Research And Application Of HEVC Multi-channel Video Parallel Coding Technique Based On TILE-Gx36 Multi-core Processor

The new generation of video coding standard,HEVC,introduces more new features in the coding module,which reduces the bit rate by twice as much as H.264/AVC,and brings about a significant increase in computational complexity.Parallel computing based on modern multi-core processor platform has become a powerful solution for high-complexity video coding,which has important research significance and application prospects.In this thesis,aiming at the high complexity of HEVC video coding,the main coding module parallel processing scheme is designed to improve the coding efficiency.The scheme is integrated into the multi-channel coding system through the core scheduling strategy to realize the parallel processing of the target scene of multichannel video..The work of the thesis is based on the TILE-Gx36 multi-core platform.The specific work is divided into the following three parts:(1)Design and implement a CTU granular parallel algorithm scheme.By analyzing the dependence and load balancing relationship between intra-frame inter-module CTUs in the video coding process,it is found that the load balancing relationship and the dependency relationship between CTUs affect the CTU coding efficiency mechanism.By designing the adjacency matrix and the inference matrix The data structure,combined with the thread pool kernel resource scheduling strategy,designed and implemented the CTU granular parallel algorithm and entropy coding improved optimization scheme,fully improved multi-core utilization,and effectively improved the parallel acceleration ratio.(2)Design and implement a parallel algorithm for CTU prediction process.Through DAG analysis,the dependence between PUs of different scale CUs in the process is predicted,and the possibility that PUs of different scales can be processed in parallel is found.Based on this,the parallel scheme of CTU in prediction module is proposed.The parallel algorithm uses priority queue and quadtree scanning technology to perform task initialization and update,and designs a PU-level parallel coding scheme in the CTU to effectively improve the coding speed of the prediction module.(3)Design and implement a parallel coding processing scheme for multi-channel video.Firstly,the two parallel schemes in which the CTU granularity parallel and the PU in the CTU are parallel are merged by the core scheduling strategy to obtain a higher parallel acceleration ratio of the single channel video;on this basis,two application scenarios are specified for the specified frame rate and real-time encoding.The kernel dynamic allocation scheme between stream and stream is designed and implemented,and multi-granular fusion parallel coding of multi-channel video is realized.In this thesis,the above three coding schemes are implemented on the Tilera multi-core platform.The performance improvement of this thesis is verified by setting the contrast algorithm.The parallel acceleration ratio of the CTU granular parallel coding scheme can reach 9 and does not affect the objective quality parameters of the video;the inter-PU parallel algorithm in the design CTU prediction process in the prediction module effectively improves the parallelism,and the relative intra 2X parallel algorithm The inter-frame local parallel algorithm has speed improvement,the intra-frame module acceleration ratio can reach 2.5,and the inter-frame module acceleration ratio can reach 2.7;the multi-channel coding system uses the nuclear dynamic allocation scheme for each target scene in the multi-path,which is more reasonable.Allocating nuclear thread resources to multiple paths has achieved the expected design goals.