[petsc-dev] parallel partitioning of uninterpolated meshes not supported

Hapla Vaclav vaclav.hapla at erdw.ethz.ch
Wed Feb 6 04:29:07 CST 2019


Bump - is this question too complex or too stupid? :-)

I'm now looking again on tests which are still failing in knepley/fix-plex-interpolate-sf-new.

There are several ones failing with
  Parallel partitioning of uninterpolated meshes not supported
So I'll just remove them for now.

One test (m_impls_plex_tests-ex18_2b) is failing just because of diff. I'll just check if the current output is correct and update the output file.

But these three
  dm_impls_plex_tests-ex18_6_tet_nsize-4interpolate-parallel_faces-2,2,2
  dm_impls_plex_tests-ex18_6_tet_nsize-4interpolate-parallel_faces-1,3,5
  dm_impls_plex_tests-ex18_6_tet_nsize-4interpolate-parallel_faces-3,4,7
fail with
  point SF contains X but not Y from its cone
Can it also be a problem of partitioning, and not interpolation?
So should we just for now disable ctetgen without interpolation and remove those tests?

If it is the case, knepley/fix-plex-interpolate-sf-new should be merged soon, just after removing the failing tests.

Thanks

Vaclav


On 31 Jan 2019, at 15:33, Hapla Vaclav <vaclav.hapla at erdw.ethz.ch<mailto:vaclav.hapla at erdw.ethz.ch>> wrote:

Matt, your commit
https://bitbucket.org/petsc/petsc/commits/ffbd61634109cd55a4e1889a4f100c5a1bdbfef1
says
Parallel partitioning of uninterpolated meshes not supported.

So should I now take it as a not-yet-implemented feature?

Does it really concern only this order of operations?
1) load sequentially
2) distribute
3) redistribute in parallel
4) interpolate in parallel

So the following three use cases
I)
1) load sequentially
2) interpolate sequentially
3) distribute
4) redistribute in parallel
II)
1) load sequntially
2) distribute
3) interpolate in parallel
4) redistribute in parallel
III)
1) load in parallel
2) interpolate in parallel
3) redistribute in parallel
can still be considered supported in general?

I mean not regarding the issues with the SF which still seem to affect parallel interpolation (https://bitbucket.org/petsc/petsc/branch/knepley/fix-plex-interpolate-sf).
BTW I now rebased this branch onto master as <https://bitbucket.org/petsc/petsc/branch/knepley/fix-plex-interpolate-sf-new> https://bitbucket.org/petsc/petsc/branch/knepley/fix-plex-interpolate-sf-new and it now needs to take that not-yet-implemented feature into account - I will update it and remind to me what cases are still failing.

Thanks in advance for clarification!

Vaclav

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20190206/fd418cb8/attachment.html>


More information about the petsc-dev mailing list