Date: Wed, 15 Aug 2001 17:52:18 +0200
From: Jesper Larsson Traeff <traff@XXXXXXXXXXXX>
To: mpi-21@XXXXXXXXXXXXX
Subject: Re: MPI-2.1 corrections, Batch 2
References: <200107052110.QAA03838@XXXXXXXXXXX>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Mailer: Mutt 0.93i
In-Reply-To: <200107052110.QAA03838@XXXXXXXXXXX>
Sender: owner-mpi-21@XXXXXXXXXXXXX
Precedence: bulk
Reply-To: mpi-21@XXXXXXXXXXXXX

On Thu, Jul 05, 2001 at 04:10:20PM -0500, Rusty Lusk wrote:
> 
> 7:  Page 120, after line 13:
> 
> MPI_REPLACE, like the other predefined operations, is defined
> only for the predefined MPI datatypes. 
> 
> Rationale: 
> The rationale for this is that, for consistency, MPI_REPLACE
> should have the same limitations as the other operations.  Extending
> it to all datatypes doesn't provide any real benefit.
> End of Rationale: 
> 

I think I disagree with the last sentence of the rationale. MPI_Accumulate
with MPI_REPLACE can be used for "simultaneous puts" to the same window 
locations with an outcome as guaranteed by the atomicity condition. I don't
see why this could not be useful also with a user-defined datatype.
Of course it is not "nice" that MPI_REPLACE is special in that respect.

(another solution could have been to have a special put-operation for
this usage; MPI_REPLACE is kind of useless for any other purpose??)

Another question: the paragraph just before (starting at line 6) to
me reads like an extension of the predefined operations: is the statement
"Each datatype argument must be a predefined datatype or a derived
datatype, where..." intended to hold just for MPI_REPLACE, or for all
predefined operations (which I thought are only defined for predefined
datatypes)?

Jesper (Larsson Traff, NEC C&C Research Laboratories, St. Augustin, GERMANY)