Skip to content

[AMDGPU] Make getAssumedAddrSpace return AS1 for pointer kernel arguments #137488

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
+8,149 −8,700
Open

[AMDGPU] Make getAssumedAddrSpace return AS1 for pointer kernel arguments #137488

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -951,6 +951,10 @@ bool AMDGPUTargetMachine::isNoopAddrSpaceCast(unsigned SrcAS,
}

unsigned AMDGPUTargetMachine::getAssumedAddrSpace(const Value *V) const {
if (auto *Arg = dyn_cast<Argument>(V);
Arg && AMDGPU::isKernelCC(Arg->getParent()))
return AMDGPUAS::GLOBAL_ADDRESS;
Comment on lines +954 to +956
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Probably need to defend against byref values. Ideally we would only use byref, in which case this needs to look at the load from the argument. But then if we're changing the IR calling convention lowering we might as well fix it to emit addrspace(1) in the first place

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we have something like ptr byref(i32) %in.byref, I suppose it is also in AS1?

Copy link
Contributor

@arsenm arsenm Apr 30, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Or 4, yes. But if we're fixing clang to do the right thing, shouldn't need to handle the flat case


const auto *LD = dyn_cast<LoadInst>(V);
if (!LD) // TODO: Handle invariant load like constant.
return AMDGPUAS::UNKNOWN_ADDRESS_SPACE;
Expand Down
23 changes: 6 additions & 17 deletions llvm/lib/Transforms/IPO/AttributorAttributes.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12593,29 +12593,18 @@ struct AAAddressSpaceImpl : public AAAddressSpace {
}

ChangeStatus updateImpl(Attributor &A) override {
unsigned FlatAS = A.getInfoCache().getFlatAddressSpace().value();
uint32_t OldAddressSpace = AssumedAddressSpace;

auto CheckAddressSpace = [&](Value &Obj) {
if (isa<UndefValue>(&Obj))
return true;
// If an argument in flat address space only has addrspace cast uses, and
// those casts are same, then we take the dst addrspace.
if (auto *Arg = dyn_cast<Argument>(&Obj)) {
if (Arg->getType()->getPointerAddressSpace() == FlatAS) {
unsigned CastAddrSpace = FlatAS;
for (auto *U : Arg->users()) {
auto *ASCI = dyn_cast<AddrSpaceCastInst>(U);
if (!ASCI)
return takeAddressSpace(Obj.getType()->getPointerAddressSpace());
if (CastAddrSpace != FlatAS &&
CastAddrSpace != ASCI->getDestAddressSpace())
return false;
CastAddrSpace = ASCI->getDestAddressSpace();
}
if (CastAddrSpace != FlatAS)
return takeAddressSpace(CastAddrSpace);
}
auto *TTI =
A.getInfoCache().getAnalysisResultForFunction<TargetIRAnalysis>(
*Arg->getParent());
unsigned AssumedAS = TTI->getAssumedAddrSpace(Arg);
if (AssumedAS != ~0U)
return takeAddressSpace(AssumedAS);
}
return takeAddressSpace(Obj.getType()->getPointerAddressSpace());
};
Expand Down
Loading