December 2011 - mpi-forum - mailman3.cels.anl.gov

Re: [Mpi-forum] MPI_Initialized, MPI_Query_thread, MPI_Is_thread_main
by Jeff Squyres 16 Dec '11

16 Dec '11

I agree that they are related, but 302 does not introduce any new issues in this arena. I'd be happy to read it, but I don't think there's room on the agenda -- it's already packed with 15 formal readings, last I checked. :-( That being said, since this is not a new problem, my $0.02 is that it could be fixed in 3.1, and therefore not be constrained by the 3.0 January deadline. On Dec 16, 2011, at 11:55 AM, Jeff Hammond wrote: > I can't make it in January. If someone can read it for me, I will write the text. > > These questions arose out of my careful reading of ticket 302, so it would make sense if you were willing to read them, since I think they are related. > > Jeff > > On Fri, Dec 16, 2011 at 10:38 AM, Jeff Squyres <jsquyres(a)cisco.com> wrote: > +1 / good points. > > Will you make a proposal about this? > > > On Dec 16, 2011, at 10:44 AM, Jeff Hammond wrote: > > > Are MPI_Initialized and MPI_Query_thread guaranteed to be thread-safe > > by all implementations? If these MPI calls cannot be made in a fully > > multithreaded context, how can a multithreaded library determine that > > it can make MPI calls in an unrestricted fashion, i.e. determine if > > MPI was initialized with MPI_THREAD_MULTIPLE? Must the multithreaded > > library mutex around these two MPI calls and assume that > > MPI_THREAD_SERIALIZED is always available? > > > > Can MPI_Is_thread_main be called from a thread that is not main when > > only MPI_THREAD_FUNNELED is supported? I recall that the funneled > > mode stipulates that only the main thread can make MPI calls and there > > is no text excepting this routine from that constraint. > > > > Thanks, > > > > Jeff > > > > -- > > Jeff Hammond > > Argonne Leadership Computing Facility > > University of Chicago Computation Institute > > jhammond(a)alcf.anl.gov / (630) 252-5381 > > http://www.linkedin.com/in/jeffhammond > > https://wiki-old.alcf.anl.gov/index.php/User:Jhammond > > _______________________________________________ > > mpi-forum mailing list > > mpi-forum(a)lists.mpi-forum.org > > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: > http://www.cisco.com/web/about/doing_business/legal/cri/ > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > > -- > Jeff Hammond > Argonne Leadership Computing Facility > University of Chicago Computation Institute > jhammond(a)alcf.anl.gov / (630) 252-5381 > http://www.linkedin.com/in/jeffhammond > https://wiki-old.alcf.anl.gov/index.php/User:Jhammond > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Tickets #286 and #305 reading
by Jim Dinan 16 Dec '11

16 Dec '11

Hi All, The collectives working group is pleased to put forth drafts of tickets #286 (MPI_Comm_create_group) and #305 (Update to MPI_Intercomm_create) for formal reading at the January meeting. Text for both tickets is included in the ticket #286 document: https://svn.mpi-forum.org/trac/mpi-forum-web/attachment/ticket/286/ticket_2… Please review the proposed changes and send and feedback. Thanks, ~Jim.

1 0

Re: [Mpi-forum] MPI_Initialized, MPI_Query_thread, MPI_Is_thread_main
by Jeff Hammond 16 Dec '11

16 Dec '11

I can't make it in January. If someone can read it for me, I will write the text. These questions arose out of my careful reading of ticket 302, so it would make sense if you were willing to read them, since I think they are related. Jeff On Fri, Dec 16, 2011 at 10:38 AM, Jeff Squyres <jsquyres(a)cisco.com> wrote: > +1 / good points. > > Will you make a proposal about this? > > > On Dec 16, 2011, at 10:44 AM, Jeff Hammond wrote: > > > Are MPI_Initialized and MPI_Query_thread guaranteed to be thread-safe > > by all implementations? If these MPI calls cannot be made in a fully > > multithreaded context, how can a multithreaded library determine that > > it can make MPI calls in an unrestricted fashion, i.e. determine if > > MPI was initialized with MPI_THREAD_MULTIPLE? Must the multithreaded > > library mutex around these two MPI calls and assume that > > MPI_THREAD_SERIALIZED is always available? > > > > Can MPI_Is_thread_main be called from a thread that is not main when > > only MPI_THREAD_FUNNELED is supported? I recall that the funneled > > mode stipulates that only the main thread can make MPI calls and there > > is no text excepting this routine from that constraint. > > > > Thanks, > > > > Jeff > > > > -- > > Jeff Hammond > > Argonne Leadership Computing Facility > > University of Chicago Computation Institute > > jhammond(a)alcf.anl.gov / (630) 252-5381 > > http://www.linkedin.com/in/jeffhammond > > https://wiki-old.alcf.anl.gov/index.php/User:Jhammond > > _______________________________________________ > > mpi-forum mailing list > > mpi-forum(a)lists.mpi-forum.org > > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: > http://www.cisco.com/web/about/doing_business/legal/cri/ > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > -- Jeff Hammond Argonne Leadership Computing Facility University of Chicago Computation Institute jhammond(a)alcf.anl.gov / (630) 252-5381 http://www.linkedin.com/in/jeffhammond https://wiki-old.alcf.anl.gov/index.php/User:Jhammond

1 0

Re: [Mpi-forum] MPI_Initialized, MPI_Query_thread, MPI_Is_thread_main
by Jeff Squyres 16 Dec '11

16 Dec '11

+1 / good points. Will you make a proposal about this? On Dec 16, 2011, at 10:44 AM, Jeff Hammond wrote: > Are MPI_Initialized and MPI_Query_thread guaranteed to be thread-safe > by all implementations? If these MPI calls cannot be made in a fully > multithreaded context, how can a multithreaded library determine that > it can make MPI calls in an unrestricted fashion, i.e. determine if > MPI was initialized with MPI_THREAD_MULTIPLE? Must the multithreaded > library mutex around these two MPI calls and assume that > MPI_THREAD_SERIALIZED is always available? > > Can MPI_Is_thread_main be called from a thread that is not main when > only MPI_THREAD_FUNNELED is supported? I recall that the funneled > mode stipulates that only the main thread can make MPI calls and there > is no text excepting this routine from that constraint. > > Thanks, > > Jeff > > -- > Jeff Hammond > Argonne Leadership Computing Facility > University of Chicago Computation Institute > jhammond(a)alcf.anl.gov / (630) 252-5381 > http://www.linkedin.com/in/jeffhammond > https://wiki-old.alcf.anl.gov/index.php/User:Jhammond > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

MPI_Initialized, MPI_Query_thread, MPI_Is_thread_main
by Jeff Hammond 16 Dec '11

16 Dec '11

Are MPI_Initialized and MPI_Query_thread guaranteed to be thread-safe by all implementations? If these MPI calls cannot be made in a fully multithreaded context, how can a multithreaded library determine that it can make MPI calls in an unrestricted fashion, i.e. determine if MPI was initialized with MPI_THREAD_MULTIPLE? Must the multithreaded library mutex around these two MPI calls and assume that MPI_THREAD_SERIALIZED is always available? Can MPI_Is_thread_main be called from a thread that is not main when only MPI_THREAD_FUNNELED is supported? I recall that the funneled mode stipulates that only the main thread can make MPI calls and there is no text excepting this routine from that constraint. Thanks, Jeff -- Jeff Hammond Argonne Leadership Computing Facility University of Chicago Computation Institute jhammond(a)alcf.anl.gov / (630) 252-5381 http://www.linkedin.com/in/jeffhammond https://wiki-old.alcf.anl.gov/index.php/User:Jhammond

1 0

Formal reading: ref-counted INIT/FINALIZE
by Jeff Squyres 16 Dec '11

16 Dec '11

Forum -- A formal reading of ticket #302 ("Reference counted MPI_INIT/MPI_FINALIZE") is on the agenda for this January. The text is based on the feedback that the Forum gave at the July 2011 Chicago meeting. Basically, there was strong favor for the idea of reference-counting INIT/FINALIZE by using the existing INIT/INIT_THREAD/FINALIZE routines (vs. making new, "safe" versions of INIT/FINALIZE). I have uploaded a PDF of the proposal to the ticket: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/302 The usual mechanism of searching for "ticket 302" in the PDF will reveal all the updated and new text. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Final Fortran #229 Version for voting
by Rolf Rabenseifner 15 Dec '11

15 Dec '11

Dear all togehter, As discussed at the last Forum meeting, I provided the Dec. 15 pdf that includes all data from the three documents used in the formal readings. For further information, please have a look at the latest attachments at https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/229 Or at the latest comment on this ticket: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/229#comment:20 Voting will be in January and March 2012. Best regards and thank you very much for all your help and patience Rolf -- Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)hlrs.de 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

RMA WG Readings
by Torsten Hoefler 13 Dec '11

13 Dec '11

Hello All, Here is a quick overview of the RMA group's proposals for the next meeting. Readings: 1a) https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/284 (Tue 2:30-4:30) (the RMA WG found that it is appropriate to prohibit MPI_WIN_LOCK_ALL(exclusive), i.e., we need to re-read with this change) 2) minor corrections to the chapter 2a) https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/309 (Tue 2-2:15) 2b) https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/308 (Tue 2:15-2:30) 2c) https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/283 (Tue 3:30-4:30) (tagged on to #284) Votes: 3) https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/270 (Mon 4-4:15) 4) https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/299 (Mon 4:15-4:30) 5) https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/298 (Mon 4:30-4:45) Thanks & All the Best, Torsten -- bash$ :(){ :|:&};: --------------------- 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

1 0

Collective WG Readings
by Torsten Hoefler 13 Dec '11

13 Dec '11

Hello All, In order to prepare for the next MPI Forum meeting readings, please consider carefully reviewing the two tickets brought forward by the collective WG: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/168 (Tue 5-6) and https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/264 (Wed 1-2) Both are ready for a reading, comments are welcome anytime. Thanks & Best, Torsten -- bash$ :(){ :|:&};: --------------------- 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

1 0

Removing MPI-1 functions
by Fab Tillier 12 Dec '11

12 Dec '11

Hi Folks, Thanks to Jeff, I've uploaded PDFs for: - ticket 278 (https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/278) to update all examples to use non-deprecated functions, and remove "This function replaces the deprecated function..." text. - ticket 303 (https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/303) to remove the "easy" MPI-1 bindings. I know these aren't on the schedule for the January meeting due to schedule constraints. I'd still appreciate any input folks have. I am going to continue working on ticket 304, "Fix the attribute disaster", and will update the mailing list when that PDF is ready for review. The tickets build on one another, so any issues with 278 would block 303, and issues with 303 would block 304, hence the request for feedback now. Note that ticket 303 includes the ticket 278 changes. I'm still hopeful that these will end up part of the 3.0 standard, as removing functions in 3.1 would be a bit awkward. Cheers, -Fab

1 0