SPARSE BLAS CSR FORMAT

SPARSE BLAS CSR FORMAT

There seems to be no consistency with the CSR format in mkl. Using the function "mkl_ddnscsr", I convert a matrix, A, to csr format. It works great, because I put that into "dfeast_scsrev" and have no problems. But when I want to use "mkl_dcsrmm" what the hell are pointers b and e? I've uploaded my code and was hoping someone could take a look. Is there a function that converts dense matrices into csr format for sparse BLAS?

Aaron.

AttachmentSize
Downloadtext/x-c++src test.cpp3.58 KB
6 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

Hello, I will research what is happening and some best known methods for what you are trying to do.

Noah,

Thank you for the help. I'll be anxiously waiting, let me know!

Aaron.

In your example matrices B and C were print transposed. I’ve implemented 2 different example with modification to represent in the correct way (in Fortran style/example-2) and case when we decided numbering elements of B and C during row (C style/example).

Attachments: 

AttachmentSize
Downloadtext/x-c++src noah-example.cpp3.76 KB
Downloadtext/x-c++src noah-example-2.cpp3.77 KB

Noah,

Thanks for the help. Did you just change "matdescra[3] = F" to "matdescra[3]=C"? I don't see anyother changes.

Aaron.

Don't forget job[2] = 1; //change to 0; in mkl_ddnscsr(job,&n,&n,A,&n,Acsr,AJ,AI,&info);

 

Leave a Comment

Please sign in to add a comment. Not a member? Join today