The Optimization Of AVS Software Decoding

The AVS video coding standard is a second-generation source coding standard owned by our country. Although some companies have developed some AVS decoding chips now, only some of these chips have come into the industrialization phase and most of these chips have use only in experiments. Additionally, these chips cost a lot to produce. Therefore, it is more practical to implement AVS software decoding before the mature of the industrialization. The executing speed is the most important indicator for AVS software decoding, so studying the optimization of AVS software decoding has great significance.This text first examine in depth the architecture of the AVS video coding standard and the principles of the important components of the AVS video coding standard, it then studies optimization for the reference AVS decoding software rm52k_r1. First, it uses the Intel? Vtune? Performace Analyzer to analyze the AVS decoding software rm52k_r1 in order to find the key points that must be optimized, then propose schems to optimize them. The key points include problems associated with meomory, decoding of k-order Exp-Golomb codes, decoding of VLC and dequantization, deblocking filtering, and sample interpolation. Problems associated with meomory can be solved by eliminating or reducing the amout of huge volume data moves or assignments. The optimization algorithm mainly uses lookup tables to optimize the decoding of k-order Exp-Golomb codes. The optimization algorithm optimizes the decoding of VLC through optimization of storage of VLC code tables. The module of dequantization must be optimized together with the VLC decoding module. The optimization algorithm utilizes the property that most coefficients before dequantization are decoded in the case of trans_coefficient less than 59, where these coefficients are very small, so that the algorithm can use the lookup table technique to reduce computation. The optimization of the deblocking filtering uses the properties of the deblocking filtering to reduce the number of conditional branches. The optimization for luma sample interpolation classify the computation into two cases, one for the case in which the entire reference blocks in reference pictures are inside the picture boundary, the other for the case in which the reference blocks in reference pictures are not entirely inside the picture boundary, to reduce the max and min macros. The latter case can be optimized using the lookup table technique, and some loop branch techniques. The optimization for chroma sample interpolation uses lookup table to eliminate multiply computation, in addition to using the techniques used by the optimization of luma sample interpolation.The end of this text first uses the Intel? Vtune? Performace Analyzer to analyze the optimized AVS decoding program with a conclusion that the performance of those key modules of AVS decoding enhance greatly. Then it uses several video sequences of different resolutions to test the optimized program with a conclusion that the techniques proposed by the text are very effective and the optimized program obviously enhances the performance of decoding relative to the original.