[petsc-dev] Mat/DM dependency

[Jed Brown]

On Tue, 5 Jan 2010 10:52:36 -0600, Dmitry Karpeev <karpeev at mcs.anl.gov> wrote:
> Automating decomposition/assembly of operators relative to a
> decomposition of identity (doi) would, in my opinion, go a long way
> towards decoupling Mat from DM.  DM would then convert their geometric
> data into a corresponding doi and Mat wouldn't need to know about DMs
> directly.

I like keeping Mat purely algebraic.  When assembly has nontrivial
semantics, I write a MyDMMatSetValuesBlockedWhatever(MyDM,Mat,...) so
that Mat doesn't see DM (since the DM is explicitly available when this
function is called, I prefer using it explicitly over patching the Mat
to retain a reference to the DM).

> That's what I'm trying to do with MatFwk -- a framework for general
> block decompositions of matrices

What granularity are you targeting with MatFwk?  I think it could be
very difficult to have good performance for very fine grains (e.g.
unassembled P1 elements).

Jed