A Proxy Modeling Method for Optimizing Sequence Selection in Multi-Object Compiler

Abstract

The invention provides a proxy modeling method for multi-objective compilation optimization sequence selection, which is a solution to the computational cost constraint problem and belongs to the field of compiler optimization. First, compile and optimize the sequence into binary codes, design fitness functions for the two optimized target code scales and running speeds respectively, generate sub-populations after selection and crossover operations, and merge with parent populations to generate new populations by fast non-dominated sorting, and finally Obtain the Pareto optimal solution set. In the search iteration process, use the compiled optimization sequence and the corresponding two target fitness values ​​to construct the proxy model, use the proxy model to calculate the approximate fitness value for the subpopulation generated by the evolution operation, and calculate the actual fitness value for the excellent solution , thereby improving the evolutionary efficiency. The invention can effectively select a compiling optimization sequence satisfying multiple objectives (such as running speed and code size) for a program to be compiled, and solve the computational cost constraint problem brought about in the iterative process.

Description

technical field [0001] The invention belongs to the technical field of compiler optimization, and relates to a proxy modeling method for the calculation cost constraint problem in multi-objective compilation optimization sequence selection. Background technique [0002] In the actual development process, the compiled target machine code needs to have relatively ideal performance, such as small executable code size and fast execution speed. However, for the compiled target machine code, code size and running speed are optimization goals that have certain conflicts. If the program running speed is pursued, the program scale will be enlarged. If the pursuit of smaller executable code size, it will affect the running speed of the program. Therefore, how to select an appropriate compilation optimization sequence for the program to be compiled so that it can achieve a trade-off between the two optimization goals of code size and running speed is a key issue. Although the compil...

AI Technical Summary

Problems solved by technology

However, if the multi-objective optimization algorithm in the heuristic search algorithm is applied to the compilation and optimization sequence selection, since a large number of fitness evaluations are required for the population individuals in the process, the actual application needs to be compiled to obtain executable code. Obtaining the code size, another optimization goal in terms of running speed, calculating the fitness value also needs to actually run the program, which will bring about considerable computational cost constraints

For example, in the non-dominated sorting multi-objective genetic algorithm, the fitness value of the compilation optimization sequence corresponding to each individual code in the population needs to be calculated in each round of iteration, that is, the compiler needs to use each optimization sequence to compile the program to be compiled once , and execute the program to obtain its running speed and code size, which will make the execution time overhead of the algorithm very huge, thus bringing about the problem of computational cost constraints

Images