An Efficient Multi-View Rasterization Architecture

Abstract

TV have been designed and built. However, these displays have received relatively little attention in the context of real-time computer graphics. We present a novel rasterization architecture that rasterizes each triangle to multiple views simultaneously. When determining which tile in which view to rasterize next, we use an efficiency measure that estimates which tile is expected to get the most hits in the texture cache. Once that tile has been rasterized, the efficiency measure is updated, and a new tile and view are selected. Our traversal algorithm provides significant reductions in the amount of texture fetches, and bandwidth gains on the order of a magnitude have been observed. We also present an approximate rasterization algorithm that avoids pixel shader evaluations for a substantial amount (up to 95%) of fragments and still maintains high image quality.