Skip to content

hipblaslt Fallback hotfix#652

Merged
alextmagro merged 4 commits into
devfrom
hipkittens_fallback_hotfix
Jun 30, 2026
Merged

hipblaslt Fallback hotfix#652
alextmagro merged 4 commits into
devfrom
hipkittens_fallback_hotfix

Conversation

@alextmagro

Copy link
Copy Markdown
Contributor

Fixes a bug where HipKittens does not fall back to hipBLASlt for unaligned mxfp8 GEMMs.

Comment thread transformer_engine/common/gemm/rocm_gemm.cu Outdated
@wenchenvincent wenchenvincent added the ci-level 3 CI test level 3 label Jun 26, 2026

@ipanfilo ipanfilo left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

looks good, just minor note. it is already approved

static_cast<int>(outputPreGelu->data.dtype),
workspace, workspaceSize, s);
}
if (!use_hipkittens || !hipkittens_gemm_complete) {

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: no need to check use_hipkittens - hipkittens_gemm_complete can only be True if use_hipkittens is True. Only use_hipkittens can be sufficient (nohipkittens_gemm_complete) , if assign it at line 2004

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I have simplified the check as you mentioned.

#ifdef USE_HIPKITTENS_GEMM

bool use_hipkittens = false;
bool use_hipkittens = false, hipkittens_gemm_complete = false;

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is unused variable

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry, missed that. Should be fixed now.

@VeeraRajasekhar VeeraRajasekhar mentioned this pull request Jun 29, 2026
13 tasks
@alextmagro alextmagro merged commit 365ff10 into dev Jun 30, 2026
5 of 9 checks passed
@alextmagro alextmagro deleted the hipkittens_fallback_hotfix branch June 30, 2026 01:19
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ci-level 3 CI test level 3

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants