NVIDIA / NVIDIA/cutlass

[FEA] A flash-attention cuteDSL kernel for sm120

Open
#2,956 2 comments 2 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

? - Needs Triage CuTe DSL feature request inactive-30d inactive-90d
Dominant language
C++
Stars
10.5k
Forks
2.1k
Avg merge
3d 11h
Merged PRs (30d)
7

Description

Which component requires the feature?

CuTe DSL

Feature Request

Dear developers,
Could we offer a cuteDSL flash attention kernel for sm120 (Blackwell_geforce).
I saw the cuteDSL flash attention kernel for ampere has already been developed, and I thought it will not be very hard for sm120 support (add TMA).
It is useful because there are few high-performance FA kernel for sm120

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 by locating the existing CuTe DSL flash-attention kernel for Ampere and the entry points for Blackwell_geforce sm120 support. Read how the Ampere kernel handles its architecture-specific operations, then determine the TMA changes needed for sm120. Done means a working sm120 CuTe DSL flash-attention kernel with evidence that it performs correctly and improves high-performance FA coverage.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp, python
Domain
machine-learning, performance
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.