[Mpi-22] New proposal: Support for large message counts

Thu Sep 4 04:59:28 CDT 2008

Good point. At face value, this seems to suggest that we should have
"extended" constructors available on a level with the traditional ones,
thus favoring the original name-shifted proposal. If I think about the
library internals in this case, however, this addition will effectively
push all of the library extent calculations to 64-bit (or more, see
below).

Another thing is that if we multiply to long values, the result may
potentially overflow a long. So, for a library that allows long counts
and long datatype extents, the internals of the library will have to be
long long. long long (128 bit) arithmetic may be rather expensive on
some CPUs. So, the 64-bit interface should probably be optional.

________________________________

From: mpi-22-bounces_at_[hidden]
[mailto:mpi-22-bounces_at_[hidden]] On Behalf Of Richard
Treumann
Sent: Wednesday, September 03, 2008 7:24 PM
To: MPI 2.2
Subject: Re: [Mpi-22] New proposal: Support for large message counts

MPI_Aint has the right number of bits to hold an address. For operations
that are address space oriented, count can be the same size as an
MPI_Aint

MPI_Offset has the right number of bits to deal with the range in a
file. For operations that are file oriented, arguments like count should
be an MPI_Offset

A problem is that MPI_Datatype constructors are intended to apply to
both and it is common for a 32 bit address space to operate on files
with 64 bit offsets. 

What type should "count" on the new "long" MPI_Datatype constructors be
in this case? The proposal says MPI_Aint but that is awkward for
datatypes as fileviews.

Dick Treumann - MPI Team 
IBM Systems & Technology Group
Dept X2ZA / MS P963 -- 2455 South Road -- Poughkeepsie, NY 12601
Tele (845) 433-7846 Fax (845) 433-8363

mpi-22-bounces_at_[hidden] wrote on 09/03/2008 12:38:35 PM:

> [image removed] 
> 
> Re: [Mpi-22] New proposal: Support for large message counts
> 
> Supalov, Alexander 
> 
> to:
> 
> MPI 2.2
> 
> 09/03/2008 12:48 PM
> 
> Sent by:
> 
> mpi-22-bounces_at_[hidden]
> 
> Please respond to "MPI 2.2"
> 
> Thanks. We may even do without changing the MPI routine names I think.
> Let's consider this:
> 
> 1) #define MPI_Count int will yield the current MPI interface.
> 2) typedef int MPI_Count will yield the new 32-bit interface, still
> backward binary compatible with 1).
> 3) typedef long MPI_Count will yield new 64-bit interface.
> 4) typedef long long MPI_Count will safely bring us into XXII century
> (or earlier).
> 
> Old applications may be rebuilt using 1) without any changes,
> dynamically linked against MPI library built using 2), and will have
to
> be rebuilt to use MPI library built using 3) or 4).
> 
> New applications will use library with the MPI_Count definitions, and,
> properly rebuilt, will work in either 32-, 64-, or 128-bit mode, using
> MPI libraries built using MPI bindings 2), 3), or 4), respectively.
> 
> By the way, should MPI_Count really be signed?
> 
> -----Original Message-----
> From: mpi-22-bounces_at_[hidden]
> [mailto:mpi-22-bounces_at_[hidden]] On Behalf Of Narasimhan,
> Kannan
> Sent: Wednesday, September 03, 2008 5:55 PM
> To: MPI 2.2
> Subject: Re: [Mpi-22] New proposal: Support for large message counts
> 
> We can certainly consider this option, since it abstracts along the
same
> lines as MPI_Offset.
> 
> Thanx!
> Kannan
> 
> -----Original Message-----
> From: mpi-22-bounces_at_[hidden]
> [mailto:mpi-22-bounces_at_[hidden]] On Behalf Of Supalov,
> Alexander
> Sent: Wednesday, September 03, 2008 10:19 AM
> To: MPI 2.2
> Subject: Re: [Mpi-22] New proposal: Support for large message counts
> 
> Dear Kannan,
> 
> Thanks. Why don't we define MPI_Count datatype instead, and make that
> grow if necessary?
> 
> Best regards.
> 
> Alexander
> 
> -----Original Message-----
> From: mpi-22-bounces_at_[hidden]
> [mailto:mpi-22-bounces_at_[hidden]] On Behalf Of Narasimhan,
> Kannan
> Sent: Friday, August 29, 2008 6:15 PM
> To: mpi-22_at_[hidden]
> Subject: [Mpi-22] New proposal: Support for large message counts
> 
> I have submitted a new 2.2 proposal to addresses the need for large
> message counts (i.e. counts greater than the size of 32-bit integer)
for
> MPI calls that communicate messages.
> 
> Please refer to
> https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/mpi22/LargeMsgCounts
> for details.
> 
> Comments and suggestions are welcome.
> 
> Thanx!
> Kannan
> 
> _______________________________________________
> mpi-22 mailing list
> mpi-22_at_[hidden]
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-22
> ---------------------------------------------------------------------
> Intel GmbH
> Dornacher Strasse 1
> 85622 Feldkirchen/Muenchen Germany
> Sitz der Gesellschaft: Feldkirchen bei Muenchen
> Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
> Registergericht: Muenchen HRB 47456 Ust.-IdNr.
> VAT Registration No.: DE129385895
> Citibank Frankfurt (BLZ 502 109 00) 600119052
> 
> This e-mail and any attachments may contain confidential material for
> the sole use of the intended recipient(s). Any review or distribution
by
> others is strictly prohibited. If you are not the intended recipient,
> please contact the sender and delete all copies.
> 
> 
> _______________________________________________
> mpi-22 mailing list
> mpi-22_at_[hidden]
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-22
> 
> _______________________________________________
> mpi-22 mailing list
> mpi-22_at_[hidden]
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-22
> ---------------------------------------------------------------------
> Intel GmbH
> Dornacher Strasse 1
> 85622 Feldkirchen/Muenchen Germany
> Sitz der Gesellschaft: Feldkirchen bei Muenchen
> Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
> Registergericht: Muenchen HRB 47456 Ust.-IdNr.
> VAT Registration No.: DE129385895
> Citibank Frankfurt (BLZ 502 109 00) 600119052
> 
> This e-mail and any attachments may contain confidential material for
> the sole use of the intended recipient(s). Any review or distribution
> by others is strictly prohibited. If you are not the intended
> recipient, please contact the sender and delete all copies.
> 
> 
> _______________________________________________
> mpi-22 mailing list
> mpi-22_at_[hidden]
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-22

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

* 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpi-forum.org/pipermail/mpi-22/attachments/20080904/1938ae4c/attachment.html>