On Thu, 24 Jul, 2014 at 3:58 PM, Eugenio Gianniti
<[email protected]> wrote:
Dear all,
I am trying to exploit the MPI features of FEniCS in the fem-fenics
package for Octave, which was introduced on the list a couple of
months ago. In this application the C++ code using DOLFIN is
dynamically loaded into Octave. Unfortunately, it seems that this way
the communication among threads is not established correctly. I am
looking for advice on how to properly handle the initialisation of
MPI and then pass the communicator on to DOLFIN.
DOLFIN checks if MPI has already been initialised. If MPI has been
initialised, then DOLFIN will be not intialised (or finalise) MPI.
Many DOLFIN objects accept an MPI communicator which, you can pass in.
This is still a work-in-progress. In other places, DOLFIN with use
MPI_COMM_WORLD.
Garth
Kind regards,
Eugenio Gianniti
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics
