Skip to content

[XPU] Declare fused GDN state mutations - #572

Merged
gc-fu merged 1 commit into
mainfrom
pr/gdn-mutation-schemas
Jul 27, 2026
Merged

[XPU] Declare fused GDN state mutations#572
gc-fu merged 1 commit into
mainfrom
pr/gdn-mutation-schemas

Conversation

@gc-fu

@gc-fu gc-fu commented Jul 27, 2026

Copy link
Copy Markdown
Contributor

Summary

  • mark conv_state and ssm_state as mutated by both fused GDN operators
  • mark output and z_out as mutated output buffers
  • declare the returned tensor as an alias of output

Scope

This PR contains only the Torch mutation/alias schemas extracted from #570. It does not include tensor validation, state-index checks, scalar-load changes, or the separate state-shift kernel.

The operator names and call signatures are unchanged. The current framework does not register fake implementations for these two GDN operators, so no framework change is required for this schema-only update.

Testing

Not run as part of PR creation. This is a dispatcher contract change and does not alter eager kernel execution.

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.

1 participant