Hyperplane Culling for Stochastic Rasterization

Abstract

We present two novel culling tests for rasterization of simultaneous depth of field and motion blur. These tests efficiently reduce the set of xyuvt samples that need to be coverage tested within a screen space tile. The first test finds linear bounds in ut- and vt-space using a separating line algorithm. We also derive a hyperplane in xyuvtspace for each triangle edge, and all samples outside of these planes are culled in our second test. Based on these tests, we present an efficient stochastic rasterizer, which has substantially higher sample test efficiency and lower arithmetic cost than previous tile-based stochastic rasterizers.