Added support for fp16 x fp8, bf16 x fp8, fp8 x fp16, and fp8 x bf16 for the V3 pipeline#6612
Open
SamiAario-AMD wants to merge 18 commits intodevelopfrom
Open
Added support for fp16 x fp8, bf16 x fp8, fp8 x fp16, and fp8 x bf16 for the V3 pipeline#6612SamiAario-AMD wants to merge 18 commits intodevelopfrom
SamiAario-AMD wants to merge 18 commits intodevelopfrom
Conversation
1 task
Contributor
Author
|
Note that the first three commits are also included in this PR: #5510 |
6c12c0d to
b532f32
Compare
6826882 to
74751db
Compare
…:get_warp_dstr_encoding
…ant_microscale_rrr_1d_64
32d7de5 to
753b287
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Motivation
Added support for fp16 x fp8, bf16 x fp8, fp8 x fp16, and fp8 x bf16 for the V3 pipeline
Technical Details
PackNumAccessas a template parameter ofWarpGemmAttributeMfmaandWarpGemmAttributeMfmaIterateK, to enable explicit choosing of a compatible warp tile distribution encoding when converting between different precisions during a transposed loadTest Plan
Added test cases to the V3 universal GEMM tests for all data type and layout combinations that were added
Test Result
Added tests pass validation on the gfx950 platform
Submission Checklist