Skip to content

[AMD] Skip mfma layout in maybeDuplicate #4170

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
zhanglx13 merged 2 commits into from
Jun 24, 2024
Merged

[AMD] Skip mfma layout in maybeDuplicate #4170

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c79231d
[AMD] Skip mfma layout in maybeDuplicate
zhanglx13 Jun 19, 2024
dca557e
Added comments
zhanglx13 Jun 21, 2024
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
6 changes: 6 additions & 0 deletions include/triton/Conversion/TritonGPUToLLVM/ElementwiseOpToLLVMBase.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,6 +88,12 @@ class ElementwiseOpConversionBase : public ConvertOpToLLVMPattern<SourceOp> {
// encoding not available
return resultVals;
Attribute baseEncoding = encoding;
if (isa<AMDMfmaEncodingAttr>(baseEncoding))
Copy link
Copy Markdown
Collaborator

@ThomasRaoux ThomasRaoux Jun 20, 2024

Choose a reason for hiding this comment

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

are sliced of Mfma okay? Shouldn't this check go line 95

Copy link
Copy Markdown
Collaborator Author

@zhanglx13 zhanglx13 Jun 20, 2024

Choose a reason for hiding this comment

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

slice of mfma is ok after #3870

Copy link
Copy Markdown
Member

@antiagainst antiagainst Jun 21, 2024

Choose a reason for hiding this comment

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

Can we add some comments on why skipping here to be clear?

// TODO: this logic seems incorrect for mfma layout. Skip for now.
// We saw mismatches for some flash-attention tests on AMD backend.
// Note that this logic works for sliced layout whose parent is
// mfma layout. Therefore, this is not combined with the following check.
return resultVals;
while (auto sliced = dyn_cast<SliceEncodingAttr>(baseEncoding))
baseEncoding = sliced.getParent();
if (isa<NvidiaMmaEncodingAttr, DotOperandEncodingAttr>(baseEncoding)) {
Expand Down