topic Hello Younghoe C., in Intel® oneAPI Math Kernel Library

Hanging on PARDISO Factorization (update MKL 11.1 update4 -> MKL 11.3)

Younghoe_C_ — Mon, 16 Nov 2015 02:50:26 GMT

Hello,

I'm using Sparse Matrix Solver in MKL.

I proceeded with the Version Update in MKL 11.1 update4 with MKL 11.3 this time.

The problem that didn't generate in an existing happens and I inquire.

The phenomenon that I stop in the Factorization process is happening after the update progress.

plz, help.

Hello Younghoe C.,

Ying_H_Intel — Mon, 16 Nov 2015 05:59:28 GMT

Hello Younghoe C.,

We run into one issue regarding OpenMP thread nested call Pardiso threading . if using MKL sequential library or mkl_set_num_threads(1) can workaround the problem. But not sure if you are in same situation.

Could you please provide us a small test case, to tell the details for example, window or linux, 32bit or 64bit, mkl:parallel or mkl:sequential etc.

Best Regards,

Ying

Hello Quote:Ying H (Intel)

Younghoe_C_ — Tue, 17 Nov 2015 02:00:00 GMT

Hello Ying H

Thank you for your answer.

I performed with the way in which you say, a normality was performed. ( mkl_set_num_threads(1) )

When should does the modification about the relevant problem and Update accomplish ?

Ying H (Intel) wrote:

Hello Younghoe C.,

We run into one issue regarding OpenMP thread nested call Pardiso threading . if using MKL sequential library or mkl_set_num_threads(1) can workaround the problem. But not sure if you are in same situation.

Could you please provide us a small test case, to tell the details for example, window or linux, 32bit or 64bit, mkl:parallel or mkl:sequential etc.

Best Regards,

Ying

Hello Younghoe C.,

Ying_H_Intel — Wed, 18 Nov 2015 01:51:24 GMT

Hello Younghoe C.,

We are still investigating the issue i mentioned. Generally, the fix should be in next release. ( and The issue seems related the OpenMP library, so we need talk with Intel Compiler team.).

Do you have other thread in your application? Regarding the mkl_set_num_threads(1) , it will tell MKL using 1 thread, in some case, it will reduce the performance, but if you have mix or nested multithread environment, it was recommended setting. it will avoid the threading conflict or overload.

If possible, could you please provide us a small test case, to tell the details for example, window or linux, 32bit or 64bit, mkl:parallel or mkl:sequential etc.

Best Regards,
Ying