Skip to content

Update SpMVOp to new cuSPARSE API and revise default CUDA architectures#97

Merged
ZedongPeng merged 5 commits into
MIT-Lu-Lab:mainfrom
ZedongPeng:update-spmvop
Jul 4, 2026
Merged

Update SpMVOp to new cuSPARSE API and revise default CUDA architectures#97
ZedongPeng merged 5 commits into
MIT-Lu-Lab:mainfrom
ZedongPeng:update-spmvop

Conversation

@ZedongPeng

Copy link
Copy Markdown
Collaborator

Summary

  • SpMVOp: adapt to the new cusparseSpMVOp API in cuSPARSE 12.8.1 (CUDA 13.3), which added a cusparseSpMVOpAlg_t parameter — pass CUSPARSE_SPMVOP_ALG2 explicitly and bump the version gate from 12703 to 12801. Older toolkits keep the classic cusparseSpMV path.
  • CMake: replace the fixed 60 70 75 80 86 89 90 default with per-toolkit architecture lists (adds Blackwell on CUDA ≥ 12.8, drops pre-Turing on CUDA 13), each with SASS for all supported archs + PTX only for the newest. CMAKE_CUDA_ARCHITECTURES / CUDAARCHS overrides still respected.
  • Docs: update README for the new SpMV backend version gate.

@ZedongPeng ZedongPeng merged commit 8599563 into MIT-Lu-Lab:main Jul 4, 2026
16 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant