dolfin team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

Anders Logg wrote:
> On Thu, Oct 23, 2008 at 04:44:22PM +0100, Garth N. Wells wrote:
> > Anders Logg wrote:
> > > On Thu, Oct 23, 2008 at 04:29:16PM +0100, Garth N. Wells wrote:
> > > > Anders Logg wrote:
> > > > > On Thu, Oct 23, 2008 at 05:24:54PM +0200, Anders Logg wrote:
> > > > > > Now that we have removed DofMapSet, how should the general assembly
> > > > > > interface look? We have lots of simple assemble() functions and they
> > > > > > all end up calling the main function which is
> > > > > >
> > > > > >   void assemble(GenericTensor& A,
> > > > > >                 const ufc::form& form,
> > > > > >                 const Array<Function*>& coefficients,
> > > > > >                 const DofMapSet& dof_map_set,
> > > > > >                 const MeshFunction<uint>* cell_domains,
> > > > > >                 const MeshFunction<uint>* exterior_facet_domains,
> > > > > >                 const MeshFunction<uint>* interior_facet_domains,
> > > > > >                 bool reset_tensor = true);
> > > > > >
> > > > > > One option would be
> > > > > >
> > > > > >   void assemble(GenericTensor& A,
> > > > > >                 const ufc::form& form,
> > > > > >                 const std::vector<std::tr1::shared_ptr<FunctionSpace> > function_spaces,
> > > > > >                 const std::vector<std::tr1::shared_ptr<Function> > coefficients,
> > > > > >                 const MeshFunction<uint>* cell_domains,
> > > > > >                 const MeshFunction<uint>* exterior_facet_domains,
> > > > > >                 const MeshFunction<uint>* interior_facet_domains,
> > > > > >                 bool reset_tensor = true);
> > > > > I meant
> > > > >
> > > > >     void assemble(GenericTensor& A,
> > > > >                   const ufc::form& form,
> > > > >                   const std::vector<FunctionSpace&> function_spaces,
> > > > >                   const std::vector<Function&> coefficients,
> > > > >                   const MeshFunction<uint>* cell_domains,
> > > > >                   const MeshFunction<uint>* exterior_facet_domains,
> > > > >                   const MeshFunction<uint>* interior_facet_domains,
> > > > >                   bool reset_tensor = true);
> > > > That look better.
> > > >
> > > > Why is
> > > >
> > > >    const std::vector<FunctionSpace&> function_spaces
> > > >
> > > > required in the assembler?
> > > >
> > > > Garth
> > > We need to get the DofMaps for the test and trial spaces. The
> > > coefficients (Functions) know their FunctionSpaces and thus their
> > > DofMaps, but the ufc::form does not now this (but dolfin::Form does).
> > Why not pass
> >
> >    std::vector<DofMap&> dof_maps
> >
> > then?
> >
> > Garth
>
> Yes, why not.
>
> Is there a reason the assembler should need to be initialized for a
> certain mesh? Do we reuse the mesh or any data between assemblies?

No.

I guess we did it to avoid passing the mesh every time we re-assembled a 
nonlinear or time dependent problem. I guess we don't need the mesh at 
all now since it's contained in the FunctionSpace. Therefore we probably 
should have

    const std::vector<FunctionSpace&> function_spaces

instead of

    const std::vector<DofMap&> dof_maps

We can have a check in the assembler that all FunctionSpaces have the 
same mesh.

Garth

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