SPM（Scratch Pad Memory）正逐渐取代Cache成为嵌入式系统中最重要的片上存储器,本课题将开展动态SPM编译优化研究，进行SPM数据和代码分配。数据分配采用基于存储器访问模式的冗余数据检测与消除方法，避免冗余数据占用SPM空间并减少数据传输代价；代码分配采用基于邻居代码热度"辐射"方法确定代码的优先级，避免SPM中代码块抖动。本课题将嵌入式系统存储层次结构、编译优化技术、局部性原理、物理世界的能量传送方式等融入到SPM编译优化研究中，有望提高SPM所在嵌入式系统的运行时间性能并降低功耗。

Scratch Pad Memory (SPM) has become more important on-chip memory in embedded system than hardware-managed Cache. In this proposal, we mainly focus on dynamic SPM management on compiler optimization to allocate data and code into SPM. For one side, in data allocation, we introduce Redundancy Check and Eliminate method based on memory access pattern, which can not only avoid redundancy data allocation, but also can reduce the cost of data transfer. For another, in code allocation, we use hot radiation of neighbor code to obtain code priority, at the same time, code tremble in SPM can be avoided. The research of this proposal can combine memory hierarchy of embedded systems, compiler optimization, principle of locality and radiation of the real world into the study of SPM optimization. The SPM management of this proposal may achieve significant run-time performance improvement and reduce the energy consumption of embedded systems.