Skip to Main Content
Diverse approaches to parallel implementation of H.264 have been proposed; however, they all share a common problem. The entropy decoder in H.264 remains mapped on a single processing element (PE). Due to the inherently sequential and context-adaptive nature of the entropy decoder, it cannot be parallelized. This renders a bottleneck to the performance of the entire decoding process. Depending on the type of the processing core and the video bit-rate, the performance of the entire decoding process is subject to the process of entropy decoding. It is, therefore, needful to research and implement new algorithmic solutions to compensate for this bottleneck, and thereby make optimal use of parallel implementation of H.264 decoder on mainstream multi-core systems. This paper presents a new CAVLC decoding method which is de-rived by constructing custom CAVLC decoding tables using dasiatable groupingpsila. Compared to the conventional dasiasequential table look-uppsila method, which requires multiple memory accesses. Our proposed method accesses the custom tables only once for the decoding of any symbol. Moreover, in our proposed method, the symbol decoding time does not depend on the symbol length and it is constant for each symbol, resulting in a nearly linear increase in computational complexity with increase in video fidelity as compared to an non linear increase in earlier proposed methods. Experimental results show that our proposed algorithm features up to 7times higher performance and 83% less memory accesses compared to conventional methods. We compare to three commonly used, state-of-the-art CAVLC algorithms, such as table look-up by sequential search, table look-up by binary search, and ldquoMoon's methodrdquo.