A Unified Graphics Rendering Pipeline for Autostereoscopic Rendering

Autostereoscopic displays require rendering a scene from multiple viewpoints. The architecture of current-generation graphics processors are still grounded in the historic evolution of monoscopic rendering. In this paper, we present a novel programmable rendering pipeline that renders to multiple viewpoints in a single pass. Our approach leverages on the computational and memory fetch coherence of rendering to multiple viewpoints to achieve significant speedup. We present an emulation of the principles of our pipeline using the current-generation GPUs and present a quantitative estimate of the benefits of our approach. We make a case for the new rendering pipeline by demonstrating its benefits for a range of applications such as autostereoscopic rendering and for shadow map computation for a scene with multiple light sources.