SkewEngine: enhancing performance of intensive calculations on regular meshes

Abstract

In various applications such as hyperspectral data manipulation, MRI data exploration, or visual basin identification in digital elevation models, performing arithmetic operations on each point of a data mesh that involve other points can lead to computationally intractable problems. This paper presents SkewEngine, a tool designed to improve the performance of intensive calculations on regular 2D or 3D data meshes, such as images, multispectral data volumes, or digital elevation models. SkewEngine addresses this problem by reorganizing the mesh in memory according to a preferred spatial direction, enabling more efficient execution of intensive calculations. It is demonstrated that SkewEngine offers significant speed improvements for a variety of test cases, suggesting its usefulness in a broader range of applications requiring intensive data processing on regular meshes.