#include "InstCombine.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/MemoryBuiltins.h"
+#include "llvm/IR/CallSite.h"
#include "llvm/IR/DataLayout.h"
-#include "llvm/Support/CallSite.h"
-#include "llvm/Support/PatternMatch.h"
+#include "llvm/IR/PatternMatch.h"
#include "llvm/Transforms/Utils/BuildLibCalls.h"
#include "llvm/Transforms/Utils/Local.h"
using namespace llvm;
using namespace PatternMatch;
+#define DEBUG_TYPE "instcombine"
+
STATISTIC(NumSimplified, "Number of library calls simplified");
/// getPromotedType - Return the specified type promoted as it would be to pass
}
Instruction *InstCombiner::SimplifyMemTransfer(MemIntrinsic *MI) {
- unsigned DstAlign = getKnownAlignment(MI->getArgOperand(0), TD);
- unsigned SrcAlign = getKnownAlignment(MI->getArgOperand(1), TD);
+ unsigned DstAlign = getKnownAlignment(MI->getArgOperand(0), DL);
+ unsigned SrcAlign = getKnownAlignment(MI->getArgOperand(1), DL);
unsigned MinAlign = std::min(DstAlign, SrcAlign);
unsigned CopyAlign = MI->getAlignment();
// 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.
// A single load+store correctly handles overlapping memory in the memmove
// case.
uint64_t Size = MemOpLength->getLimitedValue();
- assert(Size && "0-sized memory transfering should be removed already.");
+ 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();
- if (TD && SrcETy->isSized() && TD->getTypeStoreSize(SrcETy) == Size) {
+ if (DL && SrcETy->isSized() && DL->getTypeStoreSize(SrcETy) == Size) {
// The SrcETy might be something like {{{double}}} or [1 x double]. Rip
// down through these levels if so.
SrcETy = reduceToSingleValueType(SrcETy);
}
Instruction *InstCombiner::SimplifyMemSet(MemSetInst *MI) {
- unsigned Alignment = getKnownAlignment(MI->getDest(), TD);
+ unsigned Alignment = getKnownAlignment(MI->getDest(), DL);
if (MI->getAlignment() < Alignment) {
MI->setAlignment(ConstantInt::get(MI->getAlignmentType(),
Alignment, false));
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
default: break;
case Intrinsic::objectsize: {
uint64_t Size;
- if (getObjectSize(II->getArgOperand(0), Size, TD, TLI))
+ 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))))
case Intrinsic::ppc_altivec_lvx:
case Intrinsic::ppc_altivec_lvxl:
// Turn PPC lvx -> load if the pointer is known aligned.
- if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, TD) >= 16) {
+ if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, DL) >= 16) {
Value *Ptr = Builder->CreateBitCast(II->getArgOperand(0),
PointerType::getUnqual(II->getType()));
return new LoadInst(Ptr);
case Intrinsic::ppc_altivec_stvx:
case Intrinsic::ppc_altivec_stvxl:
// Turn stvx -> store if the pointer is known aligned.
- if (getOrEnforceKnownAlignment(II->getArgOperand(1), 16, TD) >= 16) {
+ if (getOrEnforceKnownAlignment(II->getArgOperand(1), 16, DL) >= 16) {
Type *OpPtrTy =
PointerType::getUnqual(II->getArgOperand(0)->getType());
Value *Ptr = Builder->CreateBitCast(II->getArgOperand(1), OpPtrTy);
case Intrinsic::x86_sse2_storeu_pd:
case Intrinsic::x86_sse2_storeu_dq:
// Turn X86 storeu -> store if the pointer is known aligned.
- if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, TD) >= 16) {
+ if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, DL) >= 16) {
Type *OpPtrTy =
PointerType::getUnqual(II->getArgOperand(1)->getType());
Value *Ptr = Builder->CreateBitCast(II->getArgOperand(0), OpPtrTy);
break;
}
+ // Constant fold <A x Bi> << Ci.
+ // FIXME: We don't handle _dq because it's a shift of an i128, but is
+ // represented in the IR as <2 x i64>. A per element shift is wrong.
+ case Intrinsic::x86_sse2_psll_d:
+ case Intrinsic::x86_sse2_psll_q:
+ case Intrinsic::x86_sse2_psll_w:
+ case Intrinsic::x86_sse2_pslli_d:
+ case Intrinsic::x86_sse2_pslli_q:
+ case Intrinsic::x86_sse2_pslli_w:
+ case Intrinsic::x86_avx2_psll_d:
+ case Intrinsic::x86_avx2_psll_q:
+ case Intrinsic::x86_avx2_psll_w:
+ case Intrinsic::x86_avx2_pslli_d:
+ case Intrinsic::x86_avx2_pslli_q:
+ 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)
+ break;
+ ConstantInt *Count;
+ if (CDV)
+ Count = cast<ConstantInt>(CDV->getElementAsConstant(0));
+ else
+ Count = CInt;
+
+ auto Vec = II->getArgOperand(0);
+ auto VT = cast<VectorType>(Vec->getType());
+ if (Count->getZExtValue() >
+ VT->getElementType()->getPrimitiveSizeInBits() - 1)
+ return ReplaceInstUsesWith(
+ CI, ConstantAggregateZero::get(Vec->getType()));
+
+ 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;
+ }
+
+ 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_sse41_pmovsxwd:
break;
}
+ case Intrinsic::x86_sse4a_insertqi: {
+ // insertqi x, y, 64, 0 can just copy y's lower bits and leave the top
+ // ones undef
+ // TODO: eventually we should lower this intrinsic to IR
+ if (auto CIWidth = dyn_cast<ConstantInt>(II->getArgOperand(2))) {
+ if (auto CIStart = dyn_cast<ConstantInt>(II->getArgOperand(3))) {
+ if (CIWidth->equalsInt(64) && CIStart->isZero()) {
+ Value *Vec = II->getArgOperand(1);
+ Value *Undef = UndefValue::get(Vec->getType());
+ const uint32_t Mask[] = { 0, 2 };
+ return ReplaceInstUsesWith(
+ CI,
+ Builder->CreateShuffleVector(
+ Vec, Undef, ConstantDataVector::get(
+ II->getContext(), ArrayRef<uint32_t>(Mask))));
+
+ } else if (auto Source =
+ dyn_cast<IntrinsicInst>(II->getArgOperand(0))) {
+ if (Source->hasOneUse() &&
+ Source->getArgOperand(1) == II->getArgOperand(1)) {
+ // If the source of the insert has only one use and it's another
+ // insert (and they're both inserting from the same vector), try to
+ // bundle both together.
+ auto CISourceWidth =
+ dyn_cast<ConstantInt>(Source->getArgOperand(2));
+ auto CISourceStart =
+ dyn_cast<ConstantInt>(Source->getArgOperand(3));
+ if (CISourceStart && CISourceWidth) {
+ unsigned Start = CIStart->getZExtValue();
+ unsigned Width = CIWidth->getZExtValue();
+ unsigned End = Start + Width;
+ unsigned SourceStart = CISourceStart->getZExtValue();
+ unsigned SourceWidth = CISourceWidth->getZExtValue();
+ unsigned SourceEnd = SourceStart + SourceWidth;
+ unsigned NewStart, NewWidth;
+ bool ShouldReplace = false;
+ if (Start <= SourceStart && SourceStart <= End) {
+ NewStart = Start;
+ NewWidth = std::max(End, SourceEnd) - NewStart;
+ ShouldReplace = true;
+ } else if (SourceStart <= Start && Start <= SourceEnd) {
+ NewStart = SourceStart;
+ NewWidth = std::max(SourceEnd, End) - NewStart;
+ ShouldReplace = true;
+ }
+
+ if (ShouldReplace) {
+ Constant *ConstantWidth = ConstantInt::get(
+ II->getArgOperand(2)->getType(), NewWidth, false);
+ Constant *ConstantStart = ConstantInt::get(
+ II->getArgOperand(3)->getType(), NewStart, false);
+ Value *Args[4] = { Source->getArgOperand(0),
+ II->getArgOperand(1), ConstantWidth,
+ ConstantStart };
+ Module *M = CI.getParent()->getParent()->getParent();
+ Value *F =
+ Intrinsic::getDeclaration(M, Intrinsic::x86_sse4a_insertqi);
+ return ReplaceInstUsesWith(CI, Builder->CreateCall(F, Args));
+ }
+ }
+ }
+ }
+ }
+ }
+ break;
+ }
+
+ case Intrinsic::x86_avx_vpermilvar_ps:
+ case Intrinsic::x86_avx_vpermilvar_ps_256:
+ case Intrinsic::x86_avx_vpermilvar_pd:
+ case Intrinsic::x86_avx_vpermilvar_pd_256: {
+ // Convert vpermil* to shufflevector if the mask is constant.
+ Value *V = II->getArgOperand(1);
+ if (auto C = dyn_cast<ConstantDataVector>(V)) {
+ unsigned Size = C->getNumElements();
+ assert(Size == 8 || Size == 4 || Size == 2);
+ uint32_t Indexes[8];
+
+ // 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;
+ }
+
+ // 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(C->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);
+ }
+ break;
+ }
+
case Intrinsic::ppc_altivec_vperm:
// Turn vperm(V1,V2,mask) -> shuffle(V1,V2,mask) if mask is a constant.
if (Constant *Mask = dyn_cast<Constant>(II->getArgOperand(2))) {
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_vst2lane:
case Intrinsic::arm_neon_vst3lane:
case Intrinsic::arm_neon_vst4lane: {
- unsigned MemAlign = getKnownAlignment(II->getArgOperand(0), TD);
+ unsigned MemAlign = getKnownAlignment(II->getArgOperand(0), DL);
unsigned AlignArg = II->getNumArgOperands() - 1;
ConstantInt *IntrAlign = dyn_cast<ConstantInt>(II->getArgOperand(AlignArg));
if (IntrAlign && IntrAlign->getZExtValue() < MemAlign) {
}
case Intrinsic::arm_neon_vmulls:
- case Intrinsic::arm_neon_vmullu: {
+ case Intrinsic::arm_neon_vmullu:
+ case Intrinsic::arm64_neon_smull:
+ case Intrinsic::arm64_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);
+ bool Zext = (II->getIntrinsicID() == Intrinsic::arm_neon_vmullu ||
+ II->getIntrinsicID() == Intrinsic::arm64_neon_umull);
VectorType *NewVT = cast<VectorType>(II->getType());
- unsigned NewWidth = NewVT->getElementType()->getIntegerBitWidth();
- if (ConstantDataVector *CV0 = dyn_cast<ConstantDataVector>(Arg0)) {
- if (ConstantDataVector *CV1 = dyn_cast<ConstantDataVector>(Arg1)) {
- VectorType* VT = cast<VectorType>(CV0->getType());
- SmallVector<Constant*, 4> NewElems;
- for (unsigned i = 0; i < VT->getNumElements(); ++i) {
- APInt CV0E =
- (cast<ConstantInt>(CV0->getAggregateElement(i)))->getValue();
- CV0E = Zext ? CV0E.zext(NewWidth) : CV0E.sext(NewWidth);
- APInt CV1E =
- (cast<ConstantInt>(CV1->getAggregateElement(i)))->getValue();
- CV1E = Zext ? CV1E.zext(NewWidth) : CV1E.sext(NewWidth);
- NewElems.push_back(
- ConstantInt::get(NewVT->getElementType(), CV0E * CV1E));
- }
- return ReplaceInstUsesWith(CI, ConstantVector::get(NewElems));
+ if (Constant *CV0 = dyn_cast<Constant>(Arg0)) {
+ if (Constant *CV1 = dyn_cast<Constant>(Arg1)) {
+ CV0 = ConstantExpr::getIntegerCast(CV0, NewVT, /*isSigned=*/!Zext);
+ CV1 = ConstantExpr::getIntegerCast(CV1, NewVT, /*isSigned=*/!Zext);
+
+ return ReplaceInstUsesWith(CI, ConstantExpr::getMul(CV0, CV1));
}
- // Couldn't simplify - cannonicalize constant to the RHS.
+ // Couldn't simplify - canonicalize constant to the RHS.
std::swap(Arg0, Arg1);
}
// Handle mul by one:
- if (ConstantDataVector *CV1 = dyn_cast<ConstantDataVector>(Arg1)) {
+ if (Constant *CV1 = dyn_cast<Constant>(Arg1))
if (ConstantInt *Splat =
- dyn_cast_or_null<ConstantInt>(CV1->getSplatValue())) {
- if (Splat->isOne()) {
- if (Zext)
- return CastInst::CreateZExtOrBitCast(Arg0, II->getType());
- // else
- return CastInst::CreateSExtOrBitCast(Arg0, II->getType());
- }
- }
- }
+ dyn_cast_or_null<ConstantInt>(CV1->getSplatValue()))
+ if (Splat->isOne())
+ return CastInst::CreateIntegerCast(Arg0, II->getType(),
+ /*isSigned=*/!Zext);
break;
}
/// passed through the varargs area, we can eliminate the use of the cast.
static bool isSafeToEliminateVarargsCast(const CallSite CS,
const CastInst * const CI,
- const DataLayout * const TD,
+ const DataLayout * const DL,
const int ix) {
if (!CI->isLosslessCast())
return false;
- // The size of ByVal arguments is derived from the type, so we
+ // The size of ByVal or InAlloca arguments is derived from the type, so we
// can't change to a type with a different size. If the size were
// passed explicitly we could avoid this check.
- if (!CS.isByValArgument(ix))
+ if (!CS.isByValOrInAllocaArgument(ix))
return true;
Type* SrcTy =
Type* DstTy = cast<PointerType>(CI->getType())->getElementType();
if (!SrcTy->isSized() || !DstTy->isSized())
return false;
- if (!TD || TD->getTypeAllocSize(SrcTy) != TD->getTypeAllocSize(DstTy))
+ if (!DL || DL->getTypeAllocSize(SrcTy) != DL->getTypeAllocSize(DstTy))
return false;
return true;
}
// Currently we're only working with the checking functions, memcpy_chk,
// mempcpy_chk, memmove_chk, memset_chk, strcpy_chk, stpcpy_chk, strncpy_chk,
// strcat_chk and strncat_chk.
-Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const DataLayout *TD) {
- if (CI->getCalledFunction() == 0) return 0;
+Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const DataLayout *DL) {
+ 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) {
// is good enough in practice and simpler than handling any number of casts.
Value *Underlying = TrampMem->stripPointerCasts();
if (Underlying != TrampMem &&
- (!Underlying->hasOneUse() || *Underlying->use_begin() != TrampMem))
- return 0;
+ (!Underlying->hasOneUse() || Underlying->user_back() != TrampMem))
+ return nullptr;
if (!isa<AllocaInst>(Underlying))
- return 0;
+ return nullptr;
- IntrinsicInst *InitTrampoline = 0;
- for (Value::use_iterator I = TrampMem->use_begin(), E = TrampMem->use_end();
- I != E; I++) {
- IntrinsicInst *II = dyn_cast<IntrinsicInst>(*I);
+ 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
int ix = FTy->getNumParams();
// See if we can optimize any arguments passed through the varargs area of
// the call.
- for (CallSite::arg_iterator I = CS.arg_begin()+FTy->getNumParams(),
+ for (CallSite::arg_iterator I = CS.arg_begin() + FTy->getNumParams(),
E = CS.arg_end(); I != E; ++I, ++ix) {
CastInst *CI = dyn_cast<CastInst>(*I);
- if (CI && isSafeToEliminateVarargsCast(CS, CI, TD, ix)) {
+ if (CI && isSafeToEliminateVarargsCast(CS, CI, DL, ix)) {
*I = CI->getOperand(0);
Changed = true;
}
// this. None of these calls are seen as possibly dead so go ahead and
// delete the instruction now.
if (CallInst *CI = dyn_cast<CallInst>(CS.getInstruction())) {
- Instruction *I = tryOptimizeCall(CI, TD);
+ Instruction *I = tryOptimizeCall(CI, DL);
// If we changed something return the result, etc. Otherwise let
// the fallthrough check.
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();
Type *OldRetTy = Caller->getType();
Type *NewRetTy = FT->getReturnType();
- if (NewRetTy->isStructTy())
- return false; // TODO: Handle multiple return values.
-
// Check to see if we are changing the return type...
if (OldRetTy != NewRetTy) {
- if (Callee->isDeclaration() &&
- // Conversion is ok if changing from one pointer type to another or from
- // a pointer to an integer of the same size.
- !((OldRetTy->isPointerTy() || !TD ||
- OldRetTy == TD->getIntPtrType(Caller->getContext())) &&
- (NewRetTy->isPointerTy() || !TD ||
- NewRetTy == TD->getIntPtrType(Caller->getContext()))))
- return false; // Cannot transform this return value.
- if (!Caller->use_empty() &&
- // void -> non-void is handled specially
- !NewRetTy->isVoidTy() && !CastInst::isCastable(NewRetTy, OldRetTy))
+ if (NewRetTy->isStructTy())
+ return false; // TODO: Handle multiple return values.
+
+ if (!CastInst::isBitCastable(NewRetTy, OldRetTy)) {
+ if (Callee->isDeclaration())
+ return false; // Cannot transform this return value.
+
+ if (!Caller->use_empty() &&
+ // void -> non-void is handled specially
+ !NewRetTy->isVoidTy())
return false; // Cannot transform this return value.
+ }
if (!CallerPAL.isEmpty() && !Caller->use_empty()) {
- AttrBuilder RAttrs = CallerPAL.getRetAttributes();
- if (RAttrs.hasAttributes(Attribute::typeIncompatible(NewRetTy)))
+ AttrBuilder RAttrs(CallerPAL, AttributeSet::ReturnIndex);
+ if (RAttrs.
+ hasAttributes(AttributeFuncs::
+ typeIncompatible(NewRetTy, AttributeSet::ReturnIndex),
+ AttributeSet::ReturnIndex))
return false; // Attribute not compatible with transformed value.
}
// the critical edge). Bail out in this case.
if (!Caller->use_empty())
if (InvokeInst *II = dyn_cast<InvokeInst>(Caller))
- for (Value::use_iterator UI = II->use_begin(), E = II->use_end();
- UI != E; ++UI)
- if (PHINode *PN = dyn_cast<PHINode>(*UI))
+ for (User *U : II->users())
+ if (PHINode *PN = dyn_cast<PHINode>(U))
if (PN->getParent() == II->getNormalDest() ||
PN->getParent() == II->getUnwindDest())
return false;
}
- unsigned NumActualArgs = unsigned(CS.arg_end()-CS.arg_begin());
+ unsigned NumActualArgs = CS.arg_size();
unsigned NumCommonArgs = std::min(FT->getNumParams(), NumActualArgs);
CallSite::arg_iterator AI = CS.arg_begin();
Type *ParamTy = FT->getParamType(i);
Type *ActTy = (*AI)->getType();
- if (!CastInst::isCastable(ActTy, ParamTy))
+ if (!CastInst::isBitCastable(ActTy, ParamTy))
return false; // Cannot transform this parameter value.
- Attribute Attrs = CallerPAL.getParamAttributes(i + 1);
- if (AttrBuilder(Attrs).
- hasAttributes(Attribute::typeIncompatible(ParamTy)))
+ if (AttrBuilder(CallerPAL.getParamAttributes(i + 1), i + 1).
+ hasAttributes(AttributeFuncs::
+ typeIncompatible(ParamTy, i + 1), i + 1))
return false; // Attribute not compatible with transformed value.
+ if (CS.isInAllocaArgument(i))
+ return false; // Cannot transform to and from inalloca.
+
// If the parameter is passed as a byval argument, then we have to have a
// sized type and the sized type has to have the same size as the old type.
- if (ParamTy != ActTy && Attrs.hasAttribute(Attribute::ByVal)) {
+ if (ParamTy != ActTy &&
+ CallerPAL.getParamAttributes(i + 1).hasAttribute(i + 1,
+ Attribute::ByVal)) {
PointerType *ParamPTy = dyn_cast<PointerType>(ParamTy);
- if (ParamPTy == 0 || !ParamPTy->getElementType()->isSized() || TD == 0)
+ if (!ParamPTy || !ParamPTy->getElementType()->isSized() || !DL)
return false;
- Type *CurElTy = cast<PointerType>(ActTy)->getElementType();
- if (TD->getTypeAllocSize(CurElTy) !=
- TD->getTypeAllocSize(ParamPTy->getElementType()))
+ Type *CurElTy = ActTy->getPointerElementType();
+ if (DL->getTypeAllocSize(CurElTy) !=
+ DL->getTypeAllocSize(ParamPTy->getElementType()))
return false;
}
-
- // Converting from one pointer type to another or between a pointer and an
- // integer of the same size is safe even if we do not have a body.
- bool isConvertible = ActTy == ParamTy ||
- (TD && ((ParamTy->isPointerTy() ||
- ParamTy == TD->getIntPtrType(Caller->getContext())) &&
- (ActTy->isPointerTy() ||
- ActTy == TD->getIntPtrType(Caller->getContext()))));
- if (Callee->isDeclaration() && !isConvertible) return false;
}
if (Callee->isDeclaration()) {
// won't be dropping them. Check that these extra arguments have attributes
// that are compatible with being a vararg call argument.
for (unsigned i = CallerPAL.getNumSlots(); i; --i) {
- if (CallerPAL.getSlot(i - 1).Index <= FT->getNumParams())
+ unsigned Index = CallerPAL.getSlotIndex(i - 1);
+ if (Index <= FT->getNumParams())
break;
- Attribute PAttrs = CallerPAL.getSlot(i - 1).Attrs;
+
// Check if it has an attribute that's incompatible with varargs.
- if (PAttrs.hasAttribute(Attribute::StructRet))
+ AttributeSet PAttrs = CallerPAL.getSlotAttributes(i - 1);
+ if (PAttrs.hasAttribute(Index, Attribute::StructRet))
return false;
}
// inserting cast instructions as necessary.
std::vector<Value*> Args;
Args.reserve(NumActualArgs);
- SmallVector<AttributeWithIndex, 8> attrVec;
+ SmallVector<AttributeSet, 8> attrVec;
attrVec.reserve(NumCommonArgs);
// Get any return attributes.
- AttrBuilder RAttrs = CallerPAL.getRetAttributes();
+ AttrBuilder RAttrs(CallerPAL, AttributeSet::ReturnIndex);
// If the return value is not being used, the type may not be compatible
// with the existing attributes. Wipe out any problematic attributes.
- RAttrs.removeAttributes(Attribute::typeIncompatible(NewRetTy));
+ RAttrs.
+ removeAttributes(AttributeFuncs::
+ typeIncompatible(NewRetTy, AttributeSet::ReturnIndex),
+ AttributeSet::ReturnIndex);
// Add the new return attributes.
if (RAttrs.hasAttributes())
- attrVec.push_back(
- AttributeWithIndex::get(AttributeSet::ReturnIndex,
- Attribute::get(FT->getContext(), RAttrs)));
+ attrVec.push_back(AttributeSet::get(Caller->getContext(),
+ AttributeSet::ReturnIndex, RAttrs));
AI = CS.arg_begin();
for (unsigned i = 0; i != NumCommonArgs; ++i, ++AI) {
Type *ParamTy = FT->getParamType(i);
+
if ((*AI)->getType() == ParamTy) {
Args.push_back(*AI);
} else {
- Instruction::CastOps opcode = CastInst::getCastOpcode(*AI,
- false, ParamTy, false);
- Args.push_back(Builder->CreateCast(opcode, *AI, ParamTy));
+ Args.push_back(Builder->CreateBitCast(*AI, ParamTy));
}
// Add any parameter attributes.
- Attribute PAttrs = CallerPAL.getParamAttributes(i + 1);
+ AttrBuilder PAttrs(CallerPAL.getParamAttributes(i + 1), i + 1);
if (PAttrs.hasAttributes())
- attrVec.push_back(AttributeWithIndex::get(i + 1, PAttrs));
+ attrVec.push_back(AttributeSet::get(Caller->getContext(), i + 1,
+ PAttrs));
}
// If the function takes more arguments than the call was taking, add them
}
// Add any parameter attributes.
- Attribute PAttrs = CallerPAL.getParamAttributes(i + 1);
+ AttrBuilder PAttrs(CallerPAL.getParamAttributes(i + 1), i + 1);
if (PAttrs.hasAttributes())
- attrVec.push_back(AttributeWithIndex::get(i + 1, PAttrs));
+ attrVec.push_back(AttributeSet::get(FT->getContext(), i + 1,
+ PAttrs));
}
}
}
- Attribute FnAttrs = CallerPAL.getFnAttributes();
- if (FnAttrs.hasAttributes())
- attrVec.push_back(AttributeWithIndex::get(AttributeSet::FunctionIndex,
- FnAttrs));
+ AttributeSet FnAttrs = CallerPAL.getFnAttributes();
+ if (CallerPAL.hasAttributes(AttributeSet::FunctionIndex))
+ attrVec.push_back(AttributeSet::get(Callee->getContext(), FnAttrs));
if (NewRetTy->isVoidTy())
Caller->setName(""); // Void type should not have a name.
const AttributeSet &NewCallerPAL = AttributeSet::get(Callee->getContext(),
- attrVec);
+ attrVec);
Instruction *NC;
if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
Value *NV = NC;
if (OldRetTy != NV->getType() && !Caller->use_empty()) {
if (!NV->getType()->isVoidTy()) {
- Instruction::CastOps opcode =
- CastInst::getCastOpcode(NC, false, OldRetTy, false);
- NV = NC = CastInst::Create(opcode, NC, OldRetTy);
+ NV = NC = CastInst::Create(CastInst::BitCast, NC, OldRetTy);
NC->setDebugLoc(Caller->getDebugLoc());
// If this is an invoke instruction, we should insert it after the first
if (!Caller->use_empty())
ReplaceInstUsesWith(*Caller, NV);
+ else if (Caller->hasValueHandle())
+ ValueHandleBase::ValueIsRAUWd(Caller, NV);
EraseInstFromFunction(*Caller);
return true;
// 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;
- Attribute NestAttr;
+ Type *NestTy = nullptr;
+ AttributeSet NestAttr;
// Look for a parameter marked with the 'nest' attribute.
for (FunctionType::param_iterator I = NestFTy->param_begin(),
E = NestFTy->param_end(); I != E; ++NestIdx, ++I)
- if (NestAttrs.getParamAttributes(NestIdx).hasAttribute(Attribute::Nest)){
+ if (NestAttrs.hasAttribute(NestIdx, Attribute::Nest)) {
// Record the parameter type and any other attributes.
NestTy = *I;
NestAttr = NestAttrs.getParamAttributes(NestIdx);
if (NestTy) {
Instruction *Caller = CS.getInstruction();
std::vector<Value*> NewArgs;
- NewArgs.reserve(unsigned(CS.arg_end()-CS.arg_begin())+1);
+ NewArgs.reserve(CS.arg_size() + 1);
- SmallVector<AttributeWithIndex, 8> NewAttrs;
+ SmallVector<AttributeSet, 8> NewAttrs;
NewAttrs.reserve(Attrs.getNumSlots() + 1);
// Insert the nest argument into the call argument list, which may
// mean appending it. Likewise for attributes.
// Add any result attributes.
- Attribute Attr = Attrs.getRetAttributes();
- if (Attr.hasAttributes())
- NewAttrs.push_back(AttributeWithIndex::get(AttributeSet::ReturnIndex,
- Attr));
+ if (Attrs.hasAttributes(AttributeSet::ReturnIndex))
+ NewAttrs.push_back(AttributeSet::get(Caller->getContext(),
+ Attrs.getRetAttributes()));
{
unsigned Idx = 1;
if (NestVal->getType() != NestTy)
NestVal = Builder->CreateBitCast(NestVal, NestTy, "nest");
NewArgs.push_back(NestVal);
- NewAttrs.push_back(AttributeWithIndex::get(NestIdx, NestAttr));
+ NewAttrs.push_back(AttributeSet::get(Caller->getContext(),
+ NestAttr));
}
if (I == E)
// Add the original argument and attributes.
NewArgs.push_back(*I);
- Attr = Attrs.getParamAttributes(Idx);
- if (Attr.hasAttributes())
- NewAttrs.push_back
- (AttributeWithIndex::get(Idx + (Idx >= NestIdx), Attr));
+ AttributeSet Attr = Attrs.getParamAttributes(Idx);
+ if (Attr.hasAttributes(Idx)) {
+ AttrBuilder B(Attr, Idx);
+ NewAttrs.push_back(AttributeSet::get(Caller->getContext(),
+ Idx + (Idx >= NestIdx), B));
+ }
++Idx, ++I;
} while (1);
}
// Add any function attributes.
- Attr = Attrs.getFnAttributes();
- if (Attr.hasAttributes())
- NewAttrs.push_back(AttributeWithIndex::get(AttributeSet::FunctionIndex,
- Attr));
+ if (Attrs.hasAttributes(AttributeSet::FunctionIndex))
+ NewAttrs.push_back(AttributeSet::get(FTy->getContext(),
+ Attrs.getFnAttributes()));
// The trampoline may have been bitcast to a bogus type (FTy).
// Handle this by synthesizing a new function type, equal to FTy
NestF->getType() == PointerType::getUnqual(NewFTy) ?
NestF : ConstantExpr::getBitCast(NestF,
PointerType::getUnqual(NewFTy));
- const AttributeSet &NewPAL = AttributeSet::get(FTy->getContext(), NewAttrs);
+ const AttributeSet &NewPAL =
+ AttributeSet::get(FTy->getContext(), NewAttrs);
Instruction *NewCaller;
if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
projects / oota-llvm.git / blobdiff