论文部分内容阅读
在H.264视频解码中,去块滤波是运算量很大的一部分.由于去块滤波过程中,数据之间存在复杂的依赖性,现有的很多去块滤波并行方案存在着并行度小、同步互斥开销大的缺点.本文结合去块滤波算法及众核处理器Godson-T的结构特性,提出了一种可以减少数据依赖的去块滤波算法并行优化方案.相对于以前的很多方法,此并行方案首先在算法上增大了并行度,减少了同步开销,同时,我们通过片上众核处理器Godson-T的硬件支持,采用计算与通信重叠等优化策略,使得优化后的算法达到了数倍的性能提升.
In H.264 video decoding, deblocking filtering is a large part of the computational complexity.Because of the complex dependence between data in the deblocking filtering process, many existing deblocking filtering parallel schemes have the disadvantages of low degree of parallelism, Synchronization and mutual exclusion overhead.According to the deblocking filtering algorithm and the structural characteristics of Godson-T, the paper proposes a parallel optimization scheme of DF algorithm which can reduce the data dependency.Compared with many previous methods, This parallel scheme firstly increases the degree of parallelism in the algorithm and reduces the synchronization overhead. At the same time, through the hardware support of Godson-T on-chip all-core processor and the optimization strategy such as overlap of computation and communication, the optimized algorithm is achieved Several times the performance improvement.