July 2008 - Mpi-21 - mailman3.cels.anl.gov

Re: [Mpi-21] A couple of bugs in 2.1
by unknown＠example.com 25 Jul '08

25 Jul '08

> * pg 332, ln 16-21: "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." Rolf, Ah, if the commutativity is the issue, then what is really wrong is the Rationale, which gives the wrong reason for the exception made for MPI_REPLACE. Of course, the "like the other predefined operations" is also wrong. Regarding MPI_MAX_OBJECT, LaTeX has automatically added it to the table of MPI constants at the end, which is also an errata item. Another very minor typo I discovered today: Last line of pg 250: "coord is not significant" should be "coords is not signficant". The name of the parameter is coords. Rajeev > -----Original Message----- > From: mpi-21-bounces_at_[hidden] > [mailto:mpi-21-bounces_at_[hidden]] On Behalf Of Rolf > Rabenseifner > Sent: Thursday, July 24, 2008 9:39 AM > To: MPI 2.1 Mailing List > Subject: Re: [Mpi-21] A couple of bugs in 2.1 > > Rajeev, > > 1) your first item (MPI_REPLACE in MPI_ACCUMULATE) should you > handle purely as an MPI-2.2 item, because this text is voted > twice as Ballot 4, Item 13. > The MPI-2.1 edited text is therefore correct. > > It should be a new MPI-2.2 discussion, because the two atomic > operations (replace by B) and (replace by C) applied to an > existing window item, are not commutative. > All other operations are commutative > (with exception with MAXLOC and MINLOC on the "location"). > > Therefore, the MPI-2.2 Forum must decide whether current > sentence "MPI_REPLACE, like the other predefined operations, > is defined only for the predefined MPI datatypes." > should read > a. "MPI_REPLACE is > defined only for the predefined MPI datatypes." > or > b. "MPI_REPLACE, like the other predefined operations, is > defined only for the MPI datatypes defined above." > > 2) Your errata to MPI_MAX_OBJECT_NAME is correct. > I have set-up the following wiki MPI-2.1 errata page: > > https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/mpi22/MPI_MA > X_OBJECT_NAME > > I linked this page in both MPI-2.1 errate summary lists, see > end of > https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/MpiTwoTwoWikiPage > > Best regards > Rolf > > > On Mon, 21 Jul 2008 21:18:13 -0500 > "Rajeev Thakur" <thakur_at_[hidden]> wrote: > > Sent it to the wrong list earlier... > > > > -----Original Message----- > >From: owner-mpi-21_at_[hidden] [mailto:owner-mpi-21_at_[hidden]] > >On > > Behalf Of Rajeev Thakur > > Sent: Monday, July 21, 2008 9:12 PM > > To: mpi-21_at_[hidden] > > Subject: [mpi-21] A couple of bugs in 2.1 > > > > I found a couple of bugs in MPI 2.1: > > > > * pg 332, ln 16-21: "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." > > > > > > That is not true. Derived datatypes are allowed in > MPI_Accumulate for > >the predefined reduction operations as explained in the > paragraph just > >above it. > > > > > > * pg 237, line 1 and line 3: MPI_MAX_OBJECT should be > >MPI_MAX_OBJECT_NAME. > > > > Accordingly, the Change Log needs to be updated for both > these items. > > > > The question is can these be fixed before the 2nd vote in > Sept.? My > >understanding of the voting process is that the reason for > having two > >separate votes in two separate meetings is to have a chance to catch > >such bugs. So it should be fixable. (Otherwise, what is the > use of the > >2nd vote?) > > > > Rajeev > > > > > > _______________________________________________ > > mpi-21 mailing list > > mpi-21_at_[hidden] > > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-21 > > Dr. Rolf Rabenseifner . . . . . . . . . .. email > rabenseifner_at_[hidden] 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) > _______________________________________________ > mpi-21 mailing list > mpi-21_at_[hidden] > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-21 >

1 0

Test
by unknown＠example.com 25 Jul '08

25 Jul '08

1 0

test - please ignore
by unknown＠example.com 25 Jul '08

25 Jul '08

Sorry - this is a test mail due to mailing list problems - please ignore Best regards Rolf

1 0

Hardcover Book MPI-2.1
by unknown＠example.com 25 Jul '08

25 Jul '08

==== Please give a short reply if you want to buy the printed ==== ==== hardcover MPI-2.1 at the MPI Forum meeting in Dublin!! ==== = I need to know the number of books that I must ship to Dublin. = ==== Please reply until Tuesday Jul. 29 directly to ==== ==== rabenseifner_at_[hidden] ==== Dear MPI Forum member, after - all the positive votes for MPI-2.1, - a positive discussion in the MPI Forum steering committee, - and a positive response from the EuroPVM/MPI'08 organizers, - the installation of the errata process for MPI-2.1, (see http://www.mpi-forum.org/docs/docs.html and MPI-2.2 wiki pages) I decided to support the MPI community with a printed hardcover MPI-2.1 book (608 pages) - with offset printing and sewn binding, A5 format, and - exact content, page and line numbers of the standard. HLRS will sell these books at cost price (17 Euro). We are not a publisher. We do not sell the book in book stores. It is a special service for large HPC centers in Euro countries. And of course for - all participants of the MPI Forum meeting in Dublin, - all participants of EuroPVM/MPI'08 in Dublin. This book and the electronic counter part are the basis for all MPI-2.2 and MPI-3.0 discussions. To get an idea, how many books I should ship to Dublin, I need to know, when you want to buy one ore more. Please reply also if you are in the European Community and you need a large number for your HPC center. (I need to know it now, because shipping starts Aug. 1.) Best regards Rolf PS: I (and the MPI Forum steering committee) know well that there is still the risk, that MPI-2.1 will never get a standard if the MPI Forum votes "NO" at next meeting. But I'm optimistic. That's my nature. PPS: To keep the price so small and to have it ready at EuroPVM/MPI, and based on the expectation that MPI-2.2 will come end 2009, I decided not to go through a normal publishing process. My apologies for any inconvenience in non-Euro countries. PPPS: I expect that this book will be the first hardcover version of MPI. Please give me details, if I'm wrong. My apologies, if you get this mail several times. The mpi-forum mail distribution made trouble. (This is my 4th trial) Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner_at_[hidden] 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)

1 0

Re: [Mpi-21] A couple of bugs in 2.1
by unknown＠example.com 24 Jul '08

24 Jul '08

Rajeev, 1) your first item (MPI_REPLACE in MPI_ACCUMULATE) should you handle purely as an MPI-2.2 item, because this text is voted twice as Ballot 4, Item 13. The MPI-2.1 edited text is therefore correct. It should be a new MPI-2.2 discussion, because the two atomic operations (replace by B) and (replace by C) applied to an existing window item, are not commutative. All other operations are commutative (with exception with MAXLOC and MINLOC on the "location"). Therefore, the MPI-2.2 Forum must decide whether current sentence "MPI_REPLACE, like the other predefined operations, is defined only for the predefined MPI datatypes." should read a. "MPI_REPLACE is defined only for the predefined MPI datatypes." or b. "MPI_REPLACE, like the other predefined operations, is defined only for the MPI datatypes defined above." 2) Your errata to MPI_MAX_OBJECT_NAME is correct. I have set-up the following wiki MPI-2.1 errata page: https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/mpi22/MPI_MAX_OBJECT_NAME I linked this page in both MPI-2.1 errate summary lists, see end of https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/MpiTwoTwoWikiPage Best regards Rolf On Mon, 21 Jul 2008 21:18:13 -0500 "Rajeev Thakur" <thakur_at_[hidden]> wrote: > Sent it to the wrong list earlier... > > -----Original Message----- >From: owner-mpi-21_at_[hidden] [mailto:owner-mpi-21_at_[hidden]] >On > Behalf Of Rajeev Thakur > Sent: Monday, July 21, 2008 9:12 PM > To: mpi-21_at_[hidden] > Subject: [mpi-21] A couple of bugs in 2.1 > > I found a couple of bugs in MPI 2.1: > > * pg 332, ln 16-21: "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." > > > That is not true. Derived datatypes are allowed in MPI_Accumulate >for the > predefined reduction operations as explained in the paragraph just >above it. > > > * pg 237, line 1 and line 3: MPI_MAX_OBJECT should be >MPI_MAX_OBJECT_NAME. > > Accordingly, the Change Log needs to be updated for both these >items. > > The question is can these be fixed before the 2nd vote in Sept.? My > understanding of the voting process is that the reason for having >two > separate votes in two separate meetings is to have a chance to catch >such > bugs. So it should be fixable. (Otherwise, what is the use of the >2nd vote?) > > Rajeev > > > _______________________________________________ > mpi-21 mailing list > mpi-21_at_[hidden] > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-21 Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner_at_[hidden] 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)

1 0

Re: [Mpi-21] A couple of bugs in 2.1
by unknown＠example.com 23 Jul '08

23 Jul '08

Rajeev, Since this is a regression from MPI 2.0, I think we need to fix this. I do not see a reason why we should not be able to fix these errors before the final vote. Is there someone that disagrees with this ? Rich On 7/21/08 10:18 PM, "Rajeev Thakur" <thakur_at_[hidden]> wrote: > Sent it to the wrong list earlier... > > -----Original Message----- > From: owner-mpi-21_at_[hidden] [mailto:owner-mpi-21_at_[hidden]] On > Behalf Of Rajeev Thakur > Sent: Monday, July 21, 2008 9:12 PM > To: mpi-21_at_[hidden] > Subject: [mpi-21] A couple of bugs in 2.1 > > I found a couple of bugs in MPI 2.1: > > * pg 332, ln 16-21: "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." > > > That is not true. Derived datatypes are allowed in MPI_Accumulate for the > predefined reduction operations as explained in the paragraph just above it. > > > * pg 237, line 1 and line 3: MPI_MAX_OBJECT should be MPI_MAX_OBJECT_NAME. > > Accordingly, the Change Log needs to be updated for both these items. > > The question is can these be fixed before the 2nd vote in Sept.? My > understanding of the voting process is that the reason for having two > separate votes in two separate meetings is to have a chance to catch such > bugs. So it should be fixable. (Otherwise, what is the use of the 2nd vote?) > > Rajeev > > > _______________________________________________ > mpi-21 mailing list > mpi-21_at_[hidden] > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-21 > *

1 0

A couple of bugs in 2.1
by unknown＠example.com 21 Jul '08

21 Jul '08

Sent it to the wrong list earlier... -----Original Message----- From: owner-mpi-21_at_[hidden] [mailto:owner-mpi-21_at_[hidden]] On Behalf Of Rajeev Thakur Sent: Monday, July 21, 2008 9:12 PM To: mpi-21_at_[hidden] Subject: [mpi-21] A couple of bugs in 2.1 I found a couple of bugs in MPI 2.1: * pg 332, ln 16-21: "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." That is not true. Derived datatypes are allowed in MPI_Accumulate for the predefined reduction operations as explained in the paragraph just above it. * pg 237, line 1 and line 3: MPI_MAX_OBJECT should be MPI_MAX_OBJECT_NAME. Accordingly, the Change Log needs to be updated for both these items. The question is can these be fixed before the 2nd vote in Sept.? My understanding of the voting process is that the reason for having two separate votes in two separate meetings is to have a chance to catch such bugs. So it should be fixable. (Otherwise, what is the use of the 2nd vote?) Rajeev

1 0

Re: [Mpi-21] Errata - Re: Updated slides from June meeting / with URL
by unknown＠example.com 09 Jul '08

09 Jul '08

Done. On Jul 9, 2008, at 3:29 AM, Rolf Rabenseifner wrote: > Jeff and all, > > I corrected the error on the "implementation-estimates" slide > (slide no. 8) in > > http://www.hlrs.de/mpi/mpi21/doc/mpi_21_at_MPIforum_2008-06_overview.pdf > and > http://www.hlrs.de/mpi/mpi21/doc/mpi_21_at_MPIforum_2008-06_overview_4to1.p… > > I always asked at the meetings about the implemementation > of the 33 MPI-2.1 issues. I explicitely stated that I do > not ask about implementation of the MPI-2.0 Dynamic Spawn chapter. > > I corrected the text on the slide to reflect this discussions > on the meetings. > > Jeff, please can you copy my both .pdf to your protocol page. > > Thanks and best regards > Rolf > > > On Thu, 03 Jul 2008 18:31:04 +0200 > "Rolf Rabenseifner" <rabenseifner_at_[hidden]> wrote: >> Jeff and all, >> I've updated my slides of June meeting at (user and pw is mpi21) >> http://www.hlrs.de/mpi/mpi21/doc/ >> See section "Protocol of June 30 - July 2, 2008 Meeting". >> Jeff, please can you copy them to the meetings web: >> http://www.hlrs.de/mpi/mpi21/doc/mpi_21_at_MPIforum_2008-06_overview.pdf >> http://www.hlrs.de/mpi/mpi21/doc/mpi_21_at_MPIforum_2008-06_overview_4to1.p… >> Best regards >> Rolf > > Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner_at_[hidden] > 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) > _______________________________________________ > mpi-21 mailing list > mpi-21_at_[hidden] > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-21 -- Jeff Squyres Cisco Systems

1 0

Errata - Re: Updated slides from June meeting / with URL
by unknown＠example.com 09 Jul '08

09 Jul '08

Jeff and all, I corrected the error on the "implementation-estimates" slide (slide no. 8) in http://www.hlrs.de/mpi/mpi21/doc/mpi_21_at_MPIforum_2008-06_overview.pdf and http://www.hlrs.de/mpi/mpi21/doc/mpi_21_at_MPIforum_2008-06_overview_4to1.p… I always asked at the meetings about the implemementation of the 33 MPI-2.1 issues. I explicitely stated that I do not ask about implementation of the MPI-2.0 Dynamic Spawn chapter. I corrected the text on the slide to reflect this discussions on the meetings. Jeff, please can you copy my both .pdf to your protocol page. Thanks and best regards Rolf On Thu, 03 Jul 2008 18:31:04 +0200 "Rolf Rabenseifner" <rabenseifner_at_[hidden]> wrote: > Jeff and all, > > I've updated my slides of June meeting at (user and pw is mpi21) > > http://www.hlrs.de/mpi/mpi21/doc/ > > > See section "Protocol of June 30 - July 2, 2008 Meeting". > > Jeff, please can you copy them to the meetings web: > > http://www.hlrs.de/mpi/mpi21/doc/mpi_21_at_MPIforum_2008-06_overview.pdf > > http://www.hlrs.de/mpi/mpi21/doc/mpi_21_at_MPIforum_2008-06_overview_4to1.p… > > > Best regards > Rolf Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner_at_[hidden] 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)

1 0

Re: [Mpi-21] Updated slides from June meeting / with URL
by unknown＠example.com 03 Jul '08

03 Jul '08

On Thu, 03 Jul 2008 18:12:33 +0200 "Rolf Rabenseifner" <rabenseifner_at_[hidden]> wrote: > Jeff and all, > > I've updated my slides of June meeting at (user and pw is mpi21) http://www.hlrs.de/mpi/mpi21/doc/ > > See section "Protocol of June 30 - July 2, 2008 Meeting". > > Jeff, please can you copy them to the meetings web: http://www.hlrs.de/mpi/mpi21/doc/mpi_21_at_MPIforum_2008-06_overview.pdf http://www.hlrs.de/mpi/mpi21/doc/mpi_21_at_MPIforum_2008-06_overview_4to1.p… > > Best regards > Rolf > > > Dr. Rolf Rabenseifner . . . . . . . . . .. email >rabenseifner_at_[hidden] > 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) > _______________________________________________ > mpi-21 mailing list > mpi-21_at_[hidden] > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-21 Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner_at_[hidden] 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)

1 0