Abstract: In the demand of real-time decoding of high-definition video stream, a multi-layer parallel algorithm is proposed. First, this algorithm distributes tasks to the processors according to the decoding function module and optimizes the load-balance between processors through the secondary partition of the function module. Then, in the loop-filter module, this algorithm processes the macro blocks parallelly according to the dependency relationship between the macro blocks. Experimental results indicate that this algorithm improves the efficiency and the parallelization of the decoder, realizes the real-time decoding of high-definition video stream on the TILEPro64 platform. This algorithm speeds up the decoding process by 10.01 at most, achieves a 80% performance increase of speedup comparing to the existing parallel algorithm.

Key words: H.264 real-time high-definition decoding, parallel decoding, parallel filter, many-core processor, TILEPro64 platform

摘要： 针对高清图像视频的实时解码需求，提出一种基于多层次并行流水架构的解码算法。该算法首先针对图像的宏块行实现基于功能模块的行级并行算法，并通过功能模块的二次划分进行核间负载均衡的优化，再针对解码过程中开销较大的滤波环节，利用宏块之间的依赖关系进行多核并行处理，对行级并行算法进行更深层次上的再优化设计。实验结果表明，该算法可以在TILEPro64平台上实现1 080P全高清码流的实时解码，实现了较高的并行加速比，最高达到10.01，和已有的并行解码算法相比，加速比提升80%。

关键词: H.264实时高清解码, 并行解码, 并行滤波, 众核处理器, TILEPro64平台