Path: utzoo!mnetor!uunet!husc6!think!ames!hao!noao!mcdsun!mcdchg!clyde!watmath!atbowler
From: atbowler@watmath.waterloo.edu (Alan T. Bowler [SDG])
Newsgroups: comp.lang.c
Subject: Re: Fast way to update grids
Message-ID: <17001@watmath.waterloo.edu>
Date: 19 Feb 88 01:56:42 GMT
References: <255@imagine.PAWL.RPI.EDU> <690005@hpfelg.HP.COM> <2212@geac.UUCP>
Reply-To: atbowler@watmath.waterloo.edu (Alan T. Bowler [SDG])
Organization: U. of Waterloo, Ontario
Lines: 22


In article <2212@geac.UUCP> daveb@geac.UUCP (David Collier-Brown) writes:
>  As
>the white book is written to allow edge-vector addressing, and at
>least one pcc-based compiler uses it, this turns out to be
>compiler-dependent.  
>  As it happens, the Waterloo (GCOS) C compiler uses edge-vectors, but
>also (if memory serves) keeps the pointed-at data contiguous to
>allow such assumptions to work nevertheless.  Something of a ``best
>of both worlds'' effect, at no extra cost to the program/programmer.

Sorry Dave, but that is just not true.  You get edge vectors only
if you declare the structure as having edgevectors.  If you declare
a multidimensional array you get the conventional contiguous chunk
of memory and addressing polynomials to access it.  Are you sure
you aren't getting confused with B (and BCPL) where edgevectors
was the only choice?

                           Alan Bowler
                           Product Support
                           Software Development Group
                           University of Waterloo