A parallel architecture of ray tracing image generation.

摘要

In computer graphics, the ray tracing algorithm can generate highly realistic images. However, it has a major disadvantage: the high computational expense associated with generation of an image. To increase the image generation speed, we present a specialized pipelined dataflow architecture for parallel processing ray tracing.;Next, we describe the image generating process on the proposed architecture and extend the application of the architecture to ray tracing image sequence generation and ray tracing stochastic surfaces. A new ray tracing image sequence generation algorithm is presented and mapped functionally to the proposed architecture. To expedite the rendering speed of two types of stochastic surfaces, we propose an intersection processor design.;Then, we present alternative implementations for some of the key components assuming the size of the target ray tracing machine is limited to a small scale implementation. Our proposed approach is compared with other techniques for improving the throughput under this constraint.;In the last part of the thesis, we study the performance of the overall architecture. To understand its performance, we model the load of the computation based on probability and model the overall architecture with a closed queueing network. In addition, we simulate the image generation task on a simulator based on the ray tracing attributes collected from ray tracing real images. The results show that a substantial speedup improvement can be achieved by the system and the hardware complexity of the system is reasonably low.;In the first part of the thesis, we present the organization of the proposed architecture. Some of the architectural issues are presented in detail along with solutions to them.