Skip to content

Conversation

@wenju-he
Copy link
Contributor

Revert bede0e9:
"[SYCL][SYCLLowerWGLocalMemoryPass] Remove implicit dependency on AlwaysInlinerPass and move to PipelineStart (#16356)"

Motivation:

  • Cleaner than manually force-inlining. In general, we should rely on the pass pipeline's ordering for inlining.
  • bede0e9's rationale - "we can't assume the backend lowers the global after AlwaysInlinerPass" - no longer has use case after a pass re-ordering in downstream target.
  • OptimizerEarlyEPCallback is the first module pass callback after AlwaysInlinerPass. This was also proposed in [SYCL] Move SYCLLowerWGLocalMemoryPass to OptimizerEarlyEPCallback #16347.

Closes #16617

…rEarlyEPCallback

Revert bede0e9:
"[SYCL][SYCLLowerWGLocalMemoryPass] Remove implicit dependency on AlwaysInlinerPass and move to PipelineStart (intel#16356)"

Motivation:
* Cleaner than manually force-inlining. In general, we should rely on
  the pass pipeline's ordering for inlining.
* bede0e9's rationale - "we can't assume the backend lowers the
  global after AlwaysInlinerPass" - no longer has use case after a
  pass re-ordering in downstream target.
* OptimizerEarlyEPCallback is the first module pass callback after
  AlwaysInlinerPass. This was also proposed in intel#16347.

Closes intel#16617
Copy link
Contributor

@bader bader left a comment

Choose a reason for hiding this comment

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

LGTM, except one comment.

Copy link
Contributor

@YuriPlyakhin YuriPlyakhin left a comment

Choose a reason for hiding this comment

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

LGTM. one nit.

@wenju-he
Copy link
Contributor Author

LGTM. one nit.

done

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.

Remove restriction that group_local_memory/group_local_memory_for_overwrite function requires to be inlined

3 participants