topic Is there a parallel implementation of mkl_sparse_d_trsv()? in Intel® oneAPI Math Kernel Library

Is there a parallel implementation of mkl_sparse_d_trsv()?

janalik__radim — Tue, 17 Jan 2017 10:38:25 GMT

Hello,
I'm using function mkl_sparse_d_trsv(). No matter how many threads I use (environment variable MKL_NUM_THREADS), the performance is always the same. So it seems to me that this function has no parallel implementation.

Can somebody confirm my assumption, please? Or am I wrong?

Thank you.

Hi Radim,

Alexander_K_Intel2 — Wed, 18 Jan 2017 07:01:08 GMT

Hi Radim,

The function mkl_sparse_d_trsv is threaded but it's scalling can significantly depend on matrix structure. Also, to calculate internal data for parallel implementation one need to call optimize() and mkl_sparse_set_sv_hint routines before

Thanks,

Alex

Thank you.

janalik__radim — Tue, 24 Jan 2017 12:39:42 GMT

Thank you.