May 2010 - mpiwg-fortran - mailman3.cels.anl.gov

Re: [MPI3 Fortran] Choice buffer woes
by Rasmussen, Craig 28 May '10

28 May '10

I might add that LANL is planning on adding the new interop functionality to gfortran so some time in the future (hopefully) we will have the chance to test out the MPI-3 Fortran bindings in gfortran. In the meantime it would nice to have a fall back position as Jeff describes. -craig On Fri, May 28, 2010 at 7:29 AM, Jeff Squyres <jsquyres(a)cisco.com> wrote: > Armed with a whiteboard and our laptops, Craig and I spent a few days in a room together earlier this week advancing the state of our prototype F08 MPI bindings implementation. ï¿½We made lots of progress (per emails from earlier this week), but were stymied on an important issue: MPI choice buffers. > > We thought that we would be able to use ifort's !DEC$ ATTRIBUTES NO_ARG_CHECK functionality to emulate the new F08-style syntax in a "close enough" manner. ï¿½Unfortunately, it seems that this functionality only passes a dope vector when the caller passes an array. ï¿½If the caller passes a scalar, only the address of that scalar is passed. ï¿½After consultation with Steve Lionel, it doesn't look like there's any way for the called function to know whether it received just an address or a dope vector. > > This is problematic for MPI. > > Does anyone know of a Fortran compiler out there that has "ignore TKR" kinds of functionality and either a) always passes a dope vector or b) somehow notifies the called function whether it got an address or a dope vector? ï¿½(either would be fine for the MPI bindings prototype) > > I mention this issue because without a working prototype, we can't take the bindings proposal to the overall MPI Forum. > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: > http://www.cisco.com/web/about/doing_business/legal/cri/ > > > _______________________________________________ > mpi3-fortran mailing list > mpi3-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran >

1 0

[MPI3 Fortran] Choice buffer woes
by Jeff Squyres 28 May '10

28 May '10

Armed with a whiteboard and our laptops, Craig and I spent a few days in a room together earlier this week advancing the state of our prototype F08 MPI bindings implementation. We made lots of progress (per emails from earlier this week), but were stymied on an important issue: MPI choice buffers. We thought that we would be able to use ifort's !DEC$ ATTRIBUTES NO_ARG_CHECK functionality to emulate the new F08-style syntax in a "close enough" manner. Unfortunately, it seems that this functionality only passes a dope vector when the caller passes an array. If the caller passes a scalar, only the address of that scalar is passed. After consultation with Steve Lionel, it doesn't look like there's any way for the called function to know whether it received just an address or a dope vector. This is problematic for MPI. Does anyone know of a Fortran compiler out there that has "ignore TKR" kinds of functionality and either a) always passes a dope vector or b) somehow notifies the called function whether it got an address or a dope vector? (either would be fine for the MPI bindings prototype) I mention this issue because without a working prototype, we can't take the bindings proposal to the overall 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: [MPI3 Fortran] MPI-2 Fortran question: 2D char array in MPI_COMM_SPAWN_MULTIPLE
by Rolf Rabenseifner 27 May '10

27 May '10

Whatever "icky" means, it is possible to write correct calls to MPI_COMM_SPAWN and MPI_COMM_SPAWN_MULTIPLE and to correctly catch the arguments by the MPI library routine. Therefore, I would not spend additional work into this interface. My opinion: it is not worth to spend to much effort in MPI Spawn. Rolf ----- Original Message ----- > From: "Craig Rasmussen" <crasmussen(a)newmexicoconsortium.org> > To: "MPI-3 Fortran working group" <mpi3-fortran(a)lists.mpi-forum.org> > Sent: Wednesday, May 26, 2010 3:52:46 PM > Subject: Re: [MPI3 Fortran] MPI-2 Fortran question: 2D char array in MPI_COMM_SPAWN_MULTIPLE > Just to clarify. Jeff and I believe the spec is "icky" (Jeff's words > not mine :-) ). There may be ways to fix make it better in MPI-3. > I'll think about it and send out an email later to see if anyone has > ideas on how to improve it. > > > -craig > > > On Tue, May 25, 2010 at 2:41 PM, Jeff Squyres < jsquyres(a)cisco.com > > wrote: > > > > On May 25, 2010, at 4:23 PM, Rolf Rabenseifner wrote: > > > MPI-2.2, p310:18-25 defines: > > > > "The argv argument argv is an array of strings containing arguments > > that are passed to > > the program. The first element of argv is the first argument passed > > to command, not, as > > is conventional in some contexts, the command itself. The argument > > list is terminated by > > NULL in C and C++ and an empty string in Fortran. > > I mis-read this last line when I read it -- now I understand what Rolf > is saying (and, in fact, it's exactly what someone said earlier that > the spec would need to be right). > > Thanks Rolf! > > > > The code fragment below is wrong because it does > > try to detect an overall second_dim_size instead of individual > > second_dim_size(i) for i=1,count. > > Also a good point -- in OMPI, we assume that all the argv arrays will > have the same number of tokens for each command. But this is not > necessarily the case. I'll fix. > > > > > Thanks! > > -- Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: > http://www.cisco.com/web/about/doing_business/legal/cri/ > > > _______________________________________________ mpi3-fortran mailing > list mpi3-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran > > > _______________________________________________ mpi3-fortran mailing > list mpi3-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran -- 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

Re: [MPI3 Fortran] Change character string declarations?
by N.M. Maclaren 27 May '10

27 May '10

On May 26 2010, Lionel, Steve wrote: > I think this is reasonable, though are you also changing other > declarations to use the :: syntax? CHARACTER(*) works too. As for kind > parameters, we've had extensive arguments about this before. For > character kinds, I suggest that leaving off the kind may be for the best > - otherwise you'll perhaps need to reference ISO_C_BINDING's C_CHAR > constant, and that seems to annoy others on this list. "Default > character" is probably safe and won't confuse users. I would put it a LOT more strongly! Character KIND is rarely used in scientific code, and I don't know whether anyone has investigated how reliably it can be mixed with default characters. Stick with the default, though providing guidelines as to what names implementations should use for the others if they provide them would be reasonable. Regards, Nick Maclaren.

1 0

Re: [MPI3 Fortran] Change character string declarations?
by Bill Long 26 May '10

26 May '10

Rasmussen, Craig wrote: > I would like to change all places in the MPI-3 standard with character > string dummy argument declarations such as, > > CHARACTER*(*) KEY, VALUE > > to > > CHARACTER(len=*) :: KEY, VALUE The len= part is maybe a little much. But I would certainly prefer character(*) over character*(*) and, similarly, character over character*1 and also character(8) over character*8 Cheers, Bill > > Comments? Would we want to add a kind parameter too? > > -craig > _______________________________________________ > mpi3-fortran mailing list > mpi3-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran -- 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

1 0

Re: [MPI3 Fortran] Change character string declarations?
by Lionel, Steve 26 May '10

26 May '10

I think this is reasonable, though are you also changing other declarations to use the :: syntax? CHARACTER(*) works too. As for kind parameters, we've had extensive arguments about this before. For character kinds, I suggest that leaving off the kind may be for the best - otherwise you'll perhaps need to reference ISO_C_BINDING's C_CHAR constant, and that seems to annoy others on this list. "Default character" is probably safe and won't confuse users. Steve -----Original Message----- From: mpi3-fortran-bounces(a)lists.mpi-forum.org [mailto:[email protected]] On Behalf Of Rasmussen, Craig Sent: Wednesday, May 26, 2010 4:41 PM To: MPI-3 Fortran working group Subject: [MPI3 Fortran] Change character string declarations? Importance: Low I would like to change all places in the MPI-3 standard with character string dummy argument declarations such as, CHARACTER*(*) KEY, VALUE to CHARACTER(len=*) :: KEY, VALUE Comments? Would we want to add a kind parameter too? -craig _______________________________________________ mpi3-fortran mailing list mpi3-fortran(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran

1 0

[MPI3 Fortran] Change character string declarations?
by Rasmussen, Craig 26 May '10

26 May '10

I would like to change all places in the MPI-3 standard with character string dummy argument declarations such as, CHARACTER*(*) KEY, VALUE to CHARACTER(len=*) :: KEY, VALUE Comments? Would we want to add a kind parameter too? -craig

1 0

Re: [MPI3 Fortran] MPI_STATUS_IGNORE (and friends)
by Supalov, Alexander 26 May '10

26 May '10

Thanks. Understood. -----Original Message----- From: mpi3-fortran-bounces(a)lists.mpi-forum.org [mailto:[email protected]] On Behalf Of Jeff Squyres Sent: Wednesday, May 26, 2010 6:56 PM To: MPI-3 Fortran working group Subject: Re: [MPI3 Fortran] MPI_STATUS_IGNORE (and friends) On May 26, 2010, at 12:49 PM, Rasmussen, Craig wrote: > There is some text on the wiki (off from the mpi-forum). It needs to be updated as it is a little out of date now. The code is a mercurial repository at http://bitbucket.org/jsquyres/mpi3-fortran (rapidly changing this week). And frequently broken, etc. It's prototype code, not yet ready for public consumption, etc. All the usual disclaimers about rapidly-changing code apply. But feel free to poke around. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ _______________________________________________ mpi3-fortran mailing list mpi3-fortran(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran --------------------------------------------------------------------- Intel GmbH Dornacher Strasse 1 85622 Feldkirchen/Muenchen Germany Sitz der Gesellschaft: Feldkirchen bei Muenchen Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer Registergericht: Muenchen HRB 47456 Ust.-IdNr. VAT Registration No.: DE129385895 Citibank Frankfurt (BLZ 502 109 00) 600119052 This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.

1 0

Re: [MPI3 Fortran] MPI_STATUS_IGNORE (and friends)
by Jeff Squyres 26 May '10

26 May '10

On May 26, 2010, at 12:49 PM, Rasmussen, Craig wrote: > There is some text on the wiki (off from the mpi-forum). It needs to be updated as it is a little out of date now. The code is a mercurial repository at http://bitbucket.org/jsquyres/mpi3-fortran (rapidly changing this week). And frequently broken, etc. It's prototype code, not yet ready for public consumption, etc. All the usual disclaimers about rapidly-changing code apply. But feel free to poke around. -- 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: [MPI3 Fortran] MPI_STATUS_IGNORE (and friends)
by Rasmussen, Craig 26 May '10

26 May '10

There is some text on the wiki (off from the mpi-forum). It needs to be updated as it is a little out of date now. The code is a mercurial repository at http://bitbucket.org/jsquyres/mpi3-fortran (rapidly changing this week). -craig On Wed, May 26, 2010 at 10:41 AM, Supalov, Alexander < alexander.supalov(a)intel.com> wrote: > Thanks. Where can one find the proposed MPI F08 bindings and the prototype > implementation you mentioned? > > -----Original Message----- > From: mpi3-fortran-bounces(a)lists.mpi-forum.org [mailto: > mpi3-fortran-bounces(a)lists.mpi-forum.org] On Behalf Of Jeff Squyres > Sent: Wednesday, May 26, 2010 6:21 PM > To: MPI-3 Fortran WG > Subject: [MPI3 Fortran] MPI_STATUS_IGNORE (and friends) > > Craig and I have been working on a prototype implementation of the proposed > MPI F08 bindings. It has been tremendously helpful to keep the proposal > ideas on track with what is possible, what is useful, etc. > > One of the things that we would like to change from the current proposal is > to remove the sentinel value constants MPI_STATUS_IGNORE and > MPI_STATUSES_IGNORE (and a few others that are similar in spirit). MPI > status structs are usually allocated and passed by the application down to > an MPI function; the MPI implementation then fills up that struct. However, > the app can pass MPI_STATUS_IGNORE (or MPI_STATUSES_IGNORE in an array > context) to indicate that they don't care about the status and don't want > the MPI implementation to fill it (them). > > Just like the C++ bindings, instead of having these constants in the F08 > bindings, we propose to have overloaded functions that simply do not have a > status parameter. For example: > > ----- > interface MPI_Test > ! Note the "status" parameter here > subroutine MPI_Test_wrapper(request, completed, status, err) > type(MPI_Request), intent(inout) :: request > logical, intent(out) :: completed > type(MPI_Status), intent(out) :: status > integer, optional, intent(out) :: err > end subroutine > > ! Note the LACK of "status" parameter here > subroutine MPI_Test_ignore_wrapper(request, completed, err) > type(MPI_Request), intent(inout) :: request > logical, intent(out) :: completed > integer, optional, intent(out) :: err > end subroutine > end interface > ----- > > Just like C++, this seems like the "natural" way to do it in the language. > It also keeps a 1:1 mapping to the MPI language neutral bindings and is > therefore not a radical departure from the MPI specification. Indeed, the > C++ bindings established a precedent for this exact case (there are 2 C++ > bindings for MPI::Test, for example -- one with a status argument and one > without). > > Thoughts? > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: > http://www.cisco.com/web/about/doing_business/legal/cri/ > > > _______________________________________________ > mpi3-fortran mailing list > mpi3-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran > --------------------------------------------------------------------- > Intel GmbH > Dornacher Strasse 1 > 85622 Feldkirchen/Muenchen Germany > Sitz der Gesellschaft: Feldkirchen bei Muenchen > Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer > Registergericht: Muenchen HRB 47456 Ust.-IdNr. > VAT Registration No.: DE129385895 > Citibank Frankfurt (BLZ 502 109 00) 600119052 > > This e-mail and any attachments may contain confidential material for > the sole use of the intended recipient(s). Any review or distribution > by others is strictly prohibited. If you are not the intended > recipient, please contact the sender and delete all copies. > > > _______________________________________________ > mpi3-fortran mailing list > mpi3-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran >

1 0