topic Sparse Matrix Operations in Intel® oneAPI Math Kernel Library

Sparse Matrix Operations

JBervel — Tue, 19 Apr 2011 13:30:42 GMT

Hello,

I am developing an application that works with sparse matrices.

I have seen mkl_?csradd(...) and mkl_?csrmultcsr(...). These functions just work with CSR format.

There are not more functions to operate with sparse matrices and return results as sparse matrices, but with different storage formats?

For example functions to add and multiply sparse matrices in COO format.

Thanks

Sparse Matrix Operations

Gennady_F_Intel — Tue, 19 Apr 2011 17:51:54 GMT

There are not more functions to operate with sparse matrices and return results as sparse matrices, but with different storage formats?

>> not yet.

but as an example, you can use saymkl_dcsrcoo() for convertion a sparse matrix in the coordinate format to CSR, then usemkl_?csradd and then convert from CSR to COO again.

--Gennady

Sparse Matrix Operations

junuylia — Thu, 21 Apr 2011 21:27:34 GMT

I'm dealing with sparse symmetric matrices, and I think the packed format works pretty well and efficient. It could do add, multipy, inversion, and find eigenvalues. However, for sparse matrices, the current library isn't good enough for me.

Sparse Matrix Operations

Gennady_F_Intel — Fri, 22 Apr 2011 04:55:06 GMT

yesmaybe for this case.it is always acompromisebetween performance andmemory consumption andusability.

Sparse Matrix Operations

Danesh_Daroui — Thu, 28 Apr 2011 16:54:11 GMT

Actually, MKL acts OK in this case. How can you be sure that the product of two sparse matrices will be sparse? In general this is not true!

D.