Case which is running with valgrind memcheck but not without

[laurent_makke]

Hi,

I am trying to run a CS 3.0 case with atmospheric and radiative stuff.
Without a valgrind memcheck, I have an error :

glibc detected.. please find the backtrace in enclosed file.

When I run it with valgrind --tool-memcheck, it works.

I have a lot of message of "Conditional Jump depends on unitialized variables" but the calcul is perfectly running. I attached also the valgrind.txt file.

Do you have any idea that could help to debug or to understand this?

Best regards,
Laurent.

[Yvan Fournier]

Hello,

Valgrind reports many errors, of which quite a few might be false positives (and the code does crash, but only at the end). Do you have similar issues with a simpler case ?
If not, the issue might be due to your XML setup. Otherwise, could you try a build with gcc, which might lead to less error messages, thus making it easier to find the useful ones.

Also, did you try running without any user subroutines (with annd without Valgrind)

Regards,

Yvan

[laurent_makke]

Hi Yvan,

"Do you have similar issues with a simpler case ?"
with the same version and the same cluster? --> I don't know
I buildt CS_3.0.7 on the CALMIP cluster in Toulouse.
I think it is the same architecture than the CCRT with an Intel Fortran Compiler.
I'll check with gcc.

"If not, the issue might be due to your XML setup"
Does that mean the xml file in DATA directory?
I made some changes in cs_gui_radiative_transfer.c and .h. Is there a link with my problem and the xml setup?

Also, did you try running without any user subroutines (with and without Valgrind)
No I didn't. This case ran with the CS_3.0.5 on year ago on a Calibre 7. I didn't have this problem.

Anyway, I will proceed step by step with your pieces of advice.

Thank you Yvan.

Laurent