Intel® oneAPI Math Kernel Library
Ask questions and share information with other developers who use Intel® Math Kernel Library.
7296 讨论

PARDISO gets stuck in num. factorization for a general complex matrix

rbunger
新分销商 II
1,046 次查看

Hi,

 

PARDISO gets stuck in the numerical factorization step, for a general complex matrix. The matrix is too big to upload it (43 MB 7z archive).

 

OneAPI versions: 2025.3.0 and 2025.3.1

 

Thanks,

 

Rainer

0 项奖励
1 解答
rbunger
新分销商 II
958 次查看

Update: I had to upload the matrix elsewhere. Intel has confirmed a bug and the fix will be in the 2026.0 release. Meanwhile two fixes are proposed:


1) use iparm[23]=1,

2) if there are even number of NUMA nodes, you could use spread-type binding by "OMP_PLACES=cores OMP_PROC_BIND=spread".


Many thanks to the Intel team & kind regards,


Rainer

 

在原帖中查看解决方案

0 项奖励
4 回复数
rbunger
新分销商 II
959 次查看

Update: I had to upload the matrix elsewhere. Intel has confirmed a bug and the fix will be in the 2026.0 release. Meanwhile two fixes are proposed:


1) use iparm[23]=1,

2) if there are even number of NUMA nodes, you could use spread-type binding by "OMP_PLACES=cores OMP_PROC_BIND=spread".


Many thanks to the Intel team & kind regards,


Rainer

 

0 项奖励
rbunger
新分销商 II
950 次查看

Option 2) has quite a large impact on the performance, in my case

0 项奖励
Fengrui
主持人
369 次查看

Thanks again for reporting this issue. The fix is available in the latest 2026.0 release. Please check.

 

Thanks,

Fengrui

rbunger
新分销商 II
346 次查看

Many thanks for the reminder. I have already performed tests and up to now everything is perfectly stable. Many thanks for your efforts!

 

Rainer

0 项奖励
回复