An Efficient Scheme for Parallelizing Fast Search Algorithmon SIMD Architecture in H.264/AVC

摘要

As the video CODECs become complexincreasingly according to the convergence of digitalequipments designed for different kinds of specificpurposes, the demand of higher performance for videoprocessing is continuously needed. The fast searchmethods are well known as a representative approachto overcome the performance bottleneck of videoencoding. But they require huge computations anddominate most of the running time of video encoding.So, a parallel processing is required to enhance theperformance. In general, the fast search methods aredifficult to be parallelized because their memoryaccess patterns are not followed by a definite form.This paper proposes a new method for parallelizing afast search method using linear SIMD architecturewith only one-to-one connection between a PE and amemory bank. The simulation results show that theproposed method can reduce nearly 20 percent ofrunning time in the motion estimation (ME).