[CUDA][cuSPARSE] Work around IMA in cuSPARSE ALG1 on SM 8.9 devices (#119610)
Originally surfaced from the discuss forum:
https://discuss.pytorch.org/t/issue-with-torch-sparse-mm-while-running-on-gpu/188669
This has been forwarded to cuSPARSE but we have not yet received a commitment on their end to fix this issue directly.
CC @ptrblck
Pull Request resolved: https://github.com/pytorch/pytorch/pull/119610
Approved by: https://github.com/jeffdaily, https://github.com/jcaip