[petsc-users] DMPlexCreateSection crashed with larger number of processors

[Matthew Knepley]

On Fri, Jul 8, 2022 at 7:41 PM Mike Michell <mi.mike1021 at gmail.com> wrote:

> As a follow-up, just "depth" default label was provided to avoid NULL
> PETScIS, and it seems working.
>

Ah, "depth" is definitely required and is produced by DMPlexStratify(). I
will put in a check for this so that you get a good error message.

  Thanks!

     Matt


> Thanks,
> Mike
>
>
>> It looks like NULL PETScIS is obtained from DMGetStratumIS() which is
>> shown in the previously attached code. Currently, "marker" label is used to
>> define default bcPointIS pointer with numBC=1 through DMGetStratumIS(), and
>> it seems that if a certain proc does not have any boundary from this
>> "marker" label, the DMPlexCreateSection() crashes.
>>
>> The PETSc's manual page for DMPlexCreateSection() says "bcPoints" is an
>> IS holding points to which each boundary condition applies. So, by
>> including no boundary for a certain distributed mesh block, it looks very
>> natural that this function crashes. How can I not to pass a NULL PETScIS in
>> this case? Is there any specific way for that (or a trick)?
>>
>> Thanks,
>> Mike
>>
>>
>>> On Wed, Jul 6, 2022 at 3:42 PM Mike Michell <mi.mike1021 at gmail.com>
>>> wrote:
>>>
>>>> I am passing NULL for IS Permutation as;
>>>>
>>>> call
>>>> DMPlexCreateSection(dm_geom,nolabel,pNumComp,pNumDof,numBC,pBcField,pBcCompIS,pBcPointIS,PETSC_NULL_IS,section,ierr)
>>>>
>>>> Should I define non-NULL IS Permutation before I call
>>>> DMPlexCreateSection() to avoid the issue?
>>>>
>>>
>>> No, the problem is not the perm argument. To be sure, we should get a
>>> stack trace with line numbers (use a debug build). To me it looked
>>> like you passed a NULL IS to DMAddBoundary().
>>>
>>>   Thanks,
>>>
>>>     Matt
>>>
>>>
>>>> Thanks,
>>>> Mike
>>>>
>>>>
>>>> On Wed, Jul 6, 2022 at 12:57 PM Mike Michell <mi.mike1021 at gmail.com>
>>>>> wrote:
>>>>>
>>>>>> Hi, I am using DMPlex for a FVM code written in Fortran that uses
>>>>>> DMPlexCreateSection() to prepare an output file in vtu format. A domain of
>>>>>> interest has around 5000 node points and with small number of processors
>>>>>> (such as nproc=3), the DMPlexCreateSection() works fine. However, if a
>>>>>> larger number of procs (such as nproc=20) is used,  DMPlexCreateSection()
>>>>>> crashes with error message below. It looks very similar problem in a
>>>>>> previous PETSc user inquiry: " [petsc-dev] DMPlexCreateSection crash
>>>>>> in a no-boundary corner case
>>>>>> <https://lists.mcs.anl.gov/pipermail/petsc-dev/2013-November/014049.html>
>>>>>> "
>>>>>>
>>>>>
>>>>> Yes, the recommendation is not to pass a NULL PetscIS to the
>>>>> AddBoundary() routine. Are you doing this?
>>>>>
>>>>>   Thanks,
>>>>>
>>>>>      Matt
>>>>>
>>>>>
>>>>>
>>>>>> When I call DMPlexCreateSection(), a nullified pointer
>>>>>> "nolabel(:)=>NULL()" is provided. What is a possible solution to this
>>>>>> problem? It looks like there is a problem with a processor that does not
>>>>>> have any boundaries. A piece of the source code is attached to show how
>>>>>> the  DMPlexCreateSection() is used in the code.
>>>>>>
>>>>>> Thanks,
>>>>>> Mike
>>>>>>
>>>>>> [16]PETSC ERROR: --------------------- Error Message
>>>>>> --------------------------------------------------------------
>>>>>> [16]PETSC ERROR: Null argument, when expecting valid pointer
>>>>>> [16]PETSC ERROR: Null Pointer: Parameter # 1
>>>>>> [16]PETSC ERROR: See https://petsc.org/release/faq/ for trouble
>>>>>> shooting.
>>>>>> [16]PETSC ERROR: Petsc Release Version 3.17.0, unknown
>>>>>> [16]PETSC ERROR: #1 ISGetLocalSize() at
>>>>>> [16]PETSC ERROR: #2 DMPlexCreateSectionBCDof() at
>>>>>> [16]PETSC ERROR: #3 DMPlexCreateSection() at
>>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> 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
>>>>>
>>>>> https://www.cse.buffalo.edu/~knepley/
>>>>> <http://www.cse.buffalo.edu/~knepley/>
>>>>>
>>>>
>>>
>>> --
>>> 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
>>>
>>> https://www.cse.buffalo.edu/~knepley/
>>> <http://www.cse.buffalo.edu/~knepley/>
>>>
>>

-- 
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

https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20220709/996df4be/attachment-0001.html>