A Parallel Graphics Machine For High-speed Ray Tracing

This paper introduces a parallel high-speed ray tracing rendering machine which exploits three levels of parallel processing: ⅰ) multiprocessing at the system level, ⅱ) macropipelining at the processing-element level, and ⅲ) instruction-level parallel processing at the stage level. This paper focuses on macropipelining, instruction-level parallel processing, and 2 rank uniform object-space subdivision. First, a ray tracing algorithm, called 3MPRT(3-stage macropipelined ray tracing), is presented to show how the ray tracing task is divided into three subtasks: as (object search), IC (intersection calculation), and SC (shading calculation). The paper then describes 3MPPE(3-stage macropipelined processing element) architecture which implements 3MPRT directly and provides VLIW processors for as and IC stages. Both 3MPRT and 3MPPE using 2 rank uniform object-space subdivision are evaluated through simulation, varying the scene characteristics and the object-space subdivision. From the simulation results, the paper concludes that a single 3MPPE can achieve speedup of 2.6 - 4.8 over non-macropipelined, non-VLIW implementations of 3MPRT.