Abstract: This paper proposes a system-level memory optimization approach，which uses LGDFG（Large Grain Data Flow Graph） model to share memory space of application programs.LGDFG model comprises nodes and edges，which denote sub-functions and producing-consuming relationship of variables respectively.After checking the LGDFG，the live time of each global variable is determined.Variables which have non-overlapping live time can share the same memory space in different periods.Finally，AC3 decoding program is employed as an example to describe the approach in detail.Implementation results show that the memory space has decreased 25.8% for AC3 decoding program.This approach has reduced the data memory size of application programs significantly.

Key words: memory optimization, data flow graph, AC3, audio

摘要： 提出了一种基于LGDFG（粗颗粒度数据流图）的系统级存储优化方法，该方法为被优化的程序建立LGDFG模型，把每个程序段作为该模型的一个节点，分析前后节点之间的数据产生消耗关系，从而得到不同数据块对应的生存周期，对于生存周期相互间不重叠的数据段可以实现存储共享。以音频AC3解码程序为例，用LGDFG模型分析了程序中不同数据块之间的产生消耗关系，从而实现不同数据块间的存储共享。试验结果表明使用该方法AC3解码程序数据存储容量减少了25.8%，有效地减少了数据存储容量。

关键词: 存储优化, 数据流图, AC3, 音频