The wrong result when using Pardiso solver to solve linear equations in parallel (OpenMP)

[Ewigstern]

Hi, Vidya,

Thank you so much for getting back to me again. I've been trying to get back to you under the post with the topic "The wrong result when using Pardiso solver to solve linear equations in parallel (OpenMP)" (link: https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/The-wrong-result-when-using-Pardiso-solver-to-solve-linear/m-p/1414332), but all my attempts have failed and I don't know why. I got in touch with Intel team, but unfortunately they haven't found a solution to my problem. I'm so sorry about that. This is the only way I can contact you, and I hope you can see this post. Here is my reply to your question:

I'm working on solving the equation: AX=B, where all columns of B can be divided into several groups. The columns in each group are completely independent. So I want to compute these columns in an OpenMP parallel way , and then solve the X corresponding to theses columns in parallel. During this parallel process, the solution time can be changed by flexibly adjusting number of threads.

I know that Pardiso can solve equations with multiple right-hand sides in parallel, but the parallelism inside Pardiso doesn't seem to be different from what I want. Also, I encountered the following confusion when using Pardiso to solve equations with multiple right-hand sides: 

1. The calculation time seems to be the same whether I use sequential MKL or parallel MKL. Furthermore, for parallel MKL, I change the number of threads by command " call mkl_set_dynamic(0)" and "call mkl_set_num_threads(p)", but the computation time is hardly affected by the number of threads. Could you tell me the reason for these phenomenons? What is the internal parallel mechanism for Pardiso to solve equations with multiple right-hand sides?

2. The Intel document says "Intel® oneAPI Math Kernel Library is thread-safe, which means that all Intel® oneAPI Math Kernel Library functions (except the LAPACK deprecated routine?lacon) work correctly during simultaneous execution by multiple threads." Does this is not work with Pardiso? "

Ps: for the convenience of replying to you, could you provide your work email address? @VidyalathaB_Intel

Thanks,

Ewigstern

[VidyalathaB_Intel]

Hi Ewigstern,

Since this is a duplicate thread of https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/The-wrong-result-when-using-Pardiso-solver-to-solve-linear/m-p/1414332, we will no longer monitor this thread. We will continue addressing this issue in the other thread. 

Regards,

Vidya.