H.265整数运动估计参考块更新的并行化设计

Paralleling design for reference block updating of integer motion estimation in H.265

针对H.265整数运动估计算法参考块更新模块中数据传输量大、运行速度慢等问题,在分析参考块间数据相关性的基础上,提出了能够减少硬件使用资源,提高运行效率的并行方案。该方案采用18×17个处理元阵列,通过相邻参考块之间3个方向数据重合的关系设计了3个缓存区,更新时根据参考块之间的关系定位缓存区,然后从外存加载相应的参考块数据。该方案中的资源占用量相比传统设计降至1/16。实验结果表明,该方案可以将数据复用率提高到98.4%,有效降低了整数运动估计算法的带宽需求。

To improve the processing time of reference block updating scheme in H.265,the paper proposes a paralleling method that can reduce hardware resources and improve the efficiency.Based on analyzing the relationship of data overlap among reference blocks,the paper uses 18 × 17 array of processing elements and three cache regions in three directions.Then selects？ cache region and loads reference block data from external memory according to data overlap direction.The resource usage of parallel scheme reaches 1/16 compared with traditional design.Experimental resuhs show that data multiplex rate of this parallel scheme reaches 98.4％,the bandwidth requirements reducing is clearly.