Sure, done, see comment #12. Thanks, Torsten On Mon, Jul 16, 2012 at 12:31:55AM +0200, Rolf Rabenseifner wrote:
Torsten,
please comment on ticket 170 why the sentence
The routine is called by all group members using the same arguments for count, datatype, op and comm.
should be wrong for MPI_SCAN and MPI_EXSCAN while this sentence is true for MPI_REDUCE and MPI_ALLREDUCE.
Without details, I would expect that this ticket 170 item is absolutely correct and should be done as a ticket 0 change to clarify which rules are valid for the scan routines. All other reductions are clear about this.
I do not expect that you want to allow that one process uses MPI_FLOAT, another MPI_DOUBLE and both use MPI_SUM. I'm not sure wether current text does not allow this. Therefore clarifying wording is needed.
I would like to see the arguments and proposals in the comments to #170, rather than only discussion at the meeting.
Best regards Rolf
----- Original Message -----
From: "Torsten Hoefler" <[email protected]> To: "Rolf Rabenseifner" <[email protected]> Cc: "George Bosilca" <[email protected]> Sent: Sunday, July 15, 2012 11:24:35 PM Subject: Re: Ticket 170, comment 9, both items Rolf,
This is in no way a ticket 0, it changes semantics and I know of a code that will break if we change this. I would at least like to see a discussion in the Forum.
Thanks, Torsten
On Sat, Jul 14, 2012 at 12:14:04PM +0200, Rolf Rabenseifner wrote:
Torsten and George,
please commit attached coll.tex and inquiry.tex as soon as possible with (Torsten, please after the coll.tex in my previous Ticket 166 mail)
Torsten: svn ci coll.tex 'Ticket 170, Item 1 - ticket-0-change marked with 0.170'
George: svn ci inquiry.tex 'Ticket 170, Item 2 - ticket-0-change marked with 0.170'
Torsten, the added sentence is the same as for MPI_REDUCE (but without root). It is not needed in MPI_ALLREDUCE due to the very strong MPI-2.2 p175:42-43. But in MPI_SCAN and EXSCAN it was missing and is now added. It is ticket-0 because it does not hurt any implementation or application and it is needed because it is undefined in MPI-2.2.
George, your change is done already partially a few lines above. Also ticket-0.
(I currently try to fix some overseen errors known for years. Ticket 170 is one of these error lists.)
Best regards Rolf
-- Dr. Rolf Rabenseifner . . . . . . . . . .. email [email protected] High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner Nobelstr. 19, D-70550 Stuttgart, Germany . (Office: Allmandring 30)
-- ### qreharg rug ebs fv crryF ------------- http://www.unixer.de/ ----- Linux is like a wigwam: no Windows, no Gates, Apache inside.
-- Dr. Rolf Rabenseifner . . . . . . . . . .. email [email protected] High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner Nobelstr. 19, D-70550 Stuttgart, Germany . (Office: Allmandring 30)
-- ### qreharg rug ebs fv crryF ------------- http://www.unixer.de/ ----- Torsten Hoefler | Performance Modeling and Simulation Lead Blue Waters Directorate | University of Illinois (UIUC) 1205 W Clark Street | Urbana, IL, 61801 NCSA Building | +01 (217) 244-7736