摘要
提出一种基于深度优先贪婪搜索(Depth First Greedy Search Partitioning,DFGSP)的分像素插值任务划分映射方法,采用任务并行的方式,按深度优先搜索节点的方式减少子任务之间的通信量。对分像素插值算法的数据流图划分后设计合理并行方案,以极大限度利用处理元为原则进行贪婪搜索,并在项目组前期研究的阵列处理器(DPR-CODEC)上加速实现。实验表明,该方法执行时间远低于两步搜索方案,与未优化的原始HEVC插值滤波器相比,硬件资源占用减少72%。
This paper proposes a task partition mapping method based on depth first greedy search partitioning(DFGSP),which uses task parallel method and depth first search node method to reduce the communication between subtasks.After dividing the data flow graph of the sub-pixel interpolation algorithm,a reasonable parallel scheme was designed,which made greedy search based on the principle of maximum utilization of processing elements.The implementation was accelerated on the array processor(DPR-CODEC)researched in the early stage of the project team.The experimental results show that the execution time of this method is much lower than the two-step search scheme.Compared with the original HEVC interpolation filter,the hardware resource consumption is reduced by 72%.
作者
胡传瞻
蒋林
朱筠
谢晓燕
王萍
杨坤
Hu Chuanzhan;Jiang Lin;Zhu Yun;Xie Xiaoyan;Wang Ping;Yang Kun(School of Computer,Xi’an University of Posts&Telecommunications,Xi’an 710121,Shaanxi,China;Laboratory of Integrated Circuit,Xi’an University of Science and Technology,Xi’an 710054,Shaanxi,China;School of Electronic Engineering,Xi’an University of Posts&Telecommunications,Xi’an 710121,Shaanxi,China)
出处
《计算机应用与软件》
北大核心
2022年第10期49-53,103,共6页
Computer Applications and Software
基金
国家自然科学基金项目(61772417,61834005,61802304,61602377)。
