Skip to content

Conversation

@Acly
Copy link
Collaborator

@Acly Acly commented Nov 29, 2025

Bug was caused by variable shadowing in flash_attn_cm2, introduced in #17186

It was only observable with coopmat2 and GGML_KQ_MASK_PAD=1

Repro:

  1. modify ggml.h and set GGML_KQ_MASK_PAD 1
  2. run test-backend-ops-o FLASH_ATTN_EXT

@Acly Acly requested a review from 0cc4m as a code owner November 29, 2025 22:17
@Acly Acly requested review from 0cc4m and jeffbolznv and removed request for 0cc4m November 29, 2025 22:17
Copy link
Collaborator

@jeffbolznv jeffbolznv left a comment

Choose a reason for hiding this comment

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

LGTM. I had initially written the new logic inside the ifs, but when cleaning it up I guess I missed this.

@github-actions github-actions bot added Vulkan Issues specific to the Vulkan backend ggml changes relating to the ggml tensor library for machine learning labels Nov 29, 2025
@pwilkin pwilkin merged commit 385c3da into ggml-org:master Nov 30, 2025
60 of 63 checks passed
@0cc4m
Copy link
Collaborator

0cc4m commented Nov 30, 2025

@pwilkin Why did you merge it? Please leave merging to the people involved in the PR or the code that it touches, especially when reviews are still open.

@pwilkin
Copy link
Collaborator

pwilkin commented Nov 30, 2025

@0cc4m sorry, I though since @jeffbolznv approved then that was the end of the review process, my bad.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ggml changes relating to the ggml tensor library for machine learning Vulkan Issues specific to the Vulkan backend

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants