Skip to content

Update Blackwell CuTeDSL helpers for Rubin compatibility - #3556

Open
xiufanl wants to merge 3 commits into
NVIDIA:mainfrom
xiufanl:fix-6676300
Open

Update Blackwell CuTeDSL helpers for Rubin compatibility#3556
xiufanl wants to merge 3 commits into
NVIDIA:mainfrom
xiufanl:fix-6676300

Conversation

@xiufanl

@xiufanl xiufanl commented Aug 27, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Update the public Blackwell CuTeDSL helper examples required by the Rubin and cute_ext examples published for the 4.8 DSL wheel.
  • Keep the change focused on the four compatibility gaps reported by NvBugs 6675927, 6676409, 6676454, and 6676300.

Fix

  • Publish the missing helpers/cli_helper.py module used by the public cute_ext examples.
  • Publish the generated 4.8 Blackwell grouped-GEMM, blockscaled grouped-GEMM, and persistent dense-GEMM examples from the authoritative DKG cutlass_ir/compiler/python sources.
  • Restore the use_cached_problem_shapes grouped scheduler contract and the dense scheduler's swizzle/raster propagation expected by Rubin subclasses.

Testing

  • Python bytecode compilation of the four changed files and affected Rubin consumers
  • Static AST contract checks for the grouped constructors and dense grid interface
  • git diff --check
  • DVS 1000193594 confirmed PR head 8f82106e and exposed stale source copies; rerun required for corrected head 2d375f51

Created by dkg-ai-bugfix agent.

@xiufanl xiufanl changed the title Fix Rubin grouped GEMM tensor helper dtype contract Update Blackwell CuTeDSL helpers for Rubin compatibility Aug 27, 2026
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