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] Add a pass to convert accumulating GEMMs to GEMMs #19587

Merged

Conversation

nirvedhmeshram
Copy link
Contributor

@nirvedhmeshram nirvedhmeshram commented Jan 2, 2025

Converts dispatches with accumulating GEMMs that are doing in place read/write to GEMM + elementwise add.
This is needed for the TileAndFuse path until we find a more permanent fix for #19546

Copy link
Contributor

@hanhanW hanhanW left a comment

Choose a reason for hiding this comment

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

I'm not sure if we really need this or not. I left some comments about the issue in #19546, please take a look.

Copy link
Contributor

@qedawkins qedawkins left a comment

Choose a reason for hiding this comment

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

This LGTM, but I'll wait to approve until we've come to a concensus on the approach.

@nirvedhmeshram nirvedhmeshram force-pushed the acc_gemm_to_elementwise branch from 7735e86 to 47c2309 Compare January 7, 2025 21:10
@nirvedhmeshram
Copy link
Contributor Author

Looks like we have consensus on #19546 that we want this pass for TileAndFuse.

Signed-off-by: Nirvedh Meshram <[email protected]>
Copy link
Member

@kuhar kuhar left a comment

Choose a reason for hiding this comment

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

Just some drive-by nits

Signed-off-by: Nirvedh Meshram <[email protected]>
Signed-off-by: Nirvedh Meshram <[email protected]>
@nirvedhmeshram nirvedhmeshram force-pushed the acc_gemm_to_elementwise branch from abc56be to b7930af Compare January 7, 2025 22:07
@nirvedhmeshram nirvedhmeshram merged commit 80cbf6b into iree-org:main Jan 7, 2025
33 checks passed
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.

4 participants