chenchenghao/DeepGEMM
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Indivisible TMA (#90)

Browse Source 
Fix indivisible shapes for TMA multicast

---------

Co-authored-by: yukuai <yukuai@deepseek.com>
Co-authored-by: Chenggang Zhao <chenggangz@deepseek.com>
This commit is contained in:
yukuai26
2025-04-23 14:55:14 +08:00
committed by GitHub
co-authored by yukuai Chenggang Zhao
parent 891f35adf5
commit 95e81b3dd6
5 changed files with 73 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+1 -1
View File
@@ -16,7 +16,7 @@ Despite its lightweight design, DeepGEMM's performance matches or exceeds expert
- [x] Shared memory swizzling for output
- [ ] Larger block size on N (up to 256)
- [x] MoE scheduler with TMA multicast compatibility
- [ ] Fix TMA multicast compatibility for indivisible shapes
- [x] Fix TMA multicast compatibility for indivisible shapes
- [ ] Weight gradient kernels for dense models
- [ ] Weight gradient kernels for MoE models
- [ ] Utility kernels for MoE models (as a pre-built CUDA library)