An Optimized Scheme for High-speed Data Interaction Based on TI-C6678 Multi-core DSP

摘要

The performance of data interaction between different memories has become a significant factor in the complex embedded systems with the huge increase of processing needs, especially between internal chip-on memory and external memory. This paper advances a constructive optimized scheme named global direct memory access (GDMA) for high-speed data interaction in the multi-core digital signal processor (DSP) systems. Fur more, we give important recommendations to actualize software programming optimization of GDMA from three aspects. This scheme is based on the key technique of enhanced direct memory access versions3 (EDMA3), quick direct memory access (QDMA) and internal direct memory access (IDMA). This scheme can enhance the peak speed of data interaction between local memories by 53.8% and 2.5% averagely for some situations using QDMA compared EDMA3. By establishing the bridge memory area, the speed from level-1 data memory to external memory is optimized by 15.3% maximumly. The multi-core DSP system can achieve the performance of high-speed data interaction at around 5 GBps to meet user expectations.