Skip to content
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.

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

[GPU] Use tile and fuse for matmul after vector distribute by default #19884

Merged
merged 2 commits into from
Feb 3, 2025

Conversation

nirvedhmeshram
Copy link
Contributor

@nirvedhmeshram nirvedhmeshram commented Feb 3, 2025

Currently some efforts such as #19854 and #19520 are ongoing to make the Tile and Fuse matmul pipeline on by default. However, these efforts are still WIP in achieving exact parity with the current default of Vector Distribute in all use cases.
This PR in the time being tries Tile and Fuse after Vector Distribute so that we can get the benefits of tile and fuse such as handling unaligned to intrinsic shape while leaving the shapes that vector distribute handles untouched.
Fixes : #19864
Progress towards: #19855

Note the for #19855 we also need #19857

Signed-off-by: Nirvedh Meshram <nirvedh@gmail.com>
Signed-off-by: Nirvedh Meshram <nirvedh@gmail.com>
@nirvedhmeshram nirvedhmeshram merged commit d661efa into iree-org:main Feb 3, 2025
43 checks passed
ita9naiwa pushed a commit to ita9naiwa/iree that referenced this pull request Feb 4, 2025
…iree-org#19884)

Currently some efforts such as
iree-org#19854 and
iree-org#19520 are ongoing to make the Tile
and Fuse matmul pipeline on by default. However, these efforts are still
WIP in achieving exact parity with the current default of Vector
Distribute in all use cases.
This PR in the time being tries Tile and Fuse after Vector Distribute so
that we can get the benefits of tile and fuse such as handling unaligned
to intrinsic shape while leaving the shapes that vector distribute
handles untouched.
Fixes :  iree-org#19864
Fixes: iree-org#19855

---------

Signed-off-by: Nirvedh Meshram <nirvedh@gmail.com>
Signed-off-by: Hyunsung Lee <ita9naiwa@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

[GPU] Generic matmul-like not lowering to MFMA by default
5 participants