Issues with converting Sparse BLAS matrices

[briancpark]

Hi, 

I am becoming pretty desperate in trying to resolve this issue. I'm currently trying to benchmark SPMV kernels with MKL for research and I'm trying to compare the different formats for Sparse matrices, namely COO, CSC, CSR, BSR. COO is easy to manipulate and read data from, so I loaded the whole Sparse Suite Matrix collection library http://sparse.tamu.edu/ in COO format in MKL Sparse. Then I try to manipulate and convert from COO to the other respective formats using the MKL Sparse API. It's successful for COO and CSR, but I cannot get CSC conversion to work properly.  BSR failed, but I realized that I have to pad the sparse matrix dimension to fit the block_size.

For CSR, I do these calls:

mkl_sparse_s_create_coo

mkl_sparse_convert_csr

And for CSC, there is no such think as convert_csc, so I tried:

mkl_sparse_s_create_coo

mkl_sparse_s_export_csc

mkl_sparse_s_create_csc

it fails at mkl_sparse_s_export_csc, and gives me status value of  SPARSE_STATUS_INVALID_VALUE.

Is there anything that I need to be aware of or any reference code that I can look off of? I followed the documentation closely but couldn't really figure out what is wrong.

[Spencer_P_Intel · Accepted solution]

CSC (column major compression) is the transpose of CSR (row major compression) format and visa-versa, so you could perform the transpose of your CSR matrix to get the "CSC" format.

So you can use https://www.intel.com/content/www/us/en/docs/onemkl/developer-reference-c/2023-1/mkl-sparse-convert-csr.html with operation = transpose.   then your output is already in "CSC" format and you can extract it as needed using mkl_sparse_?_export_csr

[ShanmukhS_Intel]

Hi,

A gentle reminder:

Has the information provided helped? Could you please let us know if there are any updates regarding your issue?

Best Regards,

Shanmukh.SS

[briancpark]

Apologies for the late response. I tried it and just got some results. Seems like this method works. Thanks for the help!

[ShanmukhS_Intel]

Hi,

It’s great to know that the issue has been resolved, in case you run into any other issues please feel free to create a new thread.

Best Regards,

Shanmukh.SS