A Vector-parallel Realization of the Marching Cubes

The Marching Cubes algorithm is a popular high-resolution isosurface extraction method used in volume data visualization. However, it is relatively computationally intensive making real-time operation on normal workstations a diicult goal when applied to large datasets. One solution is to transform the serial algorithm into a vector-parallel algorithm designed to exploit the potential computing power supplied by a supercomputer. In this paper, we present an implementation of the Marching Cubes that considers the inherent parallelism in the algorithm as well as the speciic characteristics of the pipelined CPU of a vector-parallel supercomputer (Cray C90). In our approach, we vectorize two time-consuming operations in the Marching Cubes. The rst operation is the interpolation of the intersection points between the isosurface and the cube edges. The second vectorized operation is the computation of topological equivalences for classes of intersections. In this paper, we describe the details of our parallel algorithm and present the experimental results for several typical volume datasets.