Design and Implementation of Multi Agent Simulation Library MasCUDA for Parallel Processing with GPU

Abstract

This paper presents the design and implementation of parallel processing support library, primary for multi-agent simulation with GPU (Graphical Processing Unit). GPU provides highly parallel processing environment. However, in order to develop software for GPU, high level skill and knowledge of GPU, parallel processing and GPU architecture are required, and these requirements sometimes disturb to use GPU for specific application development. In this article we will provide the library for GPU programming named MasCUDA. Users can develop their own application by their familiar language such as Ruby. GPU specific programming is hidden by MasCUDA and users need not to understand the detail of GPU programming. Our experimental evaluation proved that MasCUDA accelerates the execution speed more than 5,000 times faster than Ruby program and the number of source code with MasCUDA is approximately half of GPU specific language.