System-Level Modeling, Analysis and Code Generation: Object Recognition Case Study

International audienceOne of the most important challenges in complex embedded systems design is developing methods and tools for modeling and analyzing the behavior of application software running on multi-processor platforms. We propose a tool-supported flow for systematic and compositional construction of mixed software/hardware system models. These models are intended to represent, in an operational way, the set of timed executions of parallel application software statically mapped on a multi-processor platform. As such, system models will be used for performance analysis using simulation-based techniques as well as for code generation on specific platforms. The construction of the system model proceeds in two steps. In the first step, an abstract system model is obtained by composition and specific transformations of (1) the (untimed) model of the application software, (2) the model of the platform and (3) the mapping between them. In the second step, the abstract system model is refined into concrete system model, by including specific timing constraints for execution of the application software, according to chosen mapping on the platform. We illustrate the system model construction method and its use for performance analysis and code generation on an object recognition application provided by Hellenic Airspace Industry. This case study is build upon the HMAX models algorithm [RP99] and is looking at significant speedup factors. This paper reports results obtained on different system model configurations and used to determine the optimal implementation strategy in accordance to hardware resources