摘 要: 在JPEG2000算法中,采用5/3小波濾波來進行對圖像塊的無損壓縮變換。本文討論在DSP嵌入式平臺上對5/3小波濾波器的低功耗優化技術。本文采用TI公司的TMS320C6748 DSP處理器,詳細闡述了幾種針對DSP芯片體系結構的低功耗處理技術,主要包括循環分解、循環展開、循環嵌套、循環轉置等循環優化方法及編排流水線等方法。通過上述方法綜合應用到濾波算法的優化過程中,能夠明顯降低系統開銷。
Abstract: The JPEG2000 algorithms use 5/3 wavelet filter to realize lossless compression transformation for the image block. This article discusses the low power consumption optimization techniques for 3/5 wavelet filters based on DSP embedded platform. And elaborates several low power consumption process techniques for TMS320C6748 DSP chip architechture in details, including cycle optimization methods and arragement pipeline such as cycle decomposition,loop unrolling, loop embed, loop transposition,etc. Significantly reduces system energy cost using these methods in optimization for filters algorithm.
關鍵詞: 數字信號處理器;5/3小波變換;循環優化;低功耗
Key Words: DSP; 5/3 wavelet transform; cycle-optimization; low power
參考文獻:
[1] 羅 均, 黃尚廉, 張國彬 等. JPEG2000中可逆整數5/3小波變換及其在DSP上的實現[J]. 光電工程, 2004, 31(5): 91-93.
[2] 杜會斌, 周 旭, 張學慶, 吳曉娟. 二維離散5/3小波變換并行VLSI結構設計[J]. 綜合電子信息技術, 2006, 32(6):39-41.
[3] 陳 娟. 低功耗軟件優化技術研究[工學博士學位論文]. 國防科大研究生院, 湖南, 長沙, 2007, 1-17.
[4] 羅 剛, 郭 兵, 沈 艷, 廖海艷, 任 磊. 源程序級和算法級嵌入式軟件功耗特性的分析與優化方法研究. 計算機學報. 2009, 32(9).
[5] Hongbo Yang. Power-aware Compilation Techniques for High Performance Processors. Doctor dissertation. University of Delaware. Winter 2004.
[6] M.Kandemir, N.Vijaykrishnan, M.J.Irwin and W.Ye. Influence of compiler optimizations on system power. In proceedings of Design Automation Conference(DAC). June 2000.
[7] Kim N.S., Austin T. and Mudge T.et al. Challenges for Architectural Level Power Modeling. Power awarecomputing. Norwell, MA, USA, Kluwer Academic Publishers. p. 317-337. 2002.
[8] Texas Instruments, TMS320C6000 Programmer’s Guide [M]. 2006.
[9] Texas Instruments, TMS320C6000 Optimizing Compiler User’s Guide [M]. 2006.
作者簡介:姚天問(1988- ),男,湖北恩施,在讀工程碩士研究生,主要研究方向為軟件低功耗技術。
資助信息:國家自然科學基金資助項目,項目編號:61272146