摘要
整数变换、反整数变换、量化、反量化及之字形扫描函数的执行速度对H.264视频编码的实时性有较大影响。在支持TI公司C64x+内核系列DSP平台上对这些函数进行讨论,提出了一种新的结构以简化以上函数执行复杂度,并在此结构基础上使用C64x+内核指令进行线性汇编优化。实验结果显示调整后的结构结合线性汇编优化能更充分发挥DSP并行处理的特点,将上述耗时函数的执行速度大幅提高了38～72倍。
The executive speed of integer transform, inverse transform, quantization, inverse quantization and zig-zag scan are important to real-time H.264 video compression. In this paper, these functions are discussed on TI＇s DSP which supports C64x＋ core and a new architecture is proposed to simplify the execution process. Then the routine is optimized by C64x＋ instructions. The experimental results show that the new architecture with linear assembly optimization makes well use of DSP parallel performance. The coding efficiency of these functions is improved for 38-72 times.
出处
《电视技术》
北大核心
2010年第3期28-30,共3页
Video Engineering
基金
国家"863"计划项目
