🚀 go-pugleaf

Anthony Berno writes
> 
> In light of the release of 3.0 with bundled Renderman being just a few  
> months away, it is interesting to note that in a recent issue of Computer  
> Graphics World that the closing column discusses Silicon Graphics'  
> licensing of OpenGL, a derivative of the Iris GL, to other OS vendors.
> 
> The column discusses a few other graphics systems, such as PHIGS et al.  
> and comes to the definite conclusion that OpenGL will be THE standard for  
> high end graphics starting NOW, and that it is the BEST of all of them.
> 
> Has NeXT made a critical mistake in going with Renderman? Is the NeXT, in  
> the wake of this possible sweep of graphics systems, going to be left in  
> the dust with yet another fundamental incompatibility? I thought that  
> Renderman would, in only a few years, propel NeXT into the front ranks of  
> 3D graphics / CAD / animation. Now I'm not so sure.
> 
> Comments?
> 

This is an interesting question -- one that I've been thinking about off and on  
for the last year.  I still don't have a definitive answer.  I'll just raise a  
few interesting points for people who aren't real familiar with GL and  
RenderMan.

First of all, computer graphics can be broken down into three main areas:  
modeling, animation, and rendering.  

- Modeling is the process of mathematically describing the properties of a  
given graphical entity.  Points, vertices, connectivity, patch properties,  
color, material, etc.  

- Animation is the process of changing your model over time.  This can include  
changing the position, shape, shading characteristics, etc.

- Rendering, finally, is taking the model for each frame and generating the  
pixel data representing the scene.


GL, now called OpenGL, also (I believe) GL 5.0, is the latest and greatest  
version of Silicon Graphics "Graphics Library".  Graphics weenies like GL a lot  
because it lets them, as I like to think of it, "scrape their fingers on the  
metal" (or silicon, in this case).  GL is feature rich and offers function  
calls for most everything you need to do way spiffy, way fast graphics.   
Theoretically, it's portable among all SGI platforms, but in reality, you can  
paint yourself in a corner quickly.  GL suffers from the fact that it's a  
graphics library written for graphics hackers by graphics hackers.  No one from  
SGI would ever insult anyone's intelligence by claiming that GL was any sort of  
object oriented approach to graphics.  To write a program in GL is somewhat  
similar to writing a program on the Mac.  You learn as many calls as you can,  
then you contact your friend who already writes software for the platform, get  
a skeleton of code from them, and modify to suit your needs.  You rarely write  
a reasonable useful program using GL from scratch.  

With the upcoming Scenario package, which is SGI's object oriented layer on top  
of GL, many of the programming issues might go away, but I'm not sure of  
Scenario's relationship to OpenGL.

To begin the segue to RenderMan, it's important to realize that people who use  
SGI's for production work (pick your favorite production house, ILM, Pixar,  
PDI, Rhythm & Hues, MetroLight, etc.) never (or at least, very, very rarely)  
use GL output for final work.  They all have their own software-only renderers  
which do the final rendering.  They use GL to write programs which help them  
create models (the modelling phase I talked about above), choreagraph the  
motion of those models over time (the animation phase I talked about above),  
and finally they use GL to render their image quickly to the scene (the  
rendering phase I talked about).  In all cases, when they are finally satisfied  
with the animation they've created, they use their own software renderer to  
render the final frames.

So how does RenderMan fit in?  Well first of all,  a bit of history.  A few  
years back, for various and sundry reasons, the folks at Pixar, with the help  
and advice of most everybody who is anybody in computer graphics, defined a  
"scene description language" called RenderMan.  The basic idea of RenderMan was  
to define a language for modelling software to talk to rendering software  
concerning a photorealistic description of a scene.  The idea was that if they  
were sucessful, people could sell renderers that understood this language, and  
then modellers would only need to speak one language - RenderMan.  RenderMan  
included a really cool thing called a shading language, which was a little  
language (with a C-like syntax) that allowed you to write your own shaders.   
It's a little too complicated to go into here, but suffice it to say that much  
of the power of RenderMan is hidden in the shading language.  

It's important to make the distinction that Pixar defined the RenderMan  
interface, much as Adobe defined the Postscript interface.  Additionally, Pixar  
sells a product (bundled in 3.0) called prman, or PhotoRealistic RenderMan.   
This is a renderer which conforms to all the required portions of the RenderMan  
interface.  I believe their are other renderers out there which support a  
RenderMan interface, but don't know names off the top of my head.  Note that  
this is exactly analogous to the fact that other vendors offer Postscript RIPs  
made by people other than Adobe.


The important distinction to catch here is that gl is intended as a general  
purpose graphics library for doing fast graphics, but doesn't really have the  
hooks to do photorealistic rendering.  As time goes on, the smart folks at SGI  
have been adding functionality to get gl to asymptotically approach  
photorealistic rendering, but it's certainly not there yet.  RenderMan, on the  
other hand, was designed from the ground up to be a functional interface to a  
piece of software which, by definition, could photorealistically render a  
scene.  With the advent of Interactive RenderMan (which, as you would imagine,  
was developed on an SGI written in GL), RenderMan is pushing down,  
asymptotically approaching doing fast graphics.

The idea of NeXT's 3DKit, written by Dave Springer, is to hide much of the  
complexity of writing 3D applications on the NeXT.  The good thing is that you  
get the power of RenderMan rendering pretty much for free.  The bad part is  
that it (the realtime aspect of it) only commercially exists on the NeXT, and  
we're subject to NeXT's speeding its performance up.  With GL, you have a much  
harder time writing software, but you have the ability to go from an Indigo (GL  
in software, a machine costs ~10K for a minimal usable system) all the way to a  
4 processor, 2 pipeline SkyWriter (~450K, 1 million polys/sec per pipeline).

I've been around SGI machines since 1987, but haven't hacked them anywhere near  
as much as coworkers.  They like the power, like the flexibility.  But there  
are gaps.  Recently, some folks in my group were porting some software to our  
SkyWriter and found some obscure restrictions on texture maps.  These sorts of  
restrictions don't exist with RenderMan, but Interactive RenderMan isn't  
available on the SkyWriter.  I've been a fan of RenderMan since I saw the draft  
spec, and think NeXT did the right thing by hitching their wagon to it.  GL is  
different, but it's hard to explain to non graphicists why.  I think both of  
them have a place.

Maybe now you can see why I'm still not sure about this stuff.  It's a very  
confusing subject...

Hope the above was informative.


--

-->  Michael B. Johnson
-->  MIT Media Lab      --  Computer Graphics & Animation Group
-->  (617) 253-0663     --  wave@media-lab.media.mit.edu
-->  NeXT Mail accepted at  wave@waits.media.mit.edu