[petsc-users] Advices on creating DMPlex from custom input format

[Matthew Knepley]

On Mon, Oct 30, 2023 at 5:37 AM onur.notonur via petsc-users <
petsc-users at mcs.anl.gov> wrote:

> Hi,
>
> I hope this message finds you all in good health and high spirits.
>
> I wanted to discuss an approach problem input file reading/processing in a
> solver which is using PETSc DMPlex. In our team we have a range of solvers,
> they are not built on PETSc except this one, but they all share a common
> problem input format. This format includes essential data such as node
> coordinates, element connectivity, boundary conditions based on elements,
> and specific metadata related to the problem. I create an array for
> boundary points on each rank and utilize them in our computations, I am
> doing it hardcoded currently but I need to start reading those input files,
> But I am not sure about the approach.
>
> Here's what I have in mind:
>
>    1. - Begin by reading the node coordinates and connectivity on a
>    single core.
>    - Utilize the DMPlexCreateFromCellListPetsc() function to construct
>    the DMPlex.
>    - Distribute the mesh across processors.
>    - Proceed to read and process the boundary conditions on each
>    processor. If the global index of the boundary element corresponds to that
>    processor, we process it; otherwise, we pass.
>
> Additionally, maybe I need to reorder the mesh. In that case I think I can
> use the point permutation IS obtained from the DMPlexGetOrdering() function
> while processing boundary conditions.
>
> Also I have another approach in my mind but I don't know if it's possible:
> Read/construct DMPlex on single core including boundary conditions. Store
> BC related data in Vec or another appropriate data structure. Then
> distribute this BC holding data structure too as well as DMPlex.
>
This is by far the easier approach. If you do not have meshes that are too
big to load in serial, I would do
this. Here is what you do:

  - Read in the mesh onto 1 process
  - Mark the boundary conditions, probably with a DMLabel
  - Make a Section over the mesh indicating what data you have for BC
  - Create a Vec from this Section and fill it with boundary values
(DMCreateGlobalVector)
  - Distribute the mesh, and keep the point SF (DMPlexDIstribute)
  - Create a BC SF from the points SF (PetscSFCreateSectionSF)
  - DIstribute the BC values using the BC SF (PetscSFBcast)

  Thanks,

    Matt

> I would greatly appreciate your thoughts and any suggestions you might
> have regarding this approach. Looking forward to hearing your insights.
>
> Best regards,
>
> Onur
>


-- 
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/20231030/4b57bba6/attachment.html>