Skip to content

Add a verifier for CatOp#9996

Merged
neildhar merged 2 commits into
mainfrom
neildhar/pr9996
Apr 15, 2026
Merged

Add a verifier for CatOp#9996
neildhar merged 2 commits into
mainfrom
neildhar/pr9996

Conversation

@neildhar
Copy link
Copy Markdown
Collaborator

@neildhar neildhar commented Apr 10, 2026

@lezcano
Copy link
Copy Markdown
Contributor

lezcano commented Apr 11, 2026

@Mogball what happened with your PR to make this op just a composition of join reshape and trans?

@neildhar
Copy link
Copy Markdown
Collaborator Author

@lezcano Looks like it was reverted due to performance regressions: #8878

Comment thread lib/Dialect/TritonGPU/IR/Dialect.cpp Outdated
isExpensiveCat does not reflect the constraint we have in lowering,
which is that the number of unique result elements per thread must be
equal to the total number of unique operand elements per thread. This
means that we can sometimes fold `CatOp` into layout conversions that
have destination layouts that violate this requirement.

Rename it to `isLegalCatEncoding` to reflect that it is actually a
correctness requirement, and update it to reflect the actual constraint.
Check that the shapes and encodings of CatOp are valid and can be
lowered.
Base automatically changed from neildhar/pr9995 to main April 15, 2026 06:09
@neildhar neildhar merged commit e700f46 into main Apr 15, 2026
12 of 14 checks passed
@neildhar neildhar deleted the neildhar/pr9996 branch April 15, 2026 13:38
raymondtay pushed a commit to raymondtay/triton that referenced this pull request Apr 18, 2026
Check that the shapes and encodings of CatOp are valid and can be
lowered.
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