Ray-casting algorithm and its considerations for parallel processing optimization techniques for parallel ray-casting algorithm

This paper presents the considerations for improving the processing speed of ray-casting algorithm, which is widely used to visualize medical images, by using a parallel processing unit. The most important considerations are to exploit memory bandwidth and to maintain high occupancy. To improve memory bandwidth, cache-hit-rate should be kept high. For that, z-order cache is applied to data-load-operations and data-store-patterns are arranged to be coalesced. The tasks of the ray-casting algorithm are equally distributed into the cores of a parallel processing unit to achieve high occupancy. Furthermore, an interpolation process, which is frequently used in the sampling step of the ray-casting algorithm and has high complexity, is implemented to hardwired logic to improve processing speed. As a result of these optimizations, the processing speed was improved over four hundred times than the ray-casting algorithm without optimization.