NVIDIA / NVIDIA/CUDALibrarySamples
nvcomp: Unnecessary status synchronisation
@naveenaero is already working on this.
Since Aug 9, 2026.
- Dominant language
- Cuda
- Stars
- 2.5k
- Forks
- 478
- PR merge metrics
- No merged PRs in 30d
Description
Thank you for addressing this: https://github.com/NVIDIA/CUDALibrarySamples/issues/300
However, there are still stalls in the compute stream after the decompressions (in blue):
This seems to be some type of status synchronisation, that - if you know that the data is fine because you have compressed it yourself - shouldn't be necessary. Does the python API expose that currently?
I haven't found it.
Claude has come up with a way to circumvent it by using the C API directly, but it's not nice code: https://github.com/dxqb/OneTrainer/blob/5a4dc6a44e7a7b4887c8ec6cea6ac9a6a14b21f0/modules/util/nvcomp_util_lowlevel.py
opening this separately, because it might get lost here https://github.com/NVIDIA/CUDALibrarySamples/issues/300#issuecomment-5012111639 and this is now merged into production code: https://github.com/Nerogar/OneTrainer/pull/1630
Contributor guide
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.
Assessment
This issue has not been assessed yet.