Optimize Hevc Codec Algorithm On CUDA

With the growing demand for the standard-definition video and high-definition video, H.264video coding standard is not well positioned to meet the network and the people demand. The international organization ITU and MPEGE established JCT-VC (Joint Collaborative Team on Video Coding), specializing in the development of a new generation of video coding HEVC (High Efficiency Video Coding) who is also known as H.265, its goal is to save about50%bit-rate than the video coding standard H.264/AVC High Profile.JCT-VC convened the first meeting in2010, most of the work has been completed, and it will launche the first grass in2013. Compared to H.264/AVC performance, the coding performance of HEVC has been greatly improved.The performance improvements by increasing complexity in exchange for the cost of. Due to a high degree of complexity, in fact, the coding performance when the application is limited, it is necessary to further optimize and improve the encoding speed, the paper focuses on the GPU(Graphic Processing Unit) direction of their research to accelerate, promote the use of standard HEVC significance.In this thesis, and the video encoder in HEVC CUDA (Compute Unified Device Architecture)optimization,CUDA structure characteristics of the large amount of computation time-consuming part of such as adaptive interpolation filter(AIF) and EPZS(Enhanced Predictive Zonal Search)fast motion estimation part placed inside CUDA.EPZS algorithm implementation,we use small pieces of PU (Prediction uint) SAD(Sum of Absolute Difference) the value of synthetic PU chunk SAD value.In addition,movement HEVC estimated in the encoding unit LCU(Largest Coding Unit) can be64×64, most small pieces the size of4×8and8x4,the maximum block and the small difference is very great,movement is more intense when the video sequence when this error by examining the adjustment of the initial search point,all sub-blocks of the initial search point as close as possible to the real search point.The experimental results show that:By to modify HEVC framework and use CUDA to accelerate the AIF can significantly reduce coding time, the overall encoding time speedup of up to1.25or so (that is, the encoding time is about a25%reduction); in the the EPZS movement estimated speedup smaller aspects of movement sequences coding timespeedup about7times, the medium of movement and the larger the sequence encoding time speedup at about17times. The AIF module with two EPZS module integrated overall encoding time average acceleration ratio of1.5or above (ie, the encoding time in half).