topic iparm[35] = 1 or 2 question in Intel® oneAPI Math Kernel Library

iparm[35] = 1 or 2 question

YONGHEE_L_ — Mon, 25 Jun 2018 02:35:04 GMT

Dear Intel

I'm a user of MKL 2018 up1 parallel studio.

Usually I handle the large matrix(>50Mx50M) using schur complement.

'iparm[35] = 1' is the my general setting and it shows the schur complement matrix.

However, when I changed the setting 'iparm[35]' from 1 to 2, they shows the error message during factorization (phase=2)

Error code is -4. (not SPD)

Already, I checked iparam[29] and matrix->a[iparam[29]], but there are no zero pivot or wrong conditions for SPD.

I think, if the matrix is factorized with 'iparm[35] = 1', then 'iparm[35] = 2' should be also OK.

I ask your advice.

Thank you very much in advance!!!

Regards,
Yong-hee

Hi YongLee,

Ying_H_Intel — Mon, 25 Jun 2018 03:23:53 GMT

Hi YongLee,

Check the Bug list since MKL 2018 u1 at https://software.intel.com/en-us/articles/intel-math-kernel-library-intel-mkl-2018-bug-fixes-list

I haven't seen the exact report about the problem. But what kind of your machine memory and cpu? How about link with ILP64 or when the input size is small , does it work?

or anyway, could you please try MKL 2018 u3 if possible

Best Regards
Ying

Hi Ying,

YONGHEE_L_ — Mon, 25 Jun 2018 12:53:53 GMT

Hi Ying,
Thank you for yout quick reply.
As you mentioned, I searched the fix lists, but I can't find any related contents.
Now I'm trying to conduct my program at MKL 2018 u3.
I'll tell you the result ASAP.

Followings are my system info and my program settings
- Xeon E5-2630 v2 and 500GB memory
- Unix redhat, C++
- mtype : 2 (because this matrix is generated from passive circuit design.)
- iparm : [0]=1, [1]=3, [9]=8, [35]=1 (or 2), and [others]=0.
- Schur complement size (MxN) : 3k x 3k
- included library : "cdft_core.a", "intel_ilp64.a", intel_thread.a", "core.a", "blacs_intelmpi_ilp64.a"

Do you have any idea what might be wrong from this condition?

Best regards
Yong-hee