Automatic Generation of OpenCL Code for ARM Architectures
Universidad de La Laguna
22nd International European Conference on Parallel and Distributed Computing (Euro-Par), 2016
@article{afonso2016automatic,
title={Automatic Generation of OpenCL Code for ARM Architectures},
author={Afonso, Sergio and Acosta, Alejandro and Almeida, Francisco},
year={2016}
}
The efficient exploitation of the increasing computational capabilities of mobile devices is still a challenge. The heterogeneity of Systems on Chip (SoC) makes necessary a very specific knowledge of their hardware in order to harness their full potential. OpenCL is a well known standard for cross-platform usage of accelerator devices. We follow an annotation-based approach for solving the problem of high development cost of OpenCL programming for mobile devices. With our approach, the programmer can select from different programming models the one that offers the best performance for each section of the application. Computational results show that our automatically-generated OpenCL code can outperform Renderscript when running on the GPU of Android devices, making it the best choice for a range of parallel algorithms.
August 16, 2016 by hgpu
