November 2012 - mpi-forum - mailman3.cels.anl.gov

Meeting locations next week
by Jeff Squyres 30 Nov '12

30 Nov '12

Note that we are in multiple different Cisco buildings next week. I have put a map up on the logistics page: http://meetings.mpi-forum.org/Dec_2012_logistics.php The meeting Monday starts at 1pm in Cisco building 24 (we have not met in this building before): 510 McCarthy Blvd. Milpitas, California 95035 -- 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] Implementing and testing the new mpi_f08 module and updating the existing mpi module
by Rolf Rabenseifner 29 Nov '12

29 Nov '12

Dear Bill, good catch! TYPE(MPI_Message) must be part on the list on p677:4-13. I included the correction into the #345 errata ticket > > https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/345 Thanks and best regards Rolf ----- Original Message ----- > From: "Bill Long" <longb(a)cray.com> > To: "Rolf Rabenseifner" <rabenseifner(a)hlrs.de> > Cc: "Main MPI Forum mailing list" <mpi-forum(a)lists.mpi-forum.org>, "MPI-3 Fortran working group" > <mpi3-fortran(a)lists.mpi-forum.org>, "Pavan Balaji" <balaji(a)mcs.anl.gov>, "Craig Rasmussen" <rasmus(a)uoregon.edu>, "Jeff > Squyres" <jsquyres(a)cisco.com>, "Rajeev Thakur" <thakur(a)mcs.anl.gov>, "Dave Goodell" <goodell(a)mcs.anl.gov>, "Darius > Buntinas" <buntinas(a)mcs.anl.gov> > Sent: Thursday, November 29, 2012 12:37:40 AM > Subject: Re: Implementing and testing the new mpi_f08 module and updating the existing mpi module > Hi Rolf, > > I may be looking in the wrong pages, but in A.1.1 there are places > (bottom of page 663, for example) where it is assumed there is a > Fortran derived type named > > MPI_Message > > but the Table at the end of A.1.2 (page 677) does not list a > > TYPE(MPI_Message) > > Perhaps it is specified somewhere else, but it would be convenient if > it > were in 677 instead. > > Cheers, > Bill > > > On 11/25/12 6:16 AM, Rolf Rabenseifner wrote: > > Dear MPI implementers, > > > > I updated the ticket #345 that should be voted at the MPI Forum > > meeting. > > https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/345 > > > > - It contains a full test environment for the new mpi_f08 module > > and additional help for doing the same with the updated mpi > > module, > > see attachment > > https://svn.mpi-forum.org/trac/mpi-forum-web/attachment/ticket/345/mpi-3.0-… > > > > - It also contains all errata about the mpi_f08 interfaces that > > are known so far. The updated pdf can be found at > > https://svn.mpi-forum.org/trac/mpi-forum-web/attachment/ticket/345/mpi30%2B… > > > > The files in mpi-3.0-final-mpi-f08-interfaces.tar.gz are consistent > > with MPI-3.0 + Ticket #345. > > They are tested with the current PGI, Cray, and Intel compiler on > > our Cray XE6. > > All test are only compile and lik test, not executable tests. > > They are important to control the used argument lists in your > > mpi_f08 module > > file, because sequence, types, AND argument names must correct. > > The correctness and checking of the argument names is new with > > MPI-3.0. > > Please ignore the warnings from the Intel compiler; they occure > > because > > my dummy library is doing "less than nothing", i.e., does not return > > any > > value in the INTENT(OUT) arguments. > > > > I'm offline the next 3 days and not in San Jose. I hope, you all > > help that #345 will be done at the upcoming meeting. All further > > infos > > can be found > > - in the ticket #345 text, and > > - in the README in the tar.gz . > > > > Best regards > > Rolf > > > > > > -- > Bill Long longb(a)cray.com > Fortran Technical Support & voice: 651-605-9024 > Bioinformatics Software Development fax: 651-605-9142 > Cray Inc./Cray Plaza, Suite 210/380 Jackson St./St. Paul, MN 55101 -- 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: Room 1.307)

1 0

Callin numbers for Forum meetings
by Richard Graham 28 Nov '12

28 Nov '12

I have been requested to provide call in capabilities to two session at the upcoming meeting - the Tuesday AM FT discussion, and the Wed am FT/RMA discussion (note the time change to accommodate folks from Europe). Are there other sessions that people would want to call into ? If so, please let Jeff and me know, so that Jeff can make the appropriate arrangements. Thanks, Rich

1 0

Re: [Mpi-forum] [EXTERNAL] Wording in MPI standard
by Jeff Hammond 27 Nov '12

27 Nov '12

> I think we've done that kind of thing before (i.e., put a blanket statement in the T&C chapter, instead of fixing every individual instance). > > Do we want to take an approach like that here? Or do we want to grep/examine each instance of "should" in the standard? This is the safer route. It's possible, if not likely, that some chapter authors have used "should" in the sense of "you should brush your teeth every night" rather than "you should avoid pedestrians while driving." > I think the latter would be better, but recognize that it would be a fair amount of work (that I'm not volunteering to do). I will see how far I get on this before next week. Jeff > On Nov 27, 2012, at 10:15 AM, Jeff Hammond wrote: > >> There's a lot of ambiguity in the definitions of these words in >> different contexts. If the MPI Forum wants "should" to mean "is >> required to", then such a definition should be explicit in the text. >> >>> From http://en.wikipedia.org/wiki/Shall_and_will#Technical_specifications: >> >> "The IETF (Internet Engineering Task Force) defines shall and must as >> synonymous terms denoting absolute requirements, and should as >> denoting a somewhat flexible requirement, in RFC documents." >> >> Jeff >> >> On Tue, Nov 27, 2012 at 9:48 AM, Rajeev Thakur <thakur(a)mcs.anl.gov> wrote: >>> If you search for the word "should" in the document, it is used all over the place. And it does not mean "maybe" >>> http://dictionary.reference.com/browse/should?s=t >>> >>> Rajeev >>> >>> On Nov 27, 2012, at 8:23 AM, Pavan Balaji wrote: >>> >>>> >>>> On 11/27/2012 04:48 AM US Central Time, Jeff Hammond wrote: >>>>> Is it explicitly defined anywhere that "foo has been started" means >>>>> "MPI_I*foo has been called by the appropriate MPI rank" or "sufficient >>>>> matching has occurred such that foo can proceed without additional >>>>> explicit remote activity"? Perhaps this text would be more clear if >>>>> it were more pedantic in this respect, assuming either of my >>>>> equivalences are correct. >>>> >>>> I think this needs to be clarified -- I always tell people that it needs >>>> to be matched (even though the standard says it should have "started" -- >>>> what does that mean, if matching has started, but not completed, is it >>>> considered "started"). >>>> >>>>> As for language, while "should" isn't legally enforceable in the same >>>>> way that "must" or "shall" are, the MPI standard is not a legally >>>>> binding document and I don't think any MPI implementer wants to be >>>>> known as the jerk that exploits this loophole to create a formally >>>>> standard-compliant implementation that screws over users by violating >>>>> the principle of least surprise in important use cases such as Scott's >>>>> example. >>>> >>>> I have to agree with Scott here -- "must" or "is required to be" is a >>>> clearer way to describe it. >>>> >>>> -- Pavan >>>> >>>> -- >>>> Pavan Balaji >>>> http://www.mcs.anl.gov/~balaji >>>> _______________________________________________ >>>> mpi-forum mailing list >>>> mpi-forum(a)lists.mpi-forum.org >>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >>> >>> >>> _______________________________________________ >>> 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.alcf.anl.gov/parts/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.alcf.anl.gov/parts/index.php/User:Jhammond

1 0

Re: [Mpi-forum] [EXTERNAL] Wording in MPI standard
by William Gropp 27 Nov '12

27 Nov '12

I agree that a general cleanup of the standard to conform to established practice would be best. We tried to do that with removing the "legal/illegal" statements and correcting them to valid/invalid. Bill William Gropp Director, Parallel Computing Institute Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign On Nov 27, 2012, at 9:34 AM, Jeff Squyres wrote: > I think we've done that kind of thing before (i.e., put a blanket statement in the T&C chapter, instead of fixing every individual instance). > > Do we want to take an approach like that here? Or do we want to grep/examine each instance of "should" in the standard? > > I think the latter would be better, but recognize that it would be a fair amount of work (that I'm not volunteering to do). > > > On Nov 27, 2012, at 10:15 AM, Jeff Hammond wrote: > >> There's a lot of ambiguity in the definitions of these words in >> different contexts. If the MPI Forum wants "should" to mean "is >> required to", then such a definition should be explicit in the text. >> >>> From http://en.wikipedia.org/wiki/Shall_and_will#Technical_specifications: >> >> "The IETF (Internet Engineering Task Force) defines shall and must as >> synonymous terms denoting absolute requirements, and should as >> denoting a somewhat flexible requirement, in RFC documents." >> >> Jeff >> >> On Tue, Nov 27, 2012 at 9:48 AM, Rajeev Thakur <thakur(a)mcs.anl.gov> wrote: >>> If you search for the word "should" in the document, it is used all over the place. And it does not mean "maybe" >>> http://dictionary.reference.com/browse/should?s=t >>> >>> Rajeev >>> >>> On Nov 27, 2012, at 8:23 AM, Pavan Balaji wrote: >>> >>>> >>>> On 11/27/2012 04:48 AM US Central Time, Jeff Hammond wrote: >>>>> Is it explicitly defined anywhere that "foo has been started" means >>>>> "MPI_I*foo has been called by the appropriate MPI rank" or "sufficient >>>>> matching has occurred such that foo can proceed without additional >>>>> explicit remote activity"? Perhaps this text would be more clear if >>>>> it were more pedantic in this respect, assuming either of my >>>>> equivalences are correct. >>>> >>>> I think this needs to be clarified -- I always tell people that it needs >>>> to be matched (even though the standard says it should have "started" -- >>>> what does that mean, if matching has started, but not completed, is it >>>> considered "started"). >>>> >>>>> As for language, while "should" isn't legally enforceable in the same >>>>> way that "must" or "shall" are, the MPI standard is not a legally >>>>> binding document and I don't think any MPI implementer wants to be >>>>> known as the jerk that exploits this loophole to create a formally >>>>> standard-compliant implementation that screws over users by violating >>>>> the principle of least surprise in important use cases such as Scott's >>>>> example. >>>> >>>> I have to agree with Scott here -- "must" or "is required to be" is a >>>> clearer way to describe it. >>>> >>>> -- Pavan >>>> >>>> -- >>>> Pavan Balaji >>>> http://www.mcs.anl.gov/~balaji >>>> _______________________________________________ >>>> mpi-forum mailing list >>>> mpi-forum(a)lists.mpi-forum.org >>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >>> >>> >>> _______________________________________________ >>> 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.alcf.anl.gov/parts/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

1 0

Re: [Mpi-forum] [EXTERNAL] Wording in MPI standard
by Jeff Squyres 27 Nov '12

27 Nov '12

I think we've done that kind of thing before (i.e., put a blanket statement in the T&C chapter, instead of fixing every individual instance). Do we want to take an approach like that here? Or do we want to grep/examine each instance of "should" in the standard? I think the latter would be better, but recognize that it would be a fair amount of work (that I'm not volunteering to do). On Nov 27, 2012, at 10:15 AM, Jeff Hammond wrote: > There's a lot of ambiguity in the definitions of these words in > different contexts. If the MPI Forum wants "should" to mean "is > required to", then such a definition should be explicit in the text. > >> From http://en.wikipedia.org/wiki/Shall_and_will#Technical_specifications: > > "The IETF (Internet Engineering Task Force) defines shall and must as > synonymous terms denoting absolute requirements, and should as > denoting a somewhat flexible requirement, in RFC documents." > > Jeff > > On Tue, Nov 27, 2012 at 9:48 AM, Rajeev Thakur <thakur(a)mcs.anl.gov> wrote: >> If you search for the word "should" in the document, it is used all over the place. And it does not mean "maybe" >> http://dictionary.reference.com/browse/should?s=t >> >> Rajeev >> >> On Nov 27, 2012, at 8:23 AM, Pavan Balaji wrote: >> >>> >>> On 11/27/2012 04:48 AM US Central Time, Jeff Hammond wrote: >>>> Is it explicitly defined anywhere that "foo has been started" means >>>> "MPI_I*foo has been called by the appropriate MPI rank" or "sufficient >>>> matching has occurred such that foo can proceed without additional >>>> explicit remote activity"? Perhaps this text would be more clear if >>>> it were more pedantic in this respect, assuming either of my >>>> equivalences are correct. >>> >>> I think this needs to be clarified -- I always tell people that it needs >>> to be matched (even though the standard says it should have "started" -- >>> what does that mean, if matching has started, but not completed, is it >>> considered "started"). >>> >>>> As for language, while "should" isn't legally enforceable in the same >>>> way that "must" or "shall" are, the MPI standard is not a legally >>>> binding document and I don't think any MPI implementer wants to be >>>> known as the jerk that exploits this loophole to create a formally >>>> standard-compliant implementation that screws over users by violating >>>> the principle of least surprise in important use cases such as Scott's >>>> example. >>> >>> I have to agree with Scott here -- "must" or "is required to be" is a >>> clearer way to describe it. >>> >>> -- Pavan >>> >>> -- >>> Pavan Balaji >>> http://www.mcs.anl.gov/~balaji >>> _______________________________________________ >>> mpi-forum mailing list >>> mpi-forum(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >> >> >> _______________________________________________ >> 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.alcf.anl.gov/parts/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

Re: [Mpi-forum] [EXTERNAL] Wording in MPI standard
by Jeff Hammond 27 Nov '12

27 Nov '12

There's a lot of ambiguity in the definitions of these words in different contexts. If the MPI Forum wants "should" to mean "is required to", then such a definition should be explicit in the text. >From http://en.wikipedia.org/wiki/Shall_and_will#Technical_specifications: "The IETF (Internet Engineering Task Force) defines shall and must as synonymous terms denoting absolute requirements, and should as denoting a somewhat flexible requirement, in RFC documents." Jeff On Tue, Nov 27, 2012 at 9:48 AM, Rajeev Thakur <thakur(a)mcs.anl.gov> wrote: > If you search for the word "should" in the document, it is used all over the place. And it does not mean "maybe" > http://dictionary.reference.com/browse/should?s=t > > Rajeev > > On Nov 27, 2012, at 8:23 AM, Pavan Balaji wrote: > >> >> On 11/27/2012 04:48 AM US Central Time, Jeff Hammond wrote: >>> Is it explicitly defined anywhere that "foo has been started" means >>> "MPI_I*foo has been called by the appropriate MPI rank" or "sufficient >>> matching has occurred such that foo can proceed without additional >>> explicit remote activity"? Perhaps this text would be more clear if >>> it were more pedantic in this respect, assuming either of my >>> equivalences are correct. >> >> I think this needs to be clarified -- I always tell people that it needs >> to be matched (even though the standard says it should have "started" -- >> what does that mean, if matching has started, but not completed, is it >> considered "started"). >> >>> As for language, while "should" isn't legally enforceable in the same >>> way that "must" or "shall" are, the MPI standard is not a legally >>> binding document and I don't think any MPI implementer wants to be >>> known as the jerk that exploits this loophole to create a formally >>> standard-compliant implementation that screws over users by violating >>> the principle of least surprise in important use cases such as Scott's >>> example. >> >> I have to agree with Scott here -- "must" or "is required to be" is a >> clearer way to describe it. >> >> -- Pavan >> >> -- >> Pavan Balaji >> http://www.mcs.anl.gov/~balaji >> _______________________________________________ >> mpi-forum mailing list >> mpi-forum(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > _______________________________________________ > 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.alcf.anl.gov/parts/index.php/User:Jhammond

1 0

Re: [Mpi-forum] [EXTERNAL] Wording in MPI standard
by Rajeev Thakur 27 Nov '12

27 Nov '12

If you search for the word "should" in the document, it is used all over the place. And it does not mean "maybe" http://dictionary.reference.com/browse/should?s=t Rajeev On Nov 27, 2012, at 8:23 AM, Pavan Balaji wrote: > > On 11/27/2012 04:48 AM US Central Time, Jeff Hammond wrote: >> Is it explicitly defined anywhere that "foo has been started" means >> "MPI_I*foo has been called by the appropriate MPI rank" or "sufficient >> matching has occurred such that foo can proceed without additional >> explicit remote activity"? Perhaps this text would be more clear if >> it were more pedantic in this respect, assuming either of my >> equivalences are correct. > > I think this needs to be clarified -- I always tell people that it needs > to be matched (even though the standard says it should have "started" -- > what does that mean, if matching has started, but not completed, is it > considered "started"). > >> As for language, while "should" isn't legally enforceable in the same >> way that "must" or "shall" are, the MPI standard is not a legally >> binding document and I don't think any MPI implementer wants to be >> known as the jerk that exploits this loophole to create a formally >> standard-compliant implementation that screws over users by violating >> the principle of least surprise in important use cases such as Scott's >> example. > > I have to agree with Scott here -- "must" or "is required to be" is a > clearer way to describe it. > > -- Pavan > > -- > Pavan Balaji > http://www.mcs.anl.gov/~balaji > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum

1 0

Re: [Mpi-forum] [EXTERNAL] Wording in MPI standard
by Pavan Balaji 27 Nov '12

27 Nov '12

On 11/27/2012 04:48 AM US Central Time, Jeff Hammond wrote: > Is it explicitly defined anywhere that "foo has been started" means > "MPI_I*foo has been called by the appropriate MPI rank" or "sufficient > matching has occurred such that foo can proceed without additional > explicit remote activity"? Perhaps this text would be more clear if > it were more pedantic in this respect, assuming either of my > equivalences are correct. I think this needs to be clarified -- I always tell people that it needs to be matched (even though the standard says it should have "started" -- what does that mean, if matching has started, but not completed, is it considered "started"). > As for language, while "should" isn't legally enforceable in the same > way that "must" or "shall" are, the MPI standard is not a legally > binding document and I don't think any MPI implementer wants to be > known as the jerk that exploits this loophole to create a formally > standard-compliant implementation that screws over users by violating > the principle of least surprise in important use cases such as Scott's > example. I have to agree with Scott here -- "must" or "is required to be" is a clearer way to describe it. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Re: [Mpi-forum] [EXTERNAL] Wording in MPI standard
by Jeff Hammond 27 Nov '12

27 Nov '12

Is it explicitly defined anywhere that "foo has been started" means "MPI_I*foo has been called by the appropriate MPI rank" or "sufficient matching has occurred such that foo can proceed without additional explicit remote activity"? Perhaps this text would be more clear if it were more pedantic in this respect, assuming either of my equivalences are correct. I agree with Ron that the text seems to say what Scott wants it to say already. As for language, while "should" isn't legally enforceable in the same way that "must" or "shall" are, the MPI standard is not a legally binding document and I don't think any MPI implementer wants to be known as the jerk that exploits this loophole to create a formally standard-compliant implementation that screws over users by violating the principle of least surprise in important use cases such as Scott's example. Jeff On Tue, Nov 27, 2012 at 1:30 AM, Brightwell, Ronald <rbbrigh(a)sandia.gov> wrote: > > On Nov 26, 2012, at 2:37 PM, Scott Pakin wrote: > >> Ron, >> >> On 11/26/2012 12:18 PM, Brightwell, Ronald wrote: >>> The issue is not whether the local non-blocking operation will complete, but whether starting a local non-blocking operation is enough to satisfy the matching non-local operation. It is essentially saying that once an operation has been started and matched, completion is local. That is, completion of a send operation that has been started and satisfied by a matching receive operation (that has also been started) is not dependent on whether that receive operation has been completed. The send side doesn't have to wait for the receive side to call MPI_WAIT in order for the send to complete. >> >> Yes, I just want the standard to clarify that what you wrote above is >> guaranteed MPI behavior, not simply desirable behavior. > > I think it is guaranteed behavior. > >> >> Let me provide a little more detail on what I'm dealing with. I >> believe, but have not yet managed to prove, that a large LANL >> application is doing something like the following: >> >> Rank 0 Rank 1 >> ------ ------ >> ISSEND 1 IRECV 0 >> WAIT ALLREDUCE >> ALLREDUCE WAIT >> >> If MPI guarantees that "the send side doesn't have to wait for the >> receive side to call MPI_WAIT in order for the send to complete" >> (i.e., the non-local matching you refer to), then the preceding >> pseudocode represents a correct use of MPI. If, however, it would >> merely be "nice" for the send not to have to wait for the receive-side >> MPI_WAIT, then the preceding pseudocode represents an incorrect use of >> MPI that happens to work with all tested MPI implementations but may >> deadlock under some future implementation. > > I think this should always work. > >> >>> I'm not sure there's any semantic difference between "... the receive should complete..." versus "... the receive must complete...", but the latter does seem stronger. >> >> Some standards bodies explicitly distinguish "must" as meaning, "or >> else you don't have a compliant implementation" and "should" as >> meaning, "or else you have a technically compliant but probably poorly >> performing implementation." The key is whether users can rely on the >> stated behavior for correct application execution, as in my pseudocode >> example above. > > I don't think the Standard is usually that subtle about quality of implementation issues, so it seems to me like it should be "must". > > -Ron > > > _______________________________________________ > 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.alcf.anl.gov/parts/index.php/User:Jhammond

1 0