多核心視訊編碼器之負載平衡

Load Balancing on Multi-core Video Encoder

林俊隆(Chun-Lung Lin) ； 蔡佩璇(Pei-Hsuan Tsai) ； 馬運揚(Yun-Yang Ma) ； 王家祥(Jia-Shung Wang)

《電腦與通訊》 146期 (2012/08) Pp. 88-95

https://doi.org/10.29917/CCLTJ.201208.0013

摘要

近代視訊壓縮技術標準（例如：H.264/AVC及制定中的H.265/HEVC）採用了許多複雜的編碼工具提升壓縮效率及品質但也同時大量的增加了編碼器的運算複雜度。本篇論文研究利用多核心系統平行運算的方式將視訊編碼器平行化以達到編碼速度的提升，提出了一個以畫面片段（slice）為資料平行處理單位的平行演算法，透過改善畫面片段平行化的負載均衡，達到更高的平行化效果。其概念是針對各個即將編碼的畫面片段做運算複雜度的預估，依預測結果重新配置各片段的運算量，且利用一個提早結束編碼的機制縮短畫面片段編碼結束時間的差異，再根據過去各片段運算量的分布，動態調整下一張即將編碼影像中的畫面片段切割方式。實驗結果顯示，本篇論文所提出的方法能在使用四個CPU平行運算時，提供3.7倍至3.9倍的速度提升且僅有極少量的壓縮率及品質下降。

關鍵字

負載平衡 ； 平行化 ； 多核心視訊編碼器

並列摘要

Modern video coding standards, such as H.264/AVC and H.265/HEVC, have adopted many advanced coding tools to greatly improve coding efficiency and quality. However, these complex coding features also incur high computational complexity on video encoder. This article aims at parallelizing the H.264/AVC encoder on a multi-core system and utilizing the advantages of parallel processing to speed up the encoding procedure. This article proposes a novel slice-based parallelism algorithm, which improves the coding system performance by balancing the computing time of slices. The basic idea of the proposed algorithm is to rearrange the estimated complexity of each slice by predicting the complexity of each macroblock, reducing the variations of ending-time among slices using an early termination scheme and balancing the computing time of each slice using a dynamic adjustment of slice partition. The experimental results show that the proposed method can speed up the encoder up to 3.7 times to 3.9 times on a four-core computer system with only a small amount of coding efficiency and quality decreases.

並列關鍵字

Load Balancing ； Parallelism ； Multi-core Video Encoder