CUDA code support in Multiagent platform JADE

Abstract

This paper describes a new feature of the multiagent framework that enables the execution of GPU kernels. This brings a novel implementation of the JADE CUDA Agent that accepts PTX/CUDA-C code and runs this code on GPUs. All such agents can receive PTX code or CUDA source code via ACL messages. Beyond the integration of CUDA support, this paper is also focused on the performance measurement of the proposed solution and the evaluation of final results.