NVIDIA / NVIDIA/cutlass

[QST] Missing tcgen05.fence::before_thread_sync

Open
#2,974 5 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

? - Needs Triage question
Dominant language
C++
Stars
10.5k
Forks
2.1k
Avg merge
3d 11h
Merged PRs (30d)
7

Description

Based on the ptx docs, cross-thread ordering between async tcgen05 instructions is guaranteed by before_thread_sync and after_thread_sync. However, all the examples such as the one below only use
fence_view_async_tmem, which is tcgen05.wait.
https://github.com/NVIDIA/cutlass/blob/3f5bafb326e834b69b16fa6d10bc85b644dc1805/examples/python/CuTeDSL/blackwell/fmha_bwd.py#L2148
https://github.com/NVIDIA/cutlass/blob/3f5bafb326e834b69b16fa6d10bc85b644dc1805/examples/python/CuTeDSL/blackwell/fmha_bwd.py#L1684

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the PTX memory consistency documentation linked in the issue, then inspect the referenced examples/python/CuTeDSL/blackwell/fmha_bwd.py locations around lines 1684 and 2148. Compare their use of fence_view_async_tmem with the documented before_thread_sync and after_thread_sync ordering, and treat the issue as done when the missing support or usage is addressed and the affected examples follow the required synchronization pattern.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp, python
Domain
hpc
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
48/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.