I could also keep dijitso serial and move the mpi logic to dolfin. 1. okt. 2015 13:46 skrev "Garth N. Wells" <gn...@cam.ac.uk>:
> > On 1 October 2015 at 12:41, Martin Sandve Alnæs <marti...@simula.no> > wrote: > >> I've used it in the new dijitso module for mpi awareness in jit. I don't >> want that to depend on dolfin. I think it's possible to work around, I can >> consider other options. >> > > If you only need a few calls, I think it's fine to wrap them and to make > mpi4py a dijitso dependency in parallel. > > Garth > >> 1. okt. 2015 12:54 skrev "Garth N. Wells" <gn...@cam.ac.uk>: >> >>> >>> On 1 October 2015 at 11:33, Martin Sandve Alnæs <marti...@simula.no> >>> wrote: >>> >>>> Ok, what about requiring it iff mpi is present? >>>> >>> >>> It does mean checking before using it, which can be untidy and makes it >>> a sort of 'half-dependency'. >>> >>> Is there something in particular you want to use in mpi4py? >>> >>> Garth >>> >>> >>> >>>> 1. okt. 2015 11:59 skrev "Garth N. Wells" <gn...@cam.ac.uk>: >>>> >>>>> >>>>> On 1 October 2015 at 10:44, Martin Sandve Alnæs <marti...@simula.no> >>>>> wrote: >>>>> >>>>>> Can we make mpi4py a required dependency for fenics? Any reasons not >>>>>> to? >>>>>> >>>>>> >>>>> To keep it simple, I presume it would mean making MPI a dependency for >>>>> FEniCS (?), which is not presently the case. >>>>> >>>>> Garth >>>>> >>>>> >>>>>> Martin >>>>>> >>>>>> _______________________________________________ >>>>>> fenics mailing list >>>>>> fenics@fenicsproject.org >>>>>> http://fenicsproject.org/mailman/listinfo/fenics >>>>>> >>>>>> >>>>> >>> >
_______________________________________________ fenics mailing list fenics@fenicsproject.org http://fenicsproject.org/mailman/listinfo/fenics
