Design Space Exploration for Efficient Data Intensive Computing on SoCs

International audienceFinding efficient implementations of data intensive applications, such as radar/sonar signal and image processing, on a system-on-chip is a very challenging problem due to increasing complexity and performance requirements of such applications. One major issue is the optimization of data transfer and storage microarchitecture, which is crucial in this context. In this chapter, we propose a comprehensive method to explore the mapping of high-level representations of applications into a customizable hardware accelerator. The high-level representation is given in a language named Array-OL. The customizable architecture uses FIFO queues and a double buffering mechanism to mask the latency of data transfers and external memory access. The mapping of a high-level representation onto a given architecture is achieved by applying loop transformations in Array-OL. A method based on integer partition is used to reduce the space of explored solutions. Our proposition aims at facilitating the inference of adequate hardware realizations for data intensive applications. It is illustrated on a case study consisting in implementing a hydrophone monitoring application