> Operating on the theory that my csg was slowing down the render, I decided to
> rebuild the scene from the void up using constructive csg as much as possible.
>
> What follows is the current state of the scene, built almost entirely out of
> prisms.  The code is quite a bit less readable, but there are no difference(s)
> at all.
>
> The major speedup came when I changed the way I tiled the floors.  Rather than
> making a single large union of all the tiles on a particular level, I altered my
> macro to pass back an array of tiles which I then use to tile a given area. I
> cull tiles which fall inside areas I want kept clear using an inside(...)
> statement.  Render speed increased to "OH MY GOD THAT'S FAST".
>
> Render Time:
>    Photon Time:      No photons
>    Radiosity Time:   0 hours  0 minutes 17 seconds (17.107 seconds)
>                using 6 thread(s) with 95.544 CPU-seconds total
>    Trace Time:       0 hours  4 minutes 26 seconds (266.060 seconds)
>                using 6 thread(s) with 1535.187 CPU-seconds total
>
> The upshot seems to be a problem in bounding of difference objects where the
> base is composed of many small objects, much like you might expect from turning
> off split_unions.
>
> Regards,
> A.D.B.
>

Yes, modeling a scene differently can have a *huge* impact on the render 
time :)

Another thing that can be done:
When your tiles, or any kind of objects, are hiden under some other 
opaque object and you see them in some hole in that object, you don't 
need to use a difference nor intersection to cut them away. Makes the 
scene simpler.



Alain

Post a reply to this message