llvm-project/llvm/lib/Analysis/ConstantFolding.cpp

573 lines
21 KiB
C++

//===-- ConstantFolding.cpp - Analyze constant folding possibilities ------===//
//
// The LLVM Compiler Infrastructure
//
// This file was developed by the LLVM research group and is distributed under
// the University of Illinois Open Source License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This family of functions determines the possibility of performing constant
// folding.
//
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Function.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/MathExtras.h"
#include <cerrno>
#include <cmath>
using namespace llvm;
//===----------------------------------------------------------------------===//
// Constant Folding internal helper functions
//===----------------------------------------------------------------------===//
/// IsConstantOffsetFromGlobal - If this constant is actually a constant offset
/// from a global, return the global and the constant. Because of
/// constantexprs, this function is recursive.
static bool IsConstantOffsetFromGlobal(Constant *C, GlobalValue *&GV,
int64_t &Offset, const TargetData &TD) {
// Trivial case, constant is the global.
if ((GV = dyn_cast<GlobalValue>(C))) {
Offset = 0;
return true;
}
// Otherwise, if this isn't a constant expr, bail out.
ConstantExpr *CE = dyn_cast<ConstantExpr>(C);
if (!CE) return false;
// Look through ptr->int and ptr->ptr casts.
if (CE->getOpcode() == Instruction::PtrToInt ||
CE->getOpcode() == Instruction::BitCast)
return IsConstantOffsetFromGlobal(CE->getOperand(0), GV, Offset, TD);
// i32* getelementptr ([5 x i32]* @a, i32 0, i32 5)
if (CE->getOpcode() == Instruction::GetElementPtr) {
// Cannot compute this if the element type of the pointer is missing size
// info.
if (!cast<PointerType>(CE->getOperand(0)->getType())->getElementType()->isSized())
return false;
// If the base isn't a global+constant, we aren't either.
if (!IsConstantOffsetFromGlobal(CE->getOperand(0), GV, Offset, TD))
return false;
// Otherwise, add any offset that our operands provide.
gep_type_iterator GTI = gep_type_begin(CE);
for (unsigned i = 1, e = CE->getNumOperands(); i != e; ++i, ++GTI) {
ConstantInt *CI = dyn_cast<ConstantInt>(CE->getOperand(i));
if (!CI) return false; // Index isn't a simple constant?
if (CI->getZExtValue() == 0) continue; // Not adding anything.
if (const StructType *ST = dyn_cast<StructType>(*GTI)) {
// N = N + Offset
Offset += TD.getStructLayout(ST)->getElementOffset(CI->getZExtValue());
} else {
const SequentialType *SQT = cast<SequentialType>(*GTI);
Offset += TD.getTypeSize(SQT->getElementType())*CI->getSExtValue();
}
}
return true;
}
return false;
}
/// SymbolicallyEvaluateBinop - One of Op0/Op1 is a constant expression.
/// Attempt to symbolically evaluate the result of a binary operator merging
/// these together. If target data info is available, it is provided as TD,
/// otherwise TD is null.
static Constant *SymbolicallyEvaluateBinop(unsigned Opc, Constant *Op0,
Constant *Op1, const TargetData *TD){
// SROA
// Fold (and 0xffffffff00000000, (shl x, 32)) -> shl.
// Fold (lshr (or X, Y), 32) -> (lshr [X/Y], 32) if one doesn't contribute
// bits.
// If the constant expr is something like &A[123] - &A[4].f, fold this into a
// constant. This happens frequently when iterating over a global array.
if (Opc == Instruction::Sub && TD) {
GlobalValue *GV1, *GV2;
int64_t Offs1, Offs2;
if (IsConstantOffsetFromGlobal(Op0, GV1, Offs1, *TD))
if (IsConstantOffsetFromGlobal(Op1, GV2, Offs2, *TD) &&
GV1 == GV2) {
// (&GV+C1) - (&GV+C2) -> C1-C2, pointer arithmetic cannot overflow.
return ConstantInt::get(Op0->getType(), Offs1-Offs2);
}
}
// TODO: Fold icmp setne/seteq as well.
return 0;
}
/// SymbolicallyEvaluateGEP - If we can symbolically evaluate the specified GEP
/// constant expression, do so.
static Constant *SymbolicallyEvaluateGEP(Constant** Ops, unsigned NumOps,
const Type *ResultTy,
const TargetData *TD) {
Constant *Ptr = Ops[0];
if (!cast<PointerType>(Ptr->getType())->getElementType()->isSized())
return 0;
if (TD && Ptr->isNullValue()) {
// If this is a constant expr gep that is effectively computing an
// "offsetof", fold it into 'cast int Size to T*' instead of 'gep 0, 0, 12'
bool isFoldableGEP = true;
for (unsigned i = 1; i != NumOps; ++i)
if (!isa<ConstantInt>(Ops[i])) {
isFoldableGEP = false;
break;
}
if (isFoldableGEP) {
uint64_t Offset = TD->getIndexedOffset(Ptr->getType(),
(Value**)Ops+1, NumOps-1);
Constant *C = ConstantInt::get(TD->getIntPtrType(), Offset);
return ConstantExpr::getIntToPtr(C, ResultTy);
}
}
return 0;
}
//===----------------------------------------------------------------------===//
// Constant Folding public APIs
//===----------------------------------------------------------------------===//
/// ConstantFoldInstruction - Attempt to constant fold the specified
/// instruction. If successful, the constant result is returned, if not, null
/// is returned. Note that this function can only fail when attempting to fold
/// instructions like loads and stores, which have no constant expression form.
///
Constant *llvm::ConstantFoldInstruction(Instruction *I, const TargetData *TD) {
if (PHINode *PN = dyn_cast<PHINode>(I)) {
if (PN->getNumIncomingValues() == 0)
return Constant::getNullValue(PN->getType());
Constant *Result = dyn_cast<Constant>(PN->getIncomingValue(0));
if (Result == 0) return 0;
// Handle PHI nodes specially here...
for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingValue(i) != Result && PN->getIncomingValue(i) != PN)
return 0; // Not all the same incoming constants...
// If we reach here, all incoming values are the same constant.
return Result;
}
// Scan the operand list, checking to see if they are all constants, if so,
// hand off to ConstantFoldInstOperands.
SmallVector<Constant*, 8> Ops;
for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i)
if (Constant *Op = dyn_cast<Constant>(I->getOperand(i)))
Ops.push_back(Op);
else
return 0; // All operands not constant!
return ConstantFoldInstOperands(I, &Ops[0], Ops.size(), TD);
}
/// ConstantFoldInstOperands - Attempt to constant fold an instruction with the
/// specified opcode and operands. If successful, the constant result is
/// returned, if not, null is returned. Note that this function can fail when
/// attempting to fold instructions like loads and stores, which have no
/// constant expression form.
///
Constant *llvm::ConstantFoldInstOperands(const Instruction* I,
Constant** Ops, unsigned NumOps,
const TargetData *TD) {
unsigned Opc = I->getOpcode();
const Type *DestTy = I->getType();
// Handle easy binops first.
if (isa<BinaryOperator>(I)) {
if (isa<ConstantExpr>(Ops[0]) || isa<ConstantExpr>(Ops[1]))
if (Constant *C = SymbolicallyEvaluateBinop(I->getOpcode(), Ops[0],
Ops[1], TD))
return C;
return ConstantExpr::get(Opc, Ops[0], Ops[1]);
}
switch (Opc) {
default: return 0;
case Instruction::Call:
if (Function *F = dyn_cast<Function>(Ops[0]))
if (canConstantFoldCallTo(F))
return ConstantFoldCall(F, Ops+1, NumOps-1);
return 0;
case Instruction::ICmp:
case Instruction::FCmp:
return ConstantExpr::getCompare(cast<CmpInst>(I)->getPredicate(), Ops[0],
Ops[1]);
case Instruction::PtrToInt:
// If the input is a inttoptr, eliminate the pair. This requires knowing
// the width of a pointer, so it can't be done in ConstantExpr::getCast.
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[0])) {
if (TD && CE->getOpcode() == Instruction::IntToPtr) {
Constant *Input = CE->getOperand(0);
unsigned InWidth = Input->getType()->getPrimitiveSizeInBits();
Constant *Mask =
ConstantInt::get(APInt::getLowBitsSet(InWidth,
TD->getPointerSizeInBits()));
Input = ConstantExpr::getAnd(Input, Mask);
// Do a zext or trunc to get to the dest size.
return ConstantExpr::getIntegerCast(Input, I->getType(), false);
}
}
// FALL THROUGH.
case Instruction::IntToPtr:
case Instruction::Trunc:
case Instruction::ZExt:
case Instruction::SExt:
case Instruction::FPTrunc:
case Instruction::FPExt:
case Instruction::UIToFP:
case Instruction::SIToFP:
case Instruction::FPToUI:
case Instruction::FPToSI:
case Instruction::BitCast:
return ConstantExpr::getCast(Opc, Ops[0], DestTy);
case Instruction::Select:
return ConstantExpr::getSelect(Ops[0], Ops[1], Ops[2]);
case Instruction::ExtractElement:
return ConstantExpr::getExtractElement(Ops[0], Ops[1]);
case Instruction::InsertElement:
return ConstantExpr::getInsertElement(Ops[0], Ops[1], Ops[2]);
case Instruction::ShuffleVector:
return ConstantExpr::getShuffleVector(Ops[0], Ops[1], Ops[2]);
case Instruction::GetElementPtr:
if (Constant *C = SymbolicallyEvaluateGEP(Ops, NumOps, I->getType(), TD))
return C;
return ConstantExpr::getGetElementPtr(Ops[0], Ops+1, NumOps-1);
}
}
/// ConstantFoldLoadThroughGEPConstantExpr - Given a constant and a
/// getelementptr constantexpr, return the constant value being addressed by the
/// constant expression, or null if something is funny and we can't decide.
Constant *llvm::ConstantFoldLoadThroughGEPConstantExpr(Constant *C,
ConstantExpr *CE) {
if (CE->getOperand(1) != Constant::getNullValue(CE->getOperand(1)->getType()))
return 0; // Do not allow stepping over the value!
// Loop over all of the operands, tracking down which value we are
// addressing...
gep_type_iterator I = gep_type_begin(CE), E = gep_type_end(CE);
for (++I; I != E; ++I)
if (const StructType *STy = dyn_cast<StructType>(*I)) {
ConstantInt *CU = cast<ConstantInt>(I.getOperand());
assert(CU->getZExtValue() < STy->getNumElements() &&
"Struct index out of range!");
unsigned El = (unsigned)CU->getZExtValue();
if (ConstantStruct *CS = dyn_cast<ConstantStruct>(C)) {
C = CS->getOperand(El);
} else if (isa<ConstantAggregateZero>(C)) {
C = Constant::getNullValue(STy->getElementType(El));
} else if (isa<UndefValue>(C)) {
C = UndefValue::get(STy->getElementType(El));
} else {
return 0;
}
} else if (ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand())) {
if (const ArrayType *ATy = dyn_cast<ArrayType>(*I)) {
if (CI->getZExtValue() >= ATy->getNumElements())
return 0;
if (ConstantArray *CA = dyn_cast<ConstantArray>(C))
C = CA->getOperand(CI->getZExtValue());
else if (isa<ConstantAggregateZero>(C))
C = Constant::getNullValue(ATy->getElementType());
else if (isa<UndefValue>(C))
C = UndefValue::get(ATy->getElementType());
else
return 0;
} else if (const VectorType *PTy = dyn_cast<VectorType>(*I)) {
if (CI->getZExtValue() >= PTy->getNumElements())
return 0;
if (ConstantVector *CP = dyn_cast<ConstantVector>(C))
C = CP->getOperand(CI->getZExtValue());
else if (isa<ConstantAggregateZero>(C))
C = Constant::getNullValue(PTy->getElementType());
else if (isa<UndefValue>(C))
C = UndefValue::get(PTy->getElementType());
else
return 0;
} else {
return 0;
}
} else {
return 0;
}
return C;
}
//===----------------------------------------------------------------------===//
// Constant Folding for Calls
//
/// canConstantFoldCallTo - Return true if its even possible to fold a call to
/// the specified function.
bool
llvm::canConstantFoldCallTo(Function *F) {
switch (F->getIntrinsicID()) {
case Intrinsic::sqrt_f32:
case Intrinsic::sqrt_f64:
case Intrinsic::sqrt_f80:
case Intrinsic::sqrt_f128:
case Intrinsic::sqrt_ppcf128:
case Intrinsic::powi_f32:
case Intrinsic::powi_f64:
case Intrinsic::powi_f80:
case Intrinsic::powi_f128:
case Intrinsic::powi_ppcf128:
case Intrinsic::bswap:
case Intrinsic::ctpop:
case Intrinsic::ctlz:
case Intrinsic::cttz:
return true;
default: break;
}
const ValueName *NameVal = F->getValueName();
if (NameVal == 0) return false;
const char *Str = NameVal->getKeyData();
unsigned Len = NameVal->getKeyLength();
// In these cases, the check of the length is required. We don't want to
// return true for a name like "cos\0blah" which strcmp would return equal to
// "cos", but has length 8.
switch (Str[0]) {
default: return false;
case 'a':
if (Len == 4)
return !strcmp(Str, "acos") || !strcmp(Str, "asin") ||
!strcmp(Str, "atan");
else if (Len == 5)
return !strcmp(Str, "atan2");
return false;
case 'c':
if (Len == 3)
return !strcmp(Str, "cos");
else if (Len == 4)
return !strcmp(Str, "ceil") || !strcmp(Str, "cosf") ||
!strcmp(Str, "cosh");
return false;
case 'e':
if (Len == 3)
return !strcmp(Str, "exp");
return false;
case 'f':
if (Len == 4)
return !strcmp(Str, "fabs") || !strcmp(Str, "fmod");
else if (Len == 5)
return !strcmp(Str, "floor");
return false;
break;
case 'l':
if (Len == 3 && !strcmp(Str, "log"))
return true;
if (Len == 5 && !strcmp(Str, "log10"))
return true;
return false;
case 'p':
if (Len == 3 && !strcmp(Str, "pow"))
return true;
return false;
case 's':
if (Len == 3)
return !strcmp(Str, "sin");
if (Len == 4)
return !strcmp(Str, "sinh") || !strcmp(Str, "sqrt");
if (Len == 5)
return !strcmp(Str, "sqrtf");
return false;
case 't':
if (Len == 3 && !strcmp(Str, "tan"))
return true;
else if (Len == 4 && !strcmp(Str, "tanh"))
return true;
return false;
}
}
static Constant *ConstantFoldFP(double (*NativeFP)(double), double V,
const Type *Ty) {
errno = 0;
V = NativeFP(V);
if (errno == 0) {
if (Ty==Type::FloatTy)
return ConstantFP::get(Ty, APFloat((float)V));
else if (Ty==Type::DoubleTy)
return ConstantFP::get(Ty, APFloat(V));
else
assert(0);
}
errno = 0;
return 0;
}
static Constant *ConstantFoldBinaryFP(double (*NativeFP)(double, double),
double V, double W,
const Type *Ty) {
errno = 0;
V = NativeFP(V, W);
if (errno == 0) {
if (Ty==Type::FloatTy)
return ConstantFP::get(Ty, APFloat((float)V));
else if (Ty==Type::DoubleTy)
return ConstantFP::get(Ty, APFloat(V));
else
assert(0);
}
errno = 0;
return 0;
}
/// ConstantFoldCall - Attempt to constant fold a call to the specified function
/// with the specified arguments, returning null if unsuccessful.
Constant *
llvm::ConstantFoldCall(Function *F, Constant** Operands, unsigned NumOperands) {
const ValueName *NameVal = F->getValueName();
if (NameVal == 0) return 0;
const char *Str = NameVal->getKeyData();
unsigned Len = NameVal->getKeyLength();
const Type *Ty = F->getReturnType();
if (NumOperands == 1) {
if (ConstantFP *Op = dyn_cast<ConstantFP>(Operands[0])) {
if (Ty!=Type::FloatTy && Ty!=Type::DoubleTy)
return 0;
/// Currently APFloat versions of these functions do not exist, so we use
/// the host native double versions. Float versions are not called
/// directly but for all these it is true (float)(f((double)arg)) ==
/// f(arg). Long double not supported yet.
double V = Ty==Type::FloatTy ? (double)Op->getValueAPF().convertToFloat():
Op->getValueAPF().convertToDouble();
switch (Str[0]) {
case 'a':
if (Len == 4 && !strcmp(Str, "acos"))
return ConstantFoldFP(acos, V, Ty);
else if (Len == 4 && !strcmp(Str, "asin"))
return ConstantFoldFP(asin, V, Ty);
else if (Len == 4 && !strcmp(Str, "atan"))
return ConstantFoldFP(atan, V, Ty);
break;
case 'c':
if (Len == 4 && !strcmp(Str, "ceil"))
return ConstantFoldFP(ceil, V, Ty);
else if (Len == 3 && !strcmp(Str, "cos"))
return ConstantFoldFP(cos, V, Ty);
else if (Len == 4 && !strcmp(Str, "cosh"))
return ConstantFoldFP(cosh, V, Ty);
break;
case 'e':
if (Len == 3 && !strcmp(Str, "exp"))
return ConstantFoldFP(exp, V, Ty);
break;
case 'f':
if (Len == 4 && !strcmp(Str, "fabs"))
return ConstantFoldFP(fabs, V, Ty);
else if (Len == 5 && !strcmp(Str, "floor"))
return ConstantFoldFP(floor, V, Ty);
break;
case 'l':
if (Len == 3 && !strcmp(Str, "log") && V > 0)
return ConstantFoldFP(log, V, Ty);
else if (Len == 5 && !strcmp(Str, "log10") && V > 0)
return ConstantFoldFP(log10, V, Ty);
else if (!strcmp(Str, "llvm.sqrt.f32") ||
!strcmp(Str, "llvm.sqrt.f64")) {
if (V >= -0.0)
return ConstantFoldFP(sqrt, V, Ty);
else // Undefined
return ConstantFP::get(Ty, Ty==Type::FloatTy ? APFloat(0.0f) :
APFloat(0.0));
}
break;
case 's':
if (Len == 3 && !strcmp(Str, "sin"))
return ConstantFoldFP(sin, V, Ty);
else if (Len == 4 && !strcmp(Str, "sinh"))
return ConstantFoldFP(sinh, V, Ty);
else if (Len == 4 && !strcmp(Str, "sqrt") && V >= 0)
return ConstantFoldFP(sqrt, V, Ty);
else if (Len == 5 && !strcmp(Str, "sqrtf") && V >= 0)
return ConstantFoldFP(sqrt, V, Ty);
break;
case 't':
if (Len == 3 && !strcmp(Str, "tan"))
return ConstantFoldFP(tan, V, Ty);
else if (Len == 4 && !strcmp(Str, "tanh"))
return ConstantFoldFP(tanh, V, Ty);
break;
default:
break;
}
} else if (ConstantInt *Op = dyn_cast<ConstantInt>(Operands[0])) {
if (Len > 11 && !memcmp(Str, "llvm.bswap", 10)) {
return ConstantInt::get(Op->getValue().byteSwap());
} else if (Len > 11 && !memcmp(Str, "llvm.ctpop", 10)) {
uint64_t ctpop = Op->getValue().countPopulation();
return ConstantInt::get(Ty, ctpop);
} else if (Len > 10 && !memcmp(Str, "llvm.cttz", 9)) {
uint64_t cttz = Op->getValue().countTrailingZeros();
return ConstantInt::get(Ty, cttz);
} else if (Len > 10 && !memcmp(Str, "llvm.ctlz", 9)) {
uint64_t ctlz = Op->getValue().countLeadingZeros();
return ConstantInt::get(Ty, ctlz);
}
}
} else if (NumOperands == 2) {
if (ConstantFP *Op1 = dyn_cast<ConstantFP>(Operands[0])) {
double Op1V = Ty==Type::FloatTy ?
(double)Op1->getValueAPF().convertToFloat():
Op1->getValueAPF().convertToDouble();
if (ConstantFP *Op2 = dyn_cast<ConstantFP>(Operands[1])) {
if (Ty!=Type::FloatTy && Ty!=Type::DoubleTy)
return 0;
double Op2V = Ty==Type::FloatTy ?
(double)Op2->getValueAPF().convertToFloat():
Op2->getValueAPF().convertToDouble();
if (Len == 3 && !strcmp(Str, "pow")) {
return ConstantFoldBinaryFP(pow, Op1V, Op2V, Ty);
} else if (Len == 4 && !strcmp(Str, "fmod")) {
return ConstantFoldBinaryFP(fmod, Op1V, Op2V, Ty);
} else if (Len == 5 && !strcmp(Str, "atan2")) {
return ConstantFoldBinaryFP(atan2, Op1V, Op2V, Ty);
}
} else if (ConstantInt *Op2C = dyn_cast<ConstantInt>(Operands[1])) {
if (!strcmp(Str, "llvm.powi.f32")) {
return ConstantFP::get(Ty, APFloat((float)std::pow((float)Op1V,
(int)Op2C->getZExtValue())));
} else if (!strcmp(Str, "llvm.powi.f64")) {
return ConstantFP::get(Ty, APFloat((double)std::pow((double)Op1V,
(int)Op2C->getZExtValue())));
}
}
}
}
return 0;
}