[petsc-dev] Integrating PFLOTRAN, PETSC & SAMRAI

[Boyce Griffith]

On 6/7/11 2:42 PM, Jed Brown wrote:
> On Tue, Jun 7, 2011 at 20:41, Boyce Griffith <griffith at cims.nyu.edu
> <mailto:griffith at cims.nyu.edu>> wrote:
>
>     Nope; to keep SAMRAI stuff working like regular SAMRAI stuff, the
>     ghost values wind up in the middle of the vector.  At least with the
>     standard data types, local data storage, including ghost cells, for
>     each patch is contiguous.
>
>
> That's what I figured. Do you access multiple patches at the same time,
> or do you usually get a patch with ghost values set, compute on that
> patch, then move to the next one? If the latter, the iterator should be
> able to pull out of contiguous storage into a patch-local space.

Most restriction/prolongation/ghost cell communication operations 
involve operations on multiple patches, but these operators would not be 
well suited to use PETSc-backed storage anyway --- it could make more 
sense to recast them as scatter or as a sparse matrices.

In terms of solver stuff, everything I do is patch-by-patch.  I also 
spend a lot of my time in linear solvers, and the intention of these 
modifications would be to spend as much time as possible doing solver 
related things in PETSc, so pulling things out of / pushing things into 
different orderings might not be that big of a deal...

-- Boyce