topic Direct/PARDISO solvers in Intel® oneAPI Math Kernel Library

Direct/PARDISO solvers

kodanda — Fri, 27 Jun 2008 10:18:13 GMT

Hi,

I have been using MKL' DSS interface to deal with a indefinite matrices in the solution of A*x = y.
(1) my first question is if PARDISO is efficient compared to DSS? the length of 'x' is 60,000.

(2)In every iteration, I have been preparing A in the CSR format before passing it to the MKL subroutines, then I create the handles, create the structure, factorize, solve and delete the handles. Since the shape of my matrix A is fixed during all iterations, can I do the creation of handles(dss_create), and structure creation(dss_define_structure) and reorder (dss_reorder) only once in the very begining of iterations and use only factorization and solution during every iteration?

thanks for any inputs...

kodanda

Re: Direct/PARDISO solvers

Todd_R_Intel — Tue, 01 Jul 2008 23:17:34 GMT

(1) Intel MKL has one direct sparse solver for which we provide two interfaces: the PARDISO interface and DSS interface. The efficiency should be the same.

(2) Yes. The first part of the DSS section of the reference manual has an image (Figure 8-2 in the reference manual) that refers to just such a case.

-Todd

Re: Direct/PARDISO solvers

kodanda — Wed, 02 Jul 2008 07:57:08 GMT

thanks a lot
- kodanda