[FLASH-USERS] LaserSlab Segmentation Fault

[Klaus Weide]

On Thu, 2 Jul 2020, Descamps, Adrien wrote:

>   * I included memory_stat_freq = 1 in the parfile. Please find attached 
> the lasslab.log attached. To me, it seems that this is not a memory 
> problem as I am using 8 GB memory for each processors.

I don't see signs of a problem.

>   *   How could I know whether the segmentation fault is coming from the batch system? Please find the message error below
> 
>       60 3.0348E-12 3.0448E-13  ( 3.281E-04,  5.953E-03,   0.00    ) |  1.622E-12 9.868E+84 1.014E+88 0.4000000
>  *** Wrote checkpoint file to lasslab_hdf5_chk_0003 ****
> srun: error: sh02-01n31: task 1: Segmentation fault
> srun: Terminating job step 3490332.0
> slurmstepd: error: *** STEP 3490332.0 ON sh02-01n31 CANCELLED AT 2020-07-02T11:04:11 ***
> srun: error: sh02-01n31: task 0: Segmentation fault

I see no sign that the batch system is the culprit.

>   *   It seems to run fine with I/O. I set the maximum number of steps to 100 and it runs fine for all of it.

I assume you mean "fine without I/O" (equivalent to "fine with +noio").

It seems more and more that you are running into a problem that is 
specific to your system or library setup.

You might try running a simpler problem, like Sod or Sedov, with parallel 
IO, to see whether you have similar problems.
This would eliminate any possible effect from the HYPRE library.

You might try setup with serial IO - use +serialIO, do not use +parallelIO 
or +hdf5typeio, and --without-unit=.../LaserIO ; check the output of setup
(or later the log file) to make sure the directory IO/IOMain/hdf5/serial/PM
is included.

Klaus