>> The usual situations where manual bounding can be better are those where
>> the intersection or difference of two large objects produces a small
>> result.
>
> Understood; I've used this trick before and it can yield massive speed 
> gains in
> some circumstances.
>
>> Another factor is that the automatic bounding slabs are always aligned
>> with the axes.
>
> Presumably it follows that manual bounding shapes are transformed with 
> their
> parents and end up at angles, etc...?
>
>> Mesh objects render so
>> efficiently anyway that the gains would be unlikely to be very
>> significant unless the mesh object has a large number of large polygons.
>
> Hmm. I'll try to explain the source of my suspicions. I have a large 
> number of
> small meshes in a union (see my recent post in p.b.i). I'm using MegaPOV, 
> which
> draws bounding rectangles on the render window before it starts tracing - 
> I
> understand this is a representation of the vista buffer. All of the 
> bounding
> rectangles for my mesh objects seem to be much larger than the meshes. Is 
> this
> because of the way the vista buffer works or is it a symptom of 
> inefficient
> mesh bounding shapes?
>
> I'm sorry I don't have any screenshots - I'll try get some tonight if it 
> helps.

POV splits bounding of unions into bounding for the objects inside the
union, so if you bounded_by the union it'll almost always be less efficent,
since then it can't split the bounding of the union.

Bounded_by always adds an extra bounding test... for it to be better to
use manual bounding the automatic bounding needs to have lots of
empty space, and because automatic bounding boxes are aligned to the
axes that happens. It's also much faster to test bounding on a box that
is aligned to the axes...

Post a reply to this message