计算机工程与应用 ›› 2010, Vol. 46 ›› Issue (32): 171-176.DOI: 10.3778/j.issn.1002-8331.2010.32.048

• 图形、图像、模式识别 • 上一篇    下一篇

运动估计搜索算法的CUDA优化与实现

陈 佐1,陈 汉2,季加良1   

  1. 1.湖南大学 计算机与通信学院,长沙 410082
    2.长沙行成信息科技有限公司,长沙 410005
  • 收稿日期:2010-07-15 修回日期:2010-09-28 出版日期:2010-11-11 发布日期:2010-11-11
  • 通讯作者: 陈 佐

Optimization and realization of motion estimation search algorithm based on CUDA

CHEN Zuo1,CHEN Han2,JI Jia-liang1   

  1. 1.College of Computer and Communication,Hunan University,Changsha 410082,China
    2.Changsha Xingcheng Information Technology Co.,Ltd.,Changsha 410005,China
  • Received:2010-07-15 Revised:2010-09-28 Online:2010-11-11 Published:2010-11-11
  • Contact: CHEN Zuo

摘要: 针对H.264压缩编码中计算量大以及最为耗时的运动估计搜索算法的特点,利用图形处理器的并行优化思想,研究基于CUDA计算平台的运动估计搜索算法GEA(全域消除算法)的并行化处理方法,并对其中的并行设计、数据处理、结果反馈等关键技术问题,进行了详细论述。最后通过实验数据对算法运行效率进行对比分析。实验结果表明GPU中的GEA搜索算法运动搜索性能较之CPU中有显著提高。

关键词: 统一计算设备架构(CUDA), 运动估计, 全域消除算法(GEA), 并行计算

Abstract: Motion estimation search algorithm is one of the most time complexity parts of Video Compression Standard H.264.By making full use of the parallel processing ability of GPU,this paper proposes a parallel process method of GEA algorithm which is based on CUDA computing platform.The applications are discussed separately in details,such as the parallel design,data processing and results feedback.Finally,it gives the comparative analysis of the execution efficiency of GEA in GPU and in CPU.The result shows that GPU can improve the performance of search algorithm significantly.

Key words: Compute Unified Device Architecture(CUDA), motion estimation, Global Elimination Algorithm(GEA), parallel computing

中图分类号: