Parallelization Strategy Based on RenderScript Reductions

RenderScript is a set of tools designed by Google to support parallel processing on mobile devices with Android. This tools were designed to run on di erent processing components such as Central Processing Units (CPU), Digital Signal Processors (DSP) and Graphics Processing Units (GPU) and it allows portability between mobile electronics devices such as Tablets and Smartphones. RenderScript has a runtime that decides where and how to execute commands list in parallel, it differs in coding and abstraction problem from others platforms used as Open Computing Language (OpenCL) and Compute Uni ed Device Architecture (CUDA). However, in this new parallelization paradigm kernel is not optimized for a speci c architecture. There are not clear strategies for reduction algorithms implementation. For this reason this paper proposes several strategies for reduction algorithms implementation between vectors using RenderScript.