Skip to content

[TLE][MTHREADS] Support transposed operands in tle.wgmma - #977

Open
Kylin1207 wants to merge 3 commits into
flagos-ai:mainfrom
Kylin1207:pr/mthreads/tle/dev_wgmma
Open

[TLE][MTHREADS] Support transposed operands in tle.wgmma#977
Kylin1207 wants to merge 3 commits into
flagos-ai:mainfrom
Kylin1207:pr/mthreads/tle/dev_wgmma

Conversation

@Kylin1207

Copy link
Copy Markdown
Contributor

performance:

M N K Stages Transpose Torch (ms) Triton (ms) TLE-WS (ms) Triton vs Torch TLE-WS vs Torch TLE-WS vs Triton
8192 7168 16384 3 nn 5.952640 10.529960 6.138560 0.5653x 0.9697x 1.7154x
8192 7168 16384 3 tn 6.043200 10.282360 6.080560 0.5877x 0.9939x 1.6910x
8192 7168 16384 3 nt 6.132200 10.505520 6.108400 0.5837x 1.0039x 1.7198x
8192 7168 16384 3 tt 6.132280 10.555560 6.150320 0.5810x 0.9971x 1.7163x

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