December 2009 - mpiwg-fortran - mailman3.cels.anl.gov

Re: [MPI3 Fortran] Should new features be retrofitted to Fortran 77 bindings?
by N.M. Maclaren 14 Dec '09

14 Dec '09

On Dec 14 2009, Bill Long wrote: > >You can write an include file that will compile in either fixed or free >source form. The trick is continuation lines, where you put & in column >73 of the first line and & in column 6 of the second. Otherwise, write >in fixed form, which is just one particular choice of free form. Provided that you avoid doing certain things, most of which are perverse, such as putting spaces within keywords or variables. I agree that it's not hard to write readable dual-mode code. >The other topic of this thread has been whether a compiler named 'f77' >or some variant thereof will compile a program with f95 features. I >can't speak for all vendors, but it would seem like a waste of effort to >actually support two separate compilers, one for f77 and one for >standard Fortran. ... As you say, few vendors seem to do it nowadays. It used to be common up to a decade or so ago, but I haven't seen it in some time. Except, of course, under the sort of Linux which has gfortran, g95, f2c, g77, ratfor and any other Fortran-like compiler it can lay its hands on. My impression is that about half the vendors have a command called f77 or similar which is, as you say, the Fortran 90+ compiler with the legacy flags set. The other half don't bother. Most of the users I deal with regard the Fortran compiler command name as an system-specific incantation and wouldn't care much if it were spelled abracadabra, as long as 'man Fortran' told them its name. Regards, Nick Maclaren.

1 0

Re: [MPI3 Fortran] Should new features be retrofitted to Fortran 77 bindings?
by Bill Long 14 Dec '09

14 Dec '09

Craig Rasmussen wrote: > Bill, > > Meaning fixed form without explicit interfaces. You can write an include file that will compile in either fixed or free source form. The trick is continuation lines, where you put & in column 73 of the first line and & in column 6 of the second. Otherwise, write in fixed form, which is just one particular choice of free form. > > Some people around here would die before using any of that new-fangled stuff. If it doesn't have 6 leading characters before a statement it isn't FORTRAN. (Technically correct, since the capitalization change from FORTRAN to Fortran occurred with Fortran 90.) If there are explicit interfaces in the include file and the user never sees them, I don't see the harm. Sort of depressing that there are still people stuck so far in the past. But, modern compilers will still compile their dusty decks. The other topic of this thread has been whether a compiler named 'f77' or some variant thereof will compile a program with f95 features. I can't speak for all vendors, but it would seem like a waste of effort to actually support two separate compilers, one for f77 and one for standard Fortran. Cray avoids the issue by naming our compiler 'ftn', i.e. year-independent. IBM supplies xlf77, xlf90, xlf03, ..., but these are all the same actual compiler binary. The name does cause some different default options to be set, but the user certainly has access to the newer features without purchasing another compiler. GNU used to have g77, but that has been dropped in favor of gfortran. I believe that g77 was ancient-only, but it is hard to imaging people still using it. As I said before, a new compiler can still compile old codes, so there is no harm for users to use the new compiler even if they do not use the new features. They can rename the compiler script if it makes them feel better. At some point you have to say you are no longer going to work on new designs for inner tubes for car tires. If the question is about Fortran 2003, I believe that Cray and IBM are the only compilers available with essentially full support. Others should follow down the road. Some will probably start implementing Fortran 2008 features ahead of finishing Fortran 2003. The trailing Fortran 2003 features will likely be the object-oriented programming stuff and parameterized derived types. Neither of these has a big impact on the MPI interfaces. Cheers, Bill > > -craig > > On Dec 14, 2009, at 7:04 AM, Jeff Squyres wrote: > >> I'm talking about codes that "include 'mpif.h'". >> >> If they are compiled with compilers like "pgif77" or other "*f77" compilers, will they still allow/obey newer Fortran language constructs (namely the things that we're discussing for the F2003 bindings)? >> >> >> On Dec 11, 2009, at 5:47 PM, Bill Long wrote: >> >>> What do you mean by "Fortran 77 codes"? There are codes that are >>> written in fixed source form and others in free source form. If by F77 >>> you mean fixed source form, then that should be accommodated. >>> Otherwise, old codes that were written in the before ~ 1993 are still >>> Fortran programs and should compile with modern compilers for today's >>> incarnation of standard Fortran. Apart from supporting fixed source >>> form, what retrofit is necessary? >>> >>> Cheers, >>> Bill >>> >>> >>> >>> Craig Rasmussen wrote: >>>> Jeff raised an interesting question with me. Should new features that are being considered for MPI-3 be retrofitted if possible for Fortran 77 codes? Some users might say yes as they never intend to use modern Fortran features. On the other hand, F77 is over 30 years old and so perhaps we should move on and not put support (money) into adding new features. >>>> >>>> The new Fortran MPI-3 bindings will allow users to link to codes using F77 bindings so if users want new bindings they only have >>>> to use MPI3 locally. >>>> >>>> -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 >>> >>> >>> _______________________________________________ >>> mpi3-fortran mailing list >>> mpi3-fortran(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran >>> >> >> -- >> Jeff Squyres >> jsquyres(a)cisco.com >> >> >> _______________________________________________ >> 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] Should new features be retrofitted to Fortran 77bindings?
by Jeff Squyres 14 Dec '09

14 Dec '09

On Dec 14, 2009, at 1:52 PM, Craig Rasmussen wrote: > I'm confused, if the compiled by "f77" compilers (probably just hidden compiler options by now), of course they won't compile the MPI3 bindings. But they should be link compatible. *That's* what I mean by F77 codes -- codes that are compiled by the "*f77" compilers. Do such *f77 compilers specifically choose not to understand syntax beyond whatever was defined for Fortran beyond the 1977 version of the spec? > -craig > > > On Dec 14, 2009, at 7:04 AM, Jeff Squyres wrote: > >> I'm talking about codes that "include 'mpif.h'". >> >> If they are compiled with compilers like "pgif77" or other "*f77" compilers, will they still allow/obey newer Fortran language constructs (namely the things that we're discussing for the F2003 bindings)? >> >> >> On Dec 11, 2009, at 5:47 PM, Bill Long wrote: >> >>> What do you mean by "Fortran 77 codes"? There are codes that are >>> written in fixed source form and others in free source form. If by F77 >>> you mean fixed source form, then that should be accommodated. >>> Otherwise, old codes that were written in the before ~ 1993 are still >>> Fortran programs and should compile with modern compilers for today's >>> incarnation of standard Fortran. Apart from supporting fixed source >>> form, what retrofit is necessary? >>> >>> Cheers, >>> Bill >>> >>> >>> >>> Craig Rasmussen wrote: >>>> Jeff raised an interesting question with me. Should new features that are being considered for MPI-3 be retrofitted if possible for Fortran 77 codes? Some users might say yes as they never intend to use modern Fortran features. On the other hand, F77 is over 30 years old and so perhaps we should move on and not put support (money) into adding new features. >>>> >>>> The new Fortran MPI-3 bindings will allow users to link to codes using F77 bindings so if users want new bindings they only have >>>> to use MPI3 locally. >>>> >>>> -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 >>> >>> >>> _______________________________________________ >>> mpi3-fortran mailing list >>> mpi3-fortran(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran >>> >> >> >> -- >> Jeff Squyres >> jsquyres(a)cisco.com >> >> >> _______________________________________________ >> 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 -- Jeff Squyres jsquyres(a)cisco.com

1 0

Re: [MPI3 Fortran] Should new features be retrofitted to Fortran 77 bindings?
by Craig Rasmussen 14 Dec '09

14 Dec '09

Jeff, I'm confused, if the compiled by "f77" compilers (probably just hidden compiler options by now), of course they won't compile the MPI3 bindings. But they should be link compatible. -craig On Dec 14, 2009, at 7:04 AM, Jeff Squyres wrote: > I'm talking about codes that "include 'mpif.h'". > > If they are compiled with compilers like "pgif77" or other "*f77" compilers, will they still allow/obey newer Fortran language constructs (namely the things that we're discussing for the F2003 bindings)? > > > On Dec 11, 2009, at 5:47 PM, Bill Long wrote: > >> What do you mean by "Fortran 77 codes"? There are codes that are >> written in fixed source form and others in free source form. If by F77 >> you mean fixed source form, then that should be accommodated. >> Otherwise, old codes that were written in the before ~ 1993 are still >> Fortran programs and should compile with modern compilers for today's >> incarnation of standard Fortran. Apart from supporting fixed source >> form, what retrofit is necessary? >> >> Cheers, >> Bill >> >> >> >> Craig Rasmussen wrote: >>> Jeff raised an interesting question with me. Should new features that are being considered for MPI-3 be retrofitted if possible for Fortran 77 codes? Some users might say yes as they never intend to use modern Fortran features. On the other hand, F77 is over 30 years old and so perhaps we should move on and not put support (money) into adding new features. >>> >>> The new Fortran MPI-3 bindings will allow users to link to codes using F77 bindings so if users want new bindings they only have >>> to use MPI3 locally. >>> >>> -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 >> >> >> _______________________________________________ >> mpi3-fortran mailing list >> mpi3-fortran(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran >> > > > -- > Jeff Squyres > jsquyres(a)cisco.com > > > _______________________________________________ > mpi3-fortran mailing list > mpi3-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran

1 0

Re: [MPI3 Fortran] Should new features be retrofitted to Fortran 77 bindings?
by Craig Rasmussen 14 Dec '09

14 Dec '09

Bill, Meaning fixed form without explicit interfaces. Some people around here would die before using any of that new-fangled stuff. If it doesn't have 6 leading characters before a statement it isn't FORTRAN. -craig On Dec 14, 2009, at 7:04 AM, Jeff Squyres wrote: > I'm talking about codes that "include 'mpif.h'". > > If they are compiled with compilers like "pgif77" or other "*f77" compilers, will they still allow/obey newer Fortran language constructs (namely the things that we're discussing for the F2003 bindings)? > > > On Dec 11, 2009, at 5:47 PM, Bill Long wrote: > >> What do you mean by "Fortran 77 codes"? There are codes that are >> written in fixed source form and others in free source form. If by F77 >> you mean fixed source form, then that should be accommodated. >> Otherwise, old codes that were written in the before ~ 1993 are still >> Fortran programs and should compile with modern compilers for today's >> incarnation of standard Fortran. Apart from supporting fixed source >> form, what retrofit is necessary? >> >> Cheers, >> Bill >> >> >> >> Craig Rasmussen wrote: >>> Jeff raised an interesting question with me. Should new features that are being considered for MPI-3 be retrofitted if possible for Fortran 77 codes? Some users might say yes as they never intend to use modern Fortran features. On the other hand, F77 is over 30 years old and so perhaps we should move on and not put support (money) into adding new features. >>> >>> The new Fortran MPI-3 bindings will allow users to link to codes using F77 bindings so if users want new bindings they only have >>> to use MPI3 locally. >>> >>> -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 >> >> >> _______________________________________________ >> mpi3-fortran mailing list >> mpi3-fortran(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran >> > > > -- > Jeff Squyres > jsquyres(a)cisco.com > > > _______________________________________________ > mpi3-fortran mailing list > mpi3-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran

1 0

Re: [MPI3 Fortran] Should new features be retrofitted to Fortran 77 bindings?
by Jeff Squyres 14 Dec '09

14 Dec '09

I'm talking about codes that "include 'mpif.h'". If they are compiled with compilers like "pgif77" or other "*f77" compilers, will they still allow/obey newer Fortran language constructs (namely the things that we're discussing for the F2003 bindings)? On Dec 11, 2009, at 5:47 PM, Bill Long wrote: > What do you mean by "Fortran 77 codes"? There are codes that are > written in fixed source form and others in free source form. If by F77 > you mean fixed source form, then that should be accommodated. > Otherwise, old codes that were written in the before ~ 1993 are still > Fortran programs and should compile with modern compilers for today's > incarnation of standard Fortran. Apart from supporting fixed source > form, what retrofit is necessary? > > Cheers, > Bill > > > > Craig Rasmussen wrote: > > Jeff raised an interesting question with me. Should new features that are being considered for MPI-3 be retrofitted if possible for Fortran 77 codes? Some users might say yes as they never intend to use modern Fortran features. On the other hand, F77 is over 30 years old and so perhaps we should move on and not put support (money) into adding new features. > > > > The new Fortran MPI-3 bindings will allow users to link to codes using F77 bindings so if users want new bindings they only have > > to use MPI3 locally. > > > > -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 > > > _______________________________________________ > mpi3-fortran mailing list > mpi3-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran > -- Jeff Squyres jsquyres(a)cisco.com

1 0

Re: [MPI3 Fortran] Should new features be retrofitted to Fortran 77 bindings?
by Bill Long 11 Dec '09

11 Dec '09

What do you mean by "Fortran 77 codes"? There are codes that are written in fixed source form and others in free source form. If by F77 you mean fixed source form, then that should be accommodated. Otherwise, old codes that were written in the before ~ 1993 are still Fortran programs and should compile with modern compilers for today's incarnation of standard Fortran. Apart from supporting fixed source form, what retrofit is necessary? Cheers, Bill Craig Rasmussen wrote: > Jeff raised an interesting question with me. Should new features that are being considered for MPI-3 be retrofitted if possible for Fortran 77 codes? Some users might say yes as they never intend to use modern Fortran features. On the other hand, F77 is over 30 years old and so perhaps we should move on and not put support (money) into adding new features. > > The new Fortran MPI-3 bindings will allow users to link to codes using F77 bindings so if users want new bindings they only have > to use MPI3 locally. > > -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] Should new features be retrofitted to Fortran 77bindings?
by Jeff Squyres 11 Dec '09

11 Dec '09

To be absolutely clear on this point: we're not talking about abondoning the F77 bindings, or even F77 users. We're talking about *new* MPI-3 features that might have difficulty with F77 bindings. For such cases, should we just punt and say "this new MPI-3 feature is only available in the F2003 bindings"? Like Craig, I think that this might be a reasonable position: if you want to use the newest, latest, greatest stuff, you need to upgrade to the F2003 bindings. Don't forget that we have full interop between the F77 and F2003 bindings (being able to convert handles back and forth), so even if someone has a huge legacy F77 code, they can write one subroutine with "use mpi3" and still have it work properly with the rest of their MPI F77 code. On Dec 11, 2009, at 3:25 PM, Craig Rasmussen wrote: > Jeff raised an interesting question with me. Should new features that are being considered for MPI-3 be retrofitted if possible for Fortran 77 codes? Some users might say yes as they never intend to use modern Fortran features. On the other hand, F77 is over 30 years old and so perhaps we should move on and not put support (money) into adding new features. > > The new Fortran MPI-3 bindings will allow users to link to codes using F77 bindings so if users want new bindings they only have > to use MPI3 locally. > > -craig > > > _______________________________________________ > mpi3-fortran mailing list > mpi3-fortran(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran > -- Jeff Squyres jsquyres(a)cisco.com

1 0

[MPI3 Fortran] Should new features be retrofitted to Fortran 77 bindings?
by Craig Rasmussen 11 Dec '09

11 Dec '09

Jeff raised an interesting question with me. Should new features that are being considered for MPI-3 be retrofitted if possible for Fortran 77 codes? Some users might say yes as they never intend to use modern Fortran features. On the other hand, F77 is over 30 years old and so perhaps we should move on and not put support (money) into adding new features. The new Fortran MPI-3 bindings will allow users to link to codes using F77 bindings so if users want new bindings they only have to use MPI3 locally. -craig

1 0