// 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 Lint::visitCallSite(CallSite CS) {
Instruction &I = *CS.getInstruction();
Value *Callee = CS.getCalledValue();
visitMemoryReference(I, Callee, MemoryLocation::UnknownSize, 0, nullptr,
MemRef::Callee);
if (Function *F = dyn_cast<Function>(findValue(Callee,
/*OffsetOk=*/false))) {
Assert(CS.getCallingConv() == F->getCallingConv(),
"Undefined behavior: Caller and callee calling convention differ",
&I);
FunctionType *FT = F->getFunctionType();
unsigned NumActualArgs = CS.arg_size();
Assert(FT->isVarArg() ? FT->getNumParams() <= NumActualArgs
: FT->getNumParams() == NumActualArgs,
"Undefined behavior: Call argument count mismatches callee "
"argument count",
&I);
Assert(FT->getReturnType() == I.getType(),
"Undefined behavior: Call return type mismatches "
"callee return type",
&I);
// Check argument types (in case the callee was casted) and attributes.
// TODO: Verify that caller and callee attributes are compatible.
Function::arg_iterator PI = F->arg_begin(), PE = F->arg_end();
CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
for (; AI != AE; ++AI) {
Value *Actual = *AI;
if (PI != PE) {
Argument *Formal = &*PI++;
Assert(Formal->getType() == Actual->getType(),
"Undefined behavior: Call argument type mismatches "
"callee parameter type",
&I);
// Check that noalias arguments don't alias other arguments. This is
// not fully precise because we don't know the sizes of the dereferenced
// memory regions.
if (Formal->hasNoAliasAttr() && Actual->getType()->isPointerTy()) {
AttributeList PAL = CS.getAttributes();
unsigned ArgNo = 0;
for (CallSite::arg_iterator BI = CS.arg_begin(); BI != AE; ++BI) {
// Skip ByVal arguments since they will be memcpy'd to the callee's
// stack so we're not really passing the pointer anyway.
if (PAL.hasParamAttribute(ArgNo++, Attribute::ByVal))
continue;
if (AI != BI && (*BI)->getType()->isPointerTy()) {
AliasResult Result = AA->alias(*AI, *BI);
Assert(Result != MustAlias && Result != PartialAlias,
"Unusual: noalias argument aliases another argument", &I);
}
}
}
// Check that an sret argument points to valid memory.
if (Formal->hasStructRetAttr() && Actual->getType()->isPointerTy()) {
Type *Ty =
cast<PointerType>(Formal->getType())->getElementType();
visitMemoryReference(I, Actual, DL->getTypeStoreSize(Ty),
DL->getABITypeAlignment(Ty), Ty,
MemRef::Read | MemRef::Write);
}
}
}
}
if (CS.isCall()) {
const CallInst *CI = cast<CallInst>(CS.getInstruction());
if (CI->isTailCall()) {
const AttributeList &PAL = CI->getAttributes();
unsigned ArgNo = 0;
for (Value *Arg : CS.args()) {
// Skip ByVal arguments since they will be memcpy'd to the callee's
// stack anyway.
if (PAL.hasParamAttribute(ArgNo++, Attribute::ByVal))
continue;
Value *Obj = findValue(Arg, /*OffsetOk=*/true);
Assert(!isa<AllocaInst>(Obj),
"Undefined behavior: Call with \"tail\" keyword references "
"alloca",
&I);
}
}
}
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(&I))
switch (II->getIntrinsicID()) {
default: break;
// TODO: Check more intrinsics
case Intrinsic::memcpy: {
MemCpyInst *MCI = cast<MemCpyInst>(&I);
// TODO: If the size is known, use it.
visitMemoryReference(I, MCI->getDest(), MemoryLocation::UnknownSize,
MCI->getDestAlignment(), nullptr, MemRef::Write);
visitMemoryReference(I, MCI->getSource(), MemoryLocation::UnknownSize,
MCI->getSourceAlignment(), nullptr, MemRef::Read);
// Check that the memcpy arguments don't overlap. The AliasAnalysis API
// isn't expressive enough for what we really want to do. Known partial
// overlap is not distinguished from the case where nothing is known.
uint64_t Size = 0;
if (const ConstantInt *Len =
dyn_cast<ConstantInt>(findValue(MCI->getLength(),
/*OffsetOk=*/false)))
if (Len->getValue().isIntN(32))
Size = Len->getValue().getZExtValue();
Assert(AA->alias(MCI->getSource(), Size, MCI->getDest(), Size) !=
MustAlias,
"Undefined behavior: memcpy source and destination overlap", &I);
break;
}
case Intrinsic::memmove: {
MemMoveInst *MMI = cast<MemMoveInst>(&I);
// TODO: If the size is known, use it.
visitMemoryReference(I, MMI->getDest(), MemoryLocation::UnknownSize,
MMI->getDestAlignment(), nullptr, MemRef::Write);
visitMemoryReference(I, MMI->getSource(), MemoryLocation::UnknownSize,
MMI->getSourceAlignment(), nullptr, MemRef::Read);
break;
}
case Intrinsic::memset: {
MemSetInst *MSI = cast<MemSetInst>(&I);
// TODO: If the size is known, use it.
visitMemoryReference(I, MSI->getDest(), MemoryLocation::UnknownSize,
MSI->getDestAlignment(), nullptr, MemRef::Write);
break;
}
case Intrinsic::vastart:
Assert(I.getParent()->getParent()->isVarArg(),
"Undefined behavior: va_start called in a non-varargs function",
&I);
visitMemoryReference(I, CS.getArgument(0), MemoryLocation::UnknownSize, 0,
nullptr, MemRef::Read | MemRef::Write);
break;
case Intrinsic::vacopy:
visitMemoryReference(I, CS.getArgument(0), MemoryLocation::UnknownSize, 0,
nullptr, MemRef::Write);
visitMemoryReference(I, CS.getArgument(1), MemoryLocation::UnknownSize, 0,
nullptr, MemRef::Read);
break;
case Intrinsic::vaend:
visitMemoryReference(I, CS.getArgument(0), MemoryLocation::UnknownSize, 0,
nullptr, MemRef::Read | MemRef::Write);
break;
case Intrinsic::stackrestore:
// Stackrestore doesn't read or write memory, but it sets the
// stack pointer, which the compiler may read from or write to
// at any time, so check it for both readability and writeability.
visitMemoryReference(I, CS.getArgument(0), MemoryLocation::UnknownSize, 0,
nullptr, MemRef::Read | MemRef::Write);
break;
}
}