[rescue] ZX coolness

Tue Oct 8 01:00:37 CDT 2002

On Tue, Oct 08, 2002 at 01:44:32AM -0400, Joshua D Boyd wrote:
> On Tue, Oct 08, 2002 at 01:34:00AM -0400, Dave McGuire wrote:
> > On Monday, October 7, 2002, at 11:50 PM, Joshua D Boyd wrote:
> > >I just came across a document titled "SPARCstation10ZX and SPARCstation
> > >ZX Graphics Technology White Paper" from Sun.  According to this, the 
> > >ZX
> > >board tesselates trimmed NURBS surfaces in hardware.  Way cool.  Off 
> > >the
> > >type of my head, the only other graphics board that does that is the IR
> > >line in Onyxs.
> > 
> >   Sounds cool to me.
> > 
> >   Got time for an English translation?
> 
> Yeah, my file system is correcting the errors caused by crashing (note,
> I doubt it is linux's fault, but rather I blame it on the hardware
> makers.  Especially the video card one who refused to use standard
> drivers or userlang drivers, but instead had to supply proprietary
> drivers as a kernel module).

Um, if Linux let it happen, it is Linux's fault.  Unless you were
running everything as root, which I doubt you were.

As far as SGI's go, I think any hw that quotes Tmesh numbers is going
to have accelerated tesselation as well (at the very least).  Thus XZ
and up offer some acceleration.  

> OK.  I'm not sure what the adjective trimmed means when applied to
> NURBS.  As a major NURBS fan, Kurt might be able to answer tht.  But,
> apparently it makes NURBS far more complicated to calculate.

Google is your friend on this kind of stuff.

> So, we have a hard to calculate parametric surface (meaning that the
> surface is defined by a mathematical function rather that lots of little
> triangles or quads).  The usual way to display it would be to have

A 2D analog to this would be a Bezier curve.

> software run on the CPU that goes and turns that surface into a long
> list of triangles.  This list could be saved to be used again, as long
> as the surface doesn't change (which is going to be often if running a
> CAD program, probably), then the list has to be recalculated.  
> 
> So, what the ZX board does is rather than the host CPU calculate the
> triangulation, the board can do it itself, leaving the CPU to worry
> about other things.  The only other case of graphics hardware that I
> know of that does that is Infinate Reality (although it would surprise
> me if Impact did also).

I think you are conflating things here Josh (but I could be wrong).

An analog to trimming might be the calculation of clipping paths in
Postscript, or hidden line removal in a 2D view of a 3D object.

Tesselation is the step where a surface is turned into a bunch of
small triangles.  

Trimming the NURBS would occur in a step previous to tesselation.  You
could, with a stretch, say that tesselation is a step similar to
rasterizing a bitmap from a Postscript or other vector description of
an image.  

As such it would be perfect for an accelerated function, since:

a) sending all those triangles across a bus would be slow and you
would see immediate speedups due to reduced bus use and the data being
rendered locally
b) it is a simple function that lends itself well to being put on a chip
c) it is common in many applications and is done the same way each time

./patrick