Skip to content

[AMDGPU] Add make.buffer.rsrc to InferAddressSpaces #140770

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

Merged
merged 2 commits into from
May 20, 2025
Merged

[AMDGPU] Add make.buffer.rsrc to InferAddressSpaces #140770

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b5f7cac
[AMDGPU] Add make.buffer.rsrc to InferAddressSpaces
krzysz00 May 2, 2025
e08a2e1
Fix test typos that somehow still let things work
krzysz00 May 20, 2025
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
11 changes: 11 additions & 0 deletions llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1044,6 +1044,7 @@ bool GCNTTIImpl::collectFlatAddressOperands(SmallVectorImpl<int> &OpIndexes,
case Intrinsic::amdgcn_flat_atomic_fmax_num:
case Intrinsic::amdgcn_flat_atomic_fmin_num:
case Intrinsic::amdgcn_load_to_lds:
case Intrinsic::amdgcn_make_buffer_rsrc:
OpIndexes.push_back(0);
return true;
default:
Expand Down Expand Up @@ -1124,6 +1125,16 @@ Value *GCNTTIImpl::rewriteIntrinsicWithAddressSpace(IntrinsicInst *II,
II->setCalledFunction(NewDecl);
return II;
}
case Intrinsic::amdgcn_make_buffer_rsrc: {
Type *SrcTy = NewV->getType();
Type *DstTy = II->getType();
Module *M = II->getModule();
Function *NewDecl = Intrinsic::getOrInsertDeclaration(
M, II->getIntrinsicID(), {DstTy, SrcTy});
II->setArgOperand(0, NewV);
II->setCalledFunction(NewDecl);
Copy link
Contributor

@arsenm arsenm May 20, 2025

Choose a reason for hiding this comment

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

Not sure this readjusts the call's type correctly but I guess it must work

return II;
}
default:
return nullptr;
}
Expand Down
16 changes: 16 additions & 0 deletions llvm/test/Transforms/InferAddressSpaces/AMDGPU/mem-intrinsics.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -192,13 +192,29 @@ define amdgpu_kernel void @load_to_lds_fat_pointer_as_flat(ptr addrspace(7) %buf
ret void
}

define amdgpu_kernel void @make_buffer_rsrc_global_as_flat(ptr addrspace(1) %global, i32 %extent) {
;; NOTE: flags value not representative of real input
; CHECK-LABEL: define amdgpu_kernel void @make_buffer_rsrc_global_as_flat(
; CHECK-SAME: ptr addrspace(1) [[GLOBAL:%.*]], i32 [[EXTENT:%.*]]) {
; CHECK-NEXT: [[BUFFER_FAT_PTR:%.*]] = call ptr addrspace(7) @llvm.amdgcn.make.buffer.rsrc.p7.p1(ptr addrspace(1) [[GLOBAL]], i16 0, i32 [[EXTENT]], i32 0)
; CHECK-NEXT: store i32 [[EXTENT]], ptr addrspace(7) [[BUFFER_FAT_PTR]], align 4
; CHECK-NEXT: ret void
;
%cast = addrspacecast ptr addrspace(1) %global to ptr
%buffer.fat.ptr = call ptr addrspace(7) @llvm.amdgcn.make.buffer.rsrc.p7.p0(ptr %cast, i16 0, i32 %extent, i32 0)
store i32 %extent, ptr addrspace(7) %buffer.fat.ptr
ret void
}

Copy link
Contributor

@arsenm arsenm May 20, 2025

Choose a reason for hiding this comment

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

Maybe test vector case

Copy link
Contributor Author

@krzysz00 krzysz00 May 20, 2025

Choose a reason for hiding this comment

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

Last I checked, make.duffer.rsrc doesn't support vectors

declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i1) #1
declare void @llvm.memcpy.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1) #1
declare void @llvm.memcpy.inline.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1) #1
declare void @llvm.memcpy.p0.p3.i32(ptr nocapture writeonly, ptr addrspace(3) nocapture readonly, i32, i1) #1
declare void @llvm.memmove.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1) #1
declare void @llvm.amdgcn.load.to.lds.p0(ptr nocapture readonly, ptr addrspace(3) nocapture writeonly, i32 immarg, i32 immarg, i32 immarg) #1

declare ptr addrspace(7) @llvm.amdgcn.make.buffer.rsrc.p7.p0(ptr readnone, i16, i32, i32) #0

attributes #0 = { nounwind }
attributes #1 = { argmemonly nounwind }

Expand Down