Iterative compilation optimization method based on program characteristic analysis

Abstract

The invention discloses an iterative compilation optimization method based on program characteristic analysis. The method comprises the following steps that: S1: classifying compilation options in a compiler according to functions; S2: constructing a relational tree diagram; S3: carrying out dataset testing on a target program, and collecting and arranging testing results; S4: analyzing the testing results, and extracting the program characteristic and the calling relationship of the target program; S5: according to the program characteristics of the target program, finding a corresponding compilation option category, and forming an iterative search space; S6: in the iterative search space, taking the category as a basis to carry out local optimal search to obtain a global optimal solution; and S7: saving, and ending. By use of the method, program characteristic analysis is combined with a local search algorithm to fully consider an influence among all compilation options while the search space is greatly reduced. Compared with an existing method, the method disclosed by the invention can find an optimal compilation option sequence suitable for the target program.

Description

technical field [0001] This patent belongs to the field of information technology and computer, and in particular relates to an iterative compilation optimization method based on program feature analysis. Background technique [0002] Iterative compilation is a performance optimization method for high-performance architecture programs. It generates different versions of the program by transforming various optimization option sequences of the compiler, and selects the program version with the best performance by executing different versions of the program on the same platform. Due to the large number of compilation options in the compiler, the selection of the parameters of each compilation option, the implementation sequence and implementation times of the compilation pass, etc., a huge iterative space is generated; and the different optimization combinations formed in this iterative space Program versioning will incur a huge execution overhead. In order to reduce the over...

AI Technical Summary

Problems solved by technology

However, neither of these two methods can include each other, so only the search results corresponding to the search space can be given, and all optimization options cannot be considered.

[0006] To sum up, various optimization methods for iterative compilation are limited by their own characteristics. Machine learning methods take a lo

Images