[petsc-users] MUMPS solver crash with petsc-3.2

Hong Zhang hzhang at mcs.anl.gov
Tue Dec 6 23:16:27 CST 2011


Get updated mumps interface in petsc-dev. Run your code with
'-mat_mumps_icntl_4 4' which will display mumps info

Use a debugger to see where the program terminates.

Hong

On Tue, Dec 6, 2011 at 10:54 PM, Gong Ding <gdiso at ustc.edu> wrote:
> I can only analysis the core file at present.
> It reported
> Program terminated with signal 11, Segmentation fault.
> #0  0x000000010217319c in dmumps_462 ()
>
> Run program with -info, the output is
>
> Entering DMUMPS driver with JOB, N, NZ =   1          49            217
>
>  DMUMPS 4.10.0
> L U Solver for unsymmetric matrices
> Type of parallelism: Working host
>
>  ****** ANALYSIS STEP ********
>
>  Resetting candidate strategy to 0 because NSLAVES=1
>
>  ... Structural symmetry (in percent)=  100
>  Density: NBdense, Average, Median   =    0    3    3
>  ... No column permutation
>  Ordering based on AMF
>
> Program terminated with signal SIGSEGV, Segmentation fault.
>
> The example in src/ksp/ksp/examples/tests/ex3 run ok -- it solves a small
> linear problem.
> Also, if I reduce the matrix size to ~20x20, no crash happen.
>
> I'll try to provide more information.
>
>
>
>
>
> On Tue, Dec 6, 2011 at 12:52 AM, Gong Ding <gdiso at ustc.edu> wrote:
>>
>> Bad news.
>> MUMPS still crash on AIX/PPC.
>> I tried petsc-3.1, 3.2 and dev. All of them crash with ERROR: 0031-250
>> task 0: Segmentation fault.
>> I guess the interface of petsc-mumps has some problem, i.e. alignment.
>
>
> Please get a stack trace with the debugger.
>
>    Matt
>
>>>
>>> Yes, petsc-dev works. Please merge the update to petsc-3.2.
>>
>>
>> Parmetis 4.0 was released after PETSc 3.2 so it will not be updated to
>> work with it. In general, software projects don't update a previous release
>> on a dependent package update. Please use petsc-dev or wait until PETSc 3.3.
>
>
>
>
> --
> What most experimenters take for granted before they begin their experiments
> is infinitely more interesting than any results to which their experiments
> lead.
> -- Norbert Wiener


More information about the petsc-users mailing list