From: Andrew Vincent (amv++at++sgi.com)
Date: 05/08/2000 03:25:02
Hugh Fisher wrote:
>
> I solved my sphere picking problem rather unexpectedly.
>
> My model had some 1200 spheres, each created with pfdNewSphere
> and positioned with a pfSCS. Andrew Vincent of SGI Australia
> emailed me about fast molecule display with the advice that
> transforming tessellated spheres is Not Good. So I flattened
> and cleaned the scene graph.
Just to clarify this, my private email to Hugh said (and I'm
sure Hugh will not mind my including a snippet of his reply to me):
> > The generic rule of thumb for spheres is "never tessellate"!
> > I have a suspicion some folks on this thread are transforming
> > tessellated spheres.
>
> Er, yes, guilty. I had a SCS above each sphere.
>
> I changed the program to pfFlatten and then pdfCleanTree the
> scene graph to get rid of them. Oddly enough, this solved the
> problem with the picking code crashing. Huh?
By this I meant that transforming the hundreds or thousands
of vertices of a triangular approximation of a sphere (such as
created by pfdNew Sphere) was not the fastest way to render
spheres, rather than transforming using an SCS node
was not a good idea.
(After all, if you couldn't transform things, where would
you be? Right back where you started from?-)
I think the picking problems were related
to allocation of pfSeg space, as suggested a few days ago.
A previous vendor I worked for a long time ago had so-called
"hardware spheres", which were essentially the billboard technique
described by Bram, except they used a complicated h/w caching mechanism
and Z-buffered blits, which eventually broke for perspective
transformation at one release of hardware.
But if you're not spatializing yourself (or using pfuSpatialize or
whatever it is called), then you are behind the eight ball, so to speak.
The fastest way to draw a sphere is not to draw it at all.
Andrew
-- Andrew Vincent +61 2 6270 8702 amv++at++sgi.com 26 Brisbane Ave Barton, Canberra A.C.T. 2600 Australia Lat: -35.226874, Lng: 149.038447 Elev: 595 m Datum: WGS84
This archive was generated by hypermail 2b29 : Mon May 08 2000 - 03:24:52 PDT