topic hello, in Intel® oneAPI Math Kernel Library

dss_factor_real export factorisation

Twoot — Mon, 19 Feb 2018 15:17:41 GMT

Hello,

I am working on a method where I need the factorisation of the coefficient matrix.

Right now I call the routine dss_factor_real( handle, opt, rValues ) in C which creates a factorisation.

Can anybody tell me:

How is the factorisation stored within the Intel solver datastructure?
Is it possible to export the factorisation for usage elsewhere in the code?

Thanks in advance.

Wouter

hello,

Gennady_F_Intel — Tue, 20 Feb 2018 01:50:00 GMT

hello,

1. we don't explore some details about this internal structure and

2. there is only one option for pardiso API ( not dss ) - `to store/restore internal Pardiso's structures to/from file and stop/continue execution ...

see Intel Pardiso documentation for details.

Quote:Gennady F. (Intel)

Twoot — Tue, 20 Feb 2018 08:19:24 GMT

Gennady F. (Intel) wrote:

hello,

1. we don't explore some details about this internal structure and

2. there is only one option for pardiso API ( not dss ) - `to store/restore internal Pardiso's structures to/from file and stop/continue execution ...

see Intel Pardiso documentation for details.

Thank you for your reply.

So even if I store the Pardiso structure to file and stop execution, it would not be possible to explicitely use the factorisation in my own program?

explicitly - not possible at

Gennady_F_Intel — Wed, 21 Feb 2018 03:06:06 GMT

explicitly - not possible at this moment.

In the following

Twoot — Wed, 21 Feb 2018 11:02:00 GMT

Ok, thanks for the clarification.

In the following documentation: https://software.intel.com/en-us/articles/intel-mkl-support-to-new-functionality-schur-complement

I found that the functionality that I was trying to implement is already implemented. Could you tell me if it is possible to use this functionality with the DSS interface to PARDSIO?

If not, is there a straight-forward way to migrate from the DSS interface to the more elaborate PARDISO API?

Thanks in advance.