dolfin team mailing list archive

[Ola Skavhaug <skavhaug@xxxxxxxxx>]

Garth N. Wells skrev den 07/12-2007 følgende:
> 
> 
> Anders Logg wrote:
> > On Fri, Dec 07, 2007 at 01:52:36PM +0100, DOLFIN wrote:
> >> One or more new changesets pushed to the primary dolfin repository.
> >> A short summary of the last three changesets is included below.
> >>
> >> changeset:   3463:8951cec821a5d4f6d6f9c983cbcb836fb2c5a1b8
> >> tag:         tip
> >> parent:      3462:b8f73cb2e5a998999fb9a5ac714ebec11890f7a8
> >> parent:      3452:44e94b868ca473f050bf02d606179e0161299d12
> >> user:        "Ola Skavhaug <skavhaug@xxxxxxxxx>"
> >> date:        Fri Dec 07 12:21:11 2007 +0100
> > 
> > A short summary:
> > 
> > 1. A new base class GenericSparsityPattern has been added.
> > 
> > 2. A new base class LinearAlgebraFactory has been added.
> > 
> > 3. Implementations of the factory interface has been added for all
> > backends (PETSc and uBLAS).
> > 
> > This makes it possible to use different representations of the
> > sparsity pattern for different backends (Trilinos has its own) but
> > reuse the same interface.
> > 
> > Also, the factories can be used to propagate the choice of backend to
> > places where we need to create new matrices or vectors. For example,
> > in DiscreteFunction we do
> > 
> >   x = new Vector(n);
> > 
> > which ties the implementation of DiscreteFunction to the default
> > DOLFIN backend. Instead, one can now call the factory() function on
> > any linear algebra object (like a vector or matrix) and get a
> > singleton factory that knows how to create matrices and vectors that
> > use the same backend as the object the factory was taken from.
> > 
> > Altogether, this means that we can now add a backend for Trilinos
> > (Epetra) and users of DOLFIN can use their own backend (by
> > implementing wrappers that follow the interfaces GenericMatrix,
> > GenericVector and GenericSparsityPattern) and use those with DOLFIN.
> > 
> 
> Nice. Can we expect the merge with PyCC to accelerate now that this 
> issue has been worked out?

Yes. We already have a preliminary version of our linear algebra backend that
should work with the dolfin assembler now. The plan now is to replace most of
the C/C++ code in PyCC with constructs from dolfin, and extend the dolfin
functionality when needed instead of implementing this in PyCC. PyCC should
become an almost pure Python framework depending heavily on dolfin. We will
keep some C++ code for now, since the ode systems that we need to solve are
not efficiently solved in dolfin. Hopefully, the happy merge will not take too
long time.
 
Ola

> Garth
> 
> 
> _______________________________________________ DOLFIN-dev mailing list
> DOLFIN-dev@xxxxxxxxxx http://www.fenics.org/mailman/listinfo/dolfin-dev