Configurable System on A Chip consisting of general-purpose processor and configurable coprocessor is studied as a platform for algorithm acceleration. A simple model for quantitative evaluation ofi interface between ...
详细信息
Configurable System on A Chip consisting of general-purpose processor and configurable coprocessor is studied as a platform for algorithm acceleration. A simple model for quantitative evaluation ofi interface between processor and coprocessor is presented and applied to Xilinx Microblaze based system. Several architectural variants for coprocessor interface are described and evaluated. Efect ofi coprocessor on execution time is demonstrated by implementation ofi thee simple SW kernels – 32-bit CRC, RGB-to-YCrCb computation and IDCT. These kernels ilustrate that data transfer overhead has very high impact to execution time even when low latency SoC buses and links are used. Even with these overheads, significant speedup over SW implementation can be achieved.
暂无评论