I am going to test out two methods for optimizing complex mesh rendering.
Primarily, I am going to work on a method I have used in the past involving storing multiple polygon lists given the angle the model is viewed from. Essentially the model is stored as size separate versions: top, bottom, left, right, front, rear. If the camera is behind the object, the rear list is called et cetera. This is a crude methods but it is simple and from experience works somewhat well (75% of mesh rendered at a time using 4.5x the memory footprint). The effect is almost linear optimization (dot products are fixed CPU cost).
The second method has two branches. IE, optimization via polygon subtraction and addition with groups based on their dot product with stochastic vectors which are pre-processed and depending on how effective they are (IE how many polygons belong to them) possibly eliminated. Polygons belonging to no group will be added to a list which is rendered every time.
These methods assume the camera at no point enters the bounding volume, in this case AABB, of the object. If the camera does enter that close, a list containing all polygons will be called.
There is also possibility for SSE optimization.
Tags: graphics, mesh, mesh optimization, opengl
