Arzumanyan R.V., Sukhinov A.I. Study the feasibility of high-performance software Google VP9 decoder.

Abstract: The article is devoted to optimization and execution of parallel decoding stages of the video signal compressed in accordance with specification Google VP9. The authors in detail discuss the most time consuming stages of decoding and restoring a compressed video and study possible optimization and parallel execution of algorithms underlying such steps using both CPUs and graphics cards with general-purpose computing support. The article gives a comprehensive assessment of the characteristics of the decoding stages, including the requirements for processor and memory subsystem. The main method of the study is in carrying out a numerical experiment with the collection of information of interest and then analyzing the results. Gathering of information is implemented by modifying the source code reference codec and subsequent assembly into a software codec application. The novelty of the work lies in the fact that it carried out a comprehensive analysis of the possibility of computational methods lying in the codec based. The research evaluates the feasibility of parallel calculations, taking into account peculiarities of the target hardware (MCCPU and GPGPU). The authors performed an optimization of arithmetic decoding step taking into account the statistical characteristics of the distribution of the lengths of literals, decoded from a compressed bit stream. In this article, the authors make conclusions regarding the most computationally complex decoding stages and the possibility of their optimization and parallel implementation, and analyze differences between the described codec a competing codec N265.

Keywords: