[BUG] Memory leak in cluster pardiso solver when n=1

[tlcfem]

It seems like when n=1, although makes little sense in practice (but indeed is a valid matrix size), there is a memory leak issue in the cluster pardiso solver.

The attached reproducer is an abridged version of the example file `cl_solver_unsym_c.c`.

Compile and run the executable, the output starts with the following.

*** Error in PARDISO ( insufficient_memory) error_num= 1

*** Error in PARDISO memory allocation: MATCHING_REORDERING_DATA, allocation of 1 bytes failed

total memory wanted here: 142 kbyte



=== PARDISO: solving a real nonsymmetric system ===

Matrix checker is turned ON 

With address sanitizer on, it looks like `mm_internal_malloc` is leaking memory.

==3948==ERROR: LeakSanitizer: detected memory leaks



Direct leak of 128 byte(s) in 1 object(s) allocated from:

#0 0x7f54bcefd9c7 in malloc ../../../../src/libsanitizer/asan/asan_malloc_linux.cpp:69

#1 0x5641c65c4c1c in mm_internal_malloc (/home/runner/work/ezp/ezp/build/pardiso.memleak+0xd8c1c) (BuildId: 147c1dc7a67f9a1d99f4d13e95bd2136fa286a0f)



SUMMARY: AddressSanitizer: 128 byte(s) leaked in 1 allocation(s).

It is reproduced in a standard environment, see the following GitHub workflow.

https://github.com/TLCFEM/ezp/actions/runs/15502135308

Seems to be a memory issue?

[Chao_Y_Intel]

Hi,

Intel MKL use internal buffers to facilitate better performance.  These internal buffers may be reported as leaked memory.

Can you check this article, and see if some APIs/environment can help to remove it:
https://www.intel.com/content/www/us/en/docs/onemkl/developer-guide-linux/2025-1/avoiding-memory-leaks-in-onemkl.html

thanks,
Chao

[tlcfem]

Calling mkl_disable_fast_mm does not help, see the following workflow.

https://github.com/TLCFEM/ezp/actions/runs/15530089687

Noting the information,

*** Error in PARDISO ( insufficient_memory) error_num= 1

*** Error in PARDISO memory allocation: MATCHING_REORDERING_DATA, allocation of 1 bytes failed

total memory wanted here: 142 kbyte

It seems like some faulty parameters are passed to memory allocation, because when n>1 there is no such an error.

[Ruqiu_C_Intel]

Hi,

Thank you for submitting the issue.

We have reproduced the issue with the reproducer. We will update this post once there are further improvements.

Regards,

Ruqiu

[Fengrui]

This issue has been fixed in the oneMKL 2025.3 release. Please give it a try!