CUDA-SHAPE: A GPU-accelerated algorithm for asteroid shape modeling

MATTHIAS ENGELS

Asteroids are remnants of our early easolar system and are likely to provide answers about its origins and evolution. Accurate orbital predictions decades into the future are essential for protecting Earth from potential impact events. Asteroids will soon be attractive commercial mining targets. Asteroid shape modeling is an essential driver for this research and in addition to providing shape information, the modeling that is at the core of this research also provides information about size, composition, spin, and ephemerides. Modeling asteroid shapes from radar data uses sequential-fit inversion algorithms that are computationally complex, often taking many weeks to complete. The SHAPE algorithm has been used in all published radar-based asteroid shape research since 1994 but its slow performance is a bottleneck and three prior acceleration attempts have failed. SHAPE is comprised of many serialized independent calculations that can be parallelized at two distinct levels in SHAPE’s data structure. This work has resulted in CUDA-SHAPE, a graphical-processing-unit (GPU)-accelerated asteroid shape modeling algorithm based on SHAPE. CUDA-SHAPE uses commonly available Nvidia GPUs and the CUDA programming framework to performs up to 19.3 times faster than SHAPE on identical models while maintaining full backwards compatibility. CUDA-SHAPE also implements a CPU-hyperthreaded mode that re-uses SHAPE’s cluster computing design on a modern multi-core CPU, providing modest speed boosts of up to 5.9 times that do not require GPU hardware. The GPU-accelerated algorithm includes parallel rasterization, delay-Doppler synthesis, penalty, photometric, and reduction functions, as well as streamed frame operations. CUDA-SHAPE exploits the pixel-parallel nature of digital images and the independence of separate frames of observed data. Three scale model asteroids of increasing complexity are used to prove CUDA-SHAPE’s ability to produce shapes that are unique to the observed data, stable, and reasonable representations of the real objects. The dataset of real asteroid (341843) 2008 EV5 is used to show CUDA-SHAPE’s performance and modeling quality on realistic, noisy data and to compare it to earlier work that utilized the SHAPE algorithm to produce a shape for 2008 EV5 from the same source data and modeling approach. CUDA-SHAPE will be released to interested research communities as open-source software.

CUDA-SHAPE: A GPU-accelerated algorithm for asteroid shape modeling

Files and links (1)

Abstract

Metrics

Details