From: Dave Stampe-Psy+Eng <dstamp@watserv1.uwaterloo.ca>
Subject: Re: 386 renderer progress
Date: Thu, 9 Jan 92 00:05:27 -0500



Well, I submit that I CAN code math better than any compiler.  For example,
in one step of the scanline algorithm, I have to work with 58 to 28 bit
precision fixed point numbers.  This would be difficult, since there is
no 64 -> 64 bit divide.  However, my knowledge of the algorithm allows
me to find pairs of numbers in the subexpressions that have ratios that
can be expressed it 16 bits.  Volia-- problem solved.  You can't tell me
a compiler will know enough about the task to do that!

Also, that 800 or so lines of C in the subconverter is just an expression
of a good assembly programmer's knowlege.  It would take me 10 times
longer to create that code as it would to assemblyize the math.

Fear not, though: C versions of everything will be produced (although
perhaps not as one unit-- time pressures and all that).  So someone
else can spend the time to convert the stuff.  After all, what I'm
doing is not unique in any way: it's been done many times before.
I'm just optimizing it for one archetecture: the 386 PC.  The code will 
be valuable as an itegrated example, though.

Are you considering doing the port yourself, perhaps?

- Dave Stampe