MPI-2 chapter 7 is all about the extended collective operations -- for
each collective, it defines which process is supposed to pass in
Specifically, check out MPI-2 section 7.3.2, "Operations that Move Data"
(it's a very long section; a sub-section for each MPI collective).
> -----Original Message-----
> From: users-bounces_at_[hidden]
> [mailto:users-bounces_at_[hidden]] On Behalf Of Michael Kluskens
> Sent: Monday, March 27, 2006 10:49 AM
> To: Open MPI Users
> Subject: Re: [OMPI users] MPI_ROOT - required where/when?
> Thank you. I was using MPI_ROOT in a MPI_Bcast call from
> the parent
> process where I spawn subprocesses and it worked fine in OpenMPI but
> when it didn't work on another platform and the support engineers
> solution was to define it as zero I started looking and
> couldn't find
> anything close to what you said below in the four books I
> have. Page
> 236 of Using MPI-2 shows the correct use of MPI_ROOT but no
> explanation of why.
> On Mar 27, 2006, at 10:21 AM, Edgar Gabriel wrote:
> > MPI_ROOT is required for the rooted operations of the inter-
> > communicator
> > collectives in MPI-2, I am not aware of that you need MPI_ROOT in
> > intra-communicator collectives as defined in MPI-1.
> > Thanks
> > Edgar
> > Michael Kluskens wrote:
> >> The constant MPI_ROOT is not universally defined in all current
> >> shipping MPI implementations.
> >> Is there any MPI function/call that requires MPI_ROOT?
> >> From the complete reference it appears that MPI_ALLGATHER might be
> >> the one routine.
> >> This all relates to portability, code I write using OpenMPI has to
> >> work on other systems not using OpenMPI.
> >> Michael
> >> _______________________________________________
> >> users mailing list
> >> users_at_[hidden]
> >> http://www.open-mpi.org/mailman/listinfo.cgi/users
> > _______________________________________________
> > users mailing list
> > users_at_[hidden]
> > http://www.open-mpi.org/mailman/listinfo.cgi/users
> users mailing list