February 2013 - mpiwg-fortran - mailman3.cels.anl.gov

Re: [MPI3 Fortran] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind
by N.M. Maclaren 27 Feb '13

27 Feb '13

On Feb 27 2013, Malcolm Cohen wrote: > >>It would also allow (for example) the C library function >>isalnum() to be called from Fortran. > >Already allowed, not a problem. You just declare it with type >INTEGER(C_int) and use isalnum(...)/=0. > > The problem, such as it exists, arises if you allow LOGICAL instead of > INTEGER on the Fortran side. Thanks for saying what I meant to say, but didn't :-( I should have appended "as a LOGICAL function", but it was late at night. I fully agree with your other points, of course. > Fortran LOGICAL only has two values, .TRUE. > and .FALSE., whereas C's "boolean" use of integers has many values but > only 2 equivalence classes (true and false). Which are defined to be all non-zeroes and zero, with a canonical value for true of 1. C++'s bool is, of course, somewhere in between .... Regards, Nick Maclaren.

1 0

Re: [MPI3 Fortran] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind
by Malcolm Cohen 26 Feb '13

26 Feb '13

Nick Maclaren writes: >It would also allow (for example) the C library function >isalnum() to be called from Fortran. Already allowed, not a problem. You just declare it with type INTEGER(C_int) and use isalnum(...)/=0. The problem, such as it exists, arises if you allow LOGICAL instead of INTEGER on the Fortran side. Fortran LOGICAL only has two values, .TRUE. and .FALSE., whereas C's "boolean" use of integers has many values but only 2 equivalence classes (true and false). This is very different: a Fortran compiler might, for example, test the sign bit for logical testing. Or the bottom bit. Or do .AND. with bitwise AND (this latter is quite common in particular circumstances!). All these are perfectly valid for Fortran (I've seen 'em all), but do not interoperate well with C. This means that if you have LOGICAL on the Fortran side, either (a) the C routine always returns the same bitwise value for TRUE, and you got lucky and that bit-value is the one the Fortran compiler uses for LOGICAL, or (b) you need a wrapper that ensures the conditions in (a). Otherwise cue all kinds of incredibly obscure "incorrect results", that only occur in some cases and only with some compilers and/or some compilation options. Ugh. FWIW the NAG compiler currently uses 0 and 1 for LOGICAL, which works nicely with C relationals, and does use bitwise-and for LOGICAL .AND. operations under some circumstances. (On some ancient platforms we used 0 and -1 for compatibility with other Fortran compilers, but those machines aren't around any more.) Cheers, -- .......................Malcolm Cohen on holiday.

1 0

Re: [MPI3 Fortran] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind
by N.M. Maclaren 26 Feb '13

26 Feb '13

On Feb 27 2013, Bill Long wrote: > >If you want a name, how about MPI_LOGICAL_KIND? We already have >MPI_INTEGER_KIND for integers. Since this is peculiar to the MPI >environment, I'm not keen on adding something to Fortran. Not quite. While very few interfaces both use LOGICAL arguments and attempt to maintain Fortran and C consistency, it is applicable to any that do. It would also allow (for example) the C library function isalnum() to be called from Fortran. Whether that sort of use is important enough to be worth making a change to Fortran is another matter entirely! regards, Nick Maclaren.

1 0

Re: [MPI3 Fortran] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind
by Bill Long 26 Feb '13

26 Feb '13

On 2/26/13 6:05 PM, N.M. Maclaren wrote: > On Feb 26 2013, Craig Rasmussen wrote: >> >> I don't know what Bill will think of this but I think the problem >> should be resolved in the Fortran standard in the end so that >> >> LOGICAL(C_INT) >> >> is a legal type. This would allow mapping Fortran logical variables to >> C integers which is the type most C programmers use for logical >> variables anyway. > > An interesting thought - a VERY interesting thought! The name C_INT > wouldn't do, because kinds are specific to a type, but that's a detail. > I can't think of any reason that it wouldn't work. If you want a name, how about MPI_LOGICAL_KIND? We already have MPI_INTEGER_KIND for integers. Since this is peculiar to the MPI environment, I'm not keen on adding something to Fortran. Cheers, Bill > >> Does anyone know what size a default logical is in existing compilers? >> I assume it is the same size as a default integer. > > It had better be - that's required by the standard, and has been since > Fortran II. > > > Regards, > Nick Maclaren. > > _______________________________________________ > 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] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind
by Bill Long 26 Feb '13

26 Feb '13

On 2/26/13 5:10 PM, Craig Rasmussen wrote: > For some reason Jeff deleted me from this email list :-) (or perhaps I > removed myself one day when the mood moved me) so I haven't seen this > thread until today. > > Basically I agree with Tobias that there is a problem (from a standards > perspective) and with Nick that MPI-3 made the right choice with default > logical (following past standards as to not break existing code). I > also agree with Nick that an implementation of the MPI-3 standard can > finesse the issue a bit. I would note that this problem only exists for these call-back routines where only an interface is specified. For the routines that are actually called and supplied as part of the library, all of these 'type/kind conversion' issues can be resolved with wrappers that could be inlined away. > > I don't know what Bill will think of this but I think the problem should > be resolved in the Fortran standard in the end so that > > LOGICAL(C_INT) I have a feeling this is not what Nick had in mind. Visually jarring, at a minimum. There is no standard requirement that this is legal, even though in most cases it would as a practical matter. Most implementations will work since C_INT is just a KIND value (often = 4) that is probably legal for LOGICAL since vendors tend to use the same KIND values for LOGICAL and INTEGER of the same storage size. If C_INT is not a legal KIND value for INTEGER in your implementation, you are going to have more problems than just this issue. For an implementation that has C_INT > 0, I would be interested to know if there were ANY for which LOGICAL(C_INT) was not a legal declaration. Default LOGICAL and default INTEGER normally have the same storage size, and since MPI seems to like the idea of default INTEGER being the same as C int (even though that is not always true), it would seem to make just as much sense to leave it as default LOGICAL. > > is a legal type. This would allow mapping Fortran logical variables to > C integers which is the type most C programmers use for logical > variables anyway. Hence my earlier comment that this might have better been INTEGER in the first place. Too late now. > > Does anyone know what size a default logical is in existing compilers? Usually 32 bits, sometimes 64. > I assume it is the same size as a default integer. That is a requirement for variables in a "storage association context". Default integer, logical, and real all occupy a "numeric storage unit". Having the storage size for default integer and logical different would be extremely unusual. Cheers, Bill > > Craig Rasmussen > CAS Scientific Programmer > rasmus(a)cas.uoregon.edu <mailto:[email protected]> > > > > > On Feb 26, 2013, at 2:05 PM, Rolf Rabenseifner wrote: > >> Is TS 29113 changing something in this case? >> I.e., does TS 29113 allow LOGICAL dummy arguments in BIND(C) subroutines? >> >> Rolf >> >> ----- Original Message ----- >>> From: "Jeff Squyres (jsquyres)" <jsquyres(a)cisco.com >>> <mailto:[email protected]>> >>> To: "<longb(a)cray.com <mailto:[email protected]>>" <longb(a)cray.com >>> <mailto:[email protected]>>, "MPI-3 Fortran working group" >>> <mpi3-fortran(a)lists.mpi-forum.org >>> <mailto:[email protected]>> >>> Sent: Tuesday, February 26, 2013 7:04:09 PM >>> Subject: Re: [MPI3 Fortran] [Mpi-comments] MPI 3.0: Fortran 2008 >>> interface - issue with the LOGICAL kind >>> On Feb 26, 2013, at 11:27 AM, Bill Long <longb(a)cray.com >>> <mailto:[email protected]>> wrote: >>> >>>> In particular, the C function is not using _Bool. It would seem to >>>> be more consistent for the Fortran interface to specify the argument >>>> as >>>> >>>> INTEGER :: FLAG >>>> >>>> or, better >>>> >>>> INTEGER(c_int) :: FLAG. >>> >>> >>> FWIW, the MPI Fortran interfaces have always used LOGICAL for boolean >>> values, not INTEGER. It would be weird to change them now. >>> >>> -- >>> Jeff Squyres >>> jsquyres(a)cisco.com <mailto:[email protected]> >>> 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 >>> <mailto:[email protected]> >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran >> >> -- >> Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)hlrs.de >> <mailto:[email protected]> >> 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 >> <http://www.hlrs.de/people/rabenseifner> >> Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) >> _______________________________________________ >> mpi3-fortran mailing list >> mpi3-fortran(a)lists.mpi-forum.org <mailto:[email protected]> >> 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 > -- 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] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind
by N.M. Maclaren 26 Feb '13

26 Feb '13

On Feb 26 2013, Craig Rasmussen wrote: > > I don't know what Bill will think of this but I think the problem should > be resolved in the Fortran standard in the end so that > > LOGICAL(C_INT) > > is a legal type. This would allow mapping Fortran logical variables to C > integers which is the type most C programmers use for logical variables > anyway. An interesting thought - a VERY interesting thought! The name C_INT wouldn't do, because kinds are specific to a type, but that's a detail. I can't think of any reason that it wouldn't work. > Does anyone know what size a default logical is in existing compilers? I > assume it is the same size as a default integer. It had better be - that's required by the standard, and has been since Fortran II. Regards, Nick Maclaren.

1 0

Re: [MPI3 Fortran] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind
by Craig Rasmussen 26 Feb '13

26 Feb '13

For some reason Jeff deleted me from this email list :-) (or perhaps I removed myself one day when the mood moved me) so I haven't seen this thread until today. Basically I agree with Tobias that there is a problem (from a standards perspective) and with Nick that MPI-3 made the right choice with default logical (following past standards as to not break existing code). I also agree with Nick that an implementation of the MPI-3 standard can finesse the issue a bit. I don't know what Bill will think of this but I think the problem should be resolved in the Fortran standard in the end so that LOGICAL(C_INT) is a legal type. This would allow mapping Fortran logical variables to C integers which is the type most C programmers use for logical variables anyway. Does anyone know what size a default logical is in existing compilers? I assume it is the same size as a default integer. Craig Rasmussen CAS Scientific Programmer rasmus(a)cas.uoregon.edu On Feb 26, 2013, at 2:05 PM, Rolf Rabenseifner wrote: > Is TS 29113 changing something in this case? > I.e., does TS 29113 allow LOGICAL dummy arguments in BIND(C) subroutines? > > Rolf > > ----- Original Message ----- >> From: "Jeff Squyres (jsquyres)" <jsquyres(a)cisco.com> >> To: "<longb(a)cray.com>" <longb(a)cray.com>, "MPI-3 Fortran working group" <mpi3-fortran(a)lists.mpi-forum.org> >> Sent: Tuesday, February 26, 2013 7:04:09 PM >> Subject: Re: [MPI3 Fortran] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind >> On Feb 26, 2013, at 11:27 AM, Bill Long <longb(a)cray.com> wrote: >> >>> In particular, the C function is not using _Bool. It would seem to >>> be more consistent for the Fortran interface to specify the argument >>> as >>> >>> INTEGER :: FLAG >>> >>> or, better >>> >>> INTEGER(c_int) :: FLAG. >> >> >> FWIW, the MPI Fortran interfaces have always used LOGICAL for boolean >> values, not INTEGER. It would be weird to change them now. >> >> -- >> 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 > > -- > 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) > _______________________________________________ > 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] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind
by Bill Long 26 Feb '13

26 Feb '13

TS 29113 says nothing new about type LOGICAL arguments. They are covered in the base F2008 standard. Cheers, Bill On 2/26/13 4:05 PM, Rolf Rabenseifner wrote: > Is TS 29113 changing something in this case? > I.e., does TS 29113 allow LOGICAL dummy arguments in BIND(C) subroutines? > > Rolf > > ----- Original Message ----- >> From: "Jeff Squyres (jsquyres)" <jsquyres(a)cisco.com> >> To: "<longb(a)cray.com>" <longb(a)cray.com>, "MPI-3 Fortran working group" <mpi3-fortran(a)lists.mpi-forum.org> >> Sent: Tuesday, February 26, 2013 7:04:09 PM >> Subject: Re: [MPI3 Fortran] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind >> On Feb 26, 2013, at 11:27 AM, Bill Long <longb(a)cray.com> wrote: >> >>> In particular, the C function is not using _Bool. It would seem to >>> be more consistent for the Fortran interface to specify the argument >>> as >>> >>> INTEGER :: FLAG >>> >>> or, better >>> >>> INTEGER(c_int) :: FLAG. >> >> >> FWIW, the MPI Fortran interfaces have always used LOGICAL for boolean >> values, not INTEGER. It would be weird to change them now. >> >> -- >> 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 > -- 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] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind
by Rolf Rabenseifner 26 Feb '13

26 Feb '13

Is TS 29113 changing something in this case? I.e., does TS 29113 allow LOGICAL dummy arguments in BIND(C) subroutines? Rolf ----- Original Message ----- > From: "Jeff Squyres (jsquyres)" <jsquyres(a)cisco.com> > To: "<longb(a)cray.com>" <longb(a)cray.com>, "MPI-3 Fortran working group" <mpi3-fortran(a)lists.mpi-forum.org> > Sent: Tuesday, February 26, 2013 7:04:09 PM > Subject: Re: [MPI3 Fortran] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind > On Feb 26, 2013, at 11:27 AM, Bill Long <longb(a)cray.com> wrote: > > > In particular, the C function is not using _Bool. It would seem to > > be more consistent for the Fortran interface to specify the argument > > as > > > > INTEGER :: FLAG > > > > or, better > > > > INTEGER(c_int) :: FLAG. > > > FWIW, the MPI Fortran interfaces have always used LOGICAL for boolean > values, not INTEGER. It would be weird to change them now. > > -- > 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 -- 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: [MPI3 Fortran] [Mpi-comments] MPI 3.0: Fortran 2008 interface - issue with the LOGICAL kind
by Jeff Squyres (jsquyres) 26 Feb '13

26 Feb '13

On Feb 26, 2013, at 11:27 AM, Bill Long <longb(a)cray.com> wrote: > In particular, the C function is not using _Bool. It would seem to be more consistent for the Fortran interface to specify the argument as > > INTEGER :: FLAG > > or, better > > INTEGER(c_int) :: FLAG. FWIW, the MPI Fortran interfaces have always used LOGICAL for boolean values, not INTEGER. It would be weird to change them now. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0