Research And Implementation Of HEVC Multi-Level Decoding Algorithm Based On Tilera Multi-Core Processor

HEVC is the next generation video coding standard for high-definition and ultra-high-definition video applications, but its high computational complexity is an important obstacle faced by applications. Using the multi-core processor is an effective way to solve this problem. In this paper, we use Tilera GX36 multicore processor launched by TILERA company.And HEVC parallel video decoding technology in-depth study by the decoder key module parallel processing to achieve a multi-level parallel HEVC decoding based on Tilera multi-core processor,so as to achieve high-definition, ultra-high-definition video bit stream in real-time decoding,which not using any parallel encoding form. Our main work and innovations are as follows:1. We study and implement the decoding process entropy decoding module(ED), CTU-row pixel decoding module(CTU), deblocking filter module(DF). Achieving the frame level parallel entropy decoding algorithm, CTU row pixels parallel decoding algorithm and parallel deblocking filter algorithm respectively based on correlation analysis frame type dependency analysis and CU block data,and ensure the quality of the reconstructed image while improving decoding parallel speedup.2. We study and implement a multi-level HEVC parallel decoding algorithmin based on Tilera multi-core processor. And combining the entropy decoding module(ED), CTU-row pixel decoding module(CTU) and deblocking filter module(DF) through a parallel pipeline designed to improve multi-core utilization and reduce delay. Implementing the CTU level and frame level processing coexistence of multi-level parallel architecture HEVC decoding algorithm. A mathematical model of parallel speedup analysis based on CTU line explains the multi-core parallel speedup bottleneck.3.We inmplement platform level optimization based on Tilera multi-core processor.Optimization of multimedia applications provided by Tilera GX36 multi-core processor platform,including compiler optimization,memory optimization,instruction optimization, FeedBack optimization, to further improve the processing performance of the parallel HEVC algorithms.Experimental shows the validity of each of the parallel decoding method proposed above.And the speedup of frame level parallel entropy decoding algorithm reached 1.6, the speedup of CTU row pixels parallel decoding algorithm reached 9.5, the speedup of parallel deblocking filter algorithm reached 2.0, And the speedup of multi-level HEVC parallel decoding algorithm for decoding high-definition and ultra-high-definition video stream could reach 15.9, higher than WPPparallel method about 37%,the frame rate could reach 30 frame per second or more, achieving the goal of real-time decoding.Finally, this thesis gives a summary of our work and prospects the future research.