topic Re: PARDISO reproducibility in Intel® oneAPI Math Kernel Library

PARDISO reproducibility

nedialk — Fri, 25 Feb 2022 23:46:16 GMT

For numerical reproducibility we need to set iparm[33] > 0 before the symbolic factorization, and this iparm cannot be changed after.

Is there a way to request reproducibility for a particular phase of the solving. E.g. how to request that only the numerical factorization is reproducible? That is I want to do e.g.

iparm[33] = 4;

numerical factorization

iparm[33] = 0

Thank you.

Re:PARDISO reproducibility

ShanmukhS_Intel — Mon, 28 Feb 2022 11:59:27 GMT

Hi,

>>For numerical reproducibility we need to set iparm[33] > 0 before the symbolic factorization, and this iparm cannot be changed after.

Intel® oneAPI Math Kernel Library PARDISO reads the value of iparm[33] during the analysis phase (phase 1), so it cannot be changed later.

Optimal number of OpenMP threads for conditional numerical reproducibility (CNR) mode.

CNR mode for Intel® oneAPI Math Kernel Library PARDISO is enabled only if it is enabled for

Intel® oneAPI Math Kernel Library using the functionality described inSupport Functions for CNRand the in-core version is used.

Intel® oneAPI Math Kernel Library PARDISO determines the optimal number of OpenMP threads automatically, and produces numerically reproducible results regardless of the number of threads.

CNR mode is enabled for Intel® oneAPI Math Kernel Library

PARDISO if in-core version is used and the optimal number of OpenMP threads for

Intel® oneAPI Math Kernel Library PARDISO to rely on is defined by the value of

Iparm [33]. You can use iparm[33]to enable CNR mode independent from other

Intel® oneAPI Math Kernel Library domains. To get the best performance, set

iparm[33] to the actual number of hardware threads dedicated for

Intel® oneAPI Math Kernel Library PARDISO. Setting iparm[33] to fewer OpenMP threads than the maximum number of them in use reduces the scalability of the problem being solved. Setting iparm[33]

to more threads than are available can reduce the performance of Intel® oneAPI Math Kernel Library PARDISO.

Could you elaborate a bit regarding why iparam[33] Is required to change in your case.

Best Regards,

Shanmukh.SS

Re: PARDISO reproducibility

nedialk — Wed, 02 Mar 2022 15:00:54 GMT

>> Could you elaborate a bit regarding why iparam[33] Is required to change in your case.

I wanted to figure out which part of the solution process causes the different results I am getting. I was hoping to turn on and off

this parameter for the various parts of the solving.

Thanks,

Ned

Re: PARDISO reproducibility

ShanmukhS_Intel — Wed, 16 Mar 2022 10:41:02 GMT

Hi,

Kindly refer the below link for the usage of iparm[33] and various other iparm parameters.

https://www.intel.com/content/www/us/en/develop/documentation/onemkl-developer-reference-c/top/sparse-solver-routines/onemkl-pardiso-parallel-direct-sparse-solver-iface/pardiso-iparm-parameter.html

You cannot change the iparm[33] parameter later , because Intel® oneAPI Math Kernel Library PARDISO reads the value of iparm[33] during the analysis phase (phase 1)

Intel® oneAPI Math Kernel Library PARDISO uses C random number generator facilities during the analysis phase (phase 1)

You must take these precautions to get numerically reproducible results:

Do not alter the states of the random number generators.

Do not run multiple instances of Intel® oneAPI Math Kernel Library PARDISO in parallel in the analysis phase (phase 1).

>> I wanted to figure out which part of the solution process causes the different results I am getting.

In addition, Could you please share us a sample reproducer so that we could try it at our end?

Best Regards,

Shanmukh.SS

Re:PARDISO reproducibility

ShanmukhS_Intel — Fri, 11 Mar 2022 10:40:19 GMT

Hi,

Reminder:

Has the information provided helped? Is your issue resolved?

Kindly let us know if we could close this thread at our end.

Best Regards,

Shanmukh.SS

Re:PARDISO reproducibility

ShanmukhS_Intel — Fri, 18 Mar 2022 04:17:30 GMT

Hi,

We haven't heard back from you. We assume that your issue is resolved. If you need any additional information, please post a new question as this thread will no longer be monitored by Intel.

Best Regards,

Shanmukh.SS

Re: Re:PARDISO reproducibility

nedialk — Sat, 19 Mar 2022 03:32:32 GMT

Thank you.