On Fri, Jul 16, 2010 at 05:20:53PM -0400, Prentice Bisbal wrote:
>
>
> Eugene Loh wrote:
> > Anton Shterenlikht wrote:
> >
> >> Will this bit of code work:
> >>
> >> if (rank .eq. <somerank>) then
> >>
> >> *change var*
> >>
> >> call MPI_Bcast(var, 1, MPI_INTEGER, rank, &
> >> MPI_COMM_WORLD, errstat)
> >> end if
> >>
> >> call MPI_Barrier(MPI_COMM_WORLD, errstat)
> >>
> >> I'm using a int variable on all processors.
> >> At some point some process can change its
> >> value of this variable.
> >> I then need to let all other
> >> processes know the new value. So I tried
> >> to use MPI_Bcast for this. But I think
> >> MPI waits indefinitely at MPI_Barrier.
> >>
> >> I'm now wondering if MPI_Bcast must
> >> be executed by all processes, similar
> >> to MPI_Barrier. Is this correct?
> >>
> > All processes in the communicator must make the collective call...
> > Bcast, Barrier, Allreduce, etc.
>
> Therefore, following a Bcast with a Barrier is redundant in this example.
Sorry, just to be absolutely clear, are you saying
that even though only one process in the communicator
is calling Bcast, the call will be made on all
processes?
Or are you saying I cannot put Bcast (or any other collective call)
inside a "if (rank== ...) {} " construct?
many thanks
anton
--
Anton Shterenlikht
Room 2.6, Queen's Building
Mech Eng Dept
Bristol University
University Walk, Bristol BS8 1TR, UK
Tel: +44 (0)117 331 5944
Fax: +44 (0)117 929 4423
