Newsgroups: comp.parallel.mpi
From: zmc@uh.edu (Zhuo-Min Chen)
Subject: Something in between
Organization: University of Houston
Date: 8 May 1995 23:02:56 GMT
Message-ID: <3om7v0$7ei@masala.cc.uh.edu>

As a Scientist working on SP2, I have ran into a commincation problem,
which may be well addressed by MPI but I am not aware of. 
It is kind of Reducation which is not Totally collective, but Semi. 
I am looking for a Local Reduction which involves a subset of CPUs.
And Each subset may overlap with others. Thus this is exactly point-to-point
nor collective. By defining a communicator, I was told by someone that
this can be done rather easily. By the same token, 
Local-Broadcast   ------->    Broadcast
Local-Gather      ------>      Gather
so on. 
Anyone out there cares to response? 


