Learn how to accelerate marching cubes on the GPU by taking advantage of the GPU's high memory bandwidth and fast on-chip shared memory in a data expansion algorithm that can extract the complete iso-surface mesh from (dynamic) volume data without requiring any data transfers back to the CPU.
In this presentation, we show how ripmaps can replace Summed Area Tables (SATs) for the purpose of computing a large number of spatially varying box filter kernels throughout the input data, providing both higher accuracy and higher speed for typical use cases. For this purpose, we demonstrate an implementation of ripmap generation in CUDA C (accelerated by shared memory usage), and a texture-cache based box filter for spatially varying kernel sizes, which can be implemented in both CUDA C and graphics-based APIs (e.g. OpenGL and DirectX).
Locating connected regions in images and volumes is a substantial building block in image and volume processing pipelines. We demonstrate how the Connected Components problem strongly benefits from a new feature in the Kepler architecture, direct thread data exchange through the SHUFFLE instruction.