Newsgroups: comp.parallel.mpi
From: pvb1@Ra.MsState.Edu (Purushotham V. Bangalore)
Subject: Re: passing a structure
Organization: Mississippi State University
Date: 4 Nov 1994 15:59:07 -0600
Message-ID: <pvb1.783985339@Ra.MsState.Edu>

gmounie@ens-lyon.fr (Gregory Mounie) writes:

>In article (Purushotham V. Bangalore) writes:
>> animesh@top.cis.syr.edu (Animesh Chatterjee) writes:

>> >I need to send a structure which contains both integer and float type members.
>> >How can I do this in MPI?

>> You can either use the MPI datatype MPI_FLOAT_INT or create your own
>> datatype. There is an example on page 116 of the MPI standard document.

> Does MPI_FLOAT_INT a standard Datatype ? Does it not just a particular
> datatype for a global operation (MPI_Reduce() or MPI_Scan() ?) and so may not work for
> classic communication.

True, MPI_FLOAT_INT is not a basic datatype, it is a derived datatype.
My understanding is that MPI_FLOAT_INT is like any other derived datatype 
one would create and can be used in any of the communication operation. 
Hence I find no reason why it should not work for other communication
operations. Infact it works fine with the MPICH(ANL/MSU) and MPIF(IBM)
versions. 

Hope that helps!

Puri



