The development of multicore hardware has provided many new development opportunities for many application software algorithms. Especially, the algorithm with large calculation volume has gained a lot of room for improvement. Through the research and analysis, this paper has presented a parallel PO-Dijkstra algorithm for multicore platform which has split and parallelized the classical Dijkstra algorithm by the multi-threaded programming tool OpenMP. Experiments have shown that the speed of PO-Dijkstra algorithm has been significantly improved. According to the number of nodes, the completion time can be increased by 20–40%. Based on the improved heterogeneous dual-core simulator, the Dijkstra algorithm in Mi Bench is divided into tasks. For the G.72 encoding process, the number of running cycles using “by function” is 34% less than using “divided by data,” while the power consumption is only 83% of the latter in the same situation. Using “divide by data” will reduce the cost and management difficulty of real-time temperature. Using “divide by function” is a good choice for streaming media data. For the Dijkstra algorithm, the data is data without correlation, so using a simpler partitioning method according to the data partitioning can achieve good results. Through the simulation results and the analysis of the results of real-time power consumption, we conclude that for data such as strong data correlation of streaming media types, using “divide by function” will have better performance results; for data types where data correlation is not very strong, the effect of using “divide by data” is even better.

多核硬件的开发为许多应用软件算法提供了许多新的开发机会。特别是运算量大的算法有很大的改进空间。通过研究和分析，提出了一种并行的PO-Dijkstra多核平台算法，该算法通过多线程编程工具OpenMP对经典的Dijkstra算法进行了拆分和并行化。实验表明，PO-Dijkstra算法的速度得到了显着提高。根据节点数，完成时间可以增加20–40％。基于改进的异构双核仿真器，Mi Bench中的Dijkstra算法分为任务。对于G.72编码过程，使用“按功能”的运行周期数比使用“按数据划分”的运行周期数少34％，而在相同情况下的功耗仅为后者的83％。使用“按数据划分”将降低实时温度的成本和管理难度。使用“按功能划分”是流媒体数据的不错选择。对于Dijkstra算法，数据是没有相关性的数据，因此根据数据分区使用更简单的分区方法可以取得良好的效果。通过仿真结果和实时功耗结果分析，我们得出结论，对于流媒体类型的数据相关性强的数据，使用“按功能划分”将具有更好的性能结果；对于数据相关性不是很强的数据类型，使用“按数据划分”的效果甚至更好。