[mpi3-coll] Our MPI-2.2 items

[Torsten Hoefler]

On Tue, Jul 29, 2008 at 10:47:42AM +0200, Jesper Larsson Traeff wrote:
> 
> Hello everybody,
> 
> On Mon, Jul 28, 2008 at 02:17:50PM -0400, Torsten Hoefler wrote:
> > Hello Collectives Workgroup,
> > I just added our open MPI-2.2 items to the list. 
> >
> good. I'll try to add something, too:
> 
> MPI_Reduce_scatter_block
> Extended binary reduction operators for segmented scans
> Clarification on intercomm Barrier
great! I just added them to the Agenda which is now on:
https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/Telecon073108

> > Please review items 13-16 on
> > https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/MpiTwoTwoWikiPage
> > 
> > The deadline for things to be considered for MPI-2.2 is next Friday.
> >
> uuh - well...
yeah, I am sure that I heard this at the last meeting but I didn't find
it written down anywhere. So it might not be ``enforced'' :).

> > I dropped the partial pack/unpack from our list of open issues because
> > large datatypes can be disassembled via query functions and thus packed
> > in pieces with MPI_TYPE_GET_ENVELOPE and MPI_TYPE_GET_CONTENTS.
> >
> I disagree with this argument. While it is possible, it leaves completely
> to the user (in this case: high-level library implementer) to implement
> these pack/unpack functions efficiently, and this is not a trivial
> task at all
yes, but we could easily write a library doing this with the current
functions. It's an algorithmic problem, not an implementation problem in
my opinion (i.e., I don't see much that can be gained by doing it in the
library).

> I have been hesitant in posting a concrete proposal because the exposed
> partial packing functions should also make sense in a heterogeneous
> environment, and I'm not sure how to handle this (I'm not even sure
> MPI_Pack/Unpack do...)
oh yes. We'll talk about this on the telecon.

> Thus: I'll try to post something this week... at least, we should discuss
> about this at the teleconf.
> 
> > Please add any missing items and/or corrections to the wiki and/or feel
> > free to discuss them on the list.
> > 
> I think your fixes to the topology interface needs more
ok, please comment on the wiki or bring it up during the telecon!

Thanks & Best,
  Torsten

-- 
 bash$ :(){ :|:&};: --------------------- http://www.unixer.de/ -----
Indiana University    | http://www.indiana.edu
Open Systems Lab      | http://osl.iu.edu/
150 S. Woodlawn Ave.  | Bloomington, IN, 474045-7104 | USA
Lindley Hall Room 135 | +01 (812) 855-3608