[Enhancement] Enhance async proxy analyzing logic and GEMM dispatching #1135
Conversation
|
Important Review skippedDraft detected. Please check the settings in the CodeRabbit UI or the You can disable this status message by setting the ✨ Finishing touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
👋 Hi! Thank you for contributing to the TileLang project. Please remember to run We appreciate you taking this step! Our team will review your contribution, and we look forward to your awesome work! 🚀 |
Rachmanino
changed the title
Rachmanino
force-pushed
the
fix-fence
branch
2 times, most recently
from
4bbb03d to
d867739
Compare
- Added a mapping for GEMM instruction prefixes in `gemm.h`. - Renamed GEMM functions to include `mma_` prefix for clarity in `gemm_mma.h`, `gemm_sm70.h`, `gemm_sm90.h`, and `gemm_sp_sm80.h`. - Updated function signatures to improve consistency and readability. - Introduced new functions for MMA and WGMMA operations in `gemm_sm90.h` and `gemm_sp_sm90.h`. - Added a utility function for parsing arguments in `inject_fence_proxy.cc` to enhance async handling.
- Introduced entry kind tracking for statements to improve parent-level fence checking. - Updated logic to conditionally insert fence statements based on previous and current proxy kinds. - Added methods to set and get entry kinds for statements, enhancing the overall proxy management. - Refactored existing statement visit methods to accommodate new entry kind logic.
|
I've just refactored |
- Renamed `wmma_gemm_ss` and `wmma_gemm_rs` functions to `mma_gemm_ss` and `mma_gemm_rs` respectively for better dispatch consistency. - Added a comment to clarify the naming choice regarding the `wmma` prefix.
1 participant
gemm.h.mma_prefix for clarity ingemm_mma.h,gemm_sm70.h,gemm_sm90.h, andgemm_sp_sm80.h.gemm_sm90.handgemm_sp_sm90.h.-Enhance async handling in
inject_fence_proxy.cc.