Data-parallel visualisation using multi-dimensional transformations

Abstract

The authors show how a flexible resampling approach can be embedded within massively parallel implementations of multidimensional transformation algorithms based on one-dimensional resampling operations. They provide a consistent solution to the resampling requirements across visualization applications. Based on this framework, two applications are outlined: a surface perspective viewing algorithm with hidden-surface removal and a volume rendering algorithm. These algorithms include regular and irregular resampling requirements. The algorithm considered here is well-suited to data-parallel SIMD (single-instruction multiple-data) processing, and performance on surface and volume visualization is sufficient to achieve interactive manipulation on large SIMD arrays.<>