高性能同构多核媒体处理器

奚杰¹, 朱玥^1,2, 陈杰¹

1.中国科学院微电子研究所;2.中国科学技术大学物理学院



摘要:

为了大幅提高处理器的处理能力,设计了一款5核结构的同构多核处理器并实现了H.264在多核处理器上的并行解码.该多核处理器采用1个CPU作为主控处理器,另外的4个CPU作为受控处理器被调用,5个CPU可以同时访问1块32 KWord的共享存储器,任意2个CPU之间可以通过邮箱、信号量、硬件锁实现点对点的通讯.其中,主控处理器负责码流解析、帧内预测模式、运动向量、边界强度的计算,并把中间数据存储到共享存储器中;4个受控处理器每次取1行宏块数据处理.试验结果表明,采用行并行的解码方式,5核系统较传统的双核系统达到约3.6倍的加速比.该并行解码方法同样适用于其他的视频协议,具有广泛的通用性.

关键词:  多核  邮箱  H.264  信号量

DOI：10.11918/j.issn.0367-6234.2011.05.019

分类号:TP332

基金项目:国家高技术研究发展资助项目(2009AA011700)



High-performance homogeneous media mutli-processor

XI Jie¹, ZHU Yue^1,2, CHEN Jie¹

1.Institute of Microelectronics,Chinese Academy of Sciences,100029 Beijing,China;2.School of Physical Sciences,University of Science and Technology of China,230026 Hefei,China

Abstract:

To improve processor’s computation capability heavily,one multi-core system with 5 CPUs is designed and H.264 is decoded on it parallely.One CPU is used to lead the other four CPUs.Five CPUs share one 32 KWord SRAM.Any two CPUs can communicate point to point through mailbox,semaphore,hardware lock.When decoding H.264,the leading CPU is used to parse the bit-stream and calculate the intra prediction mode,motion vector,boundary strength.Then,the decoded data is saved to shared SRAM.Other four CPUs fetch and solve one macro-block line each time.The simulation result shows that using the line parallel decoding algorithm,the five-cores system can achieve about 3.6 times acceleration than two-cores system.The line parallel decoding algorithm can also be used on other video protocol and it is universal.

Key words:  multi-core  mailbox  H.264  semaphore