| With the exponentially growth of the data in Internet age, the digital video which shows diversified and real-time characteristics has become an important means of information acquisition for people. High Efficiency Video Coding is the new coding standard, which implements technological innovation on quad-tree coding unit and loop-filter. The standard test models use quad-tree encoding unit to predict and partition for the block, increasing flexibility and complexity of coding. Using the dataflow programming model to improve the coding performance has great value of practical application.Analysis and mining the dataflow features of predict partition module of quad-tree coding unit, separate independent computing nodes and data communication under different modes, implements data stream parallel optimization under intra and inter prediction. Loop-filter consists of the deblocking filter and the sample adaptive offset filtering. In view of the task and data parallelism of filter module, dataflow optimization of the minimum unit which contains vertical boundary filter and the horizontal boundary filter are realized. Analysis the pipeline level and each color component parallelism for sample adaptive offset on maximum coding unit, use dataflow programming language COStream to implement this module and integrate with the deblocking filter module, verify the effectiveness of dataflow programming model. Use mixed compilation mode of COStream and C++ to achieve the part which is suitable for data stream, the remaining part in test model remains the original code, then the backend code of COStream and remaining code execute program efficiently together.We choose the generic X86-64 multicore server as the experiment platform and the standard test sequences as the input of mixed model to perform the experiment and analyze programs performance. Our experimental results show the feasibility and efficiency of COStream programming language under this hybrid programming model. |
PDF Full Text Request