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

Re: [MPIWG Fortran] BIND(C) attribute of MPI callback functions
by Junchao Zhang 02 Jul '14

02 Jul '14

On Wed, Jul 2, 2014 at 7:05 AM, Bill Long <longb(a)cray.com> wrote: > > On Jul 2, 2014, at 2:11 AM, Rolf Rabenseifner <rabenseifner(a)hlrs.de> > wrote: > > > Bill, > > > > The reason for removal of BIND(C) was simple: > > Four callbacks contain LOGICAL arguments > > and we learnt that LOGICAL and BIND(C) does not work together. > > Therefore the errata is correct. > > Agreed. > > > > > I hope you find a solution for storing the callback pointer > > in a C structure and to bring it back to Fortran. > > I expect that calling the Fortran callbacks > > in a portable way may require Fortran code > > that calls the callback. > > For compilers that support the TS, there should be no problem. As noted by > Junchao, Cray and Gfortran did not have a problem. The IBM compiler > appears to have overlooked the TS change. A bug report to IBM should > resolve the issue. The fix is trivial - just disabling an error message > Have sent two bug reports regarding F08 to IBM. One is the BIND(C). The other is that XLF-15.1 does not support storage_size(). > Cheers, > Bill > > > > > > Best regards > > Rolf > > > > > > ----- Original Message ----- > >> From: "Bill Long" <longb(a)cray.com> > >> To: "MPI-WG Fortran working group" <mpiwg-fortran(a)lists.mpi-forum.org> > >> Sent: Tuesday, July 1, 2014 9:06:05 PM > >> Subject: Re: [MPIWG Fortran] BIND(C) attribute of MPI callback functions > >> > >> > >> On Jul 1, 2014, at 2:00 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> > >> wrote: > >> > >>> Hello, > >>> Errata for MPI-3.0 removed the BIND(C) attribute of MPI callback > >>> functions. But in an implementation, to pass the Fortran > >>> procedure to the backend C, I need to get its C function pointer > >>> with c_funloc, as the following code shows. When I compile it > >>> with IBM XLF 15.1, it complains with such an error "1516-203 (S) > >>> The argument to the intrinsic procedure C_FUNLOC must be a > >>> procedure with the BIND(C) attribute." In my experiments, gcc > >>> and Cray ftn are fine with it. Looking at > >>> https://gcc.gnu.org/onlinedocs/gfortran/C_005fFUNLOC.html, I > >>> find c_funloc indeed requires an interoperable function > >>> argument. Any comments? > >> > >> This requirement on the argument of C_FUNLOC was removed by TS 29113, > >> subclause 8.1 “Removed restrictions on ISO_C_BINDING module > >> procedures”. > >> > >> Cheers, > >> Bill > >> > >> > >> > >>> > >>> subroutine MPI_Comm_create_errhandler_f08(comm_errhandler_fn, > >>> errhandler, ierror) > >>> use, intrinsic :: iso_c_binding, only : c_funloc, c_int, > >>> c_funptr > >>> use :: mpi_f08, only : MPI_Errhandler, > >>> MPI_Comm_errhandler_function > >>> use :: mpi_c_interface, only : c_Errhandler, > >>> MPIR_Comm_create_errhandler_c > >>> > >>> implicit none > >>> > >>> procedure(MPI_Comm_errhandler_function) :: comm_errhandler_fn > >>> type(MPI_Errhandler), intent(out) :: errhandler > >>> integer, optional, intent(out) :: ierror > >>> > >>> type(c_funptr) :: comm_errhandler_fn_c > >>> > >>> comm_errhandler_fn_c = c_funloc(comm_errhandler_fn) !! xlf > >>> issues an error > >>> ierror_c = MPIR_Comm_create_errhandler_c(comm_errhandler_fn_c, > >>> errhandler%MPI_VAL) > >>> ... > >>> end subroutine MPI_Comm_create_errhandler_f08 > >>> > >>> --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 > >> > >> > >> _______________________________________________ > >> mpiwg-fortran mailing list > >> mpiwg-fortran(a)lists.mpi-forum.org > >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-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: Room 1.307) > > _______________________________________________ > > 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 > > > _______________________________________________ > 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] BIND(C) attribute of MPI callback functions
by N.M. Maclaren 02 Jul '14

02 Jul '14

On Jul 2 2014, Bill Long wrote: >On Jul 2, 2014, at 2:11 AM, Rolf Rabenseifner <rabenseifner(a)hlrs.de> wrote: >> >> The reason for removal of BIND(C) was simple: >> Four callbacks contain LOGICAL arguments >> and we learnt that LOGICAL and BIND(C) does not work together. I considered an extension, mentioned it, and we discussed it. The problem is virtually insoluble, because the semantic models for Fortran LOGICAL and C truth values are so different, and any attempt to make them interoperate would be a truly ghastly mess. There are outstanding issues that could be changed, but this is not one. Yes, I know that Fortran 2008 makes LOGICAL(C_BOOL) interoperate with _Bool, but I regard that as being even more of a mistake than specifying that INTEGER(C_SIZE_T) interoperates with size_t. Don't go there; REALLY don't go there :-( Regards, Nick Maclaren.

1 0

Re: [MPIWG Fortran] BIND(C) attribute of MPI callback functions
by Bill Long 02 Jul '14

02 Jul '14

On Jul 2, 2014, at 2:11 AM, Rolf Rabenseifner <rabenseifner(a)hlrs.de> wrote: > Bill, > > The reason for removal of BIND(C) was simple: > Four callbacks contain LOGICAL arguments > and we learnt that LOGICAL and BIND(C) does not work together. > Therefore the errata is correct. Agreed. > > I hope you find a solution for storing the callback pointer > in a C structure and to bring it back to Fortran. > I expect that calling the Fortran callbacks > in a portable way may require Fortran code > that calls the callback. For compilers that support the TS, there should be no problem. As noted by Junchao, Cray and Gfortran did not have a problem. The IBM compiler appears to have overlooked the TS change. A bug report to IBM should resolve the issue. The fix is trivial - just disabling an error message. Cheers, Bill > > Best regards > Rolf > > > ----- Original Message ----- >> From: "Bill Long" <longb(a)cray.com> >> To: "MPI-WG Fortran working group" <mpiwg-fortran(a)lists.mpi-forum.org> >> Sent: Tuesday, July 1, 2014 9:06:05 PM >> Subject: Re: [MPIWG Fortran] BIND(C) attribute of MPI callback functions >> >> >> On Jul 1, 2014, at 2:00 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> >> wrote: >> >>> Hello, >>> Errata for MPI-3.0 removed the BIND(C) attribute of MPI callback >>> functions. But in an implementation, to pass the Fortran >>> procedure to the backend C, I need to get its C function pointer >>> with c_funloc, as the following code shows. When I compile it >>> with IBM XLF 15.1, it complains with such an error "1516-203 (S) >>> The argument to the intrinsic procedure C_FUNLOC must be a >>> procedure with the BIND(C) attribute." In my experiments, gcc >>> and Cray ftn are fine with it. Looking at >>> https://gcc.gnu.org/onlinedocs/gfortran/C_005fFUNLOC.html, I >>> find c_funloc indeed requires an interoperable function >>> argument. Any comments? >> >> This requirement on the argument of C_FUNLOC was removed by TS 29113, >> subclause 8.1 “Removed restrictions on ISO_C_BINDING module >> procedures”. >> >> Cheers, >> Bill >> >> >> >>> >>> subroutine MPI_Comm_create_errhandler_f08(comm_errhandler_fn, >>> errhandler, ierror) >>> use, intrinsic :: iso_c_binding, only : c_funloc, c_int, >>> c_funptr >>> use :: mpi_f08, only : MPI_Errhandler, >>> MPI_Comm_errhandler_function >>> use :: mpi_c_interface, only : c_Errhandler, >>> MPIR_Comm_create_errhandler_c >>> >>> implicit none >>> >>> procedure(MPI_Comm_errhandler_function) :: comm_errhandler_fn >>> type(MPI_Errhandler), intent(out) :: errhandler >>> integer, optional, intent(out) :: ierror >>> >>> type(c_funptr) :: comm_errhandler_fn_c >>> >>> comm_errhandler_fn_c = c_funloc(comm_errhandler_fn) !! xlf >>> issues an error >>> ierror_c = MPIR_Comm_create_errhandler_c(comm_errhandler_fn_c, >>> errhandler%MPI_VAL) >>> ... >>> end subroutine MPI_Comm_create_errhandler_f08 >>> >>> --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 >> >> >> _______________________________________________ >> mpiwg-fortran mailing list >> mpiwg-fortran(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-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: Room 1.307) > _______________________________________________ > 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] BIND(C) attribute of MPI callback functions
by Rolf Rabenseifner 02 Jul '14

02 Jul '14

Bill, The reason for removal of BIND(C) was simple: Four callbacks contain LOGICAL arguments and we learnt that LOGICAL and BIND(C) does not work together. Therefore the errata is correct. I hope you find a solution for storing the callback pointer in a C structure and to bring it back to Fortran. I expect that calling the Fortran callbacks in a portable way may require Fortran code that calls the callback. Best regards Rolf ----- Original Message ----- > From: "Bill Long" <longb(a)cray.com> > To: "MPI-WG Fortran working group" <mpiwg-fortran(a)lists.mpi-forum.org> > Sent: Tuesday, July 1, 2014 9:06:05 PM > Subject: Re: [MPIWG Fortran] BIND(C) attribute of MPI callback functions > > > On Jul 1, 2014, at 2:00 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> > wrote: > > > Hello, > > Errata for MPI-3.0 removed the BIND(C) attribute of MPI callback > > functions. But in an implementation, to pass the Fortran > > procedure to the backend C, I need to get its C function pointer > > with c_funloc, as the following code shows. When I compile it > > with IBM XLF 15.1, it complains with such an error "1516-203 (S) > > The argument to the intrinsic procedure C_FUNLOC must be a > > procedure with the BIND(C) attribute." In my experiments, gcc > > and Cray ftn are fine with it. Looking at > > https://gcc.gnu.org/onlinedocs/gfortran/C_005fFUNLOC.html, I > > find c_funloc indeed requires an interoperable function > > argument. Any comments? > > This requirement on the argument of C_FUNLOC was removed by TS 29113, > subclause 8.1 “Removed restrictions on ISO_C_BINDING module > procedures”. > > Cheers, > Bill > > > > > > > subroutine MPI_Comm_create_errhandler_f08(comm_errhandler_fn, > > errhandler, ierror) > > use, intrinsic :: iso_c_binding, only : c_funloc, c_int, > > c_funptr > > use :: mpi_f08, only : MPI_Errhandler, > > MPI_Comm_errhandler_function > > use :: mpi_c_interface, only : c_Errhandler, > > MPIR_Comm_create_errhandler_c > > > > implicit none > > > > procedure(MPI_Comm_errhandler_function) :: comm_errhandler_fn > > type(MPI_Errhandler), intent(out) :: errhandler > > integer, optional, intent(out) :: ierror > > > > type(c_funptr) :: comm_errhandler_fn_c > > > > comm_errhandler_fn_c = c_funloc(comm_errhandler_fn) !! xlf > > issues an error > > ierror_c = MPIR_Comm_create_errhandler_c(comm_errhandler_fn_c, > > errhandler%MPI_VAL) > > ... > > end subroutine MPI_Comm_create_errhandler_f08 > > > > --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 > > > _______________________________________________ > mpiwg-fortran mailing list > mpiwg-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-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: Room 1.307)

1 0

Re: [MPIWG Fortran] BIND(C) attribute of MPI callback functions
by Bill Long 01 Jul '14

01 Jul '14

On Jul 1, 2014, at 2:00 PM, Junchao Zhang <jczhang(a)mcs.anl.gov> wrote: > Hello, > Errata for MPI-3.0 removed the BIND(C) attribute of MPI callback functions. But in an implementation, to pass the Fortran procedure to the backend C, I need to get its C function pointer with c_funloc, as the following code shows. When I compile it with IBM XLF 15.1, it complains with such an error "1516-203 (S) The argument to the intrinsic procedure C_FUNLOC must be a procedure with the BIND(C) attribute." In my experiments, gcc and Cray ftn are fine with it. Looking at https://gcc.gnu.org/onlinedocs/gfortran/C_005fFUNLOC.html, I find c_funloc indeed requires an interoperable function argument. Any comments? This requirement on the argument of C_FUNLOC was removed by TS 29113, subclause 8.1 “Removed restrictions on ISO_C_BINDING module procedures”. Cheers, Bill > > subroutine MPI_Comm_create_errhandler_f08(comm_errhandler_fn, errhandler, ierror) > use, intrinsic :: iso_c_binding, only : c_funloc, c_int, c_funptr > use :: mpi_f08, only : MPI_Errhandler, MPI_Comm_errhandler_function > use :: mpi_c_interface, only : c_Errhandler, MPIR_Comm_create_errhandler_c > > implicit none > > procedure(MPI_Comm_errhandler_function) :: comm_errhandler_fn > type(MPI_Errhandler), intent(out) :: errhandler > integer, optional, intent(out) :: ierror > > type(c_funptr) :: comm_errhandler_fn_c > > comm_errhandler_fn_c = c_funloc(comm_errhandler_fn) !! xlf issues an error > ierror_c = MPIR_Comm_create_errhandler_c(comm_errhandler_fn_c, errhandler%MPI_VAL) > ... > end subroutine MPI_Comm_create_errhandler_f08 > > --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

BIND(C) attribute of MPI callback functions
by Junchao Zhang 01 Jul '14

01 Jul '14

Hello, Errata for MPI-3.0 removed the BIND(C) attribute of MPI callback functions. But in an implementation, to pass the Fortran procedure to the backend C, I need to get its C function pointer with c_funloc, as the following code shows. When I compile it with IBM XLF 15.1, it complains with such an error "1516-203 (S) The argument to the intrinsic procedure C_FUNLOC must be a procedure with the BIND(C) attribute." In my experiments, gcc and Cray ftn are fine with it. Looking at https://gcc.gnu.org/onlinedocs/gfortran/C_005fFUNLOC.html, I find c_funloc indeed requires an interoperable function argument. Any comments? subroutine MPI_Comm_create_errhandler_f08(comm_errhandler_fn, errhandler, ierror) use, intrinsic :: iso_c_binding, only : c_funloc, c_int, c_funptr use :: mpi_f08, only : MPI_Errhandler, MPI_Comm_errhandler_function use :: mpi_c_interface, only : c_Errhandler, MPIR_Comm_create_errhandler_c implicit none procedure(MPI_Comm_errhandler_function) :: comm_errhandler_fn type(MPI_Errhandler), intent(out) :: errhandler integer, optional, intent(out) :: ierror type(c_funptr) :: comm_errhandler_fn_c comm_errhandler_fn_c = c_funloc(comm_errhandler_fn) !! xlf issues an error ierror_c = MPIR_Comm_create_errhandler_c(comm_errhandler_fn_c, errhandler%MPI_VAL) ... end subroutine MPI_Comm_create_errhandler_f08 --Junchao Zhang

1 0