Newsgroups: comp.parallel.pvm
Path: ukc!uknet!demon!pipex!howland.reston.ans.net!vixen.cso.uiuc.edu!milo.mcs.anl.gov!godzilla.mcs.anl.gov!gropp
From: gropp@godzilla.mcs.anl.gov (William Gropp)
Subject: Re: MPI and variable length mixed messages
Message-ID: <76524183210962@godzilla.mcs.anl.gov>
Summary: Read the CURRENT MPI draft
Keywords: MPI, derived datatype, type signature
Sender: usenet@mcs.anl.gov
Nntp-Posting-Host: godzilla.mcs.anl.gov
Organization: Math and Computer Science, Argonne National Laboratory
References: <2neog3$9vg@labss19.mathcs.emory.edu> <2nf870$35e@hawkeye.osc.edu>
Date: Fri, 1 Apr 1994 23:10:32 GMT
Lines: 15

As has already been mentioned, the CURRENT MPI draft provides a mechanism
for packing and unpacking messages consisting of arbitrary (MPI) derived
datatypes into a single buffer.  This feature was added in large part 
because of contributions during the public comment period on the need for
this functionality.  Because MPI supports derived datatypes, packing two
or more variable-length structures, each of which contains mixed data, 
into a message is relatively painless.

The February 20 version of the MPI draft has been available from netlib; 
the November version is at this point of historical interest ONLY.
A revised version will be available shortly (and the final form of 
MPI_PACK and MPI_UNPACK are slightly different from the February versions).

Bill Gropp


