An Approach To Energy Optimization Based On Frequent Pattern And Surrogate Model At GCC's Compile Time

Energy consumption is a kind of critical quality attribute of embedded software.Optimizing the software for low energy consumption is considerably significant when adhering to a strict power budget.In recent years,many approaches have been proposed to optimize energy consumption by means of selecting the best group of options ? However,neither these algorithms consider the potential multivariate interactions between compilation options,nor do they care the how time-consuming the e of energy consumption is.Consequently,it has negative effect on solution quality and convergence rate.To solve this problem,a novel approach to energy consumption optimization at GCC compile time,named FACET,is proposed in this paper.The primary contributions of this paper are as followsBuild an energy consumption evaluation process based on adaptive surrogate models.Under the consideration of building,testing,applying and updating of this surrogate model,a set of strategies are designed to improve accuracy and time consumption and these strategies are combined and applied in a clear way.As a result,a new process is built,which can both assure high accuracy of energy consumption prediction model,meanwhile,reduce the time of energy consumption evaluation.To effectively captured multivariate interactions with the appearance of high frequency and positive effect on energy consumption during evolution,frequent patterns(FPs)with annotation of energy consumption improvement is mined on the extended FP tree[15] by our algorithm.An energy consumption assessment method based on precision adaptive proxy model is proposed.Based on the construction,testing,use and update of the proxy model in the optimization process,this method designs four strategies to improve the accuracy and time gain,and further defines the process of combining these strategies.The method can maintain the high accuracy of the agent model during the optimization process,and can effectively reduce the energy consumption evaluation time.The GCC compile time energy evolution optimization algorithm is designed.The heuristic information set with the frequent compilation option pattern with energy consumption improvement annotation defines two kinds of mutation operations: “add”and “delete”.Furthermore,the energy fitness evaluation method proposed in this paper is used to calculate the individual fitness value,which can effectively improve the solution quality and convergence speed.Developed a GCC compile-time energy evolution optimization framework.The framework integrates the mining algorithm,energy evaluation process and optimization algorithm proposed in this paper organically under the eclipse platform,and is convenient for users through flexible configuration.A case study was carried out.The experimental FACET method and Tree EDA algorithm were compared in 8 different softwares in 5 different fields.The results show that the solution quality,convergence speed and multivariate correlation of compile options in the FACET method are statistically superior to the Tree EDA algorithm.The introduced proxy model has no significant statistical solution to the quality of the solution.The impact,and while maintaining a high degree of accuracy during the evolution process,also achieved a larger time gain than the actual energy consumption assessment.