PARDISO crashes with access violation in case of OpenMP parallel METIS reordering

[BZoltan]

Currently we are using Intel Fortran Compiler IFX 2025.2.1, however the crash happens with older and the newer (2025.3) versions as well.
We would like to solve sparse matrix equations with pardiso. We reproduced the crash with several Intel and AMD processors and on win10 and win11 as well.

The crash happens only if Iparm(2)=3 (parallel METIS reordering).
With Iparm(2)=2 there is no problem (METIS reordering).

Problem statement:

If we try to solve a sparse matrix with the above mentioned Iparm(2)=3 reordering type, pardiso sometimes crashes with the attached access violation.
We figured out the crash only happens if the processor usage is high from another process as well next to the actual program.

The parallel running high CPU usage program can be arbitrary. The crash can be reproduced with running several programs which call pardiso separately,
but the program running in parallel does not necessarily have to be the pardiso, it can be a random program.

This is how we reproduced the problem constantly:

In a batch file we run 20 separate exe-s with one equation solver separately one-by-one.
The 20 different exe run the pardiso solver separately. Some of the programs return with correct solution,
some of them returns with a valid error code that there is insufficient memory (which is a totally valid returning error code).
But some of the exe-s crash and show acces violation (see attached pic).

With the mentioned conditions (with arbitrary parallel running high CPU usage program) only one pardiso program could crash with the same access violation as well.

We attached the main program but the sparse matrix input files are too large to be attached. The crash can be reproduced by relatively large matrices but not by only a specific matrix.
If there is a way to attach the complete inputs (ca. 800 MB) please let us know. How can we avoid the random crash with Iparm(2)=3 reordering mode?

Thanks

[Fengrui]

Hi,

This issue has been fixed in the latest 2026.0 release. Please verify.

Thanks,

Fengrui

[BZoltan]

Dear Fengrui, 
I've checked the issue, it works now properly. You can close the thread, thank you!

[Fengrui]

Great! Thank you again for reporting this issue.