Re: Challenge 4 (matrix diagonal)

From: Dave Thompson (david.thompson1_at_worldnet.att.net)
Date: 12/28/03

  • Next message: kia: "cross chatter (was: Writing good numerical code in C)" 
    Date: Sun, 28 Dec 2003 01:49:05 GMT

    On Sun, 21 Dec 2003 02:15:36 GMT, john <john@nospam.com> wrote, in
    comp.lang.pl1 as part of a (the?) long-running David Frank flamefest;
    comp.lang.fortran added with some trepidation:

    > 4. Fortran CAN'T equivalence a vector with the diagonal of a matrix.
    >
    > dcl a(10,10) float, diag(10) defined (1sub, 1sub);
    > put (d); prints the diagonal elements of array a.

    For the specific case of diagonal, I *think* this will work:

      REAL/whatever Matrix (1:N, 1:N)
      TYPE EveryNPlus1; SEQUENCE
        REAL/whatever Take, Skip (1:N)
      END TYPE EveryNPlus1
      TYPE(EveryNPlus1) Overlay (1:N)
      EQUIVALENCE (Matrix, Overlay)

    and now you can use Overlay%Take as a vector, or I'm pretty sure even
    set an "assumed-shape" dummy or POINTER to it (F9x array POINTERs are
    full dope vectors). Overlay(N) runs off the end of the last column of
    Matrix (Fortran arrays are column-major) though Overlay%Take(N) aka
    Overlay(N)%Take doesn't, so I'm not sure this is technically legal.
    Similarly for the antidiagonal, and hyper versions; but not more
    general isub cases, which you have under a different item number.

    With F2k03? "type kind parameters" you might be able to improve the
    syntax some, but I haven't looked at it in enough detail to be sure,
    and anyway that is neither official nor implemented yet.

    - David.Thompson1 at worldnet.att.net

  • Next message: kia: "cross chatter (was: Writing good numerical code in C)"