From: Dave Stampe-Psy+Eng <dstamp@watserv1.uwaterloo.ca>
Subject: Re:  Object database rendering
Date: Thu, 9 Jan 92 17:52:07 -0500



These ideas sound like either Z-buffering or subdivision algorithms.
I looked at these early on, but they actually are quite slow on
this hardware.  The Z-buffer (or polymap) idea takes about 1.7 uS
per pixel, and 1 uS if the pixel isn't drawn.  So it's about 5x
slower than just drawing everything (taking advanage of multipixel
writes).  The Hamlin-Gear is actually the best of the subdivision
algoritms by a factor of 10 or so.

BTW, if you ignore the background clear step as a "poly", the 
average screen depth of polys is about 1.75 in a reasonably uncluttered scene.
But local areas can reach depths >>10, so if your algorithm is n**2
with depth, you may be in for a suprise.  As for the assemblyization of
the algorithm, now that I have working C code, it's not too difficult.
I expect the assembly code to be fairly compact, and and I pl do 
"incremental" conversion using inline assembler.  Just now, I am a bit
short of time is all.

- Dave Stampe