[petsc-users] Segmentation Fault in MatAssembly

[Barry Smith]

  Run with valgrind http://www.mcs.anl.gov/petsc/documentation/faq.html#valgrind


> On Feb 13, 2016, at 4:43 PM, Bikash Kanungo <bikash at umich.edu> wrote:
> 
> Hi,
> 
> I have a small (~ 2500x2500) dense matrix, say A, parallelized over 480 procs and stored as MPIAIJ instead of MPIDENSE. This is because I have to perform MatMatMult of A with an MPIDENSE matrix, say B. Since MatMatMult is not supported if both A and B are dense, I'm storing A as MPIAIJ.  The entries in A are inserted only by processor 0 through MatSetValues. Now while assembling A, I sometimes get segmentation fault. This error is random,it shows up in one run and doesn't in another. I understand that it doesn't make much sense to distribute such a small matrix over 480 procs, but what can be a possible reason for such random segmentation faults? 
> 
> Regards,
> Bikash
> 
> -- 
> Bikash S. Kanungo
> PhD Student
> Computational Materials Physics Group
> Mechanical Engineering 
> University of Michigan
>