Wiley, Concurrency and Computation: Practice and Experience, 4(26), p. 972-986, 2013
DOI: 10.1002/cpe.3076
Full text: Download
With the emergence of general-purpose computation on graphics processing units, high-level approaches that hide the conceptual complexity of the low-level Compute Unified Device Architecture and Open Computing Language platforms are the subject of active research. However, these approaches may require a trade-off in terms of achieved performance and utilisation on graphics processing units hardware and may impose algorithmic limitations. In this paper, we present and systematically evaluate the parallel performance of three implementations of the brute force, all-pairs N -body algorithm with skeletal deploy-ments based on the FastFlow, SkePU and Thrust frameworks. Our results indicate that the skeletal frame-work implementation achieves up to two orders of magnitude speed-up over serial version with a Tesla M2050 with lower implementation complexity than low-level Compute Unified Device Architecture programming.