TransformerEngine icon indicating copy to clipboard operation
TransformerEngine copied to clipboard

[PyTorch/C] Exposed Userbuffers configuration option to control comm and compute stream priorities

Open denera opened this issue 1 year ago • 0 comments

Description

Userbuffers configuration dictionary now has new "comm_priority": -1 and "gemm_priority: -1 options with default values of -1 for both.

Attn. @rachitgarg91

Type of change

  • [ ] Documentation change (change only to the documentation, either a fix or a new content)
  • [ ] Bug fix (non-breaking change which fixes an issue)
  • [x] New feature (non-breaking change which adds functionality)
  • [ ] Breaking change (fix or feature that would cause existing functionality to not work as expected)
  • [ ] Infra/Build change
  • [ ] Code refractor

Changes

  • Updated tex.UbufCommOverlap and tex.UbufP2PCommOverlap constructors with optional arguments for comm and GEMM stream priorities.
  • Updated te.module.base.initialize_ub() to read "comm_priority" and "gemm_priority" options in the config dictionary.

Checklist:

  • [x] I have read and followed the contributing guidelines
  • [x] The functionality is complete
  • [x] I have commented my code, particularly in hard-to-understand areas
  • [x] I have made corresponding changes to the documentation
  • [x] My changes generate no new warnings
  • [x] I have added tests that prove my fix is effective or that my feature works
  • [x] New and existing unit tests pass locally with my changes

denera avatar Aug 29 '24 21:08 denera