Seg fault when running the PRACE test case

[Stuberl]

Hello,

I am trying to run this benchmark (test case A) : http://www.prace-ri.eu/ueabs/#Code_Saturne

I've installed Code_Saturne 3.0.1 using these instructions and I've run the example cases successfully. But on the benchmark test case, both ./SaturnGUI dp_fcp.xml and ./runcase end with this error :

SIGSEGV signal (forbidden memory area access) intercepted!

Call stack:
1: 0x7f99d1afd860 <+0x35860> (libc.so.6)
2: 0x7f99d1b54eb6 <strlen+0x26> (libc.so.6)
3: 0x7f99d2bb599e <uicpsc_+0xcc8> (libsaturne.so.0)
4: 0x7f99d2a209d1 <cs_coal_varpos_+0xd75> (libsaturne.so.0)
5: 0x7f99d25db9e9 <ppvarp_+0x68> (libsaturne.so.0)
6: 0x7f99d251be76 <varpos_+0x7eb> (libsaturne.so.0)
7: 0x7f99d23ab219 <iniusi_+0x2b0> (libsaturne.so.0)
8: 0x7f99d23aab1a <initi1_+0xb7> (libsaturne.so.0)
9: 0x7f99d220a5fc <cs_run+0x17c> (libsaturne.so.0)
10: 0x7f99d220b3b1 <main+0x14a> (libsaturne.so.0)
11: 0x7f99d1ae83f1 <__libc_start_main+0xf1> (libc.so.6)
12: 0x55a47ddb43da <_start+0x2a> (cs_solver)
End of stack

With gdb, I found this line to be the problem :

gui/cs_gui_specific_physics.c:2656

Code: Select all

  /* IFVP2M */
  label = _scalar_name_label("solid_fuels", "Var_F1F2");
  BFT_MALLOC(vars->label[*ifvp2m -1], strlen(label)+1, char);   // <- this call to strlen crashes
  strcpy(vars->label[*ifvp2m -1], label);
  BFT_FREE(label);

Any idea ? Thanks in advance.

log_files.tar.gz

(5.87 KiB) Downloaded 189 times

[Yvan Fournier]

Hello,

In any case, I recommend using the latest bug-fix release for this branch (3.0.9) or the previous one (3.0.8 in case of issues with environment modules).

The setup seems not to use the GUI, and in any case, I doubt it is a coal combustion case, so you should not try to use the example xml file in DATA, but just recreate an xml file with the GUI to handle the exisiting mesh_input path and select the LES turbulence model (ignoring the rest), or not use the GUI at all but use the cs_user_scripts.py file in DATA (copied from DATE/REFERENCE) to set the mesh path.

Sorry, the instructions in the benchmark do not seem to be too clear on this (I believe they were written by advanced users who do not work much with the GUI).

Also, the 3.0 series will only be maintained for a few more months: when we release 5.0 next April, version 4.0 will become "old stable", and 3.0 to "retired". I do not know if there are up-to-date setups for the benchmark, but I'll try to check.

Regards,

Yvan

[Stuberl]

Hello,

Thanks for the answer. Indeed with this setting it works fine.

But is it the way the benchmark should be run ? Most of the CS benchmarks found on the internet mention they used a LES but I didn't find detailed information about the other simulation parameters. I would like to realistically compare the performance I obtain to the other benchmarks.

Thanks,

Louis Stuber