Advancing Automatic Code Generation for Agent-Based Simulations on Heterogeneous Hardware

Xiao, Jiajian and Andelfinger, Philipp and Cai, Wentong and Richmond, Paul and Knoll, Alois and Eckhoff, David (2019) Advancing Automatic Code Generation for Agent-Based Simulations on Heterogeneous Hardware. In: The International Workshop on Algorithms, Models and Tools for Parallel Computing on Heterogeneous Platforms (HeteroPar'2019), 27 Aug 2019, Göttingen, Germany. Proceedings, published by Springer, Cham, pp. 308-319.

Full text not available from this repository.
Official URL:


The performance of agent-based simulations has been shown to benefit immensely from execution on hardware accelerator devices such as graphics processing units (GPUs). Given the increasingly heterogeneous hardware platforms available to researchers, it is important to enable modellers to target multiple devices using a single model specification, and to avoid the need for in-depth knowledge of the hardware. Further, key modelling steps such as the definition of the simulation space and the specification of rules to resolve conflicts among agents should be supported in a simple and generic manner, while generating efficient code. To achieve these goals, we extend the OpenABL modelling language and code generation framework by three aspects: firstly, a new OpenCL backend enables the co-execution of arbitrary agent-based models on heterogeneous hardware. Secondly, the OpenABL language is extended to support graph-based simulation spaces. Thirdly, we specify a generic interface for specifying conflict resolution rules. In a performance comparison to the existing OpenABL backends, we show that depending on the simulation model, the opportunity for CPU-GPU co-execution enables a speedup of up to 2.0 over purely GPU-based simulation.

Item Type: Conference or Workshop Item (Paper)
Additional Information: Part of the Lecture Notes in Computer Science book series (LNCS, volume 11997) ISBN: 978-3-030-48339-5 (print) 978-3-030-48340-1 (online)