From: Jeff Squyres <jsquyres@XXXXXXXXXXXX>
Date: September 15, 2005 9:11:15 PM GMT+02:00
To: mpi-comments@XXXXXXXXXXXXX
Subject: MPI-2 errata

We found a few errata for MPI-2 this week that I didn't see listed in the docs at http://www.mpi-forum.org/docs/docs.html.

(is this the right list to send errata?)

In the extended collectives, the descriptions for MPI_ALLGATHER and MPI_ALLGATHERV make similar statements that I believe are incorrect.  Specifically, p158 lines 23-28 state:

-----
Specifically, the outcome of a call to MPI_ALLGATHER in the "in place" case is as if all processes executed n calls to

MPI_GATHER(MPI_IN_PLACE, 0, MPI_DATATYPE_NULL, recvbuf, recvcount, recvtype, root, comm)

for root = 0, ..., n - 1.
-----

A similar statement is made for MPI_ALLGATHERV on p159 lines 25-30.

By definition, only the root can use the MPI_IN_PLACE sendbuf argument, so the "as if" scenario cannot occur for different values of root -- the non-root processes need to execute MPI_GATHER (or MPI_GATHERV) with different arguments (per p154 lines 23-24, and p155 lines 27-28).

Additionally, within the last year or so, I had an e-mail conversation with Bill Gropp about language interoperability issues regarding attributes (including citing an example in MPI-2 that is wrong).  Did that get added to the list of errata?

Is the updated list of errata being published anywhere?

Thanks!

-- 
{+} Jeff Squyres
{+} The Open MPI Project
{+} http://www.open-mpi.org/