// If MemCpyInst length is 1/2/4/8 bytes then replace memcpy with
// load/store.
ConstantInt *MemOpLength = dyn_cast<ConstantInt>(MI->getArgOperand(2));
- if (MemOpLength == 0) return 0;
+ if (!MemOpLength) return nullptr;
// Source and destination pointer types are always "i8*" for intrinsic. See
// if the size is something we can handle with a single primitive load/store.
assert(Size && "0-sized memory transferring should be removed already.");
if (Size > 8 || (Size&(Size-1)))
- return 0; // If not 1/2/4/8 bytes, exit.
+ return nullptr; // If not 1/2/4/8 bytes, exit.
// Use an integer load+store unless we can find something better.
unsigned SrcAddrSp =
// dest address will be promotable. See if we can find a better type than the
// integer datatype.
Value *StrippedDest = MI->getArgOperand(0)->stripPointerCasts();
- MDNode *CopyMD = 0;
+ MDNode *CopyMD = nullptr;
if (StrippedDest != MI->getArgOperand(0)) {
Type *SrcETy = cast<PointerType>(StrippedDest->getType())
->getElementType();
ConstantInt *LenC = dyn_cast<ConstantInt>(MI->getLength());
ConstantInt *FillC = dyn_cast<ConstantInt>(MI->getValue());
if (!LenC || !FillC || !FillC->getType()->isIntegerTy(8))
- return 0;
+ return nullptr;
uint64_t Len = LenC->getLimitedValue();
Alignment = MI->getAlignment();
assert(Len && "0-sized memory setting should be removed already.");
return MI;
}
- return 0;
+ return nullptr;
}
/// visitCallInst - CallInst simplification. This mostly only handles folding
// No other transformations apply to volatile transfers.
if (MI->isVolatile())
- return 0;
+ return nullptr;
// If we have a memmove and the source operation is a constant global,
// then the source and dest pointers can't alias, so we can change this
uint64_t Size;
if (getObjectSize(II->getArgOperand(0), Size, DL, TLI))
return ReplaceInstUsesWith(CI, ConstantInt::get(CI.getType(), Size));
- return 0;
+ return nullptr;
}
case Intrinsic::bswap: {
Value *IIOperand = II->getArgOperand(0);
- Value *X = 0;
+ Value *X = nullptr;
// bswap(bswap(x)) -> x
if (match(IIOperand, m_BSwap(m_Value(X))))
uint32_t BitWidth = IT->getBitWidth();
APInt KnownZero(BitWidth, 0);
APInt KnownOne(BitWidth, 0);
- ComputeMaskedBits(II->getArgOperand(0), KnownZero, KnownOne);
+ computeKnownBits(II->getArgOperand(0), KnownZero, KnownOne);
unsigned TrailingZeros = KnownOne.countTrailingZeros();
APInt Mask(APInt::getLowBitsSet(BitWidth, TrailingZeros));
if ((Mask & KnownZero) == Mask)
uint32_t BitWidth = IT->getBitWidth();
APInt KnownZero(BitWidth, 0);
APInt KnownOne(BitWidth, 0);
- ComputeMaskedBits(II->getArgOperand(0), KnownZero, KnownOne);
+ computeKnownBits(II->getArgOperand(0), KnownZero, KnownOne);
unsigned LeadingZeros = KnownOne.countLeadingZeros();
APInt Mask(APInt::getHighBitsSet(BitWidth, LeadingZeros));
if ((Mask & KnownZero) == Mask)
uint32_t BitWidth = IT->getBitWidth();
APInt LHSKnownZero(BitWidth, 0);
APInt LHSKnownOne(BitWidth, 0);
- ComputeMaskedBits(LHS, LHSKnownZero, LHSKnownOne);
+ computeKnownBits(LHS, LHSKnownZero, LHSKnownOne);
bool LHSKnownNegative = LHSKnownOne[BitWidth - 1];
bool LHSKnownPositive = LHSKnownZero[BitWidth - 1];
if (LHSKnownNegative || LHSKnownPositive) {
APInt RHSKnownZero(BitWidth, 0);
APInt RHSKnownOne(BitWidth, 0);
- ComputeMaskedBits(RHS, RHSKnownZero, RHSKnownOne);
+ computeKnownBits(RHS, RHSKnownZero, RHSKnownOne);
bool RHSKnownNegative = RHSKnownOne[BitWidth - 1];
bool RHSKnownPositive = RHSKnownZero[BitWidth - 1];
if (LHSKnownNegative && RHSKnownNegative) {
APInt LHSKnownZero(BitWidth, 0);
APInt LHSKnownOne(BitWidth, 0);
- ComputeMaskedBits(LHS, LHSKnownZero, LHSKnownOne);
+ computeKnownBits(LHS, LHSKnownZero, LHSKnownOne);
APInt RHSKnownZero(BitWidth, 0);
APInt RHSKnownOne(BitWidth, 0);
- ComputeMaskedBits(RHS, RHSKnownZero, RHSKnownOne);
+ computeKnownBits(RHS, RHSKnownZero, RHSKnownOne);
// Get the largest possible values for each operand.
APInt LHSMax = ~LHSKnownZero;
case Intrinsic::x86_avx2_psll_w:
case Intrinsic::x86_avx2_pslli_d:
case Intrinsic::x86_avx2_pslli_q:
- case Intrinsic::x86_avx2_pslli_w: {
- // Simplify if count is constant. To 0 if > BitWidth, otherwise to shl.
+ case Intrinsic::x86_avx2_pslli_w:
+ case Intrinsic::x86_sse2_psrl_d:
+ case Intrinsic::x86_sse2_psrl_q:
+ case Intrinsic::x86_sse2_psrl_w:
+ case Intrinsic::x86_sse2_psrli_d:
+ case Intrinsic::x86_sse2_psrli_q:
+ case Intrinsic::x86_sse2_psrli_w:
+ case Intrinsic::x86_avx2_psrl_d:
+ case Intrinsic::x86_avx2_psrl_q:
+ case Intrinsic::x86_avx2_psrl_w:
+ case Intrinsic::x86_avx2_psrli_d:
+ case Intrinsic::x86_avx2_psrli_q:
+ case Intrinsic::x86_avx2_psrli_w: {
+ // Simplify if count is constant. To 0 if >= BitWidth,
+ // otherwise to shl/lshr.
auto CDV = dyn_cast<ConstantDataVector>(II->getArgOperand(1));
auto CInt = dyn_cast<ConstantInt>(II->getArgOperand(1));
if (!CDV && !CInt)
VT->getElementType()->getPrimitiveSizeInBits() - 1)
return ReplaceInstUsesWith(
CI, ConstantAggregateZero::get(Vec->getType()));
- else {
- unsigned VWidth = VT->getNumElements();
- // Get a constant vector of the same type as the first operand.
- auto VTCI = ConstantInt::get(VT->getElementType(), Count->getZExtValue());
- return BinaryOperator::CreateShl(
- Vec, Builder->CreateVectorSplat(VWidth, VTCI));
+
+ bool isPackedShiftLeft = true;
+ switch (II->getIntrinsicID()) {
+ default : break;
+ case Intrinsic::x86_sse2_psrl_d:
+ case Intrinsic::x86_sse2_psrl_q:
+ case Intrinsic::x86_sse2_psrl_w:
+ case Intrinsic::x86_sse2_psrli_d:
+ case Intrinsic::x86_sse2_psrli_q:
+ case Intrinsic::x86_sse2_psrli_w:
+ case Intrinsic::x86_avx2_psrl_d:
+ case Intrinsic::x86_avx2_psrl_q:
+ case Intrinsic::x86_avx2_psrl_w:
+ case Intrinsic::x86_avx2_psrli_d:
+ case Intrinsic::x86_avx2_psrli_q:
+ case Intrinsic::x86_avx2_psrli_w: isPackedShiftLeft = false; break;
}
- break;
+
+ unsigned VWidth = VT->getNumElements();
+ // Get a constant vector of the same type as the first operand.
+ auto VTCI = ConstantInt::get(VT->getElementType(), Count->getZExtValue());
+ if (isPackedShiftLeft)
+ return BinaryOperator::CreateShl(Vec,
+ Builder->CreateVectorSplat(VWidth, VTCI));
+
+ return BinaryOperator::CreateLShr(Vec,
+ Builder->CreateVectorSplat(VWidth, VTCI));
}
case Intrinsic::x86_sse41_pmovsxbw:
case Intrinsic::x86_avx_vpermilvar_pd_256: {
// Convert vpermil* to shufflevector if the mask is constant.
Value *V = II->getArgOperand(1);
+ unsigned Size = cast<VectorType>(V->getType())->getNumElements();
+ assert(Size == 8 || Size == 4 || Size == 2);
+ uint32_t Indexes[8];
if (auto C = dyn_cast<ConstantDataVector>(V)) {
- auto V1 = II->getArgOperand(0);
- auto V2 = UndefValue::get(V1->getType());
- auto Shuffle = Builder->CreateShuffleVector(V1, V2, C);
- return ReplaceInstUsesWith(CI, Shuffle);
+ // The intrinsics only read one or two bits, clear the rest.
+ for (unsigned I = 0; I < Size; ++I) {
+ uint32_t Index = C->getElementAsInteger(I) & 0x3;
+ if (II->getIntrinsicID() == Intrinsic::x86_avx_vpermilvar_pd ||
+ II->getIntrinsicID() == Intrinsic::x86_avx_vpermilvar_pd_256)
+ Index >>= 1;
+ Indexes[I] = Index;
+ }
+ } else if (isa<ConstantAggregateZero>(V)) {
+ for (unsigned I = 0; I < Size; ++I)
+ Indexes[I] = 0;
+ } else {
+ break;
}
- break;
+ // The _256 variants are a bit trickier since the mask bits always index
+ // into the corresponding 128 half. In order to convert to a generic
+ // shuffle, we have to make that explicit.
+ if (II->getIntrinsicID() == Intrinsic::x86_avx_vpermilvar_ps_256 ||
+ II->getIntrinsicID() == Intrinsic::x86_avx_vpermilvar_pd_256) {
+ for (unsigned I = Size / 2; I < Size; ++I)
+ Indexes[I] += Size / 2;
+ }
+ auto NewC =
+ ConstantDataVector::get(V->getContext(), makeArrayRef(Indexes, Size));
+ auto V1 = II->getArgOperand(0);
+ auto V2 = UndefValue::get(V1->getType());
+ auto Shuffle = Builder->CreateShuffleVector(V1, V2, NewC);
+ return ReplaceInstUsesWith(CI, Shuffle);
}
case Intrinsic::ppc_altivec_vperm:
bool AllEltsOk = true;
for (unsigned i = 0; i != 16; ++i) {
Constant *Elt = Mask->getAggregateElement(i);
- if (Elt == 0 ||
- !(isa<ConstantInt>(Elt) || isa<UndefValue>(Elt))) {
+ if (!Elt || !(isa<ConstantInt>(Elt) || isa<UndefValue>(Elt))) {
AllEltsOk = false;
break;
}
cast<ConstantInt>(Mask->getAggregateElement(i))->getZExtValue();
Idx &= 31; // Match the hardware behavior.
- if (ExtractedElts[Idx] == 0) {
+ if (!ExtractedElts[Idx]) {
ExtractedElts[Idx] =
Builder->CreateExtractElement(Idx < 16 ? Op0 : Op1,
Builder->getInt32(Idx&15));
case Intrinsic::arm_neon_vmulls:
case Intrinsic::arm_neon_vmullu:
- case Intrinsic::arm64_neon_smull:
- case Intrinsic::arm64_neon_umull: {
+ case Intrinsic::aarch64_neon_smull:
+ case Intrinsic::aarch64_neon_umull: {
Value *Arg0 = II->getArgOperand(0);
Value *Arg1 = II->getArgOperand(1);
// Check for constant LHS & RHS - in this case we just simplify.
bool Zext = (II->getIntrinsicID() == Intrinsic::arm_neon_vmullu ||
- II->getIntrinsicID() == Intrinsic::arm64_neon_umull);
+ II->getIntrinsicID() == Intrinsic::aarch64_neon_umull);
VectorType *NewVT = cast<VectorType>(II->getType());
if (Constant *CV0 = dyn_cast<Constant>(Arg0)) {
if (Constant *CV1 = dyn_cast<Constant>(Arg1)) {
// mempcpy_chk, memmove_chk, memset_chk, strcpy_chk, stpcpy_chk, strncpy_chk,
// strcat_chk and strncat_chk.
Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const DataLayout *DL) {
- if (CI->getCalledFunction() == 0) return 0;
+ if (!CI->getCalledFunction()) return nullptr;
if (Value *With = Simplifier->optimizeCall(CI)) {
++NumSimplified;
return CI->use_empty() ? CI : ReplaceInstUsesWith(*CI, With);
}
- return 0;
+ return nullptr;
}
static IntrinsicInst *FindInitTrampolineFromAlloca(Value *TrampMem) {
Value *Underlying = TrampMem->stripPointerCasts();
if (Underlying != TrampMem &&
(!Underlying->hasOneUse() || Underlying->user_back() != TrampMem))
- return 0;
+ return nullptr;
if (!isa<AllocaInst>(Underlying))
- return 0;
+ return nullptr;
- IntrinsicInst *InitTrampoline = 0;
+ IntrinsicInst *InitTrampoline = nullptr;
for (User *U : TrampMem->users()) {
IntrinsicInst *II = dyn_cast<IntrinsicInst>(U);
if (!II)
- return 0;
+ return nullptr;
if (II->getIntrinsicID() == Intrinsic::init_trampoline) {
if (InitTrampoline)
// More than one init_trampoline writes to this value. Give up.
- return 0;
+ return nullptr;
InitTrampoline = II;
continue;
}
if (II->getIntrinsicID() == Intrinsic::adjust_trampoline)
// Allow any number of calls to adjust.trampoline.
continue;
- return 0;
+ return nullptr;
}
// No call to init.trampoline found.
if (!InitTrampoline)
- return 0;
+ return nullptr;
// Check that the alloca is being used in the expected way.
if (InitTrampoline->getOperand(0) != TrampMem)
- return 0;
+ return nullptr;
return InitTrampoline;
}
II->getOperand(0) == TrampMem)
return II;
if (Inst->mayWriteToMemory())
- return 0;
+ return nullptr;
}
- return 0;
+ return nullptr;
}
// Given a call to llvm.adjust.trampoline, find and return the corresponding
IntrinsicInst *AdjustTramp = dyn_cast<IntrinsicInst>(Callee);
if (!AdjustTramp ||
AdjustTramp->getIntrinsicID() != Intrinsic::adjust_trampoline)
- return 0;
+ return nullptr;
Value *TrampMem = AdjustTramp->getOperand(0);
return IT;
if (IntrinsicInst *IT = FindInitTrampolineFromBB(AdjustTramp, TrampMem))
return IT;
- return 0;
+ return nullptr;
}
// visitCallSite - Improvements for call and invoke instructions.
// arguments of the call/invoke.
Value *Callee = CS.getCalledValue();
if (!isa<Function>(Callee) && transformConstExprCastCall(CS))
- return 0;
+ return nullptr;
if (Function *CalleeF = dyn_cast<Function>(Callee))
// If the call and callee calling conventions don't match, this call must
// change the callee to a null pointer.
cast<InvokeInst>(OldCall)->setCalledFunction(
Constant::getNullValue(CalleeF->getType()));
- return 0;
+ return nullptr;
}
if (isa<ConstantPointerNull>(Callee) || isa<UndefValue>(Callee)) {
if (isa<InvokeInst>(CS.getInstruction())) {
// Can't remove an invoke because we cannot change the CFG.
- return 0;
+ return nullptr;
}
// This instruction is not reachable, just remove it. We insert a store to
if (I) return EraseInstFromFunction(*I);
}
- return Changed ? CS.getInstruction() : 0;
+ return Changed ? CS.getInstruction() : nullptr;
}
// transformConstExprCastCall - If the callee is a constexpr cast of a function,
bool InstCombiner::transformConstExprCastCall(CallSite CS) {
Function *Callee =
dyn_cast<Function>(CS.getCalledValue()->stripPointerCasts());
- if (Callee == 0)
+ if (!Callee)
return false;
Instruction *Caller = CS.getInstruction();
const AttributeSet &CallerPAL = CS.getAttributes();
CallerPAL.getParamAttributes(i + 1).hasAttribute(i + 1,
Attribute::ByVal)) {
PointerType *ParamPTy = dyn_cast<PointerType>(ParamTy);
- if (ParamPTy == 0 || !ParamPTy->getElementType()->isSized() || DL == 0)
+ if (!ParamPTy || !ParamPTy->getElementType()->isSized() || !DL)
return false;
Type *CurElTy = ActTy->getPointerElementType();
// If the call already has the 'nest' attribute somewhere then give up -
// otherwise 'nest' would occur twice after splicing in the chain.
if (Attrs.hasAttrSomewhere(Attribute::Nest))
- return 0;
+ return nullptr;
assert(Tramp &&
"transformCallThroughTrampoline called with incorrect CallSite.");
const AttributeSet &NestAttrs = NestF->getAttributes();
if (!NestAttrs.isEmpty()) {
unsigned NestIdx = 1;
- Type *NestTy = 0;
+ Type *NestTy = nullptr;
AttributeSet NestAttr;
// Look for a parameter marked with the 'nest' attribute.
projects / oota-llvm.git / blobdiff