Newsgroups: comp.parallel.mpi
From: Milind Bhandarkar <milind@cs.uiuc.edu>
Subject: Re: MPI_Barrier and Collective Communications
Organization: University of Illinois
Date: Mon, 26 Feb 1996 09:48:49 -0600
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <3131D661.5B02@cs.uiuc.edu>

Eric Salo wrote:
> 
> > I suppose this means that all the processes in a group are syncronized by
> > the collective call itself
> 
> Incorrect. MPI_Barrier() is the only collective call about which any
> synchronization assumptions may be safely made. All other collective
> calls may or may not synchronize, depending on the implementation.

I suppose MPI_Barrier() is not the only such call. Calls such as MPI_Allreduce()
also have synchronization assumptions built into them.
-- 
__Milind_A._Bhandarkar______________________________________milind@cs.uiuc.edu_
(W)_____2508,DCL,1304,W.Springfield_Ave_Urbana_IL_61801_____(217)_333_5827_____
(H)_____807,S.Mattis,Apt.3,Champaign_IL_61820_______________(217)_352_1698_____
______________________________________________Cogito,Ergo_Paid_Less____________

