[Mpi3-hybridpm] [MPI Forum] #209: MPI3 Hybrid Programming: multiple endpoints per collective, option 2

[MPI Forum]

#209: MPI3 Hybrid Programming: multiple endpoints per collective, option 2
----------------------------------------------------------+-----------------
                    Reporter:  dougmill@…                 |                       Owner:  dougmill@…             
                        Type:  Enhancements to standard   |                      Status:  new                    
                    Priority:  Not ready / author rework  |                   Milestone:  2010/01/19 Atlanta, USA
                     Version:  MPI 3.0                    |                    Keywords:                         
              Implementation:  Unnecessary                |           Author_bill_gropp:  0                      
          Author_rich_graham:  0                          |           Author_adam_moody:  0                      
      Author_torsten_hoefler:  0                          |        Author_dick_treumann:  0                      
Author_jesper_larsson_traeff:  0                          |       Author_george_bosilca:  0                      
           Author_david_solt:  0                          |   Author_bronis_de_supinski:  0                      
        Author_rajeev_thakur:  0                          |         Author_jeff_squyres:  0                      
    Author_alexander_supalov:  0                          |    Author_rolf_rabenseifner:  0                      
----------------------------------------------------------+-----------------

Comment(by dougmill@…):

 A possible alternative to specifying a group would be to use some sort of
 "data identifier" or to actually formally create a shared data object for
 the data parameters and pass this data object to the allreduce instead of
 the discrete data parameters (buffer, count, datatype for source and
 destination). This data object could then be used to associate calls
 together, although it may still be necessary to add some sort of
 "participant count" parameter. something like:
 {{{
 /* 'master' thread does this: */
 MPI_Dataparams_create(sendbuf, recvbuf, 1, MPI_DOUBLE, &data_blob);
 ...
 /* each of 3 participants does this: */
 MPI_Allreduce(data_blob, 3, MPI_SUM, comm);
 }}}
 Perhaps the 'op' should also be part of the data_blob? And possibly the
 '3' (number of participants)?

-- 
Ticket URL: <https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/209#comment:1>
MPI Forum <https://svn.mpi-forum.org/>
MPI Forum