LUX: An Heterogeneous Function Composition Parallel Computer for Graphics
In this paper, we present an heterogeneous parallel computer dedicated to high realism computer graphics. A small network, with a reduced chip set, allows us to reduce rendering time by a very attractive factor. The low level mechanisms of the network are
- PDF / 149,405 Bytes
- 10 Pages / 431 x 666 pts Page_size
- 99 Downloads / 193 Views
stract. In this paper, we present an heterogeneous parallel computer dedicated to high realism computer graphics. A small network, with a reduced chip set, allows us to reduce rendering time by a very attractive factor. The low level mechanisms of the network are designed to manage the wide variety of data and algorithms used in computer graphics. Some nodes of the network may be specialized in the most time consuming parts of the algorithm and have specific data paths. Thanks to the function composition scheme, we unify both the management of specialization and of parallelism. Those mechanisms allow flexibility and easy design of programs.
1 Introduction The graphic community is in need of very high computational power to achieve high realism pictures at interactive rates. We observe that single general purpose computers can’t provide the needed performance, and general parallel computers hardly reach them, and for a high financial cost. In this paper, we present an heterogeneous parallel computer which may reach expected performances, with a reduced chip-set and this for a low financial cost. We are particularly interested in the speed up of Ray-Tracing algorithms. They simulate the propagation of light in an environment, the scene, made of geometric primitives. The most time consuming part of such algorithms is the computation of the intersection between the rays and the scene. The computation of surface lighting characteristics can also be quite expensive. The graphic community has developed many algorithms to reduce the rendering time, essentially by reducing the number of ray-object intersections, but it’s still quite high. We propose to design specialized data paths to speed-up the most time consuming parts of the algorithm. Those specialized units are interconnected through a dedicated network to form an heterogeneous parallel computer. In the second part of this paper, we briefly recall the Ray-Tracing algorithm and the way we want to accelerate it. In part 3, we show how we manage specialization and scheduling with the function composition scheme. In the two next parts (4 and 5), we precise the architecture and the load balancing strategy. At last, before the conclusion, we show the tool developed to help us to make architectural choices.
P. Amestoy et al. (Eds.): Euro-Par’99, LNCS 1685, pp. 940–949, 1999. c Springer-Verlag Berlin Heidelberg 1999
LUX: An Heterogeneous Function Composition Parallel Computer for Graphics
941
2 Motivation 2.1 Ray Tracing The Ray-Tracing algorithm simulates the propagation of light in a scene. More exactly, we follow the inverse path of light to determine which object is seen by the observer at a given pixel of the picture. A primary ray is sent, originating at the observer and passing through the considered pixel, and we compute its intersection with the scene to determine which object is visible. Once the intersection is found, we determine the amount of light at this point by sending secondary rays. We send rays to each light source to compute the intensity of direct
Data Loading...
