A Back-End for the SkePU Skeleton Programming Library targeting the Low-Power Multicore Vision Processor Myriad 2

Abstract: The SkePU skeleton programming library utilises algorithmic skeletons to offer a high-level approach for creating parallel applications. By using different back-ends, SkePU applications can run on multicore systems, GPGPU systems, and computer clusters. Myriad 2 is a low-power multicore vision processor for embedded systems, capable of running parallel applications energy-efficiently. Myriad 2 is heterogeneous, containing two different processor architectures and memories with different characteristics. In this thesis, we implement a back-end for SkePU, that allows SkePU applications to run on Myriad 2. We describe how the back-end is designed and evaluate the performance of SkePU applications running on Myriad 2. By conducting a series of benchmarks, we show that our back-end achieves enough performance to make SkePU a useful tool for creating applications for Myriad 2. We also show that SkePU applications can run more energy-efficiently on Myriad 2, compared to a GPGPU system.