/// Escape RegNode so that we can access it from child handlers. Find the call
/// to frameescape, if any, in the entry block and append RegNode to the list
/// of arguments.
int WinEHStatePass::escapeRegNode(Function &F) {
// Find the call to frameescape and extract its arguments.
IntrinsicInst *EscapeCall = nullptr;
for (Instruction &I : F.getEntryBlock()) {
IntrinsicInst *II = dyn_cast<IntrinsicInst>(&I);
if (II && II->getIntrinsicID() == Intrinsic::frameescape) {
EscapeCall = II;
break;
}
}
SmallVector<Value *, 8> Args;
if (EscapeCall) {
auto Ops = EscapeCall->arg_operands();
Args.append(Ops.begin(), Ops.end());
}
Args.push_back(RegNode);
// Replace the call (if it exists) with new one. Otherwise, insert at the end
// of the entry block.
IRBuilder<> Builder(&F.getEntryBlock(),
EscapeCall ? EscapeCall : F.getEntryBlock().end());
Builder.CreateCall(FrameEscape, Args);
if (EscapeCall)
EscapeCall->eraseFromParent();
return Args.size() - 1;
}
bool AMDGPUCodeGenPrepare::promoteUniformBitreverseToI32(
IntrinsicInst &I) const {
assert(I.getIntrinsicID() == Intrinsic::bitreverse &&
"I must be bitreverse intrinsic");
assert(needsPromotionToI32(I.getType()) &&
"I does not need promotion to i32");
IRBuilder<> Builder(&I);
Builder.SetCurrentDebugLocation(I.getDebugLoc());
Type *I32Ty = getI32Ty(Builder, I.getType());
Function *I32 =
Intrinsic::getDeclaration(Mod, Intrinsic::bitreverse, { I32Ty });
Value *ExtOp = Builder.CreateZExt(I.getOperand(0), I32Ty);
Value *ExtRes = Builder.CreateCall(I32, { ExtOp });
Value *LShrOp =
Builder.CreateLShr(ExtRes, 32 - getBaseElementBitWidth(I.getType()));
Value *TruncRes =
Builder.CreateTrunc(LShrOp, I.getType());
I.replaceAllUsesWith(TruncRes);
I.eraseFromParent();
return true;
}
void AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I) {
// Array allocations are probably not worth handling, since an allocation of
// the array type is the canonical form.
if (!I.isStaticAlloca() || I.isArrayAllocation())
return;
IRBuilder<> Builder(&I);
// First try to replace the alloca with a vector
Type *AllocaTy = I.getAllocatedType();
DEBUG(dbgs() << "Trying to promote " << I << '\n');
if (tryPromoteAllocaToVector(&I))
return;
DEBUG(dbgs() << " alloca is not a candidate for vectorization.\n");
const Function &ContainingFunction = *I.getParent()->getParent();
// FIXME: We should also try to get this value from the reqd_work_group_size
// function attribute if it is available.
unsigned WorkGroupSize = AMDGPU::getMaximumWorkGroupSize(ContainingFunction);
int AllocaSize =
WorkGroupSize * Mod->getDataLayout().getTypeAllocSize(AllocaTy);
if (AllocaSize > LocalMemAvailable) {
DEBUG(dbgs() << " Not enough local memory to promote alloca.\n");
return;
}
std::vector<Value*> WorkList;
if (!collectUsesWithPtrTypes(&I, WorkList)) {
DEBUG(dbgs() << " Do not know how to convert all uses\n");
return;
}
DEBUG(dbgs() << "Promoting alloca to local memory\n");
LocalMemAvailable -= AllocaSize;
Function *F = I.getParent()->getParent();
Type *GVTy = ArrayType::get(I.getAllocatedType(), WorkGroupSize);
GlobalVariable *GV = new GlobalVariable(
*Mod, GVTy, false, GlobalValue::InternalLinkage,
UndefValue::get(GVTy),
Twine(F->getName()) + Twine('.') + I.getName(),
nullptr,
GlobalVariable::NotThreadLocal,
AMDGPUAS::LOCAL_ADDRESS);
GV->setUnnamedAddr(true);
GV->setAlignment(I.getAlignment());
Value *TCntY, *TCntZ;
std::tie(TCntY, TCntZ) = getLocalSizeYZ(Builder);
Value *TIdX = getWorkitemID(Builder, 0);
Value *TIdY = getWorkitemID(Builder, 1);
Value *TIdZ = getWorkitemID(Builder, 2);
Value *Tmp0 = Builder.CreateMul(TCntY, TCntZ, "", true, true);
Tmp0 = Builder.CreateMul(Tmp0, TIdX);
Value *Tmp1 = Builder.CreateMul(TIdY, TCntZ, "", true, true);
Value *TID = Builder.CreateAdd(Tmp0, Tmp1);
TID = Builder.CreateAdd(TID, TIdZ);
Value *Indices[] = {
Constant::getNullValue(Type::getInt32Ty(Mod->getContext())),
TID
};
Value *Offset = Builder.CreateInBoundsGEP(GVTy, GV, Indices);
I.mutateType(Offset->getType());
I.replaceAllUsesWith(Offset);
I.eraseFromParent();
for (Value *V : WorkList) {
CallInst *Call = dyn_cast<CallInst>(V);
if (!Call) {
Type *EltTy = V->getType()->getPointerElementType();
PointerType *NewTy = PointerType::get(EltTy, AMDGPUAS::LOCAL_ADDRESS);
// The operand's value should be corrected on its own.
if (isa<AddrSpaceCastInst>(V))
continue;
// FIXME: It doesn't really make sense to try to do this for all
// instructions.
V->mutateType(NewTy);
continue;
}
IntrinsicInst *Intr = dyn_cast<IntrinsicInst>(Call);
if (!Intr) {
// FIXME: What is this for? It doesn't make sense to promote arbitrary
// function calls. If the call is to a defined function that can also be
// promoted, we should be able to do this once that function is also
// rewritten.
std::vector<Type*> ArgTypes;
for (unsigned ArgIdx = 0, ArgEnd = Call->getNumArgOperands();
ArgIdx != ArgEnd; ++ArgIdx) {
ArgTypes.push_back(Call->getArgOperand(ArgIdx)->getType());
}
Function *F = Call->getCalledFunction();
FunctionType *NewType = FunctionType::get(Call->getType(), ArgTypes,
F->isVarArg());
Constant *C = Mod->getOrInsertFunction((F->getName() + ".local").str(),
NewType, F->getAttributes());
Function *NewF = cast<Function>(C);
Call->setCalledFunction(NewF);
continue;
}
Builder.SetInsertPoint(Intr);
switch (Intr->getIntrinsicID()) {
case Intrinsic::lifetime_start:
case Intrinsic::lifetime_end:
// These intrinsics are for address space 0 only
Intr->eraseFromParent();
continue;
case Intrinsic::memcpy: {
MemCpyInst *MemCpy = cast<MemCpyInst>(Intr);
Builder.CreateMemCpy(MemCpy->getRawDest(), MemCpy->getRawSource(),
MemCpy->getLength(), MemCpy->getAlignment(),
MemCpy->isVolatile());
Intr->eraseFromParent();
continue;
}
case Intrinsic::memmove: {
MemMoveInst *MemMove = cast<MemMoveInst>(Intr);
Builder.CreateMemMove(MemMove->getRawDest(), MemMove->getRawSource(),
MemMove->getLength(), MemMove->getAlignment(),
MemMove->isVolatile());
Intr->eraseFromParent();
continue;
}
case Intrinsic::memset: {
MemSetInst *MemSet = cast<MemSetInst>(Intr);
Builder.CreateMemSet(MemSet->getRawDest(), MemSet->getValue(),
MemSet->getLength(), MemSet->getAlignment(),
MemSet->isVolatile());
Intr->eraseFromParent();
continue;
}
case Intrinsic::invariant_start:
case Intrinsic::invariant_end:
case Intrinsic::invariant_group_barrier:
Intr->eraseFromParent();
// FIXME: I think the invariant marker should still theoretically apply,
// but the intrinsics need to be changed to accept pointers with any
// address space.
continue;
case Intrinsic::objectsize: {
Value *Src = Intr->getOperand(0);
Type *SrcTy = Src->getType()->getPointerElementType();
Function *ObjectSize = Intrinsic::getDeclaration(Mod,
Intrinsic::objectsize,
{ Intr->getType(), PointerType::get(SrcTy, AMDGPUAS::LOCAL_ADDRESS) }
);
CallInst *NewCall
= Builder.CreateCall(ObjectSize, { Src, Intr->getOperand(1) });
Intr->replaceAllUsesWith(NewCall);
Intr->eraseFromParent();
continue;
}
default:
Intr->dump();
llvm_unreachable("Don't know how to promote alloca intrinsic use.");
}
}
}
// FIXME: Should try to pick the most likely to be profitable allocas first.
bool AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I, bool SufficientLDS) {
// Array allocations are probably not worth handling, since an allocation of
// the array type is the canonical form.
if (!I.isStaticAlloca() || I.isArrayAllocation())
return false;
IRBuilder<> Builder(&I);
// First try to replace the alloca with a vector
Type *AllocaTy = I.getAllocatedType();
DEBUG(dbgs() << "Trying to promote " << I << '\n');
if (tryPromoteAllocaToVector(&I, AS))
return true; // Promoted to vector.
const Function &ContainingFunction = *I.getParent()->getParent();
CallingConv::ID CC = ContainingFunction.getCallingConv();
// Don't promote the alloca to LDS for shader calling conventions as the work
// item ID intrinsics are not supported for these calling conventions.
// Furthermore not all LDS is available for some of the stages.
switch (CC) {
case CallingConv::AMDGPU_KERNEL:
case CallingConv::SPIR_KERNEL:
break;
default:
DEBUG(dbgs() << " promote alloca to LDS not supported with calling convention.\n");
return false;
}
// Not likely to have sufficient local memory for promotion.
if (!SufficientLDS)
return false;
const AMDGPUSubtarget &ST =
TM->getSubtarget<AMDGPUSubtarget>(ContainingFunction);
unsigned WorkGroupSize = ST.getFlatWorkGroupSizes(ContainingFunction).second;
const DataLayout &DL = Mod->getDataLayout();
unsigned Align = I.getAlignment();
if (Align == 0)
Align = DL.getABITypeAlignment(I.getAllocatedType());
// FIXME: This computed padding is likely wrong since it depends on inverse
// usage order.
//
// FIXME: It is also possible that if we're allowed to use all of the memory
// could could end up using more than the maximum due to alignment padding.
uint32_t NewSize = alignTo(CurrentLocalMemUsage, Align);
uint32_t AllocSize = WorkGroupSize * DL.getTypeAllocSize(AllocaTy);
NewSize += AllocSize;
if (NewSize > LocalMemLimit) {
DEBUG(dbgs() << " " << AllocSize
<< " bytes of local memory not available to promote\n");
return false;
}
CurrentLocalMemUsage = NewSize;
std::vector<Value*> WorkList;
if (!collectUsesWithPtrTypes(&I, &I, WorkList)) {
DEBUG(dbgs() << " Do not know how to convert all uses\n");
return false;
}
DEBUG(dbgs() << "Promoting alloca to local memory\n");
Function *F = I.getParent()->getParent();
Type *GVTy = ArrayType::get(I.getAllocatedType(), WorkGroupSize);
GlobalVariable *GV = new GlobalVariable(
*Mod, GVTy, false, GlobalValue::InternalLinkage,
UndefValue::get(GVTy),
Twine(F->getName()) + Twine('.') + I.getName(),
nullptr,
GlobalVariable::NotThreadLocal,
AS.LOCAL_ADDRESS);
GV->setUnnamedAddr(GlobalValue::UnnamedAddr::Global);
GV->setAlignment(I.getAlignment());
Value *TCntY, *TCntZ;
std::tie(TCntY, TCntZ) = getLocalSizeYZ(Builder);
Value *TIdX = getWorkitemID(Builder, 0);
Value *TIdY = getWorkitemID(Builder, 1);
Value *TIdZ = getWorkitemID(Builder, 2);
Value *Tmp0 = Builder.CreateMul(TCntY, TCntZ, "", true, true);
Tmp0 = Builder.CreateMul(Tmp0, TIdX);
Value *Tmp1 = Builder.CreateMul(TIdY, TCntZ, "", true, true);
Value *TID = Builder.CreateAdd(Tmp0, Tmp1);
TID = Builder.CreateAdd(TID, TIdZ);
Value *Indices[] = {
Constant::getNullValue(Type::getInt32Ty(Mod->getContext())),
TID
};
Value *Offset = Builder.CreateInBoundsGEP(GVTy, GV, Indices);
I.mutateType(Offset->getType());
I.replaceAllUsesWith(Offset);
I.eraseFromParent();
for (Value *V : WorkList) {
CallInst *Call = dyn_cast<CallInst>(V);
if (!Call) {
if (ICmpInst *CI = dyn_cast<ICmpInst>(V)) {
Value *Src0 = CI->getOperand(0);
Type *EltTy = Src0->getType()->getPointerElementType();
PointerType *NewTy = PointerType::get(EltTy, AS.LOCAL_ADDRESS);
if (isa<ConstantPointerNull>(CI->getOperand(0)))
CI->setOperand(0, ConstantPointerNull::get(NewTy));
if (isa<ConstantPointerNull>(CI->getOperand(1)))
CI->setOperand(1, ConstantPointerNull::get(NewTy));
continue;
}
// The operand's value should be corrected on its own and we don't want to
// touch the users.
if (isa<AddrSpaceCastInst>(V))
continue;
Type *EltTy = V->getType()->getPointerElementType();
PointerType *NewTy = PointerType::get(EltTy, AS.LOCAL_ADDRESS);
// FIXME: It doesn't really make sense to try to do this for all
// instructions.
V->mutateType(NewTy);
// Adjust the types of any constant operands.
if (SelectInst *SI = dyn_cast<SelectInst>(V)) {
if (isa<ConstantPointerNull>(SI->getOperand(1)))
SI->setOperand(1, ConstantPointerNull::get(NewTy));
if (isa<ConstantPointerNull>(SI->getOperand(2)))
SI->setOperand(2, ConstantPointerNull::get(NewTy));
} else if (PHINode *Phi = dyn_cast<PHINode>(V)) {
for (unsigned I = 0, E = Phi->getNumIncomingValues(); I != E; ++I) {
if (isa<ConstantPointerNull>(Phi->getIncomingValue(I)))
Phi->setIncomingValue(I, ConstantPointerNull::get(NewTy));
}
}
continue;
}
IntrinsicInst *Intr = cast<IntrinsicInst>(Call);
Builder.SetInsertPoint(Intr);
switch (Intr->getIntrinsicID()) {
case Intrinsic::lifetime_start:
case Intrinsic::lifetime_end:
// These intrinsics are for address space 0 only
Intr->eraseFromParent();
continue;
case Intrinsic::memcpy: {
MemCpyInst *MemCpy = cast<MemCpyInst>(Intr);
Builder.CreateMemCpy(MemCpy->getRawDest(), MemCpy->getDestAlignment(),
MemCpy->getRawSource(), MemCpy->getSourceAlignment(),
MemCpy->getLength(), MemCpy->isVolatile());
Intr->eraseFromParent();
continue;
}
case Intrinsic::memmove: {
MemMoveInst *MemMove = cast<MemMoveInst>(Intr);
Builder.CreateMemMove(MemMove->getRawDest(), MemMove->getDestAlignment(),
MemMove->getRawSource(), MemMove->getSourceAlignment(),
MemMove->getLength(), MemMove->isVolatile());
Intr->eraseFromParent();
continue;
}
case Intrinsic::memset: {
MemSetInst *MemSet = cast<MemSetInst>(Intr);
Builder.CreateMemSet(MemSet->getRawDest(), MemSet->getValue(),
MemSet->getLength(), MemSet->getDestAlignment(),
MemSet->isVolatile());
Intr->eraseFromParent();
continue;
}
case Intrinsic::invariant_start:
case Intrinsic::invariant_end:
case Intrinsic::invariant_group_barrier:
Intr->eraseFromParent();
// FIXME: I think the invariant marker should still theoretically apply,
// but the intrinsics need to be changed to accept pointers with any
// address space.
continue;
case Intrinsic::objectsize: {
Value *Src = Intr->getOperand(0);
Type *SrcTy = Src->getType()->getPointerElementType();
Function *ObjectSize = Intrinsic::getDeclaration(Mod,
Intrinsic::objectsize,
{ Intr->getType(), PointerType::get(SrcTy, AS.LOCAL_ADDRESS) }
);
CallInst *NewCall = Builder.CreateCall(
ObjectSize, {Src, Intr->getOperand(1), Intr->getOperand(2)});
Intr->replaceAllUsesWith(NewCall);
Intr->eraseFromParent();
continue;
}
default:
Intr->print(errs());
llvm_unreachable("Don't know how to promote alloca intrinsic use.");
}
}
return true;
}
void AMDGPUPromoteAlloca::visitAlloca(AllocaInst &I) {
IRBuilder<> Builder(&I);
// First try to replace the alloca with a vector
Type *AllocaTy = I.getAllocatedType();
DEBUG(dbgs() << "Trying to promote " << I << '\n');
if (tryPromoteAllocaToVector(&I))
return;
DEBUG(dbgs() << " alloca is not a candidate for vectorization.\n");
// FIXME: This is the maximum work group size. We should try to get
// value from the reqd_work_group_size function attribute if it is
// available.
unsigned WorkGroupSize = 256;
int AllocaSize = WorkGroupSize *
Mod->getDataLayout()->getTypeAllocSize(AllocaTy);
if (AllocaSize > LocalMemAvailable) {
DEBUG(dbgs() << " Not enough local memory to promote alloca.\n");
return;
}
std::vector<Value*> WorkList;
if (!collectUsesWithPtrTypes(&I, WorkList)) {
DEBUG(dbgs() << " Do not know how to convert all uses\n");
return;
}
DEBUG(dbgs() << "Promoting alloca to local memory\n");
LocalMemAvailable -= AllocaSize;
GlobalVariable *GV = new GlobalVariable(
*Mod, ArrayType::get(I.getAllocatedType(), 256), false,
GlobalValue::ExternalLinkage, 0, I.getName(), 0,
GlobalVariable::NotThreadLocal, AMDGPUAS::LOCAL_ADDRESS);
FunctionType *FTy = FunctionType::get(
Type::getInt32Ty(Mod->getContext()), false);
AttributeSet AttrSet;
AttrSet.addAttribute(Mod->getContext(), 0, Attribute::ReadNone);
Value *ReadLocalSizeY = Mod->getOrInsertFunction(
"llvm.r600.read.local.size.y", FTy, AttrSet);
Value *ReadLocalSizeZ = Mod->getOrInsertFunction(
"llvm.r600.read.local.size.z", FTy, AttrSet);
Value *ReadTIDIGX = Mod->getOrInsertFunction(
"llvm.r600.read.tidig.x", FTy, AttrSet);
Value *ReadTIDIGY = Mod->getOrInsertFunction(
"llvm.r600.read.tidig.y", FTy, AttrSet);
Value *ReadTIDIGZ = Mod->getOrInsertFunction(
"llvm.r600.read.tidig.z", FTy, AttrSet);
Value *TCntY = Builder.CreateCall(ReadLocalSizeY);
Value *TCntZ = Builder.CreateCall(ReadLocalSizeZ);
Value *TIdX = Builder.CreateCall(ReadTIDIGX);
Value *TIdY = Builder.CreateCall(ReadTIDIGY);
Value *TIdZ = Builder.CreateCall(ReadTIDIGZ);
Value *Tmp0 = Builder.CreateMul(TCntY, TCntZ);
Tmp0 = Builder.CreateMul(Tmp0, TIdX);
Value *Tmp1 = Builder.CreateMul(TIdY, TCntZ);
Value *TID = Builder.CreateAdd(Tmp0, Tmp1);
TID = Builder.CreateAdd(TID, TIdZ);
std::vector<Value*> Indices;
Indices.push_back(Constant::getNullValue(Type::getInt32Ty(Mod->getContext())));
Indices.push_back(TID);
Value *Offset = Builder.CreateGEP(GV, Indices);
I.mutateType(Offset->getType());
I.replaceAllUsesWith(Offset);
I.eraseFromParent();
for (std::vector<Value*>::iterator i = WorkList.begin(),
e = WorkList.end(); i != e; ++i) {
Value *V = *i;
CallInst *Call = dyn_cast<CallInst>(V);
if (!Call) {
Type *EltTy = V->getType()->getPointerElementType();
PointerType *NewTy = PointerType::get(EltTy, AMDGPUAS::LOCAL_ADDRESS);
// The operand's value should be corrected on its own.
if (isa<AddrSpaceCastInst>(V))
continue;
// FIXME: It doesn't really make sense to try to do this for all
// instructions.
V->mutateType(NewTy);
continue;
}
IntrinsicInst *Intr = dyn_cast<IntrinsicInst>(Call);
if (!Intr) {
std::vector<Type*> ArgTypes;
for (unsigned ArgIdx = 0, ArgEnd = Call->getNumArgOperands();
ArgIdx != ArgEnd; ++ArgIdx) {
ArgTypes.push_back(Call->getArgOperand(ArgIdx)->getType());
}
Function *F = Call->getCalledFunction();
FunctionType *NewType = FunctionType::get(Call->getType(), ArgTypes,
F->isVarArg());
Constant *C = Mod->getOrInsertFunction(StringRef(F->getName().str() + ".local"), NewType,
F->getAttributes());
Function *NewF = cast<Function>(C);
Call->setCalledFunction(NewF);
continue;
}
Builder.SetInsertPoint(Intr);
switch (Intr->getIntrinsicID()) {
case Intrinsic::lifetime_start:
case Intrinsic::lifetime_end:
// These intrinsics are for address space 0 only
Intr->eraseFromParent();
continue;
case Intrinsic::memcpy: {
MemCpyInst *MemCpy = cast<MemCpyInst>(Intr);
Builder.CreateMemCpy(MemCpy->getRawDest(), MemCpy->getRawSource(),
MemCpy->getLength(), MemCpy->getAlignment(),
MemCpy->isVolatile());
Intr->eraseFromParent();
continue;
}
case Intrinsic::memset: {
MemSetInst *MemSet = cast<MemSetInst>(Intr);
Builder.CreateMemSet(MemSet->getRawDest(), MemSet->getValue(),
MemSet->getLength(), MemSet->getAlignment(),
MemSet->isVolatile());
Intr->eraseFromParent();
continue;
}
default:
Intr->dump();
llvm_unreachable("Don't know how to promote alloca intrinsic use.");
}
}
}
/// HandleURoRInvokes - Handle invokes of "_Unwind_Resume_or_Rethrow" calls. The
/// "unwind" part of these invokes jump to a landing pad within the current
/// function. This is a candidate to merge the selector associated with the URoR
/// invoke with the one from the URoR's landing pad.
bool DwarfEHPrepare::HandleURoRInvokes() {
if (!EHCatchAllValue) {
EHCatchAllValue =
F->getParent()->getNamedGlobal("llvm.eh.catch.all.value");
if (!EHCatchAllValue) return false;
}
if (!SelectorIntrinsic) {
SelectorIntrinsic =
Intrinsic::getDeclaration(F->getParent(), Intrinsic::eh_selector);
if (!SelectorIntrinsic) return false;
}
SmallPtrSet<IntrinsicInst*, 32> Sels;
SmallPtrSet<IntrinsicInst*, 32> CatchAllSels;
FindAllCleanupSelectors(Sels, CatchAllSels);
if (!DT)
// We require DominatorTree information.
return CleanupSelectors(CatchAllSels);
if (!URoR) {
URoR = F->getParent()->getFunction("_Unwind_Resume_or_Rethrow");
if (!URoR) return CleanupSelectors(CatchAllSels);
}
SmallPtrSet<InvokeInst*, 32> URoRInvokes;
FindAllURoRInvokes(URoRInvokes);
SmallPtrSet<IntrinsicInst*, 32> SelsToConvert;
for (SmallPtrSet<IntrinsicInst*, 32>::iterator
SI = Sels.begin(), SE = Sels.end(); SI != SE; ++SI) {
const BasicBlock *SelBB = (*SI)->getParent();
for (SmallPtrSet<InvokeInst*, 32>::iterator
UI = URoRInvokes.begin(), UE = URoRInvokes.end(); UI != UE; ++UI) {
const BasicBlock *URoRBB = (*UI)->getParent();
if (DT->dominates(SelBB, URoRBB)) {
SelsToConvert.insert(*SI);
break;
}
}
}
bool Changed = false;
if (Sels.size() != SelsToConvert.size()) {
// If we haven't been able to convert all of the clean-up selectors, then
// loop through the slow way to see if they still need to be converted.
if (!ExceptionValueIntrinsic) {
ExceptionValueIntrinsic =
Intrinsic::getDeclaration(F->getParent(), Intrinsic::eh_exception);
if (!ExceptionValueIntrinsic)
return CleanupSelectors(CatchAllSels);
}
for (Value::use_iterator
I = ExceptionValueIntrinsic->use_begin(),
E = ExceptionValueIntrinsic->use_end(); I != E; ++I) {
IntrinsicInst *EHPtr = dyn_cast<IntrinsicInst>(*I);
if (!EHPtr || EHPtr->getParent()->getParent() != F) continue;
Changed |= PromoteEHPtrStore(EHPtr);
bool URoRInvoke = false;
SmallPtrSet<IntrinsicInst*, 8> SelCalls;
Changed |= FindSelectorAndURoR(EHPtr, URoRInvoke, SelCalls);
if (URoRInvoke) {
// This EH pointer is being used by an invoke of an URoR instruction and
// an eh.selector intrinsic call. If the eh.selector is a 'clean-up', we
// need to convert it to a 'catch-all'.
for (SmallPtrSet<IntrinsicInst*, 8>::iterator
SI = SelCalls.begin(), SE = SelCalls.end(); SI != SE; ++SI)
if (!HasCatchAllInSelector(*SI))
SelsToConvert.insert(*SI);
}
}
}
if (!SelsToConvert.empty()) {
// Convert all clean-up eh.selectors, which are associated with "invokes" of
// URoR calls, into catch-all eh.selectors.
Changed = true;
for (SmallPtrSet<IntrinsicInst*, 8>::iterator
SI = SelsToConvert.begin(), SE = SelsToConvert.end();
SI != SE; ++SI) {
IntrinsicInst *II = *SI;
// Use the exception object pointer and the personality function
// from the original selector.
CallSite CS(II);
IntrinsicInst::op_iterator I = CS.arg_begin();
IntrinsicInst::op_iterator E = CS.arg_end();
IntrinsicInst::op_iterator B = prior(E);
// Exclude last argument if it is an integer.
if (isa<ConstantInt>(B)) E = B;
// Add exception object pointer (front).
// Add personality function (next).
// Add in any filter IDs (rest).
SmallVector<Value*, 8> Args(I, E);
Args.push_back(EHCatchAllValue->getInitializer()); // Catch-all indicator.
CallInst *NewSelector =
CallInst::Create(SelectorIntrinsic, Args.begin(), Args.end(),
"eh.sel.catch.all", II);
NewSelector->setTailCall(II->isTailCall());
NewSelector->setAttributes(II->getAttributes());
NewSelector->setCallingConv(II->getCallingConv());
II->replaceAllUsesWith(NewSelector);
II->eraseFromParent();
}
}
Changed |= CleanupSelectors(CatchAllSels);
return Changed;
}
bool IntrinsicCleanerPass::runOnBasicBlock(BasicBlock &b, Module &M) {
bool dirty = false;
bool block_split=false;
#if LLVM_VERSION_CODE <= LLVM_VERSION(3, 1)
unsigned WordSize = TargetData.getPointerSizeInBits() / 8;
#else
unsigned WordSize = DataLayout.getPointerSizeInBits() / 8;
#endif
for (BasicBlock::iterator i = b.begin(), ie = b.end();
(i != ie) && (block_split == false);) {
IntrinsicInst *ii = dyn_cast<IntrinsicInst>(&*i);
// increment now since LowerIntrinsic deletion makes iterator invalid.
++i;
if(ii) {
switch (ii->getIntrinsicID()) {
case Intrinsic::vastart:
case Intrinsic::vaend:
break;
// Lower vacopy so that object resolution etc is handled by
// normal instructions.
//
// FIXME: This is much more target dependent than just the word size,
// however this works for x86-32 and x86-64.
case Intrinsic::vacopy: { // (dst, src) -> *((i8**) dst) = *((i8**) src)
Value *dst = ii->getArgOperand(0);
Value *src = ii->getArgOperand(1);
if (WordSize == 4) {
Type *i8pp = PointerType::getUnqual(PointerType::getUnqual(Type::getInt8Ty(getGlobalContext())));
Value *castedDst = CastInst::CreatePointerCast(dst, i8pp, "vacopy.cast.dst", ii);
Value *castedSrc = CastInst::CreatePointerCast(src, i8pp, "vacopy.cast.src", ii);
Value *load = new LoadInst(castedSrc, "vacopy.read", ii);
new StoreInst(load, castedDst, false, ii);
} else {
assert(WordSize == 8 && "Invalid word size!");
Type *i64p = PointerType::getUnqual(Type::getInt64Ty(getGlobalContext()));
Value *pDst = CastInst::CreatePointerCast(dst, i64p, "vacopy.cast.dst", ii);
Value *pSrc = CastInst::CreatePointerCast(src, i64p, "vacopy.cast.src", ii);
Value *val = new LoadInst(pSrc, std::string(), ii); new StoreInst(val, pDst, ii);
Value *off = ConstantInt::get(Type::getInt64Ty(getGlobalContext()), 1);
pDst = GetElementPtrInst::Create(pDst, off, std::string(), ii);
pSrc = GetElementPtrInst::Create(pSrc, off, std::string(), ii);
val = new LoadInst(pSrc, std::string(), ii); new StoreInst(val, pDst, ii);
pDst = GetElementPtrInst::Create(pDst, off, std::string(), ii);
pSrc = GetElementPtrInst::Create(pSrc, off, std::string(), ii);
val = new LoadInst(pSrc, std::string(), ii); new StoreInst(val, pDst, ii);
}
ii->removeFromParent();
delete ii;
break;
}
case Intrinsic::sadd_with_overflow:
case Intrinsic::ssub_with_overflow:
case Intrinsic::smul_with_overflow:
case Intrinsic::uadd_with_overflow:
case Intrinsic::usub_with_overflow:
case Intrinsic::umul_with_overflow: {
IRBuilder<> builder(ii->getParent(), ii);
Value *op1 = ii->getArgOperand(0);
Value *op2 = ii->getArgOperand(1);
Value *result = 0;
Value *result_ext = 0;
Value *overflow = 0;
unsigned int bw = op1->getType()->getPrimitiveSizeInBits();
unsigned int bw2 = op1->getType()->getPrimitiveSizeInBits()*2;
if ((ii->getIntrinsicID() == Intrinsic::uadd_with_overflow) ||
(ii->getIntrinsicID() == Intrinsic::usub_with_overflow) ||
(ii->getIntrinsicID() == Intrinsic::umul_with_overflow)) {
Value *op1ext =
builder.CreateZExt(op1, IntegerType::get(M.getContext(), bw2));
Value *op2ext =
builder.CreateZExt(op2, IntegerType::get(M.getContext(), bw2));
Value *int_max_s =
ConstantInt::get(op1->getType(), APInt::getMaxValue(bw));
Value *int_max =
builder.CreateZExt(int_max_s, IntegerType::get(M.getContext(), bw2));
if (ii->getIntrinsicID() == Intrinsic::uadd_with_overflow){
result_ext = builder.CreateAdd(op1ext, op2ext);
} else if (ii->getIntrinsicID() == Intrinsic::usub_with_overflow){
result_ext = builder.CreateSub(op1ext, op2ext);
} else if (ii->getIntrinsicID() == Intrinsic::umul_with_overflow){
result_ext = builder.CreateMul(op1ext, op2ext);
}
overflow = builder.CreateICmpUGT(result_ext, int_max);
} else if ((ii->getIntrinsicID() == Intrinsic::sadd_with_overflow) ||
(ii->getIntrinsicID() == Intrinsic::ssub_with_overflow) ||
(ii->getIntrinsicID() == Intrinsic::smul_with_overflow)) {
Value *op1ext =
builder.CreateSExt(op1, IntegerType::get(M.getContext(), bw2));
Value *op2ext =
builder.CreateSExt(op2, IntegerType::get(M.getContext(), bw2));
Value *int_max_s =
ConstantInt::get(op1->getType(), APInt::getSignedMaxValue(bw));
Value *int_min_s =
ConstantInt::get(op1->getType(), APInt::getSignedMinValue(bw));
Value *int_max =
builder.CreateSExt(int_max_s, IntegerType::get(M.getContext(), bw2));
Value *int_min =
builder.CreateSExt(int_min_s, IntegerType::get(M.getContext(), bw2));
if (ii->getIntrinsicID() == Intrinsic::sadd_with_overflow){
result_ext = builder.CreateAdd(op1ext, op2ext);
} else if (ii->getIntrinsicID() == Intrinsic::ssub_with_overflow){
result_ext = builder.CreateSub(op1ext, op2ext);
} else if (ii->getIntrinsicID() == Intrinsic::smul_with_overflow){
result_ext = builder.CreateMul(op1ext, op2ext);
}
overflow = builder.CreateOr(builder.CreateICmpSGT(result_ext, int_max),
builder.CreateICmpSLT(result_ext, int_min));
}
// This trunc cound be replaced by a more general trunc replacement
// that allows to detect also undefined behavior in assignments or
// overflow in operation with integers whose dimension is smaller than
// int's dimension, e.g.
// uint8_t = uint8_t + uint8_t;
// if one desires the wrapping should write
// uint8_t = (uint8_t + uint8_t) & 0xFF;
// before this, must check if it has side effects on other operations
result = builder.CreateTrunc(result_ext, op1->getType());
Value *resultStruct =
builder.CreateInsertValue(UndefValue::get(ii->getType()), result, 0);
resultStruct = builder.CreateInsertValue(resultStruct, overflow, 1);
ii->replaceAllUsesWith(resultStruct);
ii->removeFromParent();
delete ii;
dirty = true;
break;
}
case Intrinsic::dbg_value:
case Intrinsic::dbg_declare:
// Remove these regardless of lower intrinsics flag. This can
// be removed once IntrinsicLowering is fixed to not have bad
// caches.
ii->eraseFromParent();
dirty = true;
break;
case Intrinsic::trap: {
// Intrisic instruction "llvm.trap" found. Directly lower it to
// a call of the abort() function.
Function *F = cast<Function>(
M.getOrInsertFunction(
"abort", Type::getVoidTy(getGlobalContext()), NULL));
F->setDoesNotReturn();
F->setDoesNotThrow();
CallInst::Create(F, Twine(), ii);
new UnreachableInst(getGlobalContext(), ii);
ii->eraseFromParent();
dirty = true;
break;
}
case Intrinsic::objectsize: {
// We don't know the size of an object in general so we replace
// with 0 or -1 depending on the second argument to the intrinsic.
assert(ii->getNumArgOperands() == 2 && "wrong number of arguments");
Value *minArg = ii->getArgOperand(1);
assert(minArg && "Failed to get second argument");
ConstantInt *minArgAsInt = dyn_cast<ConstantInt>(minArg);
assert(minArgAsInt && "Second arg is not a ConstantInt");
assert(minArgAsInt->getBitWidth() == 1 && "Second argument is not an i1");
Value *replacement = NULL;
LLVM_TYPE_Q IntegerType *intType = dyn_cast<IntegerType>(ii->getType());
assert(intType && "intrinsic does not have integer return type");
if (minArgAsInt->isZero()) {
// min=false
replacement = ConstantInt::get(intType, -1, /*isSigned=*/true);
} else {
// min=true
replacement = ConstantInt::get(intType, 0, /*isSigned=*/false);
}
ii->replaceAllUsesWith(replacement);
ii->eraseFromParent();
dirty = true;
break;
}
default:
if (LowerIntrinsics)
IL->LowerIntrinsicCall(ii);
dirty = true;
break;
}
}
}
return dirty;
}
