Flying edges: A high-performance scalable isocontouring algorithm

Isocontouring remains one of the most widely used visualization techniques. While a plethora of important contouring algorithms have been developed over the last few decades, many were created prior to the advent of ubiquitous parallel computing systems. With the emergence of large data and parallel architectures, a rethinking of isocontouring and other visualization algorithms is necessary to take full advantage of modern computing hardware. To this end we have developed a high-performance isocontouring algorithm for structured data that is designed to be inherently scalable. Processing is performed completely independently along edges over multiple passes. This novel algorithm also employs computational trimming based on geometric reasoning to eliminate unnecessary computation, and removes the parallel bottleneck due to coincident point merging. As a result the algorithm performs well in serial or parallel execution, and supports heterogeneous parallel computation combining data parallel and shared memory approaches. Further it is capable of processing data too large to fit entirely inside GPU memory, does not suffer additional costs due to preprocessing and search structures, and is the fastest non-preprocessed isocontouring algorithm of which we are aware on shared memory, multi-core systems. The software is currently available under a permissive, open source licence in the VTK visualization system.