基于LLVM架构的图像处理程序的并行分类

Parallel Classification of Image Processing Program Based on LLVM Architecture

设计了一种针对图像处理类应用程序的并行性分类的方法,并使用工具初步实现了对程序的分类.还研究了图像处理算法的并行性.通过对图像处理算法的主要特征进行分析,得出图像处理算法包含多种并行特征的结论,分别从图像处理程序的混合结构的模型,以及程序中的依赖关系和并行特征两方面进行了分析.结合对程序的分类方式,使用LLVM编译架构初步实现了自动对程序中的程序段进行识别其中的分类模式.

According to the main characteristics of the image processing algorithm, the conclusion that the image processing algorithm contains a variety of parallel features is analyzed from the model of the mixed structure of the image processing program and the dependencies and parallel features in the program. Then, according to the analysis of the dependencies in the loop, the parallelization of the different levels of the program is found, and the loops are classified. The different categories correspond to the different execution units, and the typical image processing procedures are further analyzed.