Performance Evaluation and Optimization of Multi-core ARM and DSP Architecture

Summary

Due to the wide application areas of the embedded processors, there exist many distinct architectures that are all competitive. OMAP processor is a typical example that incorporates heterogeneous multi-core architectures into one chip. It is designed to provide best-in-class computing efficiency and flexibility. By studying the various computer architectures within the OMAP processor, we are able to investigate the advantages of each processor cores and their potential applications in terms of processing speed, and power. By focusing on the configuration of using fixed-point and floating point processing, also using different cores in OMAP, we compared the performance achieved by exploiting the DSP, NEON and ARM architectures. We also measured the power consumption of different configurations. Experimental results shown that using NEON and DSP hardware acceleration can improve the power efficiency, while compiler optimization improves performance at the cost of power consumption. The performance gains and overheads of accessing DSPLib are also analyzed. Our research results show that the overall efficiency of the application can be benefited from this advanced architecture with emphasis on utilizing their distinct features.

Architecture

Results

Screen Shot 2014-12-27 at 5.28.51 PM

 

Screen Shot 2014-12-27 at 5.29.03 PM

Screen Shot 2014-12-27 at 5.29.18 PM

Screen Shot 2014-12-27 at 5.29.42 PM

Benchmark Results

Screen Shot 2014-12-27 at 5.29.34 PM

Power Consumption

Screen Shot 2014-12-27 at 5.29.58 PM