Skip to content

Conversation

@zhaoan12-prc
Copy link
Collaborator

fix: fp8_fmha_crash

Copilot AI review requested due to automatic review settings December 30, 2025 05:50
Copy link
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

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

Pull request overview

This PR fixes a crash related to FP8 data type handling in ROCm FMHA (Fused Multi-Head Attention) operations. The fix introduces a new function getFmhaDataTypeStr that maps TYPE_FP8_E4M3 to "fp8bf16" instead of "fp8", which is the correct data type string expected by the CK (Composable Kernel) library.

  • Adds a new getFmhaDataTypeStr function specifically for FMHA operations that correctly maps FP8 to "fp8bf16"
  • Replaces all calls to getDataTypeStr with getFmhaDataTypeStr in the three FMHA methods: runCKFmha, runCKFmhaV2, and runCKFmhaMLA

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

@zhiqchen-amd zhiqchen-amd force-pushed the zhaoan/fix_fp8_fmha_crash branch from 52e636c to 13fa82f Compare December 31, 2025 02:21
Copilot AI review requested due to automatic review settings December 31, 2025 02:33
@yuzho-amd yuzho-amd force-pushed the zhaoan/fix_fp8_fmha_crash branch from 13fa82f to 321fce8 Compare December 31, 2025 02:33
Copy link
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

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

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.


💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant