A Hybrid Reconfigurable System for Parallel and Intensive Computation

Reconfigurable computing system is usually composed of general-purpose processors and reconfigurable coprocessors, featuring the flexibility of general-purpose processors and high computation performance. It is hard for traditional reconfigurable computing system to get balance between the utilization ratios of reconfigurable coprocessor and the complete mapping of algorithms. For a complex algorithm, tasks allocated to general-purpose processor are heavy and this restricts the performance of system. Meanwhile, limited bandwidth reduces computing efficient of coprocessors. In this paper, a hybrid re-configurable parallel computing system with L2-cache has been designed to enhance the independence of reconfigurable coprocessors and reduce the probability of falling into starvation for coprocessors. An FPGA-based reconfigurable computing system has been implemented, which can operate at 100MHz. DCT and MUSIC algorithms have been mapped on it. Experimental results show that designed system achieved much better performance.