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

Re: [MPI3 Fortran] What to do with mpif.h in MPI-3?
by Jeff Squyres 26 Jan '10

26 Jan '10

On Jan 26, 2010, at 11:02 AM, Torsten Hoefler wrote: > I agree to both of you, thus, we should not talk about F77 bindings in > the future. I propose to use the terms F90 bindings and "type-safe F90" > to distinguish between the old and the new bindings. We should also > check the standard and all new proposals for consistency. I was differentiating between them as follows: - mpif.h / implicit bindings - use mpi / F90-style explicit bindings - use mpi3 / proposed F03-style explicit bindings -- Jeff Squyres jsquyres(a)cisco.com

1 0

Re: [MPI3 Fortran] What to do with mpif.h in MPI-3?
by Torsten Hoefler 26 Jan '10

26 Jan '10

On Mon, Jan 25, 2010 at 11:21:49PM -0500, Jeff Squyres wrote: > On Jan 25, 2010, at 11:19 PM, Torsten Hoefler wrote: > > > > I think the key issue here is that there is no compiler in use anymore > > > that *only* supports F77 syntax (even those that have "77" in their > > > executable name). There hasn't been a only-supports-fortran-77-syntax > > > compiler for many, many years (since the early-/mid-90s? That's a > > > guess). > > That doesn't matter, if it's not in the F77 standard then we can't have > > it in the F77 bindings. > > There are no F77 bindings. There are only implicit Fortran bindings in mpif.h. Agreed. See other replies. Best, Torsten -- bash$ :(){ :|:&};: --------------------- http://www.unixer.de/ ----- Torsten Hoefler | Postdoctoral Fellow Open Systems Lab | Indiana University 150 S. Woodlawn Ave. | Bloomington, IN, 474045, USA Lindley Hall Room 135 | +01 (812) 856-0501

1 0

Re: [MPI3 Fortran] What to do with mpif.h in MPI-3?
by Torsten Hoefler 26 Jan '10

26 Jan '10

On Tue, Jan 26, 2010 at 02:10:45PM +0900, Malcolm Cohen wrote: > Torsten Hoefler wrote: >> That doesn't matter, if it's not in the F77 standard then we can't have >> it in the F77 bindings. > > You cannot be serious. I am serious ;-). We're just using the wrong term if we say F77 but mean F90 or some extended version of F77. > The so-called F77 bindings don't conform to the Fortran 77 standard, and > never did. For starters, F77 doesn't have INCLUDE files and all > procedure and variable names are limited to a maximum of 6 characters. > In the 70s and early 80s many F77 systems enforced the name length > limits. Yes. > As for "INTERGER*8", even if an F77-era Fortran compiler accepted the > syntax (many did, but quite a few did not) there was not universal > agreement on semantics; plus few 16-bit machines had big integers at all, > and the 18-bit, 24-bit, 36-bit, 48-bit and 60-bit machines quite often > didn't do any power-of-two-sized integers. INTEGER*3 and INTEGER*6 were > fun though. Yep. Best, Torsten -- bash$ :(){ :|:&};: --------------------- http://www.unixer.de/ ----- Torsten Hoefler | Postdoctoral Fellow Open Systems Lab | Indiana University 150 S. Woodlawn Ave. | Bloomington, IN, 474045, USA Lindley Hall Room 135 | +01 (812) 856-0501

1 0

Re: [MPI3 Fortran] What to do with mpif.h in MPI-3?
by Torsten Hoefler 26 Jan '10

26 Jan '10

On Tue, Jan 26, 2010 at 01:53:34PM +0000, N.M. Maclaren wrote: >> 2. In the same section, we explicitly list what in MPI is incompatible >> with both Fortran 77 (including the 6 character thing that was already >> mentioned -- you couldn't even have MPI_SEND or MPI_INIT!) and Fortran >> 90. Both lists are already non-empty. > > '_' wasn't allowed, either! > >> I think this is a pretty strong statement that we do *not* need to be >> 100% compatible with the Fortran specification as it was 30+ years ago. > > I want a Fortran II interface :-) :-) > More seriously, Fortran 77 compilers are dead. Essentially all are options > to Fortran 90 compilers, and the meaning is usually to allow all of the > features of Fortran 90 that do not break the ABI, often combined with the > detection of all the major syntactic extensions introduced in Fortran 90. > > I don't even know any significant, currently active codes that are still in > pure Fortran 77. I know at least one file in a significant code, but I agree in general that F77 is dead. I agree to both of you, thus, we should not talk about F77 bindings in the future. I propose to use the terms F90 bindings and "type-safe F90" to distinguish between the old and the new bindings. We should also check the standard and all new proposals for consistency. All the Best, Torsten -- bash$ :(){ :|:&};: --------------------- http://www.unixer.de/ ----- Torsten Hoefler | Postdoctoral Fellow Open Systems Lab | Indiana University 150 S. Woodlawn Ave. | Bloomington, IN, 474045, USA Lindley Hall Room 135 | +01 (812) 856-0501

1 0

Re: [MPI3 Fortran] What to do with mpif.h in MPI-3?
by N.M. Maclaren 26 Jan '10

26 Jan '10

On Jan 26 2010, Jeff Squyres wrote: >On Jan 25, 2010, at 11:19 PM, Torsten Hoefler wrote: > >> That doesn't matter, if it's not in the F77 standard then we can't have >> it in the F77 bindings. > >I remembered 2 other things that bothered me about this statement: > >1. In MPI-2.2:16.2.1, we say: > > "In the rest of this section, "Fortran" and "Fortran 90" shall refer > to "Fortran 90" and its successors, unless qualified." Given Fortran's attitude to compatibility, that's fine. There are more official ways to phrase that, but it's clear. > I haven't checked them all, but I would *hope* that all the places we > explicitly mention "Fortran 77" in the document are within proper context > and not globally scoped. Yes. I haven't checked, either. > 2. In the same section, we explicitly list what in MPI is incompatible > with both Fortran 77 (including the 6 character thing that was already > mentioned -- you couldn't even have MPI_SEND or MPI_INIT!) and Fortran > 90. Both lists are already non-empty. '_' wasn't allowed, either! > I think this is a pretty strong statement that we do *not* need to be > 100% compatible with the Fortran specification as it was 30+ years ago. I want a Fortran II interface :-) More seriously, Fortran 77 compilers are dead. Essentially all are options to Fortran 90 compilers, and the meaning is usually to allow all of the features of Fortran 90 that do not break the ABI, often combined with the detection of all the major syntactic extensions introduced in Fortran 90. I don't even know any significant, currently active codes that are still in pure Fortran 77. Regards, Nick Maclaren.

1 0

Re: [MPI3 Fortran] What to do with mpif.h in MPI-3?
by Jeff Squyres 26 Jan '10

26 Jan '10

On Jan 25, 2010, at 11:19 PM, Torsten Hoefler wrote: > That doesn't matter, if it's not in the F77 standard then we can't have > it in the F77 bindings. I remembered 2 other things that bothered me about this statement: 1. In MPI-2.2:16.2.1, we say: "In the rest of this section, ï¿½Fortranï¿½ and ï¿½Fortran 90ï¿½ shall refer to ï¿½Fortran 90ï¿½ and its successors, unless qualified." I haven't checked them all, but I would *hope* that all the places we explicitly mention "Fortran 77" in the document are within proper context and not globally scoped. 2. In the same section, we explicitly list what in MPI is incompatible with both Fortran 77 (including the 6 character thing that was already mentioned -- you couldn't even have MPI_SEND or MPI_INIT!) and Fortran 90. Both lists are already non-empty. I think this is a pretty strong statement that we do *not* need to be 100% compatible with the Fortran specification as it was 30+ years ago. -- Jeff Squyres jsquyres(a)cisco.com

1 0

Re: [MPI3 Fortran] What to do with mpif.h in MPI-3?
by N.M. Maclaren 26 Jan '10

26 Jan '10

On Jan 26 2010, Malcolm Cohen wrote: > > As for "INTEGER*8", even if an F77-era Fortran compiler accepted the > syntax (many did, but quite a few did not) there was not universal > agreement on semantics; plus few 16-bit machines had big integers at all, > and the 18-bit, 24-bit, 36-bit, 48-bit and 60-bit machines quite often > didn't do any power-of-two-sized integers. INTEGER*3 and INTEGER*6 were > fun though. On at least some systems, REAL*3 and INTEGER*3 were the third-smallest integer and real representations, and not the same size (or with sizes a multiple of 3). Jeff is, of course, right. INTEGER(KIND=MPI_COUNT_KIND). If it is felt to be a good idea to add to mpif.h, then the obvious solution has already been posted (I forget by who). That is a separate set of names, with (say) _c appended to the existing ones, and an implementation-defined integer type of count argument. Whether that should be done is a moot point. Regards, Nick Maclaren.

1 0

Re: [MPI3 Fortran] What to do with mpif.h in MPI-3?
by Malcolm Cohen 25 Jan '10

25 Jan '10

Torsten Hoefler wrote: > That doesn't matter, if it's not in the F77 standard then we can't have > it in the F77 bindings. You cannot be serious. The so-called F77 bindings don't conform to the Fortran 77 standard, and never did. For starters, F77 doesn't have INCLUDE files and all procedure and variable names are limited to a maximum of 6 characters. In the 70s and early 80s many F77 systems enforced the name length limits. As for "INTERGER*8", even if an F77-era Fortran compiler accepted the syntax (many did, but quite a few did not) there was not universal agreement on semantics; plus few 16-bit machines had big integers at all, and the 18-bit, 24-bit, 36-bit, 48-bit and 60-bit machines quite often didn't do any power-of-two-sized integers. INTEGER*3 and INTEGER*6 were fun though. Cheers, -- ................................Malcolm Cohen, Nihon NAG, Tokyo.

1 0

Re: [MPI3 Fortran] What to do with mpif.h in MPI-3?
by Jeff Squyres 25 Jan '10

25 Jan '10

On Jan 25, 2010, at 11:19 PM, Torsten Hoefler wrote: > > I think the key issue here is that there is no compiler in use anymore > > that *only* supports F77 syntax (even those that have "77" in their > > executable name). There hasn't been a only-supports-fortran-77-syntax > > compiler for many, many years (since the early-/mid-90s? That's a > > guess). > That doesn't matter, if it's not in the F77 standard then we can't have > it in the F77 bindings. There are no F77 bindings. There are only implicit Fortran bindings in mpif.h. > > So using (KIND=foo) syntax, even though that didn't debut until F90 or > > F95 (or whatever; I'm no Fortran expert) is perfectly acceptable here. > Not without renaming the bindings (I suppose that would require some > wording changes that I'm not totally opposed to). We already have some KIND types in the MPI Fortran bindings (MPI_AINT_KIND is my favorite :-) ). -- Jeff Squyres jsquyres(a)cisco.com

1 0

Re: [MPI3 Fortran] What to do with mpif.h in MPI-3?
by Torsten Hoefler 25 Jan '10

25 Jan '10

On Mon, Jan 25, 2010 at 11:11:54PM -0500, Jeff Squyres wrote: > On Jan 25, 2010, at 11:08 PM, Torsten Hoefler wrote: > > > D'oh, I didn't take my Fortran book on my travel but > > http://www.fortran.com/F77_std/f77_std.html doesn't list any mechanism > > to influence the size of an INTEGER. What am I missing? > > I think the key issue here is that there is no compiler in use anymore > that *only* supports F77 syntax (even those that have "77" in their > executable name). There hasn't been a only-supports-fortran-77-syntax > compiler for many, many years (since the early-/mid-90s? That's a > guess). That doesn't matter, if it's not in the F77 standard then we can't have it in the F77 bindings. > So using (KIND=foo) syntax, even though that didn't debut until F90 or > F95 (or whatever; I'm no Fortran expert) is perfectly acceptable here. Not without renaming the bindings (I suppose that would require some wording changes that I'm not totally opposed to). Best, Torsten -- bash$ :(){ :|:&};: --------------------- http://www.unixer.de/ ----- Torsten Hoefler | Postdoctoral Fellow Open Systems Lab | Indiana University 150 S. Woodlawn Ave. | Bloomington, IN, 474045, USA Lindley Hall Room 135 | +01 (812) 856-0501

1 0