Monday, 18 January 2010

How do you like your Fortran 95 interfaces?

Fortran 95 has been fully implemented for a couple of years now by all the compilers we support, so we'd like to be able to provide a Fortran 95 way of enjoying our algorithmic content. What kinds of features would you expect or like to see from such an endeavour?

At the very least, by the Mark 23 release of our standard Fortran Library all example programs will have been 'F90-ed' to have module procedures for user-supplied routines, allocation of local arrays, and a range of other cosmetic improvements. Here's one of the simpler instances of this, the example program for the Nelder–Mead routine E04CBF: e04cbfe.f90

So that's just calling the regular Fortran (77) Library interface in a more modern context. If this isn't satisfactory then what you're after are more-modern interfaces too. An uber-example of this would be for the sparse NLP solver E04VHF
INTERFACE
 SUBROUTINE nag_e04vhf(start,objadd,objrow,prob,usrfun,a,g,xlow,xupp, &
    xnames,flow,fupp,fnames,x,xstate,xmul,f,fstate,fmul,ns,ninf,sinf, &
    comm,user,error)
!         .. Implicit None Statement ..
    IMPLICIT NONE
!         .. Non-Generic Interface Blocks ..
    INTERFACE
       SUBROUTINE usrfun(status,x,f,g,cuser,iuser,ruser)
!               .. Use Statements ..
          USE nag_precisions, ONLY : wp
!               .. Scalar Arguments ..
          INTEGER, INTENT (INOUT) :: status
!               .. Array Arguments ..
          REAL (kind=wp), OPTIONAL, INTENT (INOUT) :: f(:), g(:)
          REAL (kind=wp), INTENT (INOUT) :: ruser(:)
          REAL (kind=wp), INTENT (IN) :: x(:)
          INTEGER, INTENT (INOUT) :: iuser(:)
          CHARACTER (8), INTENT (INOUT) :: cuser(:)
       END SUBROUTINE usrfun
    END INTERFACE
!         .. Structure Arguments ..
    TYPE (nag_sparse_matrix), OPTIONAL, INTENT (IN) :: a, g
    TYPE (nag_comm), INTENT (INOUT) :: comm
    TYPE (nag_error), OPTIONAL, INTENT (INOUT) :: error
    TYPE (nag_user), OPTIONAL, INTENT (INOUT) :: user
!         .. Scalar Arguments ..
    REAL (kind=wp), OPTIONAL, INTENT (IN) :: objadd
    REAL (kind=wp), OPTIONAL, INTENT (OUT) :: sinf
    INTEGER, OPTIONAL, INTENT (OUT) :: ninf
    INTEGER, OPTIONAL, INTENT (INOUT) :: ns
    INTEGER, INTENT (INOUT)      :: objrow
    INTEGER, INTENT (IN) :: start
    CHARACTER (8), OPTIONAL, INTENT (IN) :: prob
!         .. Array Arguments ..
    REAL (kind=wp), INTENT (INOUT) :: x(:)
    REAL (kind=wp), OPTIONAL, INTENT (INOUT) :: f(:), fmul(:)
    REAL (kind=wp), OPTIONAL, INTENT (IN)  :: flow(:), fupp(:), xlow(:), &
       xupp(:)
    REAL (kind=wp), OPTIONAL, INTENT (OUT) :: xmul(:)
    INTEGER, OPTIONAL, INTENT (INOUT) :: fstate(:), xstate(:)
    CHARACTER (8), OPTIONAL, INTENT (IN) :: fnames(:), xnames(:)
 END SUBROUTINE nag_e04vhf
END INTERFACE
How can we get to that point? The desirable approach for us from an engineering perspective is to use our XML technologies to generate Fortran 95 wrappers to our algorithmic base. As an example, an XML-generated interface for G01HBF could be
INTERFACE
 FUNCTION nag_multi_normal_f90(tail,a,b,xmu,sig,tol,lwk,ifail)
!         .. Use Statements ..
    USE nag_precisions, ONLY : wp
!         .. Function Return Value ..
    REAL (kind=wp)            :: nag_multi_normal_f90
!         .. Scalar Arguments ..
    REAL (kind=wp), OPTIONAL, INTENT (IN) :: tol
    INTEGER, OPTIONAL, INTENT (INOUT) :: ifail
    INTEGER, OPTIONAL, INTENT (IN) :: lwk
    CHARACTER (1), INTENT (IN) :: tail
!         .. Array Arguments ..
    REAL (kind=wp), INTENT (IN) :: a(:), b(:), sig(:,:), xmu(:)
 END FUNCTION nag_multi_normal_f90
END INTERFACE
That looks a bit like the interface for g01hb in the NAG Toolbox for MATLAB, because we'd be using similar XML technologies to generate this kind of Fortran interface as is used to generate the MATLAB interfaces.

What level of Fortran 95 interfaces would you like to see?
Are there any particular routines you feel would benefit from detailed human-driven, case-by-case design decisions (e.g., optimization, above)?
Alternatively, what do you hate about our current Fortran 77 interfaces?

No comments:

Post a Comment

NAG moderates all replies and reserves the right to not publish posts that are deemed inappropriate.