H.264计算密集型任务在可重构处理器上的映射
Mapping of computing-intensive tasks in H.264 decoding to a reconfigurable processor
朱敏 1刘雷波 1尹首一 1杨晨 1王文杰 1魏少军1
作者信息
摘要
可重构系统具有领域内灵活,性能和专用电路接近的优点,是视频解码的优秀硬件方案.然而在可重构系统上进行高清实时解码还有一定的难度,其中占80％计算量的主要是IDCT(反离散余弦变换)、MC(运动补偿)、Intra-prediction(帧内预测)、deblocking(去块效率滤波)等计算密集型任务.本文基于一款粗粒度可重构处理器,提出了上述计算密集型算法的映射方案,性能优于M.Ganesan与D.Peng在2007、2009年的方案,满足H.264高清实时解码的要求.
Abstract
Reconfigurable system is an excellent solution for video decoding,which has shown great advantages in both flexibility and high performance,reaching a satisfactory tradeoff between GPPs (General Purpose Processors) and ASICs.But high-profile 1080p real-time decoding is still a hard work,which contains computing-intensive tasks,such as IDCT (Inverse Discrete Cosine Transform),MC (Motion Compensation),Intra-prediction,deblocking.These tasks take 80％ complexity in H.264.This paper gives out a plan to map the tasks on a certain coarse-grade reconfigurable processor,called REMUS,achieving better results than Ganesan and D.Peng's works.IDCT in a micro-block executes in no more than 192 cycles,intra-prediction 389 cycles,MC 320cycles and deblocking 704 cycles,met the timing requirement of H.264 1080p high profile real-time decoding at 200MHz.
关键词
H.264/可重构处理器/计算密集型任务/算法映射Key words
H.264/reconfigurable processor/computing intensive tasks/algorithm mapping引用本文复制引用
基金项目
国家高技术研究发展计划(863计划)(2009AA011700)
国家自然科学基金(60803018)
出版年
2013
NETL
NSTL
维普
万方数据