8000 [Codegen] Propagate relayout ops before combining by Max191 · Pull Request #20901 · iree-org/iree · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

[Codegen] Propagate relayout ops before combining #20901

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

Merged
merged 2 commits into from
May 29, 2025

Conversation

Max191
Copy link
Contributor
@Max191 Max191 commented May 27, 2025

Adds propagation patterns to CombineLayoutTransformation before introducing any map_scatter ops in order to push relayout ops towards the end of the dispatch. There can be compute ops between sequences of relayout ops, which block further folding of relayout ops into the map_scatter, so the propagation can expose additional folding opportunities. Propagation is currently restricted to cases where additional pack operations will not be introduced on the operands of compute ops.

This PR will help to enable some consumer fusions, like elementwise truncations, in the data tiling fusion path.

Max191 added 2 commits May 29, 2025 18:54
Signed-off-by: Max Dawkins <max.dawkins@gmail.com>
Signed-off-by: Max Dawkins <max.dawkins@gmail.com>
@Max191 Max191 force-pushed the combine-relayout-ops-propagation branch from 3df8a6f to 3fa14b9 Compare May 29, 2025 19:12
@Max191 Max191 requested a review from hanhanW May 29, 2025 19:12
@Max191 Max191 merged commit 3f01120 into iree-org:main May 29, 2025
42 checks passed
@Max191 Max191 deleted the combine-relayout-ops-propagation branch May 29, 2025 19:46
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.

2 participants
0