Hi Roy:

Thanks for the response. I too use the PARALLEL type for the most
part. However, I also need access to the ghosts and I'm afraid I'm
interested in more than just a localize to a GHOSTED vector. I guess
I'm trying to understand the parallel data layout. I looked at how the
system.rhs and system.solution vectors were created. It looks like
they are created using Petsc's VecCreateMPI function. So, can I assume
that the following holds:

1. Each processor owns a consecutive set of global dofs.
2. The first dof owned by any processor is equal to 1 more than the
last dof owned by the previous processor.

Thanks.

Regards,

Rahul


On Wed, Dec 2, 2009 at 2:54 PM, Roy Stogner <[email protected]> wrote:
>
> On Mon, 30 Nov 2009, Rahul Sampath wrote:
>
>> Could someone please tell me how to access the ghost values (i.e. dof
>> values shared between processors) from a NumericVector sharing the
>> same parallel layout as the system.rhs vector.
>
> I'm afraid the default type for those vectors, PARALLEL, doesn't have
> any local storage for ghost values.  You'll either need to use a
> SERIAL or GHOSTED type vector instead, or localize your vector into
> one of those types.
>
> Hmm... I just noticed that everything we create with
> System::add_vector() gets created with PARALLEL type.  That's exactly
> what I want in my applications, but I see now that it might be
> suboptimal for other people.  You can reinitialize one of those
> vectors as SERIAL or GHOSTED and the new type will be preserved
> through projections, but should we add an optional argument for people
> to add non-PARALLEL vectors in the first place?
> ---
> Roy
>

------------------------------------------------------------------------------
Join us December 9, 2009 for the Red Hat Virtual Experience,
a free event focused on virtualization and cloud computing. 
Attend in-depth sessions from your desk. Your couch. Anywhere.
http://p.sf.net/sfu/redhat-sfdev2dev
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to