【摘要】 为提高龙芯平台的多媒体处理能力,分析龙芯处理器核GS464/GS464E的LoongSIMD指令系统,提出SIMD指令级优化方法。通过分析FFmpeg的开源H.264解码器的并行特征,结合龙芯处理器的结构并行特点,利用龙芯特有的性能计数器,解析影响性能的热点函数,采用手工嵌入龙芯SIMD和访存扩展等汇编指令的方式进行细粒度优化。实验结果表明,该方法使平均性能提高了50%,工作成果已推送到龙芯开源社区,增强了基于龙芯3A处理器的国产计算机在桌面多媒体应用领域的用户体验。更多还原

【Abstract】 To improve the ability of multimedia processing on Loongson platform,the LoongSIMD instruction system of GS464/GS464 ECPU core was analyzed,and the SIMD instruction level optimization method was proposed.Open source H.264 decoder of FFmpeg was finely optimized while analyzing the parallel characters of algorithms,combining Loongson distinctive architectures and performance counter,intrinsic SIMD instructions and load/store memory extension instructions were embedded in hotspot functions.The average performance is improved by 50%.This work is contributed to Loongson open source community.User experience to multimedia applications is enhanced on Loongson computers.更多还原