SoC数据映射中的快速应用程序存储分析方法  

Fast Memory Size Estimation of Application Programs for System-on-Chip Signal-to-Memory Mapping

作　　者：赵鹏[1] 李思昆[1] 

机构地区：[1]国防科学技术大学计算机学院,长沙410073

出　　处：《计算机研究与发展》2010年第2期361-369,共9页Journal of Computer Research and Development

基　　金：国家"八六三"高技术研究发展计划基金项目(2002AA1Z1480);国家自然科学基金项目(90207019;90707003)~~

摘　　要：片上系统SoC在多媒体信息处理领域中应用广泛.多媒体处理程序中频繁的循环嵌套与多维数组操作严重影响着多媒体处理SoC系统的数据传输与存储效率.根据程序各个部分的存储需求量将程序中数据映射到SoC存储层次上,这是改善SoC系统性能与功耗的必要途径.针对多媒体处理程序,提出一种面向SoC数据映射的快速存储需求量分析方法.在存储需求量分析过程中,提出并使用正交线性有界格对所操作数据的数据域进行划分,并基于相关性进行存储需求量计算,得到了较为准确的存储需求量,并大幅度减少了分析时间.System-on-chip （SoC） is comprehensively applied in the field of multimedia information processing. The application programs of multimedia information processing are abundant in nested loops and multi-dimensional signals, which greatly affects the efficiency of data transfer and storage. Therefore, the SoC signal-to-memory mapping, which pays attention to memory size of application programs and optimizes the efficiency of memory system, tends to obtain better SoC performance and declining power consumption. Concentrating on nested loops and multi-dimension signals in multimedia processing programs, a fast memory size estimation approach is presented for SoC signal- to-memory mapping. On the basis of polyhedral model and linear bounded lattice, orthogonal linear bounded lattices are put forward to partition the data domain of the multi-dimension signals, and then the memory requirement size is computed according to the data dependency of orthogonal linear bounded lattices. Orthogonal linear bounded lattices are the minimum process unit during dependency analysis and memory size computation, which can greatly reduce the analysis time and keep the estimation accuracy. The memory requirement size is taken as heuristic information and mapping criterion by the SoC signal-to-memory mapping algorithm, which helps to explore the signal-to- memory mapping space for the purpose of efficient data transfer and storage.

关 键 词：片上系统 多媒体信息处理 存储需求量分析 多面体模型 正交线性有界格 数据传输与存储优化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]