[QST]Why we use setmaxnreg? Does this change register/Occupancy?
Nobody has claimed this yet.
- Dominant language
- C++
- Stars
- 10.5k
- Forks
- 2.1k
- Avg merge
- 3d 11h
- Merged PRs (30d)
- 7
Description
setmaxnreg is a new feature since Hopper. I noticed this in cutlass:
https://github.com/NVIDIA/cutlass/blob/eee0cab26c8eedea447eb3b58b3498eeba2294da/include/cutlass/gemm/kernel/sm90_gemm_tma_warpspecialized_cooperative.hpp#L446
From above, the consumer register is 232, the producer register is 40.
Different warp can use different register number??? This will affect Occupancy.
Also, we can dynamicaly modify register during kernel running? This will affect Occupancy.
Using NCU, I find a static register number 168, not 232, not 40. Anything wrong？
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the setmaxnreg use at include/cutlass/gemm/kernel/sm90_gemm_tma_warpspecialized_cooperative.hpp around line 446, then compare the reported values with the attached NCU result. Determine how the producer and consumer register settings relate to the observed static register count and occupancy, and document the explanation for the questions raised.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp
- Domain
- performance
- Issue type
- Documentation
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100