[petsc-dev] Reduction in KSPGetVecs via DMCreateGlobalVector

[Matthew Knepley]

On Wed, Mar 26, 2014 at 7:00 AM, Jed Brown <jed at jedbrown.org> wrote:

> I don't think it's reasonable to allow this:
>
> #0  0x00007ffff5af9cc0 in PMPI_Allreduce () from
> /opt/mpich/lib/libmpich.so.12
> #1  0x00007ffff61b7436 in PetscSplitOwnership () from
> /home/jed/petsc/mpich-opt/lib/libpetscsys.so.3.04
> #2  0x00007ffff65a44b6 in PetscLayoutSetUp () from
> /home/jed/petsc/mpich-opt/lib/libpetscvec.so.3.04
> #3  0x00007ffff65ff988 in VecCreate_MPI_Private () from
> /home/jed/petsc/mpich-opt/lib/libpetscvec.so.3.04
> #4  0x00007ffff6600082 in VecCreate_MPI () from
> /home/jed/petsc/mpich-opt/lib/libpetscvec.so.3.04
> #5  0x00007ffff6610867 in VecSetType () from
> /home/jed/petsc/mpich-opt/lib/libpetscvec.so.3.04
> #6  0x00007ffff660010f in VecCreate_Standard () from
> /home/jed/petsc/mpich-opt/lib/libpetscvec.so.3.04
> #7  0x00007ffff6610867 in VecSetType () from
> /home/jed/petsc/mpich-opt/lib/libpetscvec.so.3.04
> #8  0x00007ffff6fbe98b in DMCreateGlobalVector_DA () from
> /home/jed/petsc/mpich-opt/lib/libpetscdm.so.3.04
> #9  0x00007ffff70d979e in DMCreateGlobalVector () from
> /home/jed/petsc/mpich-opt/lib/libpetscdm.so.3.04
> #10 0x00007ffff6fbe7d0 in VecDuplicate_MPI_DA () from
> /home/jed/petsc/mpich-opt/lib/libpetscdm.so.3.04
> #11 0x00007ffff660ca11 in VecDuplicate () from
> /home/jed/petsc/mpich-opt/lib/libpetscvec.so.3.04
> #12 0x00007ffff660cf37 in VecDuplicateVecs_Default () from
> /home/jed/petsc/mpich-opt/lib/libpetscvec.so.3.04
> #13 0x00007ffff660cc2e in VecDuplicateVecs () from
> /home/jed/petsc/mpich-opt/lib/libpetscvec.so.3.04
> #14 0x00007ffff7569da1 in KSPGetVecs () from
> /home/jed/petsc/mpich-opt/lib/libpetscksp.so.3.04
> #15 0x00007ffff753b403 in KSPGMRESGetNewVectors () from
> /home/jed/petsc/mpich-opt/lib/libpetscksp.so.3.04
> #16 0x00007ffff753b748 in KSPGMRESCycle () from
> /home/jed/petsc/mpich-opt/lib/libpetscksp.so.3.04
> #17 0x00007ffff753c0cc in KSPSolve_GMRES () from
> /home/jed/petsc/mpich-opt/lib/libpetscksp.so.3.04
> #18 0x00007ffff755e6b0 in KSPSolve () from
> /home/jed/petsc/mpich-opt/lib/libpetscksp.so.3.04
> #19 0x00007ffff7b9cdb0 in SNESSolve_NEWTONLS () from
> /home/jed/petsc/mpich-opt/lib/libpetscsnes.so.3.04
> #20 0x00007ffff7b73c85 in SNESSolve () from
> /home/jed/petsc/mpich-opt/lib/libpetscsnes.so.3.04
> #21 0x000000000040183a in main ()
>
> Found by breaking on PMPI_Allreduce inside
>
> ~/petsc/src/snes/examples/tutorials$ mpiexec -n 1 gdb --args ./ex5
> -da_refine 3 -ksp_max_it 5 : -n 1 ./ex5 -da_refine 3 -ksp_max_it 5
>

I think we also do this in some Vec output routines. We have to go through
and find all the
places that we can provide both sizes to Layout/IS/Vec.

  Matt

-- 
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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20140326/e2236ac7/attachment.html>