摘要
针对国产应用的性能提升,基于CPU\GPU多核技术,提出软硬件结合的并行优化策略及反距离权重(IDW)插值的并行优化算法(PIDW),优化离散数据网格化处理。针对并行处理中的线程任务分解共性难点,设计基于开放多核处理(OpenMP)与统一计算设备架构(CUDA)的线程任务分解模型(TTDM),具有线程访问安全(不越界)、计算无冗余(无重复)、计算完整(无遗漏)等特点,具有较好的计算均衡性(负载均衡)。通过国产及商用多环境实验,加速比分别是3.6和5.9,验证了PIDW算法的性能提升能力。
To improve performance of domestic application system,the parallel optimized design strategy was proposed based on CPU／GPU multi-core technology.The parallel inverse distance weighting（IDW）optimization algorithms（PIDW）were established to optimize the grid processing of discrete data.To solve the problem of thread task decomposition in parallel programming,thread task decomposition models（TTDM）were designed.Open multi processing（OpenMP）was used and unified device architecture（CUDA）was computed.The models had safety accessing in data bounds,avoiding computing problems of duplication and missing.TTDM had fine computing balancing ability in thread task.Experimental results in domestic and commercial multi environment show that the speedup of the algorithm can reach 3.6 times and 5.9 times respectively,which shows PIDW’s capability to improve performance.
作者
王家润
谢海峰
WANG Jia-run, XIE Hai-feng(The Third Basic Department, North China Institute of Computing Technology, Beijing 100083, Chin)
出处
《计算机工程与设计》
北大核心
2018年第6期1774-1781,共8页
Computer Engineering and Design
基金
十三五预研基金项目(31511070401)
