Fine Granular Parallel Algorithm for HEVC Encoding Based on Multicore Platform

Abstract

Compared with the previous standards, the coding efficiency and complexity of High Efficiency Video Coding (HEVC) have been greatly improved. Parallel encoding scheme based on CTU rows like wavefront parallel processing (WPP) and inter-frame wavefront (IFW) can efficiently reduce the encoding time of HEVC. However, due to the coding complexity of CTU within various rows may be quite different, WPP and IFW have the problem of unbalanced load among threads for parallel encoding tasks. To address this issue, in this paper, factors affecting coding efficiency are found by analyzing the data dependence and load relationship of intra- and inter-frame CTUs, and we propose a fine granular parallel strategy accordingly. In the meanwhile, refine the parallel granularity while maintaining the accuracy of symbol prediction requires additional context information in CABAC encoding, which leads to higher bit rate, and will reduce the efficiency of CABAC encoding. In order to decrease the bit rate without affecting the quality, we also making some modifications for the CABAC encoding. The proposed method is implemented on the Tilera-GX36 multicore platform. Experiment results show that our algorithm achieves up to 1.6 and 2.8 times speedup improvement compared with IFW and WPP respectively.