Re: [Opensg-core] Re: Occlusion Culling commit

[Andreas Z. <am...@gm...>]

Hi Dirk,

> 	Hi Andreas,
> 
> On Mon, 2005-12-12 at 09:25 +0100, Andreas Zieringer wrote:
> 
>>ok it is a rather simple implementation but better than nothing ;-) 
> 
> 
> That remains to be seen. ;) Have you benchmarked it? Occlusion culling
> can actually hurt performance pretty bad if it's working well.
> 
> 
>>The 
>>main goal was to improve the render speed for scenes with many pixel 
>>shaders. 
> 
> 
> I can see it making sense if few large, expensively shaded objects are
> hidden. Allen's scenes have thousands of very small expensively shaded,
> transparent, multi-passed objects.

On my slow fx5200 it is most of the time faster with the occlusion 
culling enabled. I need to test this with a faster card. BTW I'm also 
doing the front-to-back rendering. For a worst case scene with about 
100000 geometry nodes the sorting hurts a bit but it is not too bad.

> 
>>If you have any ideas how to improve the speed please tell me :-)
> 
> 
> It gets more complicated, as you have to keep multiple queries in flight
> to get around the pipeline latency, and you need to sort stuff front-to-
> back for it to actually have an effect, without spending too much time
> on sorting. There are a lot of contradicting effects that need to be
> balanced. 
> 
> Christoph is writing his PhD about it, and Dirk Staneker wrote a lot of
> papers on that topic, too, maybe they can comment a little more.
> 	Dirk

Yes I already got the source code from Christoph looks interesting I 
will try to integrate it.

Andreas