Video coding, serving as the processing of compressing and decompressing digital video signals, is the key technique that promotes the wide usage of digital videos. With the increase of video resolutions, coding complexity and applications of real-time processing, mainstream video coding standards such as H.264/AVC and HEVC encounter the problem of excessive processing time. To release the computational burden, this thesis mainly focuses on the parallel algorithm designs and GPU implementations in video coding, especially for deblocking filter, intra prediction and inter prediction. Specifically, for each algorithm, a comprehensive analysis of dependency structure is conducted to identity the crucial elements preventing the algorithm from being parallel, with which, the correspo...[ Read more ]

Video coding, serving as the processing of compressing and decompressing digital video signals, is the key technique that promotes the wide usage of digital videos. With the increase of video resolutions, coding complexity and applications of real-time processing, mainstream video coding standards such as H.264/AVC and HEVC encounter the problem of excessive processing time. To release the computational burden, this thesis mainly focuses on the parallel algorithm designs and GPU implementations in video coding, especially for deblocking filter, intra prediction and inter prediction. Specifically, for each algorithm, a comprehensive analysis of dependency structure is conducted to identity the crucial elements preventing the algorithm from being parallel, with which, the corresponding parallel algorithm is investigated. The proposed parallel algorithms achieve competitive coding efficiency with considerably higher degree of parallelism compared with previous work. In addition, the GPU implementations are well developed and optimized, by taking various issues (coalesced accesses, memory bank conflicts, hardware occupancies, etc.) into account.