Implementation And Parallelization Of H.264/AVC Decoder On TILE64 Multicore Platform

With the high development of multimedia services, video compression is facing a series of new demand and pressure. For better video data transmission and storage, a number of new video compression standards have been born, such as H.263, H.264, MPEG-4, etc., while the high complexity of video processing algorithm demands high computing capability of the processor system. The multi-core processor provides a new platform for the development of multimedia communication, parallel video processing based on multi-core processors has become a research focus of video communication.This paper focuses on H.264 decoder optimization, based on multi-core processor TILE64, using parallel programming and assembly instructions. First of all, some previous researches about parallelization of video processing are analyzed, the key technologies of H.264 and TILE64 multi-core processor system are introduced, and some key issues about parallel programming based on multi-core processor are focused. Subsequently, H.264 decoder program is ported to TILE64 multi-core development platform, and the performance of decoder on a single-core is tested. According to statistics and the analysis of each module, a new method of decoder pipeline is proposed; according to the study of motion compensation module, a novel interpolation parallel approach is proposed. Then the parallel decoder architecture is constructed. Finally, fraction-pixel interpolation module is optimized to improve the code performance, using TILE64 assembly instructions.As described above, the parallelization of H.264 decoding system is implemented based on TILE64. Compared with the serial decoder, the experiment results show that the parallel decoder can improve the speed of decoding by about 40%, while maintaining the same PSNR(Peak Signal-to-Noise Ratio), and subjective quality is good.