search

NVIDIA / cutlass

CUDA Templates for Linear Algebra Subroutines
Other
5.7k stars 978 forks source link

[QST]In tma_wasp producer's 128 threads, warp1 and warp3 are idle? #1675

Open ziyuhuang123 opened 3 months ago

ziyuhuang123 commented 3 months ago

sm90_gemm_tma_warpspecialized_cooperative

    enum class ProducerWarpRole {
      Mainloop = 0,
      Warp1 = 1,
      Epilogue = 2,
      Warp3 = 3
    };

I find usage of Mainloop and Epilogue, but no usage of warp1 and 3?

By the way, I noticed if we use 320 threads(do not use warp1 3) the occupancy will be 10, and if we use 384 threads(original cutlass method) the occupancy will be 12. Maybe because of this?

ziyuhuang123 commented 3 months ago

Well, if we would rather let 64 threads to be idle... Could we use them for consumer? Or something else?

github-actions[bot] commented 2 months ago

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.