[petsc-users] DMDACreateSection

[Matthew Knepley]

On Tue, Jan 22, 2013 at 3:34 AM, Hui Zhang <mike.hui.zhang at hotmail.com>wrote:

> Hi petsc-group,
>
> if I want to map a structured mesh onto processors by partition of mesh
> **elements**,
> can I use DMDACreate3d -> DMDACreateSection, then
> DMDACreateLocal(Global)Vector to
> give me the vector corresponding to my Q1-element d.o.f.?
>

No, this is not finished yet. I have only implemented it in serial, mostly
because no one is
asking for it.

Note that the data layout has nothing to do with partitioning. We would
still have to rewrite
those routines for an element-wise partition. This is on the list of things
to do.

   Matt


> I ask because according to the manual, the usual use of DMDA is only for
> partition of
> **nodes** to processors and the local vector is not corresponding to
> partition of elements.
> For example, consider the Q1 element on the 1D mesh, (numbers for nodes)
>
> 1--2--3--4--5
>
> Given two processors, I want the local vector after partition to reside on
> the partitioned mesh
>
> 1--2--3   3--4--5
> proc 0    proc 1
>
> The global value of d.o.f at the node 3 may be managed by proc 1, and proc
> 0 only ghosts it.
> If it is the case, proc1 does not ghost any value.
>
> This is different from the usual DMDA for finite difference stencils, in
> which the ghosting
> is symmetric, i.e. both proc0 and proc1 will ghost some values.
>
> Thanks in advance!
>
>
>
>


-- 
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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20130122/59f89ccb/attachment.html>