March 2016 - mpiwg-rma - mailman3.cels.anl.gov

Re: [mpiwg-rma] Zero byte RMA operations
by William Gropp 30 Mar '16

30 Mar '16

I agree with Rajeev - this should be allowed. Bill William Gropp Director, Parallel Computing Institute Thomas M. Siebel Chair in Computer Science Chief Scientist, NCSA University of Illinois Urbana-Champaign On Mar 30, 2016, at 11:05 AM, Thakur, Rajeev <thakur(a)mcs.anl.gov> wrote: > I would say it is allowed. I am not aware of any language disallowing it. > > Rajeev > >> On Mar 30, 2016, at 10:58 AM, Balaji, Pavan <balaji(a)anl.gov> wrote: >> >> >> Is it allowed for one to do a zero-byte RMA operation on a window of size 0? I couldn't find any wording in the standard that disallows this. If I missed something in the standard, I'd appreciate someone pointing out the page/line number specifying it. >> >> The reason is that checking for this adds an extra branch on some networks, which we are trying to avoid. >> >> Thanks, >> >> -- Pavan >> >> _______________________________________________ >> mpiwg-rma mailing list >> mpiwg-rma(a)mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma > > _______________________________________________ > mpiwg-rma mailing list > mpiwg-rma(a)mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma

1 0

Re: [mpiwg-rma] Zero byte RMA operations
by Thakur, Rajeev 30 Mar '16

30 Mar '16

I would say it is allowed. I am not aware of any language disallowing it. Rajeev > On Mar 30, 2016, at 10:58 AM, Balaji, Pavan <balaji(a)anl.gov> wrote: > > > Is it allowed for one to do a zero-byte RMA operation on a window of size 0? I couldn't find any wording in the standard that disallows this. If I missed something in the standard, I'd appreciate someone pointing out the page/line number specifying it. > > The reason is that checking for this adds an extra branch on some networks, which we are trying to avoid. > > Thanks, > > -- Pavan > > _______________________________________________ > mpiwg-rma mailing list > mpiwg-rma(a)mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma

1 0

Zero byte RMA operations
by Balaji, Pavan 30 Mar '16

30 Mar '16

Is it allowed for one to do a zero-byte RMA operation on a window of size 0? I couldn't find any wording in the standard that disallows this. If I missed something in the standard, I'd appreciate someone pointing out the page/line number specifying it. The reason is that checking for this adds an extra branch on some networks, which we are trying to avoid. Thanks, -- Pavan

1 0

Re: [mpiwg-rma] MPI RMA nit: MPI_Win_create_keyval
by Thakur, Rajeev 04 Mar '16

04 Mar '16

Yes, see Section 4.3, Canonical MPI_PACK and MPI_UNPACK "These functions read/write data to/from the buffer in the “external32” data format specified in Section 13.5.2, and calculate the size needed for packing. Their first arguments specify the data format, for future extensibility, but currently the only valid value of the datarep argument is “external32.” " It’s been 20 years :-). > On Mar 4, 2016, at 11:26 AM, Jeff Squyres (jsquyres) <jsquyres(a)cisco.com> wrote: > > That's weird. Do we have other areas in the spec where we added a whole type and parameter for something that we *might* do? (and haven't, in the past 20 years...) > > >> On Mar 4, 2016, at 12:16 PM, Thakur, Rajeev <thakur(a)mcs.anl.gov> wrote: >> >> And for future extensibility — if MPI_Win_dup ever got added to the standard, we won’t need to go back and fix MPI_Win_create_keyval. >> >> Rajeev >> >> >>> On Mar 4, 2016, at 11:04 AM, William Gropp <wgropp(a)illinois.edu> wrote: >>> >>> As Jeff H suggests, it was for symmetry with the other attributes. There is no way to invoked it in the current standard. >>> >>> Bill >>> >>> William Gropp >>> Director, Parallel Computing Institute >>> Thomas M. Siebel Chair in Computer Science >>> Chief Scientist, NCSA >>> University of Illinois Urbana-Champaign >>> >>> >>> >>> >>> >>> On Mar 4, 2016, at 10:33 AM, Jeff Hammond <jeff.science(a)gmail.com> wrote: >>> >>>> That's before my time. My guess is that it made the interface symmetric wrt other objects. >>>> >>>> Adding WG list. >>>> >>>> Sent from my iPhone >>>> >>>>> On Mar 4, 2016, at 8:07 AM, Jeff Squyres (jsquyres) <jsquyres(a)cisco.com> wrote: >>>>> >>>>> Jim / JeffH -- >>>>> >>>>> Any idea why MPI_WIN_CREATE_KEYVAL includes a win_copy_attr_fn parameter? When can that callback ever get invoked? >>>>> >>>>> -- >>>>> Jeff Squyres >>>>> jsquyres(a)cisco.com >>>>> For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ >>>>> >>>> _______________________________________________ >>>> mpiwg-rma mailing list >>>> mpiwg-rma(a)lists.mpi-forum.org >>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma >>> >>> _______________________________________________ >>> mpiwg-rma mailing list >>> mpiwg-rma(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma >> > > > -- > 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: [mpiwg-rma] MPI RMA nit: MPI_Win_create_keyval
by Thakur, Rajeev 04 Mar '16

04 Mar '16

And for future extensibility — if MPI_Win_dup ever got added to the standard, we won’t need to go back and fix MPI_Win_create_keyval. Rajeev > On Mar 4, 2016, at 11:04 AM, William Gropp <wgropp(a)illinois.edu> wrote: > > As Jeff H suggests, it was for symmetry with the other attributes. There is no way to invoked it in the current standard. > > Bill > > William Gropp > Director, Parallel Computing Institute > Thomas M. Siebel Chair in Computer Science > Chief Scientist, NCSA > University of Illinois Urbana-Champaign > > > > > > On Mar 4, 2016, at 10:33 AM, Jeff Hammond <jeff.science(a)gmail.com> wrote: > >> That's before my time. My guess is that it made the interface symmetric wrt other objects. >> >> Adding WG list. >> >> Sent from my iPhone >> >>> On Mar 4, 2016, at 8:07 AM, Jeff Squyres (jsquyres) <jsquyres(a)cisco.com> wrote: >>> >>> Jim / JeffH -- >>> >>> Any idea why MPI_WIN_CREATE_KEYVAL includes a win_copy_attr_fn parameter? When can that callback ever get invoked? >>> >>> -- >>> Jeff Squyres >>> jsquyres(a)cisco.com >>> For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ >>> >> _______________________________________________ >> mpiwg-rma mailing list >> mpiwg-rma(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma > > _______________________________________________ > mpiwg-rma mailing list > mpiwg-rma(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma

1 0

Re: [mpiwg-rma] MPI RMA nit: MPI_Win_create_keyval
by William Gropp 04 Mar '16

04 Mar '16

As Jeff H suggests, it was for symmetry with the other attributes. There is no way to invoked it in the current standard. Bill William Gropp Director, Parallel Computing Institute Thomas M. Siebel Chair in Computer Science Chief Scientist, NCSA University of Illinois Urbana-Champaign On Mar 4, 2016, at 10:33 AM, Jeff Hammond <jeff.science(a)gmail.com> wrote: > That's before my time. My guess is that it made the interface symmetric wrt other objects. > > Adding WG list. > > Sent from my iPhone > >> On Mar 4, 2016, at 8:07 AM, Jeff Squyres (jsquyres) <jsquyres(a)cisco.com> wrote: >> >> Jim / JeffH -- >> >> Any idea why MPI_WIN_CREATE_KEYVAL includes a win_copy_attr_fn parameter? When can that callback ever get invoked? >> >> -- >> Jeff Squyres >> jsquyres(a)cisco.com >> For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ >> > _______________________________________________ > mpiwg-rma mailing list > mpiwg-rma(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma

1 0

Re: [mpiwg-rma] MPI RMA nit: MPI_Win_create_keyval
by Jeff Hammond 04 Mar '16

04 Mar '16

That's before my time. My guess is that it made the interface symmetric wrt other objects. Adding WG list. Sent from my iPhone > On Mar 4, 2016, at 8:07 AM, Jeff Squyres (jsquyres) <jsquyres(a)cisco.com> wrote: > > Jim / JeffH -- > > Any idea why MPI_WIN_CREATE_KEYVAL includes a win_copy_attr_fn parameter? When can that callback ever get invoked? > > -- > 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: [mpiwg-rma] notes from RMA working group meeting at the March 2016 Forum meeting
by Jeff Hammond 01 Mar '16

01 Mar '16

On Tue, Mar 1, 2016 at 12:43 AM, Jeff Hammond <jeff.science(a)gmail.com> wrote: > > > On Mon, Feb 29, 2016 at 7:58 PM, Thakur, Rajeev <thakur(a)mcs.anl.gov> > wrote: > >> I have added these notes from today’s meeting to the bottom of the agenda >> page: >> https://github.com/mpiwg-rma/rma-issues/wiki/Draft-Agenda-for-MPI-RMA-Worki… >> . >> >> Rajeev >> >> >> Issues moved to new wiki >> ----------------------- >> #24: Jeff and Pavan will fix the ticket >> > > Will wait to discuss with Pavan when he shows up at Forum. He will at > some point, right? > > >> #23: Jeff will fix the ticket >> > > Will happen soon. > > >> >> Tickets in the old wiki >> —————————--- >> #416: Jeff will move to new wiki >> > > https://github.com/mpi-forum/mpi-issues/issues/36 > > >> #434: Jeff will move to new wiki >> > > https://github.com/mpi-forum/mpi-issues/issues/37 > > >> #436: Jeff will move to new wiki >> > > https://github.com/mpi-forum/mpi-issues/issues/38 > > >> #438: Jeff will move to new wiki >> > > https://github.com/mpi-forum/mpi-issues/issues/39 > > >> #439: Needs discussion. Jim will move to new wiki. >> #459: Jeff will either close or move to new wiki >> > > I will close. I am content that it can be optimized inside of > implementations. > > >> #475: Needs discussion and possible migration to new wiki >> >> > AR: Jim but not immediately. > > >> Nonblocking RMA epochs paper: Pavan or Bill should create a ticket. >> >> Shared memory (#456): No resolution >> > > I will think about this when I work up MPI_WIN_SYNC proposals. Hubert > gave some very valuable guidance on Fortran today and I think I have > something that might help Rolf. > > >> >> 10. Old Google docs spreadsheet is done. >> >> 11. Nathan will create a ticket. >> >> MPI_Win_sync: Lot of discussion. Jeff will create one or more tickets on >> Win_sync. >> > https://github.com/mpi-forum/mpi-issues/issues/40 https://github.com/mpi-forum/mpi-issues/issues/41 https://github.com/mpi-forum/mpi-issues/issues/42 Jeff > > > -- > Jeff Hammond > jeff.science(a)gmail.com > http://jeffhammond.github.io/ > -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/

1 0

Re: [mpiwg-rma] notes from RMA working group meeting at the March 2016 Forum meeting
by Jeff Hammond 01 Mar '16

01 Mar '16

On Mon, Feb 29, 2016 at 7:58 PM, Thakur, Rajeev <thakur(a)mcs.anl.gov> wrote: > I have added these notes from today’s meeting to the bottom of the agenda > page: > https://github.com/mpiwg-rma/rma-issues/wiki/Draft-Agenda-for-MPI-RMA-Worki… > . > > Rajeev > > > Issues moved to new wiki > ----------------------- > #24: Jeff and Pavan will fix the ticket > Will wait to discuss with Pavan when he shows up at Forum. He will at some point, right? > #23: Jeff will fix the ticket > Will happen soon. > > Tickets in the old wiki > —————————--- > #416: Jeff will move to new wiki > https://github.com/mpi-forum/mpi-issues/issues/36 > #434: Jeff will move to new wiki > https://github.com/mpi-forum/mpi-issues/issues/37 > #436: Jeff will move to new wiki > https://github.com/mpi-forum/mpi-issues/issues/38 > #438: Jeff will move to new wiki > https://github.com/mpi-forum/mpi-issues/issues/39 > #439: Needs discussion. Jim will move to new wiki. > #459: Jeff will either close or move to new wiki > I will close. I am content that it can be optimized inside of implementations. > #475: Needs discussion and possible migration to new wiki > > AR: Jim but not immediately. > Nonblocking RMA epochs paper: Pavan or Bill should create a ticket. > > Shared memory (#456): No resolution > I will think about this when I work up MPI_WIN_SYNC proposals. Hubert gave some very valuable guidance on Fortran today and I think I have something that might help Rolf. > > 10. Old Google docs spreadsheet is done. > > 11. Nathan will create a ticket. > > MPI_Win_sync: Lot of discussion. Jeff will create one or more tickets on > Win_sync. > This is still in-progress. Jeff -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/

1 0

notes from RMA working group meeting at the March 2016 Forum meeting
by Thakur, Rajeev 29 Feb '16

29 Feb '16

I have added these notes from today’s meeting to the bottom of the agenda page: https://github.com/mpiwg-rma/rma-issues/wiki/Draft-Agenda-for-MPI-RMA-Worki…. Rajeev Issues moved to new wiki ----------------------- #24: Jeff and Pavan will fix the ticket #23: Jeff will fix the ticket Tickets in the old wiki —————————--- #416: Jeff will move to new wiki #434: Jeff will move to new wiki #436: Jeff will move to new wiki #438: Jeff will move to new wiki #439: Needs discussion. Jim will move to new wiki. #459: Jeff will either close or move to new wiki #475: Needs discussion and possible migration to new wiki Nonblocking RMA epochs paper: Pavan or Bill should create a ticket. Shared memory (#456): No resolution 10. Old Google docs spreadsheet is done. 11. Nathan will create a ticket. MPI_Win_sync: Lot of discussion. Jeff will create one or more tickets on Win_sync.

1 0