High performance computing on Android devices -- a case study

TL;DR

This paper evaluates high performance computing techniques on Android devices, focusing on GPU utilization via OpenCL, demonstrating significant speedups with SIMD and OpenCL implementations for computationally intensive tasks.

Contribution

It compares various parallel programming paradigms on Android, highlighting the potential of OpenCL and SIMD for high performance computing on low-power devices.

Findings

OpenCL can be used on Android tablets despite lack of official support.

Significant speedups achieved with SIMD and OpenCL implementations.

GPU-based computing on Android can rival high-speed CPUs.

Abstract

High performance computing for low power devices can be useful to speed up calculations on processors that use a lower clock rate than computers for which energy efficiency is not an issue. In this trial, different high performance techniques for Android devices have been compared, with a special focus on the use of the GPU. Although not officially supported, the OpenCL framework can be used on Android tablets. For the comparison of the different parallel programming paradigms, a benchmark was chosen that could be implemented easily with all frameworks. The Mandelbrot algorithm is computationally intensive and has very few input and output operations. The algorithm has been implemented in Java, C, C with assembler, C with SIMD assembler, C with OpenCL and scalar instructions and C with OpenCL and vector instructions. The implementations have been tested for all architectures currently supported by Android. High speedups can be achieved using SIMD and OpenCL, although the implementation is not straightforward for either one. Apps that use the GPU must account for the fact that they can be suspended by the user at any moment. In using the OpenCL framework on the GPU of Android devices, a computational power comparable to those of modern high speed CPUs can be made available to the software developer.