On 2013-12-07 17:13, Jan Blechta wrote:
On Fri, 06 Dec 2013 17:13:12 +0000
"Garth N. Wells" <[email protected]> wrote:
On 2013-12-06 13:54, Jan Blechta wrote:
> I'd like to introduce support for communicator being supplied by
> user, stored within mesh and used by other objects based on the
> mesh. There are few applications of this. Mikael has one being
> rather non-trivial, I
> think.
>
Any object that requires an MPI communicator should take the
communicator as an argument. Our present design is way too simple,
and is flawed.
Can you explain me, for example, what should be comm stored within
FunctionSpace instance (different from comm of underlying Mesh) useful
for?
A FunctionSpace probably doesn't require a communicator.
Garth
Jan
> To achieve this it will require some changes in the interface,
> roughly sketched:
>
> - MPICommunicator::MPICommunicator()
> + MPICommunicator::MPICommunicator(const MPI_Comm&
> comm=MPI_COMM_WORLD)
>
> + MPICommunicator::MPICommunicator(const MPICommunicator& comm)
>
> merging classes MPI, MPINonbocking into MPICommunicator
>
> + Mesh::Mesh(..., shared_ptr<MPICommunicator>
> comm=MPICommunicator())
>
> + UnitIntervalMesh(..., comm), ...
>
Seems a bit complicated to me.
As a starting point, I would suggest changing the interface to all
the utility functions in dolfin/common/MPI.h to accept an MPI
communicator. The next stage is to propagate this upwards. For
simplicity and backwards compatibility we should have two interfaces
at the top level interfaces - one that assumes the default
communicator, and one that takes the communicator as an argument.
Garth
> Importat question is which type of comm to reaveal to user in C++
> DOLFIN, pyDOLFIN to make it as much as possibly compatible with
> other libraries, like boost, PETSc, mpi4py, petsc4py.
>
> Jan
> _______________________________________________
> fenics mailing list
> [email protected]
> http://fenicsproject.org/mailman/listinfo/fenics
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics
