Path: utzoo!attcan!uunet!decwrl!apple!bionet!ames!uhccux!virtue!comp.vuw.ac.nz!munnari.oz.au!mudla!ok
From: ok@mudla.cs.mu.OZ.AU (Richard O'Keefe)
Newsgroups: comp.lang.fortran
Subject: Re: What is the FORTRAN for ?
Keywords: FORTRAN, stupidity
Message-ID: <4923@munnari.oz.au>
Date: 27 Jul 90 12:53:50 GMT
References: <1990Jul25.174153.16896@ecn.purdue.edu> <14448@lambda.UUCP> <7404@drydock.cray.com>
Sender: news@cs.mu.oz.au
Lines: 29

In article <7404@drydock.cray.com>, jot@drydock.cray.com (Otto Tennant) writes:
> One of the key facctors is multi-dimensional array processing.
> In many languages, a multi-dimensional array is not a "first-class"
> object.  Instead, a two dimensional array is a vector of vectors,
> and so on.

> This means that there is a preferred direction for vecttorization.

Ahem.  Not to knock Fortran, _but_ arrays are not first-class objects
in Fortran 77, not like they are in Lisp or APL or Pop or Scheme.  It
has historically been the case that Fortran users _knew_ that "arrays
are stored in column-major order", and if you have arrays which are
large compared with your system's page size, it _still_ pays you to
access Fortran arrays in the "easy" direction, e.g.
	DO 10 ICOL = 1, N
	    DO 20 IROW = 1, M
		use A(IROW, ICOL)	!IROW varies "faster"
    20	    CONTINUE
    10	CONTINUE

What if any difference this makes to vectorisation depends on how
well your system handles non-unity strides.  I _have_ encountered
a compiler which could vectorise
	DO 10 ICOL = 1, N
	    DO 20 IROW = 1, M
		A(IROW,ICOL) = B(IROW,ICOL) + C(IROW,ICOL)
    20	    CONTINUE
    10	CONTINUE
but if you switched the nesting of the DO statements it wouldn't.