Path: utzoo!attcan!uunet!decwrl!sgi!shinobu!odin!texas.esd.sgi.com!robert
From: robert@texas.esd.sgi.com (Robert Skinner)
Newsgroups: comp.graphics
Subject: Re: A-Buffer source (and plotting lots of triangles)?
Message-ID: <10760@odin.corp.sgi.com>
Date: 23 Jul 90 17:40:57 GMT
References: <1990Jul16.225553.2623@maths.tcd.ie> <1264@idunno.Princeton.EDU> <10491@odin.corp.sgi.com> <11657@hydra.gatech.EDU>
Sender: news@odin.corp.sgi.com
Reply-To: robert@sgi.com
Organization: Silicon Graphics Inc., Entry Systems Division
Lines: 66


In article <11657@hydra.gatech.EDU>, gg10@prism.gatech.EDU (Gregory L.
Galloway) writes:
|> In article <10491@odin.corp.sgi.com>, robert@texas.esd.sgi.com
(Robert Skinner) writes:
|> > I implemented a renderer with a Reyes-like front end, and two scan 
|> > conversion back ends:  scanline Z buffer with 4x4 or 8x8
|> > stochastically jittered subsamples,  an alpha buffer with 4x4 or
|> > 8x8 masks.
|> > 
|> > Robert Skinner
|> > robert@sgi.com
|> 
|> Cook, et al., said that the Reyes system worked really well when splitting
|> patches, but not polygons.  

that's not how I interpreted the statement:

	"Polygons in general do not have a natural coordinate system
	for dicing.  This is fine in our case, because bicubic patches
	are our most common primitive, and we hardly ever use
	polygons."

I think the first sentence refers to polygons with 5 or more vertices,
in which case I agree.  How do you split or dice a pentagon?  I
believe the second sentence implies that Pixar almost always uses 
more sophisticated primitives than polygons.  I also got the
*impression* that polygons didn't render much faster than bicubic
patches, so why bother with polygon limitations?

|> How about just simple triangles?  Has anyone
|> tried writing a rendering system based on Reyes which just uses simple
|> midpoint subdivision (as in fractals but w/o the random displacement)?  

that is basically what I did.  I rendered triangles and quadrilaterals
and did simple midpoint subdivision (I didn't do anything special for
bow-tie quads.)  

|> I'm
|> curious to know how the system performs compared to a simple Z-buffer w/ 
|> straight scan-conversion and no stochastic sampling.

well, you are going to pay the price for the splitting and dicing that
Reyes does that a plain vanilla Z-buffer doesn't.  In addition, the the
straight scan-conversion algorithm could use coherency effectively on
large polygons.  Reyes micropolygons are much smaller, so you would
lose some there too.

I think Reyes is a beautiful system.  Its clean and simple.  I plan on
re-implementing my Reyes-like renderer in C++ to take advantage of the
(sort of) object-oriented style.  But I think it sacrifices some 
speed for the great God of realism and the potential for parallelism.  

I think the trade-off is worth it.  

|> 
|> Thanks in advance,
|> 
|> Greg Galloway
|> gg10@prism.gatech.edu

Robert Skinner
robert@sgi.com

You will encounter several species of small, furry animals,
 gathered together in a cave and grooving with a pict