Quote:
|
Originally Posted by smogo
Quote:
|
Checking to see if a point lies within a square requires two subtractions and two comparisons. Proper vector distance calculations for radial inclusion usually require 3 multiplies an addition a subtraction and a comparison.I haven't looked at the code for this lately, but it does still do a vector distance calc, doesn't it?
|
the actual code does box-checking in AICheckCloseArrgo as far as i know, thus abs(x1-x2)<dist && abs(y1-y2) <dist.
In the Terrain code, it is exact-distance calc. However this is little overhead, as :
- those in the corner (pass square test, fail exact test) are not checked against lately in AICheckCloseArrgo), thus time saved in the end
- event though caching uses cpu (about 3% in very heavy zones), it wipes-out a lot of tests afterwards. |
Yeah, I remember when Trump added that box filter. I think it _adds_ to computation time instead of reducing it. My suggestion was that we rely solely on it, instead of using it as merely a filter. Maybe I'm out of my element, but bounding volumes seem like a very forward-looking solution. They are extremely fast and map nodes can be mapped onto them in a fairly straight-forward manner (Kay-Kajiya bounding trees, maybe?). i.e... you can just lop off portions of your bounding square to reflect walls which nets you line-of-sight calculations for free.