Based On The General-purpose Programmable Gpu Video Codec - Architecture, Algorithms And Implementation,

With the development of digital TV and networks, commodity PC centered amusement fashion for digital home becomes a mainstream trend. How to implement low cost real-time encoding-decoding for high definition video under existing PC system without upgrading hardware, has come into a number of researchists' notice.The appearance of programmable GPU (Graphics Processing Unit) provides a possible approach for solving the above problem. As the core processor of graphics card, GPU is traditionally in charge of 3D rendering. While, researchers discover GPU can take on some other tasks besides graphics rendering because of its programmable ability. General scientific calculation for GPU has become one of hot spots rapidly in computer science field. However, the study on applying GPU in video coding system is still very rare so far.This thesis presents a scheme that uses GPU as an effect co-processor for accelerating video encoding-decoding. For fully utilizing the powerful calculation capacity of GPU, we develop a parallel architecture for codec based on GPU+CPU system. We also advance a series of video coding algorithms that are fit for GPU. The results of experiments show our scheme can improve the encoding-decoding speed considerably compared with conventional codec for high definition video without upgrade existing hardware.The thesis briefly introduces the development of high definition video technique and the operation principle of programmable GPU in chapter one and two. The third chapter detailed discusses our MPEG2 encoder based on GPU. The chapter concretely analyzes its special parallel architecture, original ATSMVP, MCSAD and ZB-LMES algorithms and Bi-threads programming. Meanwhile, in the chapter, we analyze and compare the performance between our encoder and conventional encoder based on the future of encoding speed and SNR. The forth chapter describes our MPEG2 decoder for GPU in detail. The chapter expounds our hierarchical parallel architecture of decoder, original multiple channel algorithms and a load-balancing algorithm between GPU and CPU. Moreover, we compare the performance between our new decoder and tradition decoder in the forth chapter.Through the results of experiments and serious analysis, we come to the conclusion that the parallel codec based on GPU+CPU system has more high encoding-decoding speed for high definition video compared with conventional structured codec that has been optimized by SSE2. Meanwhile the new codec still can keep the high coding gain.