Page 1 of 1

Floating Point exception arises in different CS version

Posted: Wed Aug 31, 2016 6:22 pm
by Oscar
Hello,

I have a case which runs perfectly fine on my computer in CS v_4.0 but when I attempt to run the same code on a cluster which has CS v_4.0.5 installed I get the floating point error described in listing:

Code: Select all

SIGFPE signal (floating point exception) intercepted!

Call stack:
   1: 0x40467a     <cs_user_boundary_conditions_+0x5c6> (cs_solver)
   2: 0x7f21b1017656 <tridim_+0x34b6>                 (libsaturne.so.0)
   3: 0x7f21b0ef61c1 <caltri_+0x2b41>                 (libsaturne.so.0)
   4: 0x7f21b0ed8025 <cs_run+0x3e5>                   (libsaturne.so.0)
   5: 0x7f21b0ed8165 <main+0x115>                     (libsaturne.so.0)
   6: 0x7f21af1bbd5d <__libc_start_main+0xfd>         (libc.so.6)
   7: 0x403789     <>                               (cs_solver)
End of stack

so something is going wrong when specifying the boundary conditions that is particular to this code_saturne version?


I have attached the listing and all the source files in the zipped directory.

Re: Floating Point exception arises in different CS version

Posted: Wed Aug 31, 2016 8:32 pm
by Yvan Fournier
Hello,

In your cs_user_boundary_conditions.f90 file, you commented the definition of Area, so you probably use it undefined, which might explain the error.

Also, you computation of Ubulk_verif is valid only when running on a single process. In parallel, you would need to add a call to "parsom" (see examples under Doxygen) to have a valid value on all ranks.

Regards,

Yvan

Re: Floating Point exception arises in different CS version

Posted: Thu Sep 01, 2016 12:16 pm
by Oscar
Hi Yvan,

Thanks that solved the issue and now it is running. As for the parsom, I still am able to run the case in parallel the way it already is. Is that unexpected?

kind regards,

Oscar