Re: [petsc-dev] DMPlexCreateMedFromFile broken

Thu, 09 Nov 2017 08:23:30 -0800


7. 11. 2017 v 19:51, Matthew Knepley <knep...@gmail.com>:

On Tue, Nov 7, 2017 at 11:44 AM, Vaclav Hapla <vaclav.ha...@erdw.ethz.ch> wrote:
Nico Schlömer just introduced MED support into his meshio tool, based on my issue request. When I convert blockcylinder-50.exo directly to the MED format with meshio, the result can be loaded into GMSH and looks reasonable. But DMPlexCreateMedFromFile denies to open it with the following error:

_MEDmeshAdvancedRd236.c [285] : Erreur de valeur invalide du filtre 
_MEDmeshAdvancedRd236.c [285] : 
_MEDmeshAdvancedRd236.c [286] : geotype = 0
_MEDmeshAdvancedRd236.c [286] : meshname = "mesh0"
_MEDmeshAdvancedRd236.c [286] : _datagroupname2 = "NOE"
_MEDmeshAdvancedRd236.c [287] : (*_filter).storagemode = 2
[0]PETSC ERROR: #1 DMPlexCreateMedFromFile() line 86 in /scratch/petsc-dev/src/dm/impls/plex/plexmed.c
[0]PETSC ERROR: #2 DMPlexCreateFromFile() line 2815 in /scratch/petsc-dev/src/dm/impls/plex/plexcreate.c
[0]PETSC ERROR: #3 main() line 38 in /scratch/petsc-dev/src/dm/impls/plex/examples/tutorials/ex5.c

Can you see whether it's a bug in DMPlexCreateMedFromFile or the MED file is broken?

It appears to be an MED incompatibility


We should ask Nico about it.

I tried to load the MED file from meshio (attached) by the MED-supplied test $PETSC_ARCH/externalpackages/med-3.2.0/tests/c/test5.c, and it didn't complain. I compared it with DMPlexCreateMedFromFile, replaced MED_COMPACT_STMODE flag by MED_GLOBAL_PFLMODE in MEDfilterBlockOfEntityCr calls (patch attached), and it started to work magically.

But I have no idea what's the meaning of this flag - I found no documentation of the MED library.

Note also that your knepley/fix-plex-med-orientation fix is still needed so I mean this should be applied generally - Nico's meshio is independent of GMSH.

Vaclav

Attachment: blockcylinder-50.med
Description: Binary data

Attachment: plexmed.patch
Description: Binary data



  Thanks,

     Matt
 
Thanks

Vaclav


5. 11. 2017 v 18:48, Matthew Knepley <knep...@gmail.com>:

On Thu, Nov 2, 2017 at 12:08 PM, Vaclav Hapla <vaclav.ha...@erdw.ethz.ch> wrote:
It seems that DMPlexCreateMedFromFile leaves out some mesh elements. I found it out when investigating why ParMetis redistribution crashes.

I attach the datafile $PETSC_DIR/share/petsc/datafiles/meshes/blockcylinder-50.exo converted to GMSH and MED format.
The conversion EXO to GMSH was done by meshio (github.com/nschloe/meshio), and GMSH to MED by GMSH itself.

I did:

cd $PETSC_DIR/src/dm/impls/plex/examples/tutorials
make ex5
FILE=blockcylinder-50.exo && ./ex5 -filename $FILE -dm_view hdf5:$FILE.h5 && $PETSC_DIR/bin/petsc_gen_xdmf.py $FILE.h5
FILE=blockcylinder-50.msh && ./ex5 -filename $FILE -dm_view hdf5:$FILE.h5 && $PETSC_DIR/bin/petsc_gen_xdmf.py $FILE.h5
FILE=blockcylinder-50.med && ./ex5 -filename $FILE -dm_view hdf5:$FILE.h5 && $PETSC_DIR/bin/petsc_gen_xdmf.py $FILE.h5

While the output from blockcylinder-50.exo and blockcylinder-50.msh looks OK, that from blockcylinder-50.med is corrupted.

I also attach my HDF5/XDMF outputs and screenshots from ParaView.

It appears that MED, at least coming from GMsh, inverts hexes just like GMsh does. I have fixed this in branch

  knepley/fix-plex-med-orientation

If you run your test in this, everything should be fine.

Michael, can you check whether this is a general fix, or it only applied to MED from GMsh?

  Thanks,

     Matt
 

Vaclav










-- 
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





-- 
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


Reply via email to