Skip to content

[CK Tile] Mixed precision fp16 bf16 x fp8 for non transposed B#5649

Closed
SamiAario-AMD wants to merge 33 commits intodevelopfrom
users/samaario/mixed-prec-fp16-bf16-x-fp8-non-transposed-B
Closed

[CK Tile] Mixed precision fp16 bf16 x fp8 for non transposed B#5649
SamiAario-AMD wants to merge 33 commits intodevelopfrom
users/samaario/mixed-prec-fp16-bf16-x-fp8-non-transposed-B

Conversation

@SamiAario-AMD
Copy link
Copy Markdown
Contributor

Motivation

  • Add tests for bf/fp16 x fp8 and fp8 x bf/fp16 for non-transposed B
  • Add DetermineWarpPrecType, a templated struct that uses pattern-matching to determine the A and B warp GEMM types

This depends on #5510, and is therefore opened as a draft.

Technical Details

Test Plan

Test Result

Submission Checklist

@SamiAario-AMD SamiAario-AMD self-assigned this Mar 20, 2026
@SamiAario-AMD SamiAario-AMD requested a review from a team as a code owner March 20, 2026 12:39
@SamiAario-AMD SamiAario-AMD changed the title Users/samaario/mixed prec fp16 bf16 x fp8 non transposed b [CK Tile] Mixed precision fp16 bf16 x fp8 for non transposed B Mar 20, 2026
@SamiAario-AMD SamiAario-AMD force-pushed the users/samaario/mixed-prec-fp16-bf16-x-fp8-non-transposed-B branch 2 times, most recently from 924b17b to 54fd180 Compare March 24, 2026 14:20
@SamiAario-AMD SamiAario-AMD force-pushed the users/samaario/mixed-prec-fp16-bf16-x-fp8-non-transposed-B branch from 54fd180 to 8055ae8 Compare March 25, 2026 08:20
@SamiAario-AMD SamiAario-AMD force-pushed the users/samaario/mixed-prec-fp16-bf16-x-fp8-non-transposed-B branch from 8055ae8 to 7ff231f Compare March 27, 2026 07:45
@SamiAario-AMD SamiAario-AMD force-pushed the users/samaario/mixed-prec-fp16-bf16-x-fp8-non-transposed-B branch from 7ff231f to 84b498f Compare March 30, 2026 08:20
@SamiAario-AMD SamiAario-AMD force-pushed the users/samaario/mixed-prec-fp16-bf16-x-fp8-non-transposed-B branch from e7aa726 to b4d6af5 Compare April 2, 2026 08:39
@SamiAario-AMD SamiAario-AMD force-pushed the users/samaario/mixed-prec-fp16-bf16-x-fp8-non-transposed-B branch from b4d6af5 to 29d3b81 Compare April 7, 2026 07:20
@aosewski aosewski marked this pull request as draft April 14, 2026 11:26
…tion encodings so that row permutations are not needed
@SamiAario-AMD SamiAario-AMD force-pushed the users/samaario/mixed-prec-fp16-bf16-x-fp8-non-transposed-B branch from 29d3b81 to 26ce440 Compare April 16, 2026 14:48
@SamiAario-AMD
Copy link
Copy Markdown
Contributor Author

I am closing this PR because its functionality is covered by #6612

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

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant