Re: [grt-talk] Some suggestions.

[Nikodemus Siivola]

On Thu, 8 May 2003, Anton N. Mescheryakov wrote:

Sorry for terseness, I'm in a bit of a horry.

> Ditto. Let's go and we'll see. BTW, did anyone run a profiler over grt?

Yes. Most of the current inline declaratios, plus some other things are
result of profiling.

> Why *less*, BTW? I've thought it's *equal*.

POVRay is quite a bit more than just a classical raytracer: it
includes stochastic radiosity and backwards raytracing. And as a raytracer
it's more than nice. It's superlative.

My only quibbles with POVRay are it's restrictive license wich precludes
doing several interesting things with it, and the abomination of a
language.

> 1. It looks like grt lacks some interesting feature called "scene
> graph". That is, scene objects seems to be thrown in the single layer
> without transformation groups and so on.

Spatial coherence will be exploitable in 0.2: Scene graph == CSG tree, and
we will also have an octree.

> 2. CSG stuff is really good indeed, besides one severe limitation: it's
> two-way only. I think that it may be resolved in conjuntion with more
> robust occlusion/visibliliy testing.

??? I'm sorry, but I don't understand. Can you please explain. Also, the
CSG is still immature.

> 3. I'm somewhat new in raytracing internals, as you could see. So one
> thing is really strange for me: is it absolutely nescessary to transform
> ray->object space for each test rather than transform object->ray space
> once per traversal?

Good question. I *think* so. Bounding tests are done in ray space, so can
help us avoid unnecessary transformations. Also, currently it might be more
effective to do it the way you suggest, but octree should bring the number
of intersection tests towrds O(log N).

But it bears to keep your point in mind. I'll think about this.

Cheers,

  -- Nikodemus