Newsgroups: comp.parallel.mpi
From: Enric Guinovart Gracia <enricg@ac.upc.es>
Subject: Re: Concurrency doubts
Organization: Universitat Politecnica de Catalunya (UPC)
Date: Fri, 06 Sep 1996 13:12:57 +0100
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <32301549.34DB@ac.upc.es>

Eric Salo wrote:
> 
> > : Not quite. It is not a violation of the standard for an implementation
> > : to be thread-unsafe. The intent was merely that MPI not get in the way
> > : of multi-threaded implementations should any arise. So, for example,
> > : MPI has no notion of a global error code, nor of a "current" pack/unpack
> > : buffer. And so on.
> >
> > Is that for MPI-1 or 2?
> 
> That was the philosophy of MPI-1. MPI-2 may or may not address threads
> specifically. The current state seems to be (someone please correct me
> if I have this wrong) that MPI-2 will not mandate thread support but
> will possibly define certain behavior that must be met by those
> implementations that do support threads.

Through books I have read, I have reached the following conclusion and
perhaps my point of view is not the correct one:

  I think that the fact of thread supporting it is not a MPI-standard
problem, it is an implementation one. It seems to me that MPI-standard
fixed the bases for thread-supporting at specifycation level, but it is
an implementation problem to do it nor not. Thus, MPI-2-standard must
remain in this way ...

  But the main problem is that MPI implementation and thread library
must work hand-in-hand. Thus, both pachages are very relationed.

  The question I posted, arised from the fact that I did not be able to
find a partner for MPI-F on SP2 ...

I keep on searching ...

--- Enric

