January 2009 - Mpi-22 - mailman3.cels.anl.gov

Re: [Mpi-22] Please review tickets
by unknown＠example.com 29 Jan '09

29 Jan '09

I think the best way might be to search for your Trac ID and/or email address using the "search" box. On the result page of the search, if you turn up more info than you need, you can limit the search to just tickets. On Jan 29, 2009, at 2:33 PM, Richard Treumann wrote: > Jeff - Do you know of a way for a Forum member to pull a list of > tickets he is obliged to review? > > 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 01/29/2009 12:57:39 PM: > > > [image removed] > > > > [Mpi-22] Please review tickets > > > > Jeff Squyres > > > > to: > > > > MPI 2.2 > > > > 01/29/2009 01:01 PM > > > > Sent by: > > > > mpi-22-bounces_at_[hidden] > > > > Please respond to "MPI 2.2" > > > > Hi all. Don't forget that we are about 1.5 weeks away from the San > > Jose meeting. > > > > If you are a reviewer, please review your MPI-2.2 tickets so that we > > can discuss them at the meeting. > > > > Thanks! > > > > -- > > Jeff Squyres > > Cisco Systems > > > > _______________________________________________ > > 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 -- Jeff Squyres Cisco Systems

1 0

Re: [Mpi-22] Please review tickets
by unknown＠example.com 29 Jan '09

29 Jan '09

Jeff - Do you know of a way for a Forum member to pull a list of tickets he is obliged to review? 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 01/29/2009 12:57:39 PM: > [image removed] > > [Mpi-22] Please review tickets > > Jeff Squyres > > to: > > MPI 2.2 > > 01/29/2009 01:01 PM > > Sent by: > > mpi-22-bounces_at_[hidden] > > Please respond to "MPI 2.2" > > Hi all. Don't forget that we are about 1.5 weeks away from the San > Jose meeting. > > If you are a reviewer, please review your MPI-2.2 tickets so that we > can discuss them at the meeting. > > Thanks! > > -- > Jeff Squyres > Cisco Systems > > _______________________________________________ > mpi-22 mailing list > mpi-22_at_[hidden] > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-22 *

1 0

Please review tickets
by unknown＠example.com 29 Jan '09

29 Jan '09

Hi all. Don't forget that we are about 1.5 weeks away from the San Jose meeting. If you are a reviewer, please review your MPI-2.2 tickets so that we can discuss them at the meeting. Thanks! -- Jeff Squyres Cisco Systems

1 0

Re: [Mpi-22] [MPI Forum] #55: MPI-2.1 Cross-language attribute example is wrong
by unknown＠example.com 27 Jan '09

27 Jan '09

Probably the least damaging thing to do now is to make them MPI_Fints - this (probably) will have no affect in any current implementation, and is consistent with most of the text. I agree that it would be good to deprecate these values, even if we don't replace them in 2.2 . Bill On Jan 27, 2009, at 5:08 PM, Jeff Squyres wrote: > On Jan 27, 2009, at 5:56 PM, William Gropp wrote: > >> Eek! Not for 2.2 ! Maybe a historical advice to users .... > > There are two issues: > > 1. The current text about the C type for predefined attributes is > erroneous. > > p487:46 through 488-3 says that predefined attributes were set by "a > Fortran call" (it doesn't say which Fortran call), but then says the > result in C is of type int. There is no Fortran attribute call that > sets an attribute value that is exactly equivalent to a C int. That's > why MPI-2 created MPI_Fint, right? > > So I think that we must do *something* about this issue. > > 2. Deprecating the old attribute names now (as you suggested) is not a > Bad Thing. Right now, we have a discontinuity of deprecations: some > functions are deprecated, but values that are supposedly set by those > functions are *not* deprecated. If we deprecate those values now and > provide replacements, it opens the door in MPI-3 (or later) to > actually whack them from the spec and only use the new names. > Specifically: deprecating the old names doesn't mean that we have to > break user apps for MPI-2.2. > > -- > Jeff Squyres > Cisco Systems > > _______________________________________________ > mpi-22 mailing list > mpi-22_at_[hidden] > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-22 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-22] [MPI Forum] #55: MPI-2.1 Cross-language attribute example is wrong
by unknown＠example.com 27 Jan '09

27 Jan '09

On Jan 27, 2009, at 5:56 PM, William Gropp wrote: > Eek! Not for 2.2 ! Maybe a historical advice to users .... There are two issues: 1. The current text about the C type for predefined attributes is erroneous. p487:46 through 488-3 says that predefined attributes were set by "a Fortran call" (it doesn't say which Fortran call), but then says the result in C is of type int. There is no Fortran attribute call that sets an attribute value that is exactly equivalent to a C int. That's why MPI-2 created MPI_Fint, right? So I think that we must do *something* about this issue. 2. Deprecating the old attribute names now (as you suggested) is not a Bad Thing. Right now, we have a discontinuity of deprecations: some functions are deprecated, but values that are supposedly set by those functions are *not* deprecated. If we deprecate those values now and provide replacements, it opens the door in MPI-3 (or later) to actually whack them from the spec and only use the new names. Specifically: deprecating the old names doesn't mean that we have to break user apps for MPI-2.2. -- Jeff Squyres Cisco Systems

1 0

Re: [Mpi-22] [MPI Forum] #55: MPI-2.1 Cross-language attribute example is wrong
by unknown＠example.com 27 Jan '09

27 Jan '09

Eek! Not for 2.2 ! Maybe a historical advice to users .... Bill On Jan 27, 2009, at 4:49 PM, Jeff Squyres wrote: > On Jan 27, 2009, at 5:35 PM, William Gropp wrote: > >> On the last point, the real answer is that since the attributes >> behave as if they were set by the now deprecated functions, they >> should themselves be deprecated, since we must not change their >> behavior. This is also the source of the int/MPI_Fint issue - there >> was no MPI_Fint in MPI-1. > > Wow, the scope of this ticket keeps expanding. :-) > > I see the following attributes in OMPI's mpi.h: > > /* MPI-1 */ > MPI_TAG_UB, > MPI_HOST, > MPI_IO, > MPI_WTIME_IS_GLOBAL, > > /* MPI-2 */ > MPI_APPNUM, > MPI_LASTUSEDCODE, > MPI_UNIVERSE_SIZE, > MPI_WIN_BASE, > MPI_WIN_SIZE, > MPI_WIN_DISP_UNIT, > > Excluding MPI_WIN_BASE (which, IIRC, is the only address-sized > attribute in this list), if we deprecate these names, I assume we'll > simply replace them with new names that behave as if they were set > from Fortran MPI_COMM_SET_ATTR. Specifically, the sizes of the values > will be such that you have to access them with > INTEGER(KIND=MPI_ADDRESS_KIND) and (MPI_Aint*). > > Is that what you're thinking? If so, how about s/MPI_/MPI_ATTR_/ in > all the names (perhaps also creating an alias for MPI_WIN_BASE -> > MPI_ATTR_WIN_BASE just for symmetry)? I'm open to suggestions for new > names. > > I'd almost prefer to make that a separate (but related) ticket -- the > issues in #55 are already quite tangled... (i.e., leave out any > mention of the predefined attributes from #55 and have a new ticket > for those issues) Sound reasonable? > > -- > Jeff Squyres > Cisco Systems > > _______________________________________________ > mpi-22 mailing list > mpi-22_at_[hidden] > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-22 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-22] [MPI Forum] #55: MPI-2.1 Cross-language attribute example is wrong
by unknown＠example.com 27 Jan '09

27 Jan '09

On Jan 27, 2009, at 5:35 PM, William Gropp wrote: > On the last point, the real answer is that since the attributes > behave as if they were set by the now deprecated functions, they > should themselves be deprecated, since we must not change their > behavior. This is also the source of the int/MPI_Fint issue - there > was no MPI_Fint in MPI-1. Wow, the scope of this ticket keeps expanding. :-) I see the following attributes in OMPI's mpi.h: /* MPI-1 */ MPI_TAG_UB, MPI_HOST, MPI_IO, MPI_WTIME_IS_GLOBAL, /* MPI-2 */ MPI_APPNUM, MPI_LASTUSEDCODE, MPI_UNIVERSE_SIZE, MPI_WIN_BASE, MPI_WIN_SIZE, MPI_WIN_DISP_UNIT, Excluding MPI_WIN_BASE (which, IIRC, is the only address-sized attribute in this list), if we deprecate these names, I assume we'll simply replace them with new names that behave as if they were set from Fortran MPI_COMM_SET_ATTR. Specifically, the sizes of the values will be such that you have to access them with INTEGER(KIND=MPI_ADDRESS_KIND) and (MPI_Aint*). Is that what you're thinking? If so, how about s/MPI_/MPI_ATTR_/ in all the names (perhaps also creating an alias for MPI_WIN_BASE -> MPI_ATTR_WIN_BASE just for symmetry)? I'm open to suggestions for new names. I'd almost prefer to make that a separate (but related) ticket -- the issues in #55 are already quite tangled... (i.e., leave out any mention of the predefined attributes from #55 and have a new ticket for those issues) Sound reasonable? -- Jeff Squyres Cisco Systems

1 0

Re: [Mpi-22] [MPI Forum] #55: MPI-2.1 Cross-language attribute example is wrong
by unknown＠example.com 27 Jan '09

27 Jan '09

On the last point, the real answer is that since the attributes behave as if they were set by the now deprecated functions, they should themselves be deprecated, since we must not change their behavior. This is also the source of the int/MPI_Fint issue - there was no MPI_Fint in MPI-1. Bill On Jan 27, 2009, at 4:09 PM, MPI Forum wrote: > #55: MPI-2.1 Cross-language attribute example is wrong > ----------------------------------- > +---------------------------------------- > Reporter: jsquyres | Owner: jsquyres > Type: Correction to standard | Status: new > Priority: Waiting for reviews | Milestone: 2008/12/15 Menlo > Park > Version: MPI 2.2 | Resolution: > Keywords: | > ----------------------------------- > +---------------------------------------- > > Comment(by jsquyres): > > I am re-working this proposal and will post a new one soon. Some > things > that have been consistently missed by multiple reviewers: > > * The original example is '''definitely wrong'''. > MPI_COMM_PUT_ATTR does > not exist; the correct call is MPI_COMM_SET_ATTR. > * The original example is '''definitely wrong'''. Example B sets an > INTEGER(KIND=MPI_ADDRESS_KIND) value, and therefore the output C > type must > be MPI_Aint* (not (int*)) -- regardless of the behavior of predefined > attributes. > * If predefined attributes are supposed to behave as if they were > set via > the (deprecated!) Fortran MPI-1 function, then the value when > retrieved in > C (e.g., MPI_TAG_UB) should actually be of type MPI_Fint, not int. I > think that the text in MPI-2.1 p488:3 is wrong on this point (i.e., > s/int/MPI_Fint/). > > Per the last point, I think it is pretty good luck that existing MPI > apps > work when using (int*) instead of (MPI_Fint*) -- indeed, most > compilers > have the default sizeof(INTEGER) == sizeof(int). But that does not > have > to be the case; MPI_Fint is technically more correct. > > Also per the last point, why are we specifying that predefined > attributes > behave as if they were set by deprecated functions? That seems > strange. > > -- > Ticket URL: <https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/55#comment:21 > > > MPI Forum <https://svn.mpi-forum.org/> > 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-22] [MPI Forum] #96: MPI_OP_CREATE and associativity
by unknown＠example.com 19 Jan '09

19 Jan '09

On Tue, Jan 13, 2009 at 09:23:20AM +0100, Jesper Larsson Traeff wrote: > maybe I misunderstand you, but I don't see that allowing operators to > be explicitly non-associative would give you more freedom for optimizations. > Doesn't your arguments above show that there are rather less than for > associative ops? But maybe it is better to discuss this at the next meeting? just for the other readers: we clarified the discussion during the last Collectives Workinggroup Teleconference. And decided to update ticket #96 and merge ticket #95 into it. For details see: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/96 > If your argument is: "there are natural (user-defined) operators in app > A, B and C that are mathematically not associative, and users want to perform > reductions with these" - then I agree that MPI is missing something. yes, exactly. Such operators are MPI_SUM of MPI_FLOAT and friends. > Maybe an "MPI_Op_create_nonassoc(...)"? (I also don't think an extra function > for setting the associativity of an already defined function fits with MPI; > although a function like MPI_Op_use_some_special_canonical_order(MPI_Op f), > where f could be a user as well as a predefined MPI_Op, > could perhaps make sense, and would allow to control the evaluation order > each time MPI_Reduce/MPI_Allreduce/... is called, but then such a function > would either have to be collective or users would be required to use it > consistently, ... that is, many complications) exactly, this is why the whole discussion was moved from the context of MPI-2.2 to MPI-3 into the collective working group. > Note also that non-associativity does put some non-trivial burden on > implementers: for non-associative (with or without hyphen?) operators, > a special algorithm that does the reduction in the sequential order is needed I don't see a big/non-trivial problem. I copied the collective working group list in order to take further discussion off the MPI-2.2 list. All the Best, Torsten -- bash$ :(){ :|:&};: --------------------- http://www.unixer.de/ ----- Torsten Hoefler | Postdoctoral Researcher Open Systems Lab | Indiana University 150 S. Woodlawn Ave. | Bloomington, IN, 474045, USA Lindley Hall Room 135 | +01 (812) 855-3608 *

1 0

Re: [Mpi-22] [MPI Forum] #96: MPI_OP_CREATE and associativity
by unknown＠example.com 13 Jan '09

13 Jan '09

On Mon, Jan 12, 2009 at 02:05:47PM -0500, Torsten Hoefler wrote: > > > > yes, they can still be treated as associative (for all internal types), > > > but thet user should have a chance to define non-associative reductions. > > > > > In the apps you refer to, I guess they often want an MPI_SUM on an MPI_FLOAT > > done in some particular order with a particular bracketing. Thus, what > > you need is some ability to control each individual MPI_Reduce call? > yes, but I guess controlling each call is not necessary. However, > controlling the order of operations might be necessary to ensure > correctness. But I don't know a particular app, so the whole argument is > weak. I just want to discuss the issue and either understand why we > don't have this feature or discuss its addition. My point is that there > are (limited) optimization possibilities and that we limit ourselves > enourmously by enforcing the same reduction order for all float/double > operations and all user-defined operations in MPI-2.1. If associativity > would be an explicit property of a user-defined operation, then we could > apply dynamic optimizations based on arrival time, which we can't right > now (side-note: we can do this for associative ops, such as all ops on > non-fp types). > maybe I misunderstand you, but I don't see that allowing operators to be explicitly non-associative would give you more freedom for optimizations. Doesn't your arguments above show that there are rather less than for associative ops? But maybe it is better to discuss this at the next meeting? If your argument is: "there are natural (user-defined) operators in app A, B and C that are mathematically not associative, and users want to perform reductions with these" - then I agree that MPI is missing something. Maybe an "MPI_Op_create_nonassoc(...)"? (I also don't think an extra function for setting the associativity of an already defined function fits with MPI; although a function like MPI_Op_use_some_special_canonical_order(MPI_Op f), where f could be a user as well as a predefined MPI_Op, could perhaps make sense, and would allow to control the evaluation order each time MPI_Reduce/MPI_Allreduce/... is called, but then such a function would either have to be collective or users would be required to use it consistently, ... that is, many complications) Note also that non-associativity does put some non-trivial burden on implementers: for non-associative (with or without hyphen?) operators, a special algorithm that does the reduction in the sequential order is needed Jesper

1 0