Optimization And Implement Of Hevc High Definition Real-time Encoder Based On Multi-core Platform

Video information is a really important type of digital information carrier.Video transmission has permeated in every aspect of our life.With rapid development of video content and video industry,as well as 4K ultra high definition video and 3D video become more and more popular,the video industry chain are suffering more and more pressure.Higher requirements of future video markets has been getting out of H.264/AVC compression capabilities.In this context,new video coding standard HEVC(High Efficiency Video Coding)was born.The target of HEVC is to achieve 50% bitrate reduction compared with predecessor H.264/AVC.The cost of higher compression efficiency is the rapid increase of computing complexity.HEVC is mainly developed for higher definition video.They both contribute to the obstacles of real-time HEVC encoder implement.Video Coding contains lots of data-intensive computing,which means multiple data performs the same arithmetic calculation and can be viewed as a kind of vector calculation.There exits strong parallel characteristics among data.SIMD(Single Instruction Multiple data)is a type of instruction set designed for parallel data processing.In this thesis,we further studied SIMD technology and optimized sub-pixel interpolation and transform modules,which can greatly improve speed.In sub-pixel interpolation,we designed simplified filter in motion estimation by SIMD,which can significantly improve speed with little compression efficiency loss.In transform module,in order to overcome parallel limit caused by large bit width,we proposed an implementation of rounding the low bits of data by SIMD,which can significantly accelerate speed with ignorable coding efficiency loss.In this thesis,we first give a brief review of HEVC key technologies and main modules.We study the detail of Blocked-based Hybrid Video Coding Framework of HEVC.In the codes of our implemented HEVC encoder,sub-pixel interpolation and transform module compensates large part of compression complexity.We deeply studied the key modules of high complexity and proposed algorithm optimization and SIMD optimization methods.These methods greatly improved encoding speed and reach real-time encoding requirements.After all,we build a high definition HEVC real-time encoder based on TILE-GX36 multi-core platform.This encoding system has a good performance.