Optimization Of Opus Decoders In Arm Cortex-A8

Currently, with the rapid development of embedded systems, various applications ofmultimedia technology focus on mobile platforms, most of which are using ARMprocessors. Due to the limited performance of the ARM processors, the optimization of theaudio decoder is important for the mobile platform.ARM Cortex-A8is an ARM processor, which is based on architecture ARMv7.Itsunique NEON coprocessor makes it possible to deal with audio, video and graphicsparallelly. Opus is a latest audio coding algorithm, which is a perfect blend of the speechcoding algorithm SILK and the music coding algorithm CELT. Compared to the otheraudio coding standards, such as AAC, OGG, Opus has higher compression ratio andexcellent quality. Now Opus gets more and more attentions in the Internet, mobilemultimedia and other areas. Optimization of Opus decoder on the Cortex-A8has broadprospects.Cortex-A8with NEON coprocessor is based on the ARMv7instruction set. There aretwo pipelines in the Cortex-A8processor, which is including13stages traditional ARMpipeline and10stages NEON co-processing unit pipeline. With the analysis of the twopipelines, some practical ARM pipeline strategies which are applied to the NEONoptimization of Opus decoder are mentioned.After analyzing the Opus decoding process and the algorithm of some complexmodules, some modules that is suitable for using parallel processing mechanism areoptimized by using NEON pipeline. For the other modules, the optimization is justdirectly to use the ARMv7instruction set. Finally, the optimized decoder is tested by opensource codecs FFmpeg on iPad1. Test results indicate that NEON optimization can makethe whole performance of SILK decoding increase about60%and the whole performanceof CELT decoding increase about78%.