摘要
为了提高指令装载效率以达到实时编码,本文提出两项H.264编码的指令Cache优化措施:一是调整编码过程以避免宏块编码循环体内的指令码数量大于指令Cache容量,为此提出一种新的编码过程,把通常编码过程的一个帧层宏块编码循环拆分成三个Slice层宏块编码循环,同时运动估计与模式选择的过程与结果不变;二是恰当地压缩代码,即根据新编码过程三个Slice层宏块编码循环体内代码长度和指令Cache缺失的不同情况,分别侧重于压缩代码与展开代码这两种截然相反的优化措施中的一种.
In order to improve the efficiency of instruction loading into DSP, two techniques of instruction cache optimization on H. 264 encoding are proposed in this paper. As the first technique, adjust encoding procedure to prevent the quantity of the instruction codes inside marcoblock encoding loop over the capacity of instruction cache. Thus a new encoding procedure is proposed which fragments the conventional encoding procedure from one rnaeroblock encoding loop at the frame level into three macroblock encoding loops at the slice level, while the results of motion estimation and encoding mode decision are unchanged. As the second technique, compact source code suitably. That is, emphasize adaptively either source code compacting or unrolling according to the length of source codes as well as the quantity of instruction cache miss.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2008年第8期1615-1619,共5页
Acta Electronica Sinica