#include "llvm/Target/TargetData.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/MathExtras.h"
+#include "llvm/ADT/SmallPtrSet.h"
#include <cstring>
using namespace llvm;
assert(V && "No Value?");
assert(Depth <= MaxDepth && "Limit Search Depth");
unsigned BitWidth = Mask.getBitWidth();
- assert((V->getType()->isIntOrIntVector() || isa<PointerType>(V->getType())) &&
- "Not integer or pointer type!");
+ assert((V->getType()->isIntOrIntVectorTy() || V->getType()->isPointerTy())
+ && "Not integer or pointer type!");
assert((!TD ||
TD->getTypeSizeInBits(V->getType()->getScalarType()) == BitWidth) &&
- (!V->getType()->isIntOrIntVector() ||
+ (!V->getType()->isIntOrIntVectorTy() ||
V->getType()->getScalarSizeInBits() == BitWidth) &&
KnownZero.getBitWidth() == BitWidth &&
KnownOne.getBitWidth() == BitWidth &&
unsigned SrcBitWidth;
// Note that we handle pointer operands here because of inttoptr/ptrtoint
// which fall through here.
- if (isa<PointerType>(SrcTy))
+ if (SrcTy->isPointerTy())
SrcBitWidth = TD->getTypeSizeInBits(SrcTy);
else
SrcBitWidth = SrcTy->getScalarSizeInBits();
}
case Instruction::BitCast: {
const Type *SrcTy = I->getOperand(0)->getType();
- if ((SrcTy->isInteger() || isa<PointerType>(SrcTy)) &&
+ if ((SrcTy->isIntegerTy() || SrcTy->isPointerTy()) &&
// TODO: For now, not handling conversions like:
// (bitcast i64 %x to <2 x i32>)
- !isa<VectorType>(I->getType())) {
+ !I->getType()->isVectorTy()) {
ComputeMaskedBits(I->getOperand(0), Mask, KnownZero, KnownOne, TD,
Depth+1);
return;
}
case Instruction::SRem:
if (ConstantInt *Rem = dyn_cast<ConstantInt>(I->getOperand(1))) {
- APInt RA = Rem->getValue();
- if (RA.isPowerOf2() || (-RA).isPowerOf2()) {
- APInt LowBits = RA.isStrictlyPositive() ? (RA - 1) : ~RA;
+ APInt RA = Rem->getValue().abs();
+ if (RA.isPowerOf2()) {
+ APInt LowBits = RA - 1;
APInt Mask2 = LowBits | APInt::getSignBit(BitWidth);
ComputeMaskedBits(I->getOperand(0), Mask2, KnownZero2, KnownOne2, TD,
Depth+1);
- // If the sign bit of the first operand is zero, the sign bit of
- // the result is zero. If the first operand has no one bits below
- // the second operand's single 1 bit, its sign will be zero.
+ // The low bits of the first operand are unchanged by the srem.
+ KnownZero = KnownZero2 & LowBits;
+ KnownOne = KnownOne2 & LowBits;
+
+ // If the first operand is non-negative or has all low bits zero, then
+ // the upper bits are all zero.
if (KnownZero2[BitWidth-1] || ((KnownZero2 & LowBits) == LowBits))
- KnownZero2 |= ~LowBits;
+ KnownZero |= ~LowBits;
- KnownZero |= KnownZero2 & Mask;
+ // If the first operand is negative and not all low bits are zero, then
+ // the upper bits are all one.
+ if (KnownOne2[BitWidth-1] && ((KnownOne2 & LowBits) != 0))
+ KnownOne |= ~LowBits;
+
+ KnownZero &= Mask;
+ KnownOne &= Mask;
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
}
///
unsigned llvm::ComputeNumSignBits(Value *V, const TargetData *TD,
unsigned Depth) {
- assert((TD || V->getType()->isIntOrIntVector()) &&
+ assert((TD || V->getType()->isIntOrIntVectorTy()) &&
"ComputeNumSignBits requires a TargetData object to operate "
"on non-integer values!");
const Type *Ty = V->getType();
switch (Operator::getOpcode(V)) {
default: break;
case Instruction::SExt:
- Tmp = TyBits-cast<IntegerType>(U->getOperand(0)->getType())->getBitWidth();
+ Tmp = TyBits - U->getOperand(0)->getType()->getScalarSizeInBits();
return ComputeNumSignBits(U->getOperand(0), TD, Depth+1) + Tmp;
case Instruction::AShr:
Tmp2 = ComputeNumSignBits(U->getOperand(1), TD, Depth+1);
if (Tmp2 == 1) return 1;
- return std::min(Tmp, Tmp2)-1;
- break;
+ return std::min(Tmp, Tmp2)-1;
case Instruction::Sub:
Tmp2 = ComputeNumSignBits(U->getOperand(1), TD, Depth+1);
// is, at worst, one more bit than the inputs.
Tmp = ComputeNumSignBits(U->getOperand(0), TD, Depth+1);
if (Tmp == 1) return 1; // Early out.
- return std::min(Tmp, Tmp2)-1;
- break;
+ return std::min(Tmp, Tmp2)-1;
+
+ case Instruction::PHI: {
+ PHINode *PN = cast<PHINode>(U);
+ // Don't analyze large in-degree PHIs.
+ if (PN->getNumIncomingValues() > 4) break;
+
+ // Take the minimum of all incoming values. This can't infinitely loop
+ // because of our depth threshold.
+ Tmp = ComputeNumSignBits(PN->getIncomingValue(0), TD, Depth+1);
+ for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i) {
+ if (Tmp == 1) return Tmp;
+ Tmp = std::min(Tmp,
+ ComputeNumSignBits(PN->getIncomingValue(i), TD, Depth+1));
+ }
+ return Tmp;
+ }
+
case Instruction::Trunc:
// FIXME: it's tricky to do anything useful for this, but it is an important
// case for targets like X86.
assert(V && "No Value?");
assert(Depth <= MaxDepth && "Limit Search Depth");
- assert(V->getType()->isInteger() && "Not integer or pointer type!");
+ assert(V->getType()->isIntegerTy() && "Not integer or pointer type!");
const Type *T = V->getType();
switch (I->getOpcode()) {
default: break;
- case Instruction::SExt: {
+ case Instruction::SExt:
if (!LookThroughSExt) return false;
// otherwise fall through to ZExt
- }
- case Instruction::ZExt: {
+ case Instruction::ZExt:
return ComputeMultiple(I->getOperand(0), Base, Multiple,
LookThroughSExt, Depth+1);
- }
case Instruction::Shl:
case Instruction::Mul: {
Value *Op0 = I->getOperand(0);
}
Value *Mul0 = NULL;
- Value *Mul1 = NULL;
- bool M0 = ComputeMultiple(Op0, Base, Mul0,
- LookThroughSExt, Depth+1);
- bool M1 = ComputeMultiple(Op1, Base, Mul1,
- LookThroughSExt, Depth+1);
-
- if (M0) {
- if (isa<Constant>(Op1) && isa<Constant>(Mul0)) {
- // V == Base * (Mul0 * Op1), so return (Mul0 * Op1)
- Multiple = ConstantExpr::getMul(cast<Constant>(Mul0),
- cast<Constant>(Op1));
- return true;
- }
+ if (ComputeMultiple(Op0, Base, Mul0, LookThroughSExt, Depth+1)) {
+ if (Constant *Op1C = dyn_cast<Constant>(Op1))
+ if (Constant *MulC = dyn_cast<Constant>(Mul0)) {
+ if (Op1C->getType()->getPrimitiveSizeInBits() <
+ MulC->getType()->getPrimitiveSizeInBits())
+ Op1C = ConstantExpr::getZExt(Op1C, MulC->getType());
+ if (Op1C->getType()->getPrimitiveSizeInBits() >
+ MulC->getType()->getPrimitiveSizeInBits())
+ MulC = ConstantExpr::getZExt(MulC, Op1C->getType());
+
+ // V == Base * (Mul0 * Op1), so return (Mul0 * Op1)
+ Multiple = ConstantExpr::getMul(MulC, Op1C);
+ return true;
+ }
if (ConstantInt *Mul0CI = dyn_cast<ConstantInt>(Mul0))
if (Mul0CI->getValue() == 1) {
}
}
- if (M1) {
- if (isa<Constant>(Op0) && isa<Constant>(Mul1)) {
- // V == Base * (Mul1 * Op0), so return (Mul1 * Op0)
- Multiple = ConstantExpr::getMul(cast<Constant>(Mul1),
- cast<Constant>(Op0));
- return true;
- }
+ Value *Mul1 = NULL;
+ if (ComputeMultiple(Op1, Base, Mul1, LookThroughSExt, Depth+1)) {
+ if (Constant *Op0C = dyn_cast<Constant>(Op0))
+ if (Constant *MulC = dyn_cast<Constant>(Mul1)) {
+ if (Op0C->getType()->getPrimitiveSizeInBits() <
+ MulC->getType()->getPrimitiveSizeInBits())
+ Op0C = ConstantExpr::getZExt(Op0C, MulC->getType());
+ if (Op0C->getType()->getPrimitiveSizeInBits() >
+ MulC->getType()->getPrimitiveSizeInBits())
+ MulC = ConstantExpr::getZExt(MulC, Op0C->getType());
+
+ // V == Base * (Mul1 * Op0), so return (Mul1 * Op0)
+ Multiple = ConstantExpr::getMul(MulC, Op0C);
+ return true;
+ }
if (ConstantInt *Mul1CI = dyn_cast<ConstantInt>(Mul1))
if (Mul1CI->getValue() == 1) {
if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(I))
// sqrt(-0.0) = -0.0, no other negative results are possible.
if (II->getIntrinsicID() == Intrinsic::sqrt)
- return CannotBeNegativeZero(II->getOperand(1), Depth+1);
+ return CannotBeNegativeZero(II->getArgOperand(0), Depth+1);
if (const CallInst *CI = dyn_cast<CallInst>(I))
if (const Function *F = CI->getCalledFunction()) {
if (F->getName() == "fabsl") return true;
if (F->getName() == "sqrt" || F->getName() == "sqrtf" ||
F->getName() == "sqrtl")
- return CannotBeNegativeZero(CI->getOperand(1), Depth+1);
+ return CannotBeNegativeZero(CI->getArgOperand(0), Depth+1);
}
}
if (idx_begin == idx_end)
return V;
// We have indices, so V should have an indexable type
- assert((isa<StructType>(V->getType()) || isa<ArrayType>(V->getType()))
+ assert((V->getType()->isStructTy() || V->getType()->isArrayTy())
&& "Not looking at a struct or array?");
assert(ExtractValueInst::getIndexedType(V->getType(), idx_begin, idx_end)
&& "Invalid indices for type?");
/// GetConstantStringInfo - This function computes the length of a
/// null-terminated C string pointed to by V. If successful, it returns true
/// and returns the string in Str. If unsuccessful, it returns false.
-bool llvm::GetConstantStringInfo(Value *V, std::string &Str, uint64_t Offset,
+bool llvm::GetConstantStringInfo(const Value *V, std::string &Str,
+ uint64_t Offset,
bool StopAtNul) {
// If V is NULL then return false;
if (V == NULL) return false;
// Look through bitcast instructions.
- if (BitCastInst *BCI = dyn_cast<BitCastInst>(V))
+ if (const BitCastInst *BCI = dyn_cast<BitCastInst>(V))
return GetConstantStringInfo(BCI->getOperand(0), Str, Offset, StopAtNul);
// If the value is not a GEP instruction nor a constant expression with a
// GEP instruction, then return false because ConstantArray can't occur
// any other way
- User *GEP = 0;
- if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(V)) {
+ const User *GEP = 0;
+ if (const GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(V)) {
GEP = GEPI;
- } else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
+ } else if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
if (CE->getOpcode() == Instruction::BitCast)
return GetConstantStringInfo(CE->getOperand(0), Str, Offset, StopAtNul);
if (CE->getOpcode() != Instruction::GetElementPtr)
// Make sure the index-ee is a pointer to array of i8.
const PointerType *PT = cast<PointerType>(GEP->getOperand(0)->getType());
const ArrayType *AT = dyn_cast<ArrayType>(PT->getElementType());
- if (AT == 0 || AT->getElementType() != Type::getInt8Ty(V->getContext()))
+ if (AT == 0 || !AT->getElementType()->isIntegerTy(8))
return false;
// Check to make sure that the first operand of the GEP is an integer and
// has value 0 so that we are sure we're indexing into the initializer.
- ConstantInt *FirstIdx = dyn_cast<ConstantInt>(GEP->getOperand(1));
+ const ConstantInt *FirstIdx = dyn_cast<ConstantInt>(GEP->getOperand(1));
if (FirstIdx == 0 || !FirstIdx->isZero())
return false;
// into the array. If this occurs, we can't say anything meaningful about
// the string.
uint64_t StartIdx = 0;
- if (ConstantInt *CI = dyn_cast<ConstantInt>(GEP->getOperand(2)))
+ if (const ConstantInt *CI = dyn_cast<ConstantInt>(GEP->getOperand(2)))
StartIdx = CI->getZExtValue();
else
return false;
StopAtNul);
}
- if (MDString *MDStr = dyn_cast<MDString>(V)) {
- Str = MDStr->getString();
- return true;
- }
-
// The GEP instruction, constant or instruction, must reference a global
// variable that is a constant and is initialized. The referenced constant
// initializer is the array that we'll use for optimization.
- GlobalVariable* GV = dyn_cast<GlobalVariable>(V);
+ const GlobalVariable* GV = dyn_cast<GlobalVariable>(V);
if (!GV || !GV->isConstant() || !GV->hasDefinitiveInitializer())
return false;
- Constant *GlobalInit = GV->getInitializer();
+ const Constant *GlobalInit = GV->getInitializer();
// Handle the ConstantAggregateZero case
if (isa<ConstantAggregateZero>(GlobalInit)) {
}
// Must be a Constant Array
- ConstantArray *Array = dyn_cast<ConstantArray>(GlobalInit);
- if (Array == 0 ||
- Array->getType()->getElementType() != Type::getInt8Ty(V->getContext()))
+ const ConstantArray *Array = dyn_cast<ConstantArray>(GlobalInit);
+ if (Array == 0 || !Array->getType()->getElementType()->isIntegerTy(8))
return false;
// Get the number of elements in the array
// to in the array.
Str.reserve(NumElts-Offset);
for (unsigned i = Offset; i != NumElts; ++i) {
- Constant *Elt = Array->getOperand(i);
- ConstantInt *CI = dyn_cast<ConstantInt>(Elt);
+ const Constant *Elt = Array->getOperand(i);
+ const ConstantInt *CI = dyn_cast<ConstantInt>(Elt);
if (!CI) // This array isn't suitable, non-int initializer.
return false;
if (StopAtNul && CI->isZero())
// The array isn't null terminated, but maybe this is a memcpy, not a strcpy.
return true;
}
+
+// These next two are very similar to the above, but also look through PHI
+// nodes.
+// TODO: See if we can integrate these two together.
+
+/// GetStringLengthH - If we can compute the length of the string pointed to by
+/// the specified pointer, return 'len+1'. If we can't, return 0.
+static uint64_t GetStringLengthH(Value *V, SmallPtrSet<PHINode*, 32> &PHIs) {
+ // Look through noop bitcast instructions.
+ if (BitCastInst *BCI = dyn_cast<BitCastInst>(V))
+ return GetStringLengthH(BCI->getOperand(0), PHIs);
+
+ // If this is a PHI node, there are two cases: either we have already seen it
+ // or we haven't.
+ if (PHINode *PN = dyn_cast<PHINode>(V)) {
+ if (!PHIs.insert(PN))
+ return ~0ULL; // already in the set.
+
+ // If it was new, see if all the input strings are the same length.
+ uint64_t LenSoFar = ~0ULL;
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
+ uint64_t Len = GetStringLengthH(PN->getIncomingValue(i), PHIs);
+ if (Len == 0) return 0; // Unknown length -> unknown.
+
+ if (Len == ~0ULL) continue;
+
+ if (Len != LenSoFar && LenSoFar != ~0ULL)
+ return 0; // Disagree -> unknown.
+ LenSoFar = Len;
+ }
+
+ // Success, all agree.
+ return LenSoFar;
+ }
+
+ // strlen(select(c,x,y)) -> strlen(x) ^ strlen(y)
+ if (SelectInst *SI = dyn_cast<SelectInst>(V)) {
+ uint64_t Len1 = GetStringLengthH(SI->getTrueValue(), PHIs);
+ if (Len1 == 0) return 0;
+ uint64_t Len2 = GetStringLengthH(SI->getFalseValue(), PHIs);
+ if (Len2 == 0) return 0;
+ if (Len1 == ~0ULL) return Len2;
+ if (Len2 == ~0ULL) return Len1;
+ if (Len1 != Len2) return 0;
+ return Len1;
+ }
+
+ // If the value is not a GEP instruction nor a constant expression with a
+ // GEP instruction, then return unknown.
+ User *GEP = 0;
+ if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(V)) {
+ GEP = GEPI;
+ } else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
+ if (CE->getOpcode() != Instruction::GetElementPtr)
+ return 0;
+ GEP = CE;
+ } else {
+ return 0;
+ }
+
+ // Make sure the GEP has exactly three arguments.
+ if (GEP->getNumOperands() != 3)
+ return 0;
+
+ // Check to make sure that the first operand of the GEP is an integer and
+ // has value 0 so that we are sure we're indexing into the initializer.
+ if (ConstantInt *Idx = dyn_cast<ConstantInt>(GEP->getOperand(1))) {
+ if (!Idx->isZero())
+ return 0;
+ } else
+ return 0;
+
+ // If the second index isn't a ConstantInt, then this is a variable index
+ // into the array. If this occurs, we can't say anything meaningful about
+ // the string.
+ uint64_t StartIdx = 0;
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(GEP->getOperand(2)))
+ StartIdx = CI->getZExtValue();
+ else
+ return 0;
+
+ // The GEP instruction, constant or instruction, must reference a global
+ // variable that is a constant and is initialized. The referenced constant
+ // initializer is the array that we'll use for optimization.
+ GlobalVariable* GV = dyn_cast<GlobalVariable>(GEP->getOperand(0));
+ if (!GV || !GV->isConstant() || !GV->hasInitializer() ||
+ GV->mayBeOverridden())
+ return 0;
+ Constant *GlobalInit = GV->getInitializer();
+
+ // Handle the ConstantAggregateZero case, which is a degenerate case. The
+ // initializer is constant zero so the length of the string must be zero.
+ if (isa<ConstantAggregateZero>(GlobalInit))
+ return 1; // Len = 0 offset by 1.
+
+ // Must be a Constant Array
+ ConstantArray *Array = dyn_cast<ConstantArray>(GlobalInit);
+ if (!Array || !Array->getType()->getElementType()->isIntegerTy(8))
+ return false;
+
+ // Get the number of elements in the array
+ uint64_t NumElts = Array->getType()->getNumElements();
+
+ // Traverse the constant array from StartIdx (derived above) which is
+ // the place the GEP refers to in the array.
+ for (unsigned i = StartIdx; i != NumElts; ++i) {
+ Constant *Elt = Array->getOperand(i);
+ ConstantInt *CI = dyn_cast<ConstantInt>(Elt);
+ if (!CI) // This array isn't suitable, non-int initializer.
+ return 0;
+ if (CI->isZero())
+ return i-StartIdx+1; // We found end of string, success!
+ }
+
+ return 0; // The array isn't null terminated, conservatively return 'unknown'.
+}
+
+/// GetStringLength - If we can compute the length of the string pointed to by
+/// the specified pointer, return 'len+1'. If we can't, return 0.
+uint64_t llvm::GetStringLength(Value *V) {
+ if (!V->getType()->isPointerTy()) return 0;
+
+ SmallPtrSet<PHINode*, 32> PHIs;
+ uint64_t Len = GetStringLengthH(V, PHIs);
+ // If Len is ~0ULL, we had an infinite phi cycle: this is dead code, so return
+ // an empty string as a length.
+ return Len == ~0ULL ? 1 : Len;
+}