cusparseSpSM supported sparse matrix formats

Accelerated Computing GPU-Accelerated Libraries
1

Hello,

In the docs for the cusparseSpSM() function (v12.8 at the time of writing), it is said that only CSR and COO sparse matrix formats are supported. Similar issue is with the cusparseSpSV() function.

Why is CSC not supported?

From what I know, supporting CSC is as simple as supporting CSR, you just need to switch the transpose flag (because CSR nontrans = CSC trans, and CSR trans = CSC nontrans), and switch the uplo flag.

So, why do I have to do this workaround myself?

I see that many BLAS-like libraries have this deficiency, not only cusparse. There must be a larger reason for this, which I just don’t understand. I asked a similar question on the Intel MKL forum, without any response.

Thanks,

Jakub

2

Hi @jhomola. Thanks for the feedback. We agree that the routines should support CSC for users’ convenience. We’ll add the support in next releases.

3

Perfect, thanks :)

4

Also, would you mind sharing your use case with us? What’s your application? How does cusparseSpSM help you in your computation?

5

Sure. Simply said, I use it to compute a Schur complement of a sparse matrix.

I wrote a little about the matrices I use in this post.

We recently had a coference paper accepted about the topic, you can read the preprint on arxiv.

Feel free to ask more details if needed.

6

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.