Manta Interactive Ray Tracer Development Mailing List

Text archives Help


Re: [MANTA] Re: Manta Problem


Chronological Thread 
  • From: Dirk Reiners <dirk@infiscape.com>
  • To: manta@sci.utah.edu
  • Subject: Re: [MANTA] Re: Manta Problem
  • Date: Fri, 11 May 2007 10:09:59 -0500

        Hi Abe,

Abe Stephens wrote:
Manta supports all of those things (geometric primitives, instances, acceleration structures; BVH's are used more than kdtrees), except that the renderer itself doesn't contain any logic of how to split a scene into different acceleration structures.

I've viewed structure optimization as more the responsibility of the scene graph since it contains information about all of the geometry and can make a global decision. I don't think it's necessary for the renderer itself to be aware of the scene graph library at all.

Do you agree? If every rendering backend has some notion of a bvh, kdtree, and a grid, could the scene graph intelligently decide by itself how to chop up the scene?

It could, although in most cases the scenegraph is used for structuring the scene more at a logical rather than an acceleration-based level, i.e. the App decides which structuring makes the most sense. Most scenegraphs have to ability to do further splits, but often the apps decide not to use them because they need the original objects.

But IMHO the main difference between a regular scenegraph and a hierarchy for RT is the level it's done at. To get good OpenGL performance you want large objects (thousands of polygons), to get good RT performance you need to the hierarchy to split down to the triangle level (or close to it). A 1:1 mapping from sg to rt hierarchy won't give you any kind of good acceleration.

Unless I misunderstood and manta internally splits objects further down. In any case, a good hierarchy builder is IMHO a critical component of a ray tracing system...

My $.02

        Dirk





Archive powered by MHonArc 2.6.16.

Top of page