[FLASH-USERS] LaserSlab Segmentation Fault

[Klaus Weide]

On Tue, 23 Jun 2020, Descamps, Adrien wrote:

> I am running into the same segmentation fault when trying to run the LaserSlab example.

Adrien,

It doesn't seem to be exactly the same fault - reporting the exact 
messages on standard output and/or standard error would be useful.

> Please find attached the las slab.log file attached

The log file had output from several runs, and it is not quite clear what 
happened in those runs.

Some runs end with a line logging

    [IO_writeCheckpoint] open: type=checkpoint name=lasslab_hdf5_chk_0000

others with

    step: n=1 ...

> Does the problem happen even when you run on one proc?
> 
> It worked fine on one processor
> 

It doesn't look like that in the log file.

> Does the command
>   ldd flash4
> show anything interesting?
> 
> linux-vdso.so.1 =>  (0x00007fff4db51000)
> libhdf5.so.103 => not found

There is something fishy here.

Maybe the version of the HDF5 library that flash sees at run time 
disagrees with that that was seen at compile time.


> Did you configure and build AND INSTALL the HYPRE library yourself?
> Which options were used to configure the HYPRE library?
> Was HYPRE compiled with the same compiler version(s) (for C and Fortran)
> that you are using for compiling FLASH?
> 
> I configured and install hypre myself.


I saw no indication that your runs got to the point were HYPRE routines 
would be called.

To simplify things during debugging, I recommend

 * building FLASH without OpenMP support
   (remove -fopenmp from compiler flags.)
 * building a version of HYPRE without --with-openmp or LDFLAGS=-fopenmp
 * setting environment variable OMP_NUM_THREADS=1 .
 
Klaus