code_saturne User's Forum

Hi everybody,

                   I thought I had installed correctly CodeSaturne on Ubuntu 10.04 (32 bit), but when I try to run the first guided example (Simple_Junction) I get the error:

SIGSEGV signal (forbidden memory area access) intercepted!

Call stack:
   1: 0x2f5400     ?                                (?)
   2: 0x40f3ee     <+0x7b3ee>                       (libsaturne.so.0)
   3: 0x412809     <+0x7e809>                       (libsaturne.so.0)
   4: 0x415e6d     <uiclim_+0x24cd>                 (libsaturne.so.0)
   5: 0x58c991     <tridim_+0x3331>                 (libsaturne.so.0)
   6: 0x3cedc2     <caltri_+0x3df2>                 (libsaturne.so.0)
   7: 0x3b14e6     <cs_run+0x846>                   (libsaturne.so.0)
   8: 0x3b1845     <main+0x255>                     (libsaturne.so.0)
   9: 0x60d4bd6    <__libc_start_main+0xe6>         (libc.so.6)
  10: 0x80485f1    <>                               (cs_solver)
End of stack

The all listing is attached.

Has anyone an idea on how could I fix it ?

Thank you very much for your attention

                                         JP Voyat

Hello,
I did not reproduce the error running Code_Saturne 2.0.2 on Ubuntu 11.04 64 bit, so you may try upgrading to 2.0.2, though uiclim did not change between 2.0.1 and 2.0.2. If you can compile and install Code_Saturne in debug mode, running "addr2line" on your stack trace above would give us the exact line where the code crashed, which may help (running it under Valgrind would also give info).
For your info, usclim is where boundary conditions defined by the GUI are applied.
In any case, running the code under Valgrind (setting VALGRIND=valgrind in the runcase, if the Valgrind tool is installed on your Ubuntu machine) may provide information even in your current build, so the output from that would be interesting.
Othewise, I can try to setup Ubuntu 10.4 in 32 bit mode in a virtual machine and check if I reproduce the error, but I won't have time for this before a few days, so if you can have more info (i.e. a debug build on you side, it would be helpful.
Also, how did you install the code ? Automatic installer, manual install, ...? Did you use tutorial data from 2.0.1 or from an earlier beta or rc version ?
Best regards,
  Yvan

Hi, I installed valgrind, modified the runcase, but the output seems to be just the same ...
I installed code-saturne manually. I used the installing tutorial for the 2.02 version ... I didn't notice that earlier.
I downloaded the last version of c-s, I'm going to try to install it as soon as possible.
Thank you very much for your attention
                       JP Voyat

I tried to use the installer. No chance ...
I got with the attached log.
I tried to modify the setup file to use an installed version of the hdf5, but the installation keeps starting from the new hdf5 installation ...
Ha e a nice day
                     JP

Hello,
Could you post your installer setup file ? You should not need to install hdf5 manually, as it is available as a Debian/Ubuntu package (but do not forget to install the -dev part of the package, for headers).
Best regards,
  Yvan

Hi.

I understood that I should put "...     yes       no " in order not to install the optional packages that I already installed through the package manager.

I tried the installation again and after some more changes I could install all but the ncs ... it says I don't have a good fvm version. This is strange, since the installer just installed without problems the 0.15.3 version of fvm.

I attach the relative logs to help you to help me !

Thank you very much.

                        JP

I could not post all the files in one reply ... (I got an error)

I could not post all the files in one reply ... (I got an error)

I could not post all the files in one reply ... (I got an error)

Hello,

In your setup file, you used /usr/bin instead of /usr/lib for HDF5, which you should correct, but this is not the cause of the problem.

According to the config.log from the NCS package, the problem seems related to different MPI options between NCS and FVM. Did you change the value of MPICC between tests ? in this case, a fresh reinstall with the final options seems in order.

I would also recommend either defining mpiCompC = mpicc (line 46) and  mpi = None no no (line 156), or

mpiCompC = None, and mpi = /usr/lib/openmpi no no , but avoid defining both mpiCompC and MPI paths (especially in case you have multiple MPI libraries on your Ubuntu machine).

Code_Saturne should work fine with either MPICH2 or OpenMPI, but if you have both installed, you must be careful to use only one of the 2, and not a mix (i.e.) ensure that the mpiexec command matches the mpicc command, for example.

Best regards,

  Yvan