NVIDIA / NVIDIA/cutlass

[BUG] EpilogueTileAuto doesn't work when tile shape N={48, 80, 96, 112, ...} on SM100

Open
#2,268 3 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

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

Description

For non-blockscaled blackwell gemm, theoretically all multiple number of 16 as ctaN = {16, 32, 48, 64, 80, ...} should be supported as the UMMA specification.

However, I get compilation error with those CtaNs other than 16, 32, 64, 96, 128, 192, 256.

For example, when I set the tile shape of the example 70_blackwell_fp16_gemm.cu to <_64, _48, _64>, I encounter some long cryptic error that appears to be related to shape divisibility.

PR #2220 addresses this kind of issue with Hopper kernels, so I will come up with the similar approach to this issue soon.

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 tile-shape configuration in example 70_blackwell_fp16_gemm.cu and reproduce the compilation failure using CtaN=48. Read PR #2220 for the related Hopper approach, then compare the supported and failing Blackwell shapes. Done means non-blockscaled Blackwell GEMM compiles for the specified multiples of 16, including CtaN=48, 80, and 112.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.