topic Could you please direct me to in Intel® oneAPI Math Kernel Library

Speed up performance for a given matrix structure

Dinesh_S_ — Mon, 27 Nov 2017 18:48:08 GMT

If the logical structure (not actual values) of the matrix remains same, is there a way to avoid symbolic factorization stage for problems requiring large number of of linear system solutions?

Regards

Dinesh

you may set maxfct ( maximum

Gennady_F_Intel — Tue, 28 Nov 2017 03:28:40 GMT

you may set maxfct ( maximum #of factors with identical sparsity structure) accordingly your case, make symbolic factorizations only once and then do numerical factorization maxfct times. we have a couple of examples show these cases.

Could you please direct me to

Dinesh_S_ — Tue, 28 Nov 2017 15:38:58 GMT

Could you please direct me to an example to achieve this?

My problem involve solving many linear systems (to obtain solution to non linear system)

For example, I would like to solve

A1 x1 = b1

then

A2 x2 = b2

where A2 depends on x1, but the sparsity of A1 and A2 are the same (however I do not have A2 until I solve for x1)

Regards

Dinesh