Hello Mike,
Thanks a lot, good questions and suggestions here.
> - any ideas on exactly where you're bottlenecked?
>
As I said, if I set nodemask to 0 on the cables the draw time goes way
down. So I'm pretty sure the bottleneck is there (in other words, it's
not how I build the geometry, because that's being done anyways; it's
not the physics either). Though more specifically than that I can't say.
> - if perception of cable width is critical, how are you ensuring the quads
> are orthogonal to the viewer? Billboards, adjusting the quad verts,
> viewer/quad relationship is fixed, etc?
>
Sorry, I wasn't very clear. It isn't billboards, and I can see that it
sounded like that the way I described it. We're building "cylinders"
with quad strips. So we'll give a subdivision level (number of quads for
a segment, for example 6). And then the cable is a set of segments, each
of which consists of 6 quads in this case. Since in this case, the cable
with the most segments has about 20 of them (120 quads), I don't quite
see why it's so slow. Of course, it might be the display list rebuild
time, but I can't switch them off for some reason...
> - quadstrips are converted to tristrips in the driver; perhaps there's
> advantage in dispatching tristrips directly.
>
I thought of this too, but didn't think it could be significant. If you
think it might be something to try, I'll try it out and see.
> - Ideally, and your target hardware supports it, suggest dispatch linestrip
> and tessellate each linesegment into a pair of tris in a geometry shader. A
> vertex shader could take care of texcoord computation given cable/segment
> lengths.
>
I haven't had much chance of using geometry shaders yet, unfortunately.
I'd really have to demonstrate a large incentive to use them (though the
hardware we ship to clients will likely support them anyways). It could
also be used if the hardware supports it, with other fallbacks if not.
We'll see in the next few months.
Anyways, thanks a lot, I'm at least making progress and have a few
things to try out.
J-S
--
______________________________________________________
Jean-Sebastien Guay [EMAIL PROTECTED]
http://www.cm-labs.com/
http://whitestar02.webhop.org/
_______________________________________________
osg-users mailing list
osg-users@lists.openscenegraph.org
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
