 |
I'm still a bit scary that we might have situations where mapper puts e.g. a barrel into the wall and it should be visible in both rooms. The suggested optimization would definitely break such a case. Such cases should be rare, but it still sucks to fix them all the time... Instead, I decided to improve the algorithm for searching the areas covered by an entity. Right now we check oriented bounding box against BSP tree. I decided to check BVH trees of static models instead, since they should give slightly better bounds. This did not work out: I did not receive any noticeable improvement in Perilious Refuge regarding the number of entities drawn in view. The only difference that I found was BVH code giving MORE refs, and it turned out to be because it uses "<=" instead of "<" comparison in box-vs-plane check =) Despite my worries about performance of such a check (I planned to enable it only for the very start of the level), it actually worked pretty fast even globally. Still, it is a little bit slower because I changed "common children" important optimization to several areas in a node, which only slowed things down. Perhaps the reason for this is that most of our models are very low-poly and BVH has little effect. Perhaps that's because the whole issue of static entity leaking a wall almost never happens. Really, this is the problem only when entity is within room A, but it's oriented bounding box ix so large that it touches room B behind a wall. Anyway, I decided to not commit this. It is a big change which obviously gives nothing =( |
- Anonymous
