There is sparse version of BLAS in intel MKL, for example doing multiplication of matrix and vector, we could use
mkl_?coogemv
Computes matrix-vector product of a sparse general matrix stored in the coordinate format with one-based indexing
But I heard that multiplication is much more efficient when doing in CSR storage.
However to generate CSR storage sparse matrix is a headache, it is much easy and understandable to generate a COO storage sparse matrix.
What I want to know is that: Will MKL change the COO storage to CSR storage internally when using coogemv
? The documentation didn't say.