Hmm, actually the texture cache isn't relevant because it would be invalidated on frames that are important here. Yes I'm speculating then because the performance hit is now GPU-specific.
Another speculation is that performing occlusion tests at all is fast enough to justify the effort. That's an assumption made in Mir right now. But I know from previous work it doesn't always hold true.
Hmm, actually the texture cache isn't relevant because it would be invalidated on frames that are important here. Yes I'm speculating then because the performance hit is now GPU-specific.
Another speculation is that performing occlusion tests at all is fast enough to justify the effort. That's an assumption made in Mir right now. But I know from previous work it doesn't always hold true.