[mpi3-coll] array parameters in nonblocking collectives

[Graham, Richard L.]

The only issue I would have with this is that this forces the users to keep these objects in scope for the duration of the communication, which sort of defeats the intent of the nonblocking collectives.  Having said that, the same issue holds for the data, and I would not suggest that we keep a copy of the data, so I realize I am advocating an inconsistent view point, so this may not be so good an idea.

Rich


On 8/1/09 11:33 AM, "Anthony Skjellum" <tony at runtimecomputing.com> wrote:

It appears to me that mpi should temporarily  "control"(own) the passed by reference data structures from the point the call is made to the time of completion... Users should accept this as part of the contact and shouldn't assume they own this data during the call's lifetime... From a state transition diagram perspective, ownership passes to and from mpi at ifunction and wait call, respectively.


Tony Skjellum, PhD
RunTime Computing Solutions, LLC
tony at runtimecomputing.com
direct: +1-205-314-3595
cell:    +1-205-807-4968

-----Original Message-----
From: "Rajeev Thakur" <thakur at mcs.anl.gov>

Date: Sat, 1 Aug 2009 10:18:41
To: 'MPI-3 Collective Subgroup Discussions'<mpi3-coll at lists.mpi-forum.org>
Subject: Re: [mpi3-coll] array parameters in nonblocking collectives


_______________________________________________
mpi3-coll mailing list
mpi3-coll at lists.mpi-forum.org
http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-coll


_______________________________________________
mpi3-coll mailing list
mpi3-coll at lists.mpi-forum.org
http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-coll