一种基于VLIW架构的高效DCT实现方法  

An efficient method to implement DCT algorithm on VLIW architectures

作　　者：周晶[1] 张涛[1] 孙张明[1] 冯超[1] 王晓辉[1] 蒯英杰 胡美丽[1] 高敏[1] 

机构地区：[1]天津大学电子信息工程学院,天津300072

出　　处：《信息技术》2015年第11期1-3,6,共4页Information Technology

基　　金：国家自然科学基金资助(61350009)

摘　　要：文中提出了一种在VLIW架构DSP上计算AVS视频标准中DCT的方法。在对DCT变换矩阵进行分解的基础上,利用复数乘法实现矩阵乘法计算,并通过合理组织数据,实现了变换矩阵打包系数的复用,减少了寄存器占用,使得算法更适于进行循环展开和软件流水,从而实现更高的并行度,执行速度得到有效提升。在计算效率上,文中提出的计算方法比AVS标准中的快速算法提高了4.28倍,并且比现有方法的计算耗时减少了31.1%。In this paper, a method to compute DCT defined in AVS on VLIW DSPs is presented. Complex multiplications are employed to implement the matrix multiplications based on the decomposition of IDCT transform matrix. To reduce register pressure, reuse of packed transform matrix coefficients is also achieved with rational organization of data, so that the proposed method is more suitable for loop unrolling and software pipelining. Eventually a higher ILP is achieved and the computation efficiency is improved. When implemented on VLIW DSPs, the proposed method saves 31.1% computation time compared with the existing ones. And it is 4.28 times faster than the fastalgorithm in AVS standard.

关 键 词：AVS视频标准 离散余弦变换 超长指令字 数字信号处理器 复数乘法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]