一种基于遗传算法的编译器参数自动调优方法
By using a compiler parameter automatic tuning method based on genetic algorithms, the problem of traditional compiler optimization option combinations being difficult to adapt to different platforms is solved. It achieves the best performance selection under the user-defined option set, improving compilation efficiency and adaptability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2023-01-09
- Publication Date
- 2026-07-17
AI Technical Summary
Traditional methods of manually selecting compiler optimization option combinations are difficult to achieve optimal performance across different platforms and program behaviors, and existing compiler default option combinations are often not suitable for specific applications.
An automatic compiler parameter tuning method based on genetic algorithms is adopted. By modifying the tuning option configuration file and the compilation and execution script file, and combining the Xgboost cost model for iterative compilation, the optimal combination of compilation options is automatically selected, realizing the separation of the compilation process and the execution process, and supporting user-defined compilation option sets.
It improves the efficiency of obtaining compilation parameter combinations, saves compilation and runtime time, achieves optimal performance for user-specific goals, has good versatility and adaptability, and does not require modification of user programs or compiler code.
Smart Images
Figure CN116204191B_ABST