Embedded Real-Time Application Prototyping Using a Hybrid Multiprocessing Platform

Modern embedded systems require both high performance computing capability and low power/energy consumption in order to meet ever increasing customer demands. Recently, several semiconductor manufacturers have provided a new hybrid multiprocessing model which couples an FPGA with programmable processors. This constellation of processing elements in SoC platforms certainly provides needed performance while still keeping flexibility. However, several questions are yet to be answered to understand and use the platform better: What will the development flow look like to handle both software programmable processors and hardware blocks? What are the key points to map real-time applications to such platforms to achieve an optimal trade-off between performance and power? To address those issues, this paper presents a case-study of developing and mapping embedded applications to a real-life Lyrtech prototyping board consisting of a TI DSP and a Xilinx Virtex-4 FPGA. A customized development flow for programming DSP and FPGA was used in our work. The measured performance and power results for different mappings of benchmarks are demonstrated. The observations that we made are also discussed to provide some empirical guidelines on using hybrid multiprocessing platforms.