[petsc-users] Creating dmplex in every rank and redistributing

[Matthew Knepley]

On Mon, Jul 11, 2022 at 12:05 PM Prateek Gupta <prateekgupta1709 at gmail.com>
wrote:

> Thank you. In the meantime, can you please let me know what these are?
> They look like hash maps to me but I am not sure what functions like
>
> PetscHSetIGetElems are for and would be really helpful if you could maybe pass on the old documentation information
>
> We are using khash (). The PETSc wrapper is here:

  https://gitlab.com/petsc/petsc/-/blob/main/include/petsc/private/hashset.h

and we make the integer hash here


https://gitlab.com/petsc/petsc/-/blob/main/include/petsc/private/hashseti.h

You can see from the khash docs what those functions do. GetElems() reads
out everything in the hash set, updating the offset into the given array.

  Thanks,

     Matt


> Thank you.
> Sincerely,
> Prateek Gupta, PhD
>
>
> On Mon, Jul 11, 2022 at 5:19 PM Matthew Knepley <knepley at gmail.com> wrote:
>
>> On Mon, Jul 11, 2022 at 5:54 AM Prateek Gupta <prateekgupta1709 at gmail.com>
>> wrote:
>>
>>> Thanks Matt.
>>> While looking into the function, I noted PetscHSetI and related
>>> functions but couldn't find their docs (404 error). Have these functions
>>> and data types deprecated?
>>>
>>
>> No. These are a bit complicated because they are made by the preprocessor
>> from a 3rd party package.
>>
>> We broke the links when we moved all the documentation to Sphinx. I will
>> see if I can fix them.
>>
>>   Thanks,
>>
>>       Matt
>>
>>
>>> Thank you.
>>> Sincerely,
>>> Prateek Gupta
>>>
>>>
>>> On Mon, Jul 4, 2022 at 6:09 PM Matthew Knepley <knepley at gmail.com>
>>> wrote:
>>>
>>>> On Mon, Jul 4, 2022 at 2:29 AM Prateek Gupta <
>>>> prateekgupta1709 at gmail.com> wrote:
>>>>
>>>>> Hi,
>>>>>
>>>>> Using dmplex, I am trying to create an example where I can start with
>>>>> a poor distribution of an unstructured mesh (reading from a file in
>>>>> parallel) and then use redistribution to optimize it.
>>>>>
>>>>
>>>> You can call DMPlexDistribute() on any mesh, including a parallel one.
>>>>
>>>>
>>>>> I know that I can call ReBalanceSharedPoints on an already created
>>>>> distribution from dmplexDistribute. But is it possible to initialize a
>>>>> dmplex from each rank (each rank initializes its own chunk of the mesh) and
>>>>> then call this function?
>>>>>
>>>>
>>>> Yes, this is done in
>>>>
>>>>
>>>> https://petsc.org/main/docs/manualpages/DMPLEX/DMPlexCreateFromCellListParallelPetsc/
>>>>
>>>> It is not trivial to do by hand, so I would look at that code first if
>>>> you want to do that.
>>>>
>>>>
>>>>> Most of the numbering in dmplex DAG representation is local, but while
>>>>> reading from a file in parallel, I only have access to global numbering of
>>>>> nodes. Do I need to reassign this to a local numbering? Is there a
>>>>> datastructure within petsc that can help with this?
>>>>>
>>>>>
>>>>> Thank you.
>>>>> Sincerely,
>>>>> Prateek Gupta, PhD
>>>>>
>>>> --
>>>> 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/20220711/ac8c294b/attachment.html>