March 2014 - mpiwg-fortran - mailman3.cels.anl.gov

Re: [MPIWG Fortran] Question about an argument of MPI_Testall
by Junchao Zhang 12 Mar '14

12 Mar '14

That makes sense. I found MPI_STATUS_IGNORE and MPI_STATUSES_IGNORE are a subtle issue in binding. Thanks, Jeff. --Junchao Zhang On Wed, Mar 12, 2014 at 10:30 AM, Jeff Squyres (jsquyres) < jsquyres(a)cisco.com> wrote: > Rolf R. just reminded me off-list what the real reason is: > MPI_STATUSES_IGNORE. > > We can't know what the (count) will be, but you still must be able to pass > MPI_STATUSES_IGNORE through that dummy argument. Hence, it must be (*), > not (count). > > Keep in mind that the spec prohibits you from having a special type for > MPI_STATUSES_IGNORE. I.e., it must be Type(MPI_Status) so that tools can > know how to intercept it properly. > > Make sense? > > > On Mar 12, 2014, at 5:25 AM, N.M. Maclaren <nmm1(a)cam.ac.uk> wrote: > > > On Mar 12 2014, Malcolm Cohen wrote: > >> > >> That reason is bogus, if the dummy argument is declared to have "count" > elements the actual argument is required to have at least "count" elements, > but this need not be exact. The Fortran standard states > >> > >> "The rank and shape of the actual argument need not agree with the rank > and shape of the dummy argument, but the number of elements in the dummy > argument shall not exceed the number of elements in the element sequence of > the actual argument." > >> > >> (F2008 12.5.1.11) > > > > And that has been true since Fortran II - I have just checked that it > > was explicitly stated there, too. > > > > > > Regards, > > Nick Maclaren. > > > > _______________________________________________ > > mpiwg-fortran mailing list > > mpiwg-fortran(a)lists.mpi-forum.org > > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-fortran > > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: > http://www.cisco.com/web/about/doing_business/legal/cri/ > > _______________________________________________ > mpiwg-fortran mailing list > mpiwg-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-fortran >

1 0

Re: [MPIWG Fortran] Question about an argument of MPI_Testall
by William Gropp 12 Mar '14

12 Mar '14

Should this be an optional argument in that case, and eliminate MPI_STATUS_IGNORE and MPI_STATUSES_IGNORE, as was done for the C++ interface? Bill William Gropp Director, Parallel Computing Institute Thomas M. Siebel Chair in Computer Science University of Illinois Urbana-Champaign On Mar 12, 2014, at 10:30 AM, Jeff Squyres (jsquyres) wrote: > Rolf R. just reminded me off-list what the real reason is: MPI_STATUSES_IGNORE. > > We can't know what the (count) will be, but you still must be able to pass MPI_STATUSES_IGNORE through that dummy argument. Hence, it must be (*), not (count). > > Keep in mind that the spec prohibits you from having a special type for MPI_STATUSES_IGNORE. I.e., it must be Type(MPI_Status) so that tools can know how to intercept it properly. > > Make sense? > > > On Mar 12, 2014, at 5:25 AM, N.M. Maclaren <nmm1(a)cam.ac.uk> wrote: > >> On Mar 12 2014, Malcolm Cohen wrote: >>> >>> That reason is bogus, if the dummy argument is declared to have "count" elements the actual argument is required to have at least "count" elements, but this need not be exact. The Fortran standard states >>> >>> "The rank and shape of the actual argument need not agree with the rank and shape of the dummy argument, but the number of elements in the dummy argument shall not exceed the number of elements in the element sequence of the actual argument." >>> >>> (F2008 12.5.1.11) >> >> And that has been true since Fortran II - I have just checked that it >> was explicitly stated there, too. >> >> >> Regards, >> Nick Maclaren. >> >> _______________________________________________ >> mpiwg-fortran mailing list >> mpiwg-fortran(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-fortran > > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ > > _______________________________________________ > mpiwg-fortran mailing list > mpiwg-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-fortran

1 0

Re: [MPIWG Fortran] Question about an argument of MPI_Testall
by Jeff Squyres (jsquyres) 12 Mar '14

12 Mar '14

Rolf R. just reminded me off-list what the real reason is: MPI_STATUSES_IGNORE. We can't know what the (count) will be, but you still must be able to pass MPI_STATUSES_IGNORE through that dummy argument. Hence, it must be (*), not (count). Keep in mind that the spec prohibits you from having a special type for MPI_STATUSES_IGNORE. I.e., it must be Type(MPI_Status) so that tools can know how to intercept it properly. Make sense? On Mar 12, 2014, at 5:25 AM, N.M. Maclaren <nmm1(a)cam.ac.uk> wrote: > On Mar 12 2014, Malcolm Cohen wrote: >> >> That reason is bogus, if the dummy argument is declared to have "count" elements the actual argument is required to have at least "count" elements, but this need not be exact. The Fortran standard states >> >> "The rank and shape of the actual argument need not agree with the rank and shape of the dummy argument, but the number of elements in the dummy argument shall not exceed the number of elements in the element sequence of the actual argument." >> >> (F2008 12.5.1.11) > > And that has been true since Fortran II - I have just checked that it > was explicitly stated there, too. > > > Regards, > Nick Maclaren. > > _______________________________________________ > mpiwg-fortran mailing list > mpiwg-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-fortran -- 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 Fortran] Question about an argument of MPI_Testall
by N.M. Maclaren 12 Mar '14

12 Mar '14

On Mar 12 2014, Malcolm Cohen wrote: > > That reason is bogus, if the dummy argument is declared to have "count" > elements the actual argument is required to have at least "count" > elements, but this need not be exact. The Fortran standard states > > "The rank and shape of the actual argument need not agree with the rank > and shape of the dummy argument, but the number of elements in the dummy > argument shall not exceed the number of elements in the element sequence > of the actual argument." > >(F2008 12.5.1.11) And that has been true since Fortran II - I have just checked that it was explicitly stated there, too. Regards, Nick Maclaren.

1 0

Re: [MPIWG Fortran] Question about an argument of MPI_Testall
by Malcolm Cohen 11 Mar '14

11 Mar '14

That reason is bogus, if the dummy argument is declared to have "count" elements the actual argument is required to have at least "count" elements, but this need not be exact. The Fortran standard states "The rank and shape of the actual argument need not agree with the rank and shape of the dummy argument, but the number of elements in the dummy argument shall not exceed the number of elements in the element sequence of the actual argument." (F2008 12.5.1.11) Cheers, -----Original Message----- From: Jeff Squyres (jsquyres) Date: 平成 26年3月12日 1:21 To: MPI-WG Fortran working group Subject: Re: [MPIWG Fortran] Question about an argument of MPI_Testall The WG chose not to do this, because Rolf and others indicated that it is not uncommon to have a large status of arrays -- i.e., something >=count statuses. If you use array_of_statuses(count), then you can only pass in an array of *exactly* size "count" -- and that doesn't fit what users tend to do for convenience. On Mar 11, 2014, at 12:03 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> wrote: > Hello, > > P60 of MPI3 reads: > > MPI_Testall(count, array_of_requests, flag, array_of_statuses, ierror) > INTEGER, INTENT(IN) :: count > > TYPE(MPI_Request), INTENT(INOUT) :: array_of_requests(count) > > LOGICAL, INTENT(OUT) :: flag > > TYPE(MPI_Status) :: array_of_statuses(*) > > INTEGER, OPTIONAL, INTENT(OUT) :: ierror > > Should array_of_statuses be declared as TYPE(MPI_Status) :: > array_of_statuses(count)? > > Thanks > > --Junchao Zhang > _______________________________________________ > mpiwg-fortran mailing list > mpiwg-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-fortran -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ _______________________________________________ mpiwg-fortran mailing list mpiwg-fortran(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-fortran -- ................................Malcolm Cohen, Nihon NAG, Tokyo.

1 0

Re: [MPIWG Fortran] Question about an argument of MPI_Testall
by Jeff Squyres (jsquyres) 11 Mar '14

11 Mar '14

The WG chose not to do this, because Rolf and others indicated that it is not uncommon to have a large status of arrays -- i.e., something >=count statuses. If you use array_of_statuses(count), then you can only pass in an array of *exactly* size "count" -- and that doesn't fit what users tend to do for convenience. On Mar 11, 2014, at 12:03 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> wrote: > Hello, > > P60 of MPI3 reads: > > MPI_Testall(count, array_of_requests, flag, array_of_statuses, ierror) > INTEGER, INTENT(IN) :: count > > TYPE(MPI_Request), INTENT(INOUT) :: array_of_requests(count) > > LOGICAL, INTENT(OUT) :: flag > > TYPE(MPI_Status) :: array_of_statuses(*) > > INTEGER, OPTIONAL, INTENT(OUT) :: ierror > > Should array_of_statuses be declared as TYPE(MPI_Status) :: array_of_statuses(count)? > > Thanks > > --Junchao Zhang > _______________________________________________ > mpiwg-fortran mailing list > mpiwg-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-fortran -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Question about an argument of MPI_Testall
by Junchao Zhang 11 Mar '14

11 Mar '14

Hello, P60 of MPI3 reads: MPI_Testall(count, array_of_requests, flag, array_of_statuses, ierror) INTEGER, INTENT(IN) :: count TYPE(MPI_Request), INTENT(INOUT) :: array_of_requests(count) LOGICAL, INTENT(OUT) :: flag TYPE(MPI_Status) :: array_of_statuses(*) INTEGER, OPTIONAL, INTENT(OUT) :: ierror Should array_of_statuses be declared as TYPE(MPI_Status) :: array_of_statuses(count)? Thanks --Junchao Zhang

1 0

Re: [MPIWG Fortran] Assumed-size array in MPI3
by N.M. Maclaren 10 Mar '14

10 Mar '14

On Mar 10 2014, Bill Long wrote: > > I agree with Steve. In these sorts of calls the actual size of the array > is available from somewhere else - either another argument or information > in the communicator. This is the common situation when calling C. I don't, but only in the slant of my comment. I would say that assumed size arrays should be used ONLY when calling facilities that use a C-like interface (which obviously includes MPI 2.2). While it would be nice to say that programs should always use the new MPI 3.0 interfaces, currently their availability is patchy and unreliable. So the book over-simplifies, but I agree with its recommendation for pure Fortran code. There are other reasons to use assumed size, but they are mostly pretty arcane. Regards, Nick.

1 0

Re: [MPIWG Fortran] Assumed-size array in MPI3
by Bill Long 10 Mar '14

10 Mar '14

I agree with Steve. In these sorts of calls the actual size of the array is available from somewhere else - either another argument or information in the communicator. This is the common situation when calling C. Cheers, Bill On Mar 10, 2014, at 11:06 AM, Junchao Zhang <jczhang(a)mcs.anl.gov> wrote: > Hello, > I read a Fortran textbook, which says "Assumed-size dummy arrays are a holdover from earlier versions of Fortran. They should never be used in any new programs". > But I find MPI3 has interfaces like > MPI_Alltoallv(sendbuf, sendcounts, sdispls, sendtype, recvbuf, recvcounts, rdispls, recvtype, comm, ierror) > ... > INTEGER, INTENT(IN) :: sendcounts(*), sdispls(*), recvcounts(*), rdispls(*) > ... > > I was wondering why MPI3 still uses it. > Thanks. > > --Junchao Zhang > _______________________________________________ > mpiwg-fortran mailing list > mpiwg-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-fortran Bill Long longb(a)cray.com Fortran Technical Suport & 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: [MPIWG Fortran] Assumed-size array in MPI3
by Lionel, Steve 10 Mar '14

10 Mar '14

Don't believe everything you read. In the context of interfaces to non-Fortran procedures, assumed-size dummy arrays are almost always exactly what you want. I would agree that for an all-Fortran application, I would discourage their use. Steve From: mpiwg-fortran [mailto:[email protected]] On Behalf Of Junchao Zhang Sent: Monday, March 10, 2014 12:07 PM To: MPI-WG Fortran working group Subject: [MPIWG Fortran] Assumed-size array in MPI3 Hello, I read a Fortran textbook, which says "Assumed-size dummy arrays are a holdover from earlier versions of Fortran. They should never be used in any new programs". But I find MPI3 has interfaces like MPI_Alltoallv(sendbuf, sendcounts, sdispls, sendtype, recvbuf, recvcounts, rdispls, recvtype, comm, ierror) ... INTEGER, INTENT(IN) :: sendcounts(*), sdispls(*), recvcounts(*), rdispls(*) ... I was wondering why MPI3 still uses it. Thanks. --Junchao Zhang

1 0