Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cornell!batcomputer!riley
From: riley@batcomputer.tn.cornell.edu (Daniel S. Riley)
Newsgroups: comp.graphics
Subject: Re: MTV-Raytracer on ATARI ST
Message-ID: <7511@batcomputer.tn.cornell.edu>
Date: 7 Mar 89 17:50:21 GMT
References: <1171@laura.UUCP>
Reply-To: riley@tcgould.tn.cornell.edu (Daniel S. Riley)
Organization: Cornell Theory Center, Cornell University, Ithaca NY
Lines: 24

In article <1171@laura.UUCP> wagener@unidocv.UUCP (Roland Wagener) writes:
>I have ported the MTV-Raytracer on the ATARI ST using the Turbo-C-
>Compiler. The Program works fine and it uses about 3 hours CPU-Time
>to create the Balls-Picture in 320x200 Resolution.

>But there is a bug somewhere in the program. There are white spots in
>all reflecting surfaces. This bug does not apear on a IBM-PC programmed
>with Zortech-C++. But the PC needs 5 hours for a 200x200-Picture ...

This sounds like a floating point precision problem.  I've been playing
with a number of ray tracers, including MTV, on my Amiga.  I've seen
white spots and other sorts of splotches if I use the Motorola ffp format
floating point routines, which are single precision (32 bit) only.  They 
go away if I use IEEE math libraries (all calculations done with 64 bit
doubles).  3 hours cpu for a 320x200 picture on an 8 MHz 68000 sounds like
single precision to me, but I don't know the ST or Turbo-C that well.

I suppose there must be papers on controlling round-off errors in
ray tracing algorithms, but none of the ray-tracers I've seen make any
special efforts in that regard.  Of course, all the ones I have source
code to are meant to be clean and simple, not fast and convoluted...:-)

-Dan Riley (riley@tcgould.tn.cornell.edu, cornell!batcomputer!riley)
-Wilson Lab, Cornell U.