topic Hi, in Intel® oneAPI Math Kernel Library

Threaded preconditioners (dcsrilu0 and dcsrilut)

Kostas_M_Intel — Thu, 29 Jan 2009 12:37:40 GMT

Dear developers. are the routines dcsrilu0() and dcsrilut() multithreaded, are there any plans on their parallelization?

Re: Threaded preconditioners (dcsrilu0 and dcsrilut)

Sergey_G_Intel — Fri, 30 Jan 2009 03:23:19 GMT

Hi Kostas,

The routines for preconditioners are not multithreaded. If you believe, it is important for you to have these routines multithreaded, please file a QUAD via https://premier.intel.com and provide motivation for you request. Thank you!

Regards,
Sergey G

7 years later, I checked with

Andrey_Vladimirov — Tue, 05 Jul 2016 23:13:19 GMT

7 years later, I checked with MKL 11.3.1, and dcsrilu0 runs with 1 thread on a matrix with 70,000 equations and 2,000,000 non-zero elements. Is this routine still single-threaded, or is there a flag/environment variable to enable multi-threading in LU preconditioners?

Hi,

Alexander_K_Intel2 — Wed, 06 Jul 2016 07:34:33 GMT

Hi,

ILU0 and ILUt implemented sequentially because parallel implementation provided a bit different matrix. However triangular solver for these matrices became parallel starting from MKL 11.3

Thanks

Alex

Thanks!

Andrey_Vladimirov — Wed, 06 Jul 2016 18:06:50 GMT

Thanks!