Inconsistent COSMO/Sigma-profile results when running GAMESS with OpenMP enabled.

[RhayWenn]

Greetings!

I need to use GAMESS to perform COSMO calculations and obtain sigma-profiles.
I have compiled and installed GAMESS (15 JUL 2024, R2 Patch 1) on both WSL (Ubuntu 22.04) and a VirtualBox VM running Ubuntu 22.04.

I tested both the "serial" and the "sockets" installations, using OpenMP support (GMS_OPENMP=true in install.info),
with the OpenBLAS library compiled properly to support OMP (make INTERFACE64=1 -j4 USE_OPENMP=1).

All compilation and linking steps completed successfully without errors.

However, the main issue is that even very simple molecules (e.g., methane) fail to converge during geometry optimization,
even when the input already contains a pre-relaxed geometry. Additionally, the sigma-profiles are completely inconsistent —
the total energy is not minimized and sigma values appear to be flipped or incorrect.

I noticed that both the geometry optimization and sigma-profiles work correctly when I recompile GAMESS without OpenMP support.
I came to this conclusion by comparing the outputs and sigma-profiles with those generated by other versions of GAMESS.

What might be causing this issue? Can you help me?