownNegative;
switch (Pred) {
break;
case ICmpInst::ICMP_SGT:
case ICmpInst::ICMP_SGE:
- ComputeSignBit(RHS, KnownNonNegative, KnownNegative, Q.TD);
+ ComputeSignBit(LHS, KnownNonNegative, KnownNegative, Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT);
if (!KnownNonNegative)
break;
// fall-through
return getTrue(ITy);
case ICmpInst::ICMP_SLT:
case ICmpInst::ICMP_SLE:
- ComputeSignBit(RHS, KnownNonNegative, KnownNegative, Q.TD);
+ ComputeSignBit(LHS, KnownNonNegative, KnownNegative, Q.DL,
+ 0, Q.AT, Q.CxtI, Q.DT);
if (!KnownNonNegative)
break;
// fall-through
return getTrue(ITy);
}
+ // handle:
+ // CI2 << X == CI
+ // CI2 << X != CI
+ //
+ // where CI2 is a power of 2 and CI isn't
+ if (auto *CI = dyn_cast<ConstantInt>(RHS)) {
+ const APInt *CI2Val, *CIVal = &CI->getValue();
+ if (LBO && match(LBO, m_Shl(m_APInt(CI2Val), m_Value())) &&
+ CI2Val->isPowerOf2()) {
+ if (!CIVal->isPowerOf2()) {
+ // CI2 << X can equal zero in some circumstances,
+ // this simplification is unsafe if CI is zero.
+ //
+ // We know it is safe if:
+ // - The shift is nsw, we can't shift out the one bit.
+ // - The shift is nuw, we can't shift out the one bit.
+ // - CI2 is one
+ // - CI isn't zero
+ if (LBO->hasNoSignedWrap() || LBO->hasNoUnsignedWrap() ||
+ *CI2Val == 1 || !CI->isZero()) {
+ if (Pred == ICmpInst::ICMP_EQ)
+ return ConstantInt::getFalse(RHS->getContext());
+ if (Pred == ICmpInst::ICMP_NE)
+ return ConstantInt::getTrue(RHS->getContext());
+ }
+ }
+ if (CIVal->isSignBit() && *CI2Val == 1) {
+ if (Pred == ICmpInst::ICMP_UGT)
+ return ConstantInt::getFalse(RHS->getContext());
+ if (Pred == ICmpInst::ICMP_ULE)
+ return ConstantInt::getTrue(RHS->getContext());
+ }
+ }
+ }
+
if (MaxRecurse && LBO && RBO && LBO->getOpcode() == RBO->getOpcode() &&
LBO->getOperand(1) == RBO->getOperand(1)) {
switch (LBO->getOpcode()) {
// Simplify comparisons of related pointers using a powerful, recursive
// GEP-walk when we have target data available..
- if (Q.TD && LHS->getType()->isPointerTy() && RHS->getType()->isPointerTy())
- if (Constant *C = computePointerICmp(*Q.TD, Pred, LHS, RHS))
+ if (LHS->getType()->isPointerTy())
+ if (Constant *C = computePointerICmp(Q.DL, Q.TLI, Pred, LHS, RHS))
return C;
if (GetElementPtrInst *GLHS = dyn_cast<GetElementPtrInst>(LHS)) {
}
}
+ // If a bit is known to be zero for A and known to be one for B,
+ // then A and B cannot be equal.
+ if (ICmpInst::isEquality(Pred)) {
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
+ uint32_t BitWidth = CI->getBitWidth();
+ APInt LHSKnownZero(BitWidth, 0);
+ APInt LHSKnownOne(BitWidth, 0);
+ computeKnownBits(LHS, LHSKnownZero, LHSKnownOne, Q.DL, /*Depth=*/0, Q.AT,
+ Q.CxtI, Q.DT);
+ const APInt &RHSVal = CI->getValue();
+ if (((LHSKnownZero & RHSVal) != 0) || ((LHSKnownOne & ~RHSVal) != 0))
+ return Pred == ICmpInst::ICMP_EQ
+ ? ConstantInt::getFalse(CI->getContext())
+ : ConstantInt::getTrue(CI->getContext());
+ }
+ }
+
// If the comparison is with the result of a select instruction, check whether
// comparing with either branch of the select always yields the same value.
if (isa<SelectInst>(LHS) || isa<SelectInst>(RHS))
if (Value *V = ThreadCmpOverPHI(Pred, LHS, RHS, Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
Value *llvm::SimplifyICmpInst(unsigned Predicate, Value *LHS, Value *RHS,
- const DataLayout *TD,
+ const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyICmpInst(Predicate, LHS, RHS, Query (TD, TLI, DT),
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ Instruction *CxtI) {
+ return ::SimplifyICmpInst(Predicate, LHS, RHS, Query (DL, TLI, DT, AT, CxtI),
RecursionLimit);
}
if (Constant *CLHS = dyn_cast<Constant>(LHS)) {
if (Constant *CRHS = dyn_cast<Constant>(RHS))
- return ConstantFoldCompareInstOperands(Pred, CLHS, CRHS, Q.TD, Q.TLI);
+ return ConstantFoldCompareInstOperands(Pred, CLHS, CRHS, Q.DL, Q.TLI);
// If we have a constant, make sure it is on the RHS.
std::swap(LHS, RHS);
if (Value *V = ThreadCmpOverPHI(Pred, LHS, RHS, Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
Value *llvm::SimplifyFCmpInst(unsigned Predicate, Value *LHS, Value *RHS,
- const DataLayout *TD,
+ const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyFCmpInst(Predicate, LHS, RHS, Query (TD, TLI, DT),
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyFCmpInst(Predicate, LHS, RHS, Query (DL, TLI, DT, AT, CxtI),
RecursionLimit);
}
unsigned MaxRecurse) {
// select true, X, Y -> X
// select false, X, Y -> Y
- if (ConstantInt *CB = dyn_cast<ConstantInt>(CondVal))
- return CB->getZExtValue() ? TrueVal : FalseVal;
+ if (Constant *CB = dyn_cast<Constant>(CondVal)) {
+ if (CB->isAllOnesValue())
+ return TrueVal;
+ if (CB->isNullValue())
+ return FalseVal;
+ }
// select C, X, X -> X
if (TrueVal == FalseVal)
if (isa<UndefValue>(FalseVal)) // select C, X, undef -> X
return TrueVal;
- return 0;
+ if (const auto *ICI = dyn_cast<ICmpInst>(CondVal)) {
+ Value *X;
+ const APInt *Y;
+ ICmpInst::Predicate Pred = ICI->getPredicate();
+ if (ICmpInst::isEquality(Pred) &&
+ match(ICI->getOperand(0), m_And(m_Value(X), m_APInt(Y))) &&
+ match(ICI->getOperand(1), m_Zero())) {
+ const APInt *C;
+ // (X & Y) == 0 ? X & ~Y : X --> X
+ // (X & Y) != 0 ? X & ~Y : X --> X & ~Y
+ if (FalseVal == X && match(TrueVal, m_And(m_Specific(X), m_APInt(C))) &&
+ *Y == ~*C)
+ return Pred == ICmpInst::ICMP_EQ ? FalseVal : TrueVal;
+ // (X & Y) == 0 ? X : X & ~Y --> X & ~Y
+ // (X & Y) != 0 ? X : X & ~Y --> X
+ if (TrueVal == X && match(FalseVal, m_And(m_Specific(X), m_APInt(C))) &&
+ *Y == ~*C)
+ return Pred == ICmpInst::ICMP_EQ ? FalseVal : TrueVal;
+
+ if (Y->isPowerOf2()) {
+ // (X & Y) == 0 ? X | Y : X --> X | Y
+ // (X & Y) != 0 ? X | Y : X --> X
+ if (FalseVal == X && match(TrueVal, m_Or(m_Specific(X), m_APInt(C))) &&
+ *Y == *C)
+ return Pred == ICmpInst::ICMP_EQ ? TrueVal : FalseVal;
+ // (X & Y) == 0 ? X : X | Y --> X
+ // (X & Y) != 0 ? X : X | Y --> X | Y
+ if (TrueVal == X && match(FalseVal, m_Or(m_Specific(X), m_APInt(C))) &&
+ *Y == *C)
+ return Pred == ICmpInst::ICMP_EQ ? TrueVal : FalseVal;
+ }
+ }
+ if (Pred == ICmpInst::ICMP_SLT && match(ICI->getOperand(1), m_Zero())) {
+ // (X < 0) ? X : X | SignBit --> X | SignBit
+ if (TrueVal == ICI->getOperand(0) &&
+ match(FalseVal, m_Or(m_Specific(TrueVal), m_SignBit())))
+ return FalseVal;
+ // (X < 0) ? X | SignBit : X --> X
+ if (FalseVal == ICI->getOperand(0) &&
+ match(TrueVal, m_Or(m_Specific(FalseVal), m_SignBit())))
+ return FalseVal;
+ // (X < 0) ? X & INT_MAX : X --> X & INT_MAX
+ if (FalseVal == ICI->getOperand(0) &&
+ match(TrueVal, m_And(m_Specific(FalseVal), m_MaxSignedValue())))
+ return TrueVal;
+ // (X < 0) ? X : X & INT_MAX --> X
+ if (TrueVal == ICI->getOperand(0) &&
+ match(FalseVal, m_And(m_Specific(TrueVal), m_MaxSignedValue())))
+ return TrueVal;
+ }
+ if (Pred == ICmpInst::ICMP_SGT && match(ICI->getOperand(1), m_AllOnes())) {
+ // (X > -1) ? X : X | SignBit --> X
+ if (TrueVal == ICI->getOperand(0) &&
+ match(FalseVal, m_Or(m_Specific(TrueVal), m_SignBit())))
+ return TrueVal;
+ // (X > -1) ? X | SignBit : X --> X | SignBit
+ if (FalseVal == ICI->getOperand(0) &&
+ match(TrueVal, m_Or(m_Specific(FalseVal), m_SignBit())))
+ return TrueVal;
+ // (X > -1) ? X & INT_MAX : X --> X
+ if (FalseVal == ICI->getOperand(0) &&
+ match(TrueVal, m_And(m_Specific(FalseVal), m_MaxSignedValue())))
+ return FalseVal;
+ // (X > -1) ? X : X & INT_MAX --> X & INT_MAX
+ if (TrueVal == ICI->getOperand(0) &&
+ match(FalseVal, m_And(m_Specific(TrueVal), m_MaxSignedValue())))
+ return FalseVal;
+ }
+ }
+
+ return nullptr;
}
Value *llvm::SimplifySelectInst(Value *Cond, Value *TrueVal, Value *FalseVal,
- const DataLayout *TD,
+ const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifySelectInst(Cond, TrueVal, FalseVal, Query (TD, TLI, DT),
- RecursionLimit);
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifySelectInst(Cond, TrueVal, FalseVal,
+ Query (DL, TLI, DT, AT, CxtI), RecursionLimit);
}
/// SimplifyGEPInst - Given operands for an GetElementPtrInst, see if we can
/// fold the result. If not, this returns null.
static Value *SimplifyGEPInst(ArrayRef<Value *> Ops, const Query &Q, unsigned) {
// The type of the GEP pointer operand.
- PointerType *PtrTy = dyn_cast<PointerType>(Ops[0]->getType());
- // The GEP pointer operand is not a pointer, it's a vector of pointers.
- if (!PtrTy)
- return 0;
+ PointerType *PtrTy = cast<PointerType>(Ops[0]->getType()->getScalarType());
+ unsigned AS = PtrTy->getAddressSpace();
// getelementptr P -> P.
if (Ops.size() == 1)
return Ops[0];
- if (isa<UndefValue>(Ops[0])) {
- // Compute the (pointer) type returned by the GEP instruction.
- Type *LastType = GetElementPtrInst::getIndexedType(PtrTy, Ops.slice(1));
- Type *GEPTy = PointerType::get(LastType, PtrTy->getAddressSpace());
+ // Compute the (pointer) type returned by the GEP instruction.
+ Type *LastType = GetElementPtrInst::getIndexedType(PtrTy, Ops.slice(1));
+ Type *GEPTy = PointerType::get(LastType, AS);
+ if (VectorType *VT = dyn_cast<VectorType>(Ops[0]->getType()))
+ GEPTy = VectorType::get(GEPTy, VT->getNumElements());
+
+ if (isa<UndefValue>(Ops[0]))
return UndefValue::get(GEPTy);
- }
if (Ops.size() == 2) {
// getelementptr P, 0 -> P.
- if (ConstantInt *C = dyn_cast<ConstantInt>(Ops[1]))
- if (C->isZero())
- return Ops[0];
- // getelementptr P, N -> P if P points to a type of zero size.
- if (Q.TD) {
- Type *Ty = PtrTy->getElementType();
- if (Ty->isSized() && Q.TD->getTypeAllocSize(Ty) == 0)
+ if (match(Ops[1], m_Zero()))
+ return Ops[0];
+
+ Type *Ty = PtrTy->getElementType();
+ if (Q.DL && Ty->isSized()) {
+ Value *P;
+ uint64_t C;
+ uint64_t TyAllocSize = Q.DL->getTypeAllocSize(Ty);
+ // getelementptr P, N -> P if P points to a type of zero size.
+ if (TyAllocSize == 0)
return Ops[0];
+
+ // The following transforms are only safe if the ptrtoint cast
+ // doesn't truncate the pointers.
+ if (Ops[1]->getType()->getScalarSizeInBits() ==
+ Q.DL->getPointerSizeInBits(AS)) {
+ auto PtrToIntOrZero = [GEPTy](Value *P) -> Value * {
+ if (match(P, m_Zero()))
+ return Constant::getNullValue(GEPTy);
+ Value *Temp;
+ if (match(P, m_PtrToInt(m_Value(Temp))))
+ if (Temp->getType() == GEPTy)
+ return Temp;
+ return nullptr;
+ };
+
+ // getelementptr V, (sub P, V) -> P if P points to a type of size 1.
+ if (TyAllocSize == 1 &&
+ match(Ops[1], m_Sub(m_Value(P), m_PtrToInt(m_Specific(Ops[0])))))
+ if (Value *R = PtrToIntOrZero(P))
+ return R;
+
+ // getelementptr V, (ashr (sub P, V), C) -> Q
+ // if P points to a type of size 1 << C.
+ if (match(Ops[1],
+ m_AShr(m_Sub(m_Value(P), m_PtrToInt(m_Specific(Ops[0]))),
+ m_ConstantInt(C))) &&
+ TyAllocSize == 1ULL << C)
+ if (Value *R = PtrToIntOrZero(P))
+ return R;
+
+ // getelementptr V, (sdiv (sub P, V), C) -> Q
+ // if P points to a type of size C.
+ if (match(Ops[1],
+ m_SDiv(m_Sub(m_Value(P), m_PtrToInt(m_Specific(Ops[0]))),
+ m_SpecificInt(TyAllocSize))))
+ if (Value *R = PtrToIntOrZero(P))
+ return R;
+ }
}
}
// Check to see if this is constant foldable.
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
if (!isa<Constant>(Ops[i]))
- return 0;
+ return nullptr;
return ConstantExpr::getGetElementPtr(cast<Constant>(Ops[0]), Ops.slice(1));
}
-Value *llvm::SimplifyGEPInst(ArrayRef<Value *> Ops, const DataLayout *TD,
+Value *llvm::SimplifyGEPInst(ArrayRef<Value *> Ops, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyGEPInst(Ops, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyGEPInst(Ops, Query (DL, TLI, DT, AT, CxtI), RecursionLimit);
}
/// SimplifyInsertValueInst - Given operands for an InsertValueInst, see if we
return Agg;
}
- return 0;
+ return nullptr;
}
Value *llvm::SimplifyInsertValueInst(Value *Agg, Value *Val,
ArrayRef<unsigned> Idxs,
- const DataLayout *TD,
+ const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyInsertValueInst(Agg, Val, Idxs, Query (TD, TLI, DT),
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyInsertValueInst(Agg, Val, Idxs,
+ Query (DL, TLI, DT, AT, CxtI),
RecursionLimit);
}
static Value *SimplifyPHINode(PHINode *PN, const Query &Q) {
// If all of the PHI's incoming values are the same then replace the PHI node
// with the common value.
- Value *CommonValue = 0;
+ Value *CommonValue = nullptr;
bool HasUndefInput = false;
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
Value *Incoming = PN->getIncomingValue(i);
continue;
}
if (CommonValue && Incoming != CommonValue)
- return 0; // Not the same, bail out.
+ return nullptr; // Not the same, bail out.
CommonValue = Incoming;
}
// instruction, we cannot return X as the result of the PHI node unless it
// dominates the PHI block.
if (HasUndefInput)
- return ValueDominatesPHI(CommonValue, PN, Q.DT) ? CommonValue : 0;
+ return ValueDominatesPHI(CommonValue, PN, Q.DT) ? CommonValue : nullptr;
return CommonValue;
}
static Value *SimplifyTruncInst(Value *Op, Type *Ty, const Query &Q, unsigned) {
if (Constant *C = dyn_cast<Constant>(Op))
- return ConstantFoldInstOperands(Instruction::Trunc, Ty, C, Q.TD, Q.TLI);
+ return ConstantFoldInstOperands(Instruction::Trunc, Ty, C, Q.DL, Q.TLI);
- return 0;
+ return nullptr;
}
-Value *llvm::SimplifyTruncInst(Value *Op, Type *Ty, const DataLayout *TD,
+Value *llvm::SimplifyTruncInst(Value *Op, Type *Ty, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyTruncInst(Op, Ty, Query (TD, TLI, DT), RecursionLimit);
+ const DominatorTree *DT,
+ AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyTruncInst(Op, Ty, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
//=== Helper functions for higher up the class hierarchy.
if (Constant *CLHS = dyn_cast<Constant>(LHS))
if (Constant *CRHS = dyn_cast<Constant>(RHS)) {
Constant *COps[] = {CLHS, CRHS};
- return ConstantFoldInstOperands(Opcode, LHS->getType(), COps, Q.TD,
+ return ConstantFoldInstOperands(Opcode, LHS->getType(), COps, Q.DL,
Q.TLI);
}
if (Value *V = ThreadBinOpOverPHI(Opcode, LHS, RHS, Q, MaxRecurse))
return V;
- return 0;
+ return nullptr;
}
}
Value *llvm::SimplifyBinOp(unsigned Opcode, Value *LHS, Value *RHS,
- const DataLayout *TD, const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyBinOp(Opcode, LHS, RHS, Query (TD, TLI, DT), RecursionLimit);
+ const DataLayout *DL, const TargetLibraryInfo *TLI,
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyBinOp(Opcode, LHS, RHS, Query (DL, TLI, DT, AT, CxtI),
+ RecursionLimit);
}
/// SimplifyCmpInst - Given operands for a CmpInst, see if we can
}
Value *llvm::SimplifyCmpInst(unsigned Predicate, Value *LHS, Value *RHS,
- const DataLayout *TD, const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyCmpInst(Predicate, LHS, RHS, Query (TD, TLI, DT),
+ const DataLayout *DL, const TargetLibraryInfo *TLI,
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyCmpInst(Predicate, LHS, RHS, Query (DL, TLI, DT, AT, CxtI),
RecursionLimit);
}
+static bool IsIdempotent(Intrinsic::ID ID) {
+ switch (ID) {
+ default: return false;
+
+ // Unary idempotent: f(f(x)) = f(x)
+ case Intrinsic::fabs:
+ case Intrinsic::floor:
+ case Intrinsic::ceil:
+ case Intrinsic::trunc:
+ case Intrinsic::rint:
+ case Intrinsic::nearbyint:
+ case Intrinsic::round:
+ return true;
+ }
+}
+
template <typename IterTy>
static Value *SimplifyIntrinsic(Intrinsic::ID IID, IterTy ArgBegin, IterTy ArgEnd,
const Query &Q, unsigned MaxRecurse) {
+ // Perform idempotent optimizations
+ if (!IsIdempotent(IID))
+ return nullptr;
+
+ // Unary Ops
+ if (std::distance(ArgBegin, ArgEnd) == 1)
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(*ArgBegin))
+ if (II->getIntrinsicID() == IID)
+ return II;
+
+ return nullptr;
}
template <typename IterTy>
Function *F = dyn_cast<Function>(V);
if (!F)
- return 0;
+ return nullptr;
+
+ if (unsigned IID = F->getIntrinsicID())
+ if (Value *Ret =
+ SimplifyIntrinsic((Intrinsic::ID) IID, ArgBegin, ArgEnd, Q, MaxRecurse))
+ return Ret;
if (!canConstantFoldCallTo(F))
- return 0;
+ return nullptr;
SmallVector<Constant *, 4> ConstantArgs;
ConstantArgs.reserve(ArgEnd - ArgBegin);
for (IterTy I = ArgBegin, E = ArgEnd; I != E; ++I) {
Constant *C = dyn_cast<Constant>(*I);
if (!C)
- return 0;
+ return nullptr;
ConstantArgs.push_back(C);
}
}
Value *llvm::SimplifyCall(Value *V, User::op_iterator ArgBegin,
- User::op_iterator ArgEnd, const DataLayout *TD,
+ User::op_iterator ArgEnd, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyCall(V, ArgBegin, ArgEnd, Query(TD, TLI, DT),
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyCall(V, ArgBegin, ArgEnd, Query(DL, TLI, DT, AT, CxtI),
RecursionLimit);
}
Value *llvm::SimplifyCall(Value *V, ArrayRef<Value *> Args,
- const DataLayout *TD, const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return ::SimplifyCall(V, Args.begin(), Args.end(), Query(TD, TLI, DT),
- RecursionLimit);
+ const DataLayout *DL, const TargetLibraryInfo *TLI,
+ const DominatorTree *DT, AssumptionTracker *AT,
+ const Instruction *CxtI) {
+ return ::SimplifyCall(V, Args.begin(), Args.end(),
+ Query(DL, TLI, DT, AT, CxtI), RecursionLimit);
}
/// SimplifyInstruction - See if we can compute a simplified version of this
/// instruction. If not, this returns null.
-Value *llvm::SimplifyInstruction(Instruction *I, const DataLayout *TD,
+Value *llvm::SimplifyInstruction(Instruction *I, const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
+ const DominatorTree *DT,
+ AssumptionTracker *AT) {
Value *Result;
switch (I->getOpcode()) {
default:
- Result = ConstantFoldInstruction(I, TD, TLI);
+ Result = ConstantFoldInstruction(I, DL, TLI);
break;
case Instruction::FAdd:
Result = SimplifyFAddInst(I->getOperand(0), I->getOperand(1),
- I->getFastMathFlags(), TD, TLI, DT);
+ I->getFastMathFlags(), DL, TLI, DT, AT, I);
break;
case Instruction::Add:
Result = SimplifyAddInst(I->getOperand(0), I->getOperand(1),
cast<BinaryOperator>(I)->hasNoSignedWrap(),
cast<BinaryOperator>(I)->hasNoUnsignedWrap(),
- TD, TLI, DT);
+ DL, TLI, DT, AT, I);
break;
case Instruction::FSub:
Result = SimplifyFSubInst(I->getOperand(0), I->getOperand(1),
- I->getFastMathFlags(), TD, TLI, DT);
+ I->getFastMathFlags(), DL, TLI, DT, AT, I);
break;
case Instruction::Sub:
Result = SimplifySubInst(I->getOperand(0), I->getOperand(1),
cast<BinaryOperator>(I)->hasNoSignedWrap(),
cast<BinaryOperator>(I)->hasNoUnsignedWrap(),
- TD, TLI, DT);
+ DL, TLI, DT, AT, I);
break;
case Instruction::FMul:
Result = SimplifyFMulInst(I->getOperand(0), I->getOperand(1),
- I->getFastMathFlags(), TD, TLI, DT);
+ I->getFastMathFlags(), DL, TLI, DT, AT, I);
break;
case Instruction::Mul:
- Result = SimplifyMulInst(I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ Result = SimplifyMulInst(I->getOperand(0), I->getOperand(1),
+ DL, TLI, DT, AT, I);
break;
case Instruction::SDiv:
- Result = SimplifySDivInst(I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ Result = SimplifySDivInst(I->getOperand(0), I->getOperand(1),
+ DL, TLI, DT, AT, I);
break;
case Instruction::UDiv:
- Result = SimplifyUDivInst(I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ Result = SimplifyUDivInst(I->getOperand(0), I->getOperand(1),
+ DL, TLI, DT, AT, I);
break;
case Instruction::FDiv:
- Result = SimplifyFDivInst(I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ Result = SimplifyFDivInst(I->getOperand(0), I->getOperand(1),
+ DL, TLI, DT, AT, I);
break;
case Instruction::SRem:
- Result = SimplifySRemInst(I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ Result = SimplifySRemInst(I->getOperand(0), I->getOperand(1),
+ DL, TLI, DT, AT, I);
break;
case Instruction::URem:
- Result = SimplifyURemInst(I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ Result = SimplifyURemInst(I->getOperand(0), I->getOperand(1),
+ DL, TLI, DT, AT, I);
break;
case Instruction::FRem:
- Result = SimplifyFRemInst(I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ Result = SimplifyFRemInst(I->getOperand(0), I->getOperand(1),
+ DL, TLI, DT, AT, I);
break;
case Instruction::Shl:
Result = SimplifyShlInst(I->getOperand(0), I->getOperand(1),
cast<BinaryOperator>(I)->hasNoSignedWrap(),
cast<BinaryOperator>(I)->hasNoUnsignedWrap(),
- TD, TLI, DT);
+ DL, TLI, DT, AT, I);
break;
case Instruction::LShr:
Result = SimplifyLShrInst(I->getOperand(0), I->getOperand(1),
cast<BinaryOperator>(I)->isExact(),
- TD, TLI, DT);
+ DL, TLI, DT, AT, I);
break;
case Instruction::AShr:
Result = SimplifyAShrInst(I->getOperand(0), I->getOperand(1),
cast<BinaryOperator>(I)->isExact(),
- TD, TLI, DT);
+ DL, TLI, DT, AT, I);
break;
case Instruction::And:
- Result = SimplifyAndInst(I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ Result = SimplifyAndInst(I->getOperand(0), I->getOperand(1),
+ DL, TLI, DT, AT, I);
break;
case Instruction::Or:
- Result = SimplifyOrInst(I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ Result = SimplifyOrInst(I->getOperand(0), I->getOperand(1), DL, TLI, DT,
+ AT, I);
break;
case Instruction::Xor:
- Result = SimplifyXorInst(I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ Result = SimplifyXorInst(I->getOperand(0), I->getOperand(1),
+ DL, TLI, DT, AT, I);
break;
case Instruction::ICmp:
Result = SimplifyICmpInst(cast<ICmpInst>(I)->getPredicate(),
- I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ I->getOperand(0), I->getOperand(1),
+ DL, TLI, DT, AT, I);
break;
case Instruction::FCmp:
Result = SimplifyFCmpInst(cast<FCmpInst>(I)->getPredicate(),
- I->getOperand(0), I->getOperand(1), TD, TLI, DT);
+ I->getOperand(0), I->getOperand(1),
+ DL, TLI, DT, AT, I);
break;
case Instruction::Select:
Result = SimplifySelectInst(I->getOperand(0), I->getOperand(1),
- I->getOperand(2), TD, TLI, DT);
+ I->getOperand(2), DL, TLI, DT, AT, I);
break;
case Instruction::GetElementPtr: {
SmallVector<Value*, 8> Ops(I->op_begin(), I->op_end());
- Result = SimplifyGEPInst(Ops, TD, TLI, DT);
+ Result = SimplifyGEPInst(Ops, DL, TLI, DT, AT, I);
break;
}
case Instruction::InsertValue: {
InsertValueInst *IV = cast<InsertValueInst>(I);
Result = SimplifyInsertValueInst(IV->getAggregateOperand(),
IV->getInsertedValueOperand(),
- IV->getIndices(), TD, TLI, DT);
+ IV->getIndices(), DL, TLI, DT, AT, I);
break;
}
case Instruction::PHI:
- Result = SimplifyPHINode(cast<PHINode>(I), Query (TD, TLI, DT));
+ Result = SimplifyPHINode(cast<PHINode>(I), Query (DL, TLI, DT, AT, I));
break;
case Instruction::Call: {
CallSite CS(cast<CallInst>(I));
Result = SimplifyCall(CS.getCalledValue(), CS.arg_begin(), CS.arg_end(),
- TD, TLI, DT);
+ DL, TLI, DT, AT, I);
break;
}
case Instruction::Trunc:
- Result = SimplifyTruncInst(I->getOperand(0), I->getType(), TD, TLI, DT);
+ Result = SimplifyTruncInst(I->getOperand(0), I->getType(), DL, TLI, DT,
+ AT, I);
break;
}
/// This routine returns 'true' only when *it* simplifies something. The passed
/// in simplified value does not count toward this.
static bool replaceAndRecursivelySimplifyImpl(Instruction *I, Value *SimpleV,
- const DataLayout *TD,
+ const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
+ const DominatorTree *DT,
+ AssumptionTracker *AT) {
bool Simplified = false;
SmallSetVector<Instruction *, 8> Worklist;
// If we have an explicit value to collapse to, do that round of the
// simplification loop by hand initially.
if (SimpleV) {
- for (Value::use_iterator UI = I->use_begin(), UE = I->use_end(); UI != UE;
- ++UI)
- if (*UI != I)
- Worklist.insert(cast<Instruction>(*UI));
+ for (User *U : I->users())
+ if (U != I)
+ Worklist.insert(cast<Instruction>(U));
// Replace the instruction with its simplified value.
I->replaceAllUsesWith(SimpleV);
I = Worklist[Idx];
// See if this instruction simplifies.
- SimpleV = SimplifyInstruction(I, TD, TLI, DT);
+ SimpleV = SimplifyInstruction(I, DL, TLI, DT, AT);
if (!SimpleV)
continue;
// Stash away all the uses of the old instruction so we can check them for
// recursive simplifications after a RAUW. This is cheaper than checking all
// uses of To on the recursive step in most cases.
- for (Value::use_iterator UI = I->use_begin(), UE = I->use_end(); UI != UE;
- ++UI)
- Worklist.insert(cast<Instruction>(*UI));
+ for (User *U : I->users())
+ Worklist.insert(cast<Instruction>(U));
// Replace the instruction with its simplified value.
I->replaceAllUsesWith(SimpleV);
}
bool llvm::recursivelySimplifyInstruction(Instruction *I,
- const DataLayout *TD,
+ const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
- return replaceAndRecursivelySimplifyImpl(I, 0, TD, TLI, DT);
+ const DominatorTree *DT,
+ AssumptionTracker *AT) {
+ return replaceAndRecursivelySimplifyImpl(I, nullptr, DL, TLI, DT, AT);
}
bool llvm::replaceAndRecursivelySimplify(Instruction *I, Value *SimpleV,
- const DataLayout *TD,
+ const DataLayout *DL,
const TargetLibraryInfo *TLI,
- const DominatorTree *DT) {
+ const DominatorTree *DT,
+ AssumptionTracker *AT) {
assert(I != SimpleV && "replaceAndRecursivelySimplify(X,X) is not valid!");
assert(SimpleV && "Must provide a simplified value.");
- return replaceAndRecursivelySimplifyImpl(I, SimpleV, TD, TLI, DT);
+ return replaceAndRecursivelySimplifyImpl(I, SimpleV, DL, TLI, DT, AT);
}
projects / oota-llvm.git / blobdiff