Hey, I didn’t know which subforum in here I should post this to but since this is a rendering related problem I figured I’d post it in here, so here goes:
I’ve been profiling a project lately and just noticed that Occlusion Culling seems to take absurdly large amounts of time to finish. Now, I don’t know if this is
a bug or anything, but it seems to be bottlenecking the entire rendering process and I’m not sure what measures I could take against that.
If there’s anyone who might be able to help me out with this, I’d be really grateful because I’m kind of at the point where I don’t know at all what to do now.
Cheers!
Start your editor with -opengl and profile again to see whether there are any changes. It is a known issue that the GPU profiler shows inaccurate timings under DX11, and they often appear in “HZB Setup Mipmap” and “Prepass”.
Make sure you’re using cull distance volumes for your level. Otherwise, you’re spending a lot of wasted time making the GPU check their visibility status.
Alright, I’ll do both those things! Thanks for your answers!
How exactly does the culling process work? Is there a broad phase (culling volume) and then a narrow phase, like with collision detection? Or does the cull distance volume completely replace occlusion queries?
Roughly, first object is frustum culled (determined if it is inside camera view). If so, the object’s depth is rendered by gpu and compared to depth buffer from previous frame.
Applicably, cull distance volume is applied before GPU occlusion querry.
Is it possible to have a volume where everything inside is rendered (and occlusion culled when necessary), but anything outside the volume is culled ?