Approaches that exploit vector-parallelism for three rendering and volume visualization techniques

Abstract

Direct volume rendering using volume ray-casting and indirect volume rendering using the Marching Cubes isosurface extraction are popular techniques for volume visualization. Surface ray-tracing is a popular graphics technique for rendering scenes composed of well-defined surface primitives. However, these techniques are relatively computationally intensive. Thus, near-real-time computational performance is a difficult goal. This paper presents approaches to these rendering and volume visualization techniques that are tuned for efficient performance on a vector-parallel supercomputer. The approaches decompose and reconstruct the techniques to exploit inherent data parallelism and the specific characteristics of the CPU. Experimental results for several datasets are also exhibited.