Skip to content

[mlir][nvvm] Introduce fence.mbarrier.init #74058

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 4 commits into from
Dec 6, 2023
Merged

[mlir][nvvm] Introduce fence.mbarrier.init #74058

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
9f35504
[mlir][nvvm] Introduce `fence.mbarrier.init`
grypp Dec 1, 2023
34e29b2
add descripton
grypp Dec 1, 2023
e86f790
Merge branch 'main' into fence-mbarrier
grypp Dec 4, 2023
c5d6688
fix typo in test
grypp Dec 6, 2023
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
16 changes: 16 additions & 0 deletions mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -471,6 +471,22 @@ def NVVM_SetMaxRegisterOp : NVVM_PTXBuilder_Op<"setmaxregister"> {
let hasVerifier = 1;
}

def NVVM_FenceMbarrierInitOp : NVVM_PTXBuilder_Op<"fence.mbarrier.init"> {
Copy link
Contributor

Choose a reason for hiding this comment

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

I would like to get your thoughts on:
Should we name the Op as "fence.mbarrier_init" ?
(All other places below use the "_" version).

Either ways, change looks good to me.

Copy link
Member Author

Choose a reason for hiding this comment

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

As far as I can see, MLIR prefers '_' more than '.' . I think I need to change all the OPs to that. I'll do that in follow up PR. It's great that you catch this. So we will have the consistency.

Copy link
Contributor

Choose a reason for hiding this comment

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

Sure, thank you!

let arguments = (ins );
let description = [{
Fence operation that applies on the prior nvvm.mbarrier.init
[For more information, see PTX ISA]
(https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#parallel-synchronization-and-communication-instructions-membar)
}];

let assemblyFormat = "attr-dict";
let extraClassDefinition = [{
std::string $cppClass::getPtx() {
return std::string("fence.mbarrier_init.release.cluster;");
}
}];
}

def ShflKindBfly : I32EnumAttrCase<"bfly", 0>;
def ShflKindUp : I32EnumAttrCase<"up", 1>;
def ShflKindDown : I32EnumAttrCase<"down", 2>;
Expand Down
9 changes: 9 additions & 0 deletions mlir/test/Conversion/NVVMToLLVM/nvvm-to-llvm.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -629,6 +629,15 @@ func.func @cp_bulk_commit() {
nvvm.cp.async.bulk.commit.group
func.return
}


// -----

func.func @fence_mbarrier_init() {
//CHECK: llvm.inline_asm has_side_effects asm_dialect = att "fence.mbarrier_init.release.cluster;"
nvvm.fence.mbarrier.init
func.return
}
// -----

func.func @fence_proxy() {
Expand Down