I'd like to know what's the most efficient way of doing frustum culling using the programmable pipeline. I mean, if I understand correctly, following the method described here: Geometric Aproach (by the way the only method described there that worked for me some time ago), functions like glGetFloatv(GL_MODELVIEW_MATRIX, ...) are not valid anymore, as the final vertex position is computed in shader stage. Do I have to compute the frustum planes on the client side for every bounding box transformation I have to check before rendering?
The idea of frustum culling is to prevent polygons from being sent to the GPU in the first place, those polygons you already know that will be culled after the vertex shader. So idea is to prevent the vertex shader from transforming those polygons. Using shaders or not, the best way is to keep track of the frustum planes on client side, and traverse the scene graph (could be hierarchical tree or just a list) and cull objects that lay outside the frustum, and don't use
glGetFloatv or it's equivalent it is not efficient as it will copy the data from the GPU. You can use feedback buffers instead.