摘要
在多媒体系统的系统集成芯片(SoC)中,从系统集成芯片工作实时性要求,应用程序和数据尽可能存放在片上存储或Cache,执行方便,处理速度快,就要使用大量的存储部件,使得存储部件的面积和功耗占到整个芯片的很大部分。为了减少片上存储部件,则部分程序和数据移到片外存储,在执行时轮流调进到芯片内,势必增加I/O的开销。因此如何使设计优化是软硬件协同设计中的一个问题。本文以MPEG2集成解码芯片中音频存储优化为例给出了系统集成芯片存储优化的一些方法。包括通过LGDFG(LargeGrainDataFlowGraph)模型分析改变程序结构,共享数据空间,改变数据类型以及添加片上SRAM并减少片上Cache容量从而减少系统存储消耗等。这些方法显著地减少系统的存储消耗,降低系统芯片的面积和功耗。
In most of multimedia systems-on-chip (SoC), in order to maintain the real-time requirement, application program and data are preferably stored in on-chip memory or Cache for fast execution and better performance, so large memory is often used and the area and power of the memory units usually hold the largest percentage of the whole chip. In order to reduce the on-chip memory space, if parts of program and data are stored in off-chip memory and are loaded into on-chip memory during execution time, this would increase the I/O overhead. Therefore how to optimize the design is a problem in hardware/software co-design. In this paper, by analyzing the optimization of audio decoding in MPEG2 decoding chip, some approaches are given to optimize the memory usage of SoC. These approaches include changing the structure of the AC3 decoding program by analyzing the LGDFG (Large Grain Data Flow Graph) models, sharing the data memory, changing the data type and exploring the on-chip SRAM memory to reduce the memory space of the system. These approaches have reduced the area and power of the whole chip significantly.
出处
《电路与系统学报》
CSCD
北大核心
2005年第6期45-49,共5页
Journal of Circuits and Systems
基金
国家自然科学基金资助项目(69972043)
国家高技术研究发展计划863基金资助项目(2002AA1Z1140)