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

Re: [Mpi-forum] Sparse / neighborhood topology ticket
by Jeff Squyres 08 Feb '11

08 Feb '11

On Feb 8, 2011, at 11:17 AM, Torsten Hoefler wrote: > I feel like we need more reviews for this proposal. It is pretty short > and obvious ... The use of the word "obvious" here makes me smile. :-) -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Re: [Mpi-forum] Sparse / neighborhood topology ticket
by Torsten Hoefler 08 Feb '11

08 Feb '11

On Tue, Feb 08, 2011 at 05:27:10AM -0800, Jeff Squyres wrote: > FYI: I talked with Torsten last night after the meeting and raised the > point that the overall Forum wasn't presented with the final text of > the ticket 2 weeks prior to the meeting. Torsten immediately agreed > and suggested that we reset the clock and have a(nother) first reading > next meeting -- including the minor changes suggested from last night. I agree. I received a couple of more comments from Jeff and Fab (thanks!). All complaints are small issues (no change in semantics), however, the sum of those deserve a clock reset. I will not be able to fix them today either, so we should read again at the next meeting (not sure if we actually want a full reading or if we consider this a waste of time). I feel like we need more reviews for this proposal. It is pretty short and obvious but I'd like to use the same care as we used for the nonblocking collectives. Please review! 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

Sparse / neighborhood topology ticket
by Jeff Squyres 08 Feb '11

08 Feb '11

FYI: I talked with Torsten last night after the meeting and raised the point that the overall Forum wasn't presented with the final text of the ticket 2 weeks prior to the meeting. Torsten immediately agreed and suggested that we reset the clock and have a(nother) first reading next meeting -- including the minor changes suggested from last night. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Sparse Collective Proposal Updated
by Torsten Hoefler 07 Feb '11

07 Feb '11

Hello Forum, I received numerous comments on the sparse collectives proposal. Special thanks go to Rolf who designed an excellent example and provided many fixes to the chapter. I also decided to rename the collective from "Sparse" collectives to "Sparse Neighborhood" collectives and all functions are now called MPI_Neighbor_{alltoall,...}. The updated proposal is attached to: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/258 It would be great to receive more reviews. Any comments are appreciated! 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

Meeting today at 1pm
by Jeff Squyres 07 Feb '11

07 Feb '11

We are meeting in the same building at Cisco as last time: McCarthy Ranch building #3 (same big room, too). It's the 3rd building in the group of Cisco buildings at 155 McCarthy Ranch, Milipitas, CA, 95035. No badges are needed to get into the Cisco building, but if you're out in the lobby and need an escort in, call me -- I've written my phone number on the white board in the lobby. Please try to arrive around meeting start or break times. Thanks! -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Re: [Mpi-forum] MPI_UNDEFINED
by Jeff Squyres 04 Feb '11

04 Feb '11

On Feb 4, 2011, at 3:51 PM, N.M. Maclaren wrote: > In any case, it isn't a problem. The first paragraph of A.1.1 states > that a const int may be a literal integer constant. Perfect -- I missed that. Thanks! -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Re: [Mpi-forum] MPI_UNDEFINED
by N.M. Maclaren 04 Feb '11

04 Feb '11

On Feb 4 2011, Jeff Squyres wrote: > Question about MPI_UNDEFINED (and all the constants in the top table in > MPI-2.2 p515): its type in C is listed as (const int) or unnamed (enum). > But in Open MPI, we have this value in mpi.h as: > > #define MPI_UNDEFINED -32766 > >So does MPICH2. > > A #define of an integer constant will work just fine with other const > int's, but it technically isn't a const int or unnamed enum. Er, technically there isn't any such animal as a const int type! const is a type qualifier, that has effect only on lvalues (and which is ignored on rvalues). > Are we wrong? Or should I propose a ticket to change the wording of this > table from: > > C type: const int (or unnamed enum) > >to > > C type: const int, unnamed enum, or integer constant That isn't any better - an integer constant isn't a type, and you need to be specific about what you mean by one of those if you are nitpicking, anyway - C has several different integer constant concepts. In any case, it isn't a problem. The first paragraph of A.1.1 states that a const int may be a literal integer constant. While the wording isn't ideal, there are a LOT of places where the MPI specification uses dubious (but not misleading) Fortran and C terminology. Regards, Nick Maclaren.

1 0

Re: [Mpi-forum] MPI_UNDEFINED
by William Gropp 04 Feb '11

04 Feb '11

This should be a proper ticket - its not "minor" in the way we define minor. Bill On Feb 4, 2011, at 2:14 PM, Jeff Squyres wrote: > Is this a "minor" issue such that it could be done in 1 reading, or > should I do a proper ticket + formal reading + 2 votes? > > (either is fine with me; I'm just asking for clarification) > > > On Feb 4, 2011, at 3:07 PM, William Gropp wrote: > >> I recommend a change to the standard - I'm sure the Forum intended >> integer constants to be valid here. >> >> Bill >> >> On Feb 4, 2011, at 1:18 PM, Jeff Squyres wrote: >> >>> Question about MPI_UNDEFINED (and all the constants in the top >>> table in MPI-2.2 p515): its type in C is listed as (const int) or >>> unnamed (enum). But in Open MPI, we have this value in mpi.h as: >>> >>> #define MPI_UNDEFINED -32766 >>> >>> So does MPICH2. >>> >>> A #define of an integer constant will work just fine with other >>> const int's, but it technically isn't a const int or unnamed enum. >>> >>> Are we wrong? Or should I propose a ticket to change the wording >>> of this table from: >>> >>> C type: const int (or unnamed enum) >>> >>> to >>> >>> C type: const int, unnamed enum, or integer constant >>> >>> (the same issue doesn't exist for the C++ constants -- even though >>> the point is moot with them, since they're deprecated -- because >>> they're in the MPI:: namespace, and therefore really should be >>> const int's or unnamed enum's) >>> >>> -- >>> 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 >> >> William Gropp >> Deputy Director for Research >> Institute for Advanced Computing Applications and Technologies >> Paul and Cynthia Saylor Professor of Computer Science >> University of Illinois Urbana-Champaign >> >> >> >> >> _______________________________________________ >> 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 William Gropp Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign

1 0

Re: [Mpi-forum] MPI_UNDEFINED
by Jeff Squyres 04 Feb '11

04 Feb '11

Is this a "minor" issue such that it could be done in 1 reading, or should I do a proper ticket + formal reading + 2 votes? (either is fine with me; I'm just asking for clarification) On Feb 4, 2011, at 3:07 PM, William Gropp wrote: > I recommend a change to the standard - I'm sure the Forum intended integer constants to be valid here. > > Bill > > On Feb 4, 2011, at 1:18 PM, Jeff Squyres wrote: > >> Question about MPI_UNDEFINED (and all the constants in the top table in MPI-2.2 p515): its type in C is listed as (const int) or unnamed (enum). But in Open MPI, we have this value in mpi.h as: >> >> #define MPI_UNDEFINED -32766 >> >> So does MPICH2. >> >> A #define of an integer constant will work just fine with other const int's, but it technically isn't a const int or unnamed enum. >> >> Are we wrong? Or should I propose a ticket to change the wording of this table from: >> >> C type: const int (or unnamed enum) >> >> to >> >> C type: const int, unnamed enum, or integer constant >> >> (the same issue doesn't exist for the C++ constants -- even though the point is moot with them, since they're deprecated -- because they're in the MPI:: namespace, and therefore really should be const int's or unnamed enum's) >> >> -- >> 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 > > William Gropp > Deputy Director for Research > Institute for Advanced Computing Applications and Technologies > Paul and Cynthia Saylor Professor of Computer Science > University of Illinois Urbana-Champaign > > > > > _______________________________________________ > 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

Re: [Mpi-forum] MPI_UNDEFINED
by William Gropp 04 Feb '11

04 Feb '11

I recommend a change to the standard - I'm sure the Forum intended integer constants to be valid here. Bill On Feb 4, 2011, at 1:18 PM, Jeff Squyres wrote: > Question about MPI_UNDEFINED (and all the constants in the top table > in MPI-2.2 p515): its type in C is listed as (const int) or unnamed > (enum). But in Open MPI, we have this value in mpi.h as: > > #define MPI_UNDEFINED -32766 > > So does MPICH2. > > A #define of an integer constant will work just fine with other > const int's, but it technically isn't a const int or unnamed enum. > > Are we wrong? Or should I propose a ticket to change the wording of > this table from: > > C type: const int (or unnamed enum) > > to > > C type: const int, unnamed enum, or integer constant > > (the same issue doesn't exist for the C++ constants -- even though > the point is moot with them, since they're deprecated -- because > they're in the MPI:: namespace, and therefore really should be const > int's or unnamed enum's) > > -- > 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 William Gropp Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign

1 0