[petsc-dev] Integrating PFLOTRAN, PETSC & SAMRAI

[Boyce Griffith]

On 6/6/11 4:45 PM, Jed Brown wrote:
> On Mon, Jun 6, 2011 at 22:41, Boyce Griffith <griffith at cims.nyu.edu
> <mailto:griffith at cims.nyu.edu>> wrote:
>
>     At least in my code, and I expect in Bobby's too, all Mat and PC
>     objects are just shells.  So the only non-PETSc objects are Vec
>     wrapper objects.
>
>
> You are still supposed to call MatAssemblyEnd() which does this.

Yes, I understand.

I don't have any problems with Mat state --- I was just indicating that 
it would be nice not to have to worry about screwing up the state stuff.

> How do the SAMRAI wrappers get access to Vec's internals to modify it?
> That is, how do they turn a Vec into some mutable structure?

Each PETSc Vec just holds a pointer to the SAMRAI vector.

I do something like this:

     Vec petsc_vector;

     ierr = VecCreate(comm, &petsc_vector); IBTK_CHKERRQ(ierr);

     petsc_vector->petscnative= PETSC_FALSE;
     petsc_vector->map->n = 0;
     petsc_vector->map->N = 0;
     petsc_vector->map->bs = 1;

     d_petsc_vector->ops->duplicate =
         PETScSAMRAIVectorReal::VecDuplicate_SAMRAI;
     d_petsc_vector->ops->duplicatevecs = VecDuplicateVecs_Default;
     d_petsc_vector->ops->destroyvecs = VecDestroyVecs_Default;
     d_petsc_vector->ops->dot = VecDot_SAMRAI;
     d_petsc_vector->ops->mdot = VecMDot_SAMRAI;

     ....

     d_petsc_vector->data = /* pointer to SAMRAI vector object */;

where the various function pointers set in vec->ops are static functions 
that call the appropriate SAMRAI methods.

If there is a better way to do this, I'm happy to change this code, 
which I have never liked much to begin with.  :-)

-- Boyce