Yeah, that was my next question -- how much would it really affect OSG? If you're
transform or fill or shader bound, it's not gonna make much difference. I guess it might
affect the Cull stage, if you're bound up there.
It all depends upon the scene graph you are trying to render.
Town models I have are all CPU bandwidth bound, unless you really crank up the antialiasing it hard to hit fill. Transform bound apps are very rare, but tend to be bandwidth and dispatch bound before they ever get to be genuinely bound by vertex processor performace. There specialist apps which break rule this though ;-)
Fill rate bound apps tend to be the ones with complex shaders, high overlay or heavy texture read demands (i.e volume rendering).
Only if you are CPU bandwidth bound will any of these datastructure reordering be of any significance, even then its unlikely to be a huge difference, 20% difference I think would be a major achievement for this type of optimization. Typically you can achive this type of gain much more easily with better design of your scene graph.
However, it doesn't mean that making improving isn't usful, but we must be aware of getting too suckered into work that seems worthy, but in the end is never going to make a huge difference, we arn't going to see a double in performance by just jigging data structures.
Robert.
_______________________________________________ osg-users mailing list [email protected] http://openscenegraph.net/mailman/listinfo/osg-users http://www.openscenegraph.org/
