Design And Implementation Of A Many-core Software Mapping Tool

For the many-core processors,which are integrated with many cores,it is essential to get the appropriate supports of the fundamental software tools to release their parallel computing power completely,or to improve their lifetime reliability effectively.In order to meet this requirement,a many-core software mapping tool was implemented after some relevant software mapping algorithms were studied and designed.The specific work of this thesis is as follows:(1)Based on the idea of dynamic programming,a software mapping algorithm for throughput single-objective optimization was presented.By dynamically adjusting the expected value of task subgraph workload in the process of tasks allocating,this algorithm effectively improved the imbalance of subgraph workload caused by the granularity of task nodes,and achieved the purpose of improving the system throughput.Experimental results showed that this algorithm can effectively improve the throughput of many-core systems compared with the existing algorithms.(2)Based on the idea of mean value gray wolf algorithm,an approximate algorithm of throughput single-objective optimization for software mapping was gained.The purpose of dynamic programming algorithm is to search the exact optimal solution.If the state space is large,it will take a lot of time.The software mapping algorithm based on the idea of mean value gray wolf algorithm can get the approximate optimal solution to meet the needs in a limited number of iterations.(3)Considering the EM effect,the lifetime reliability model of many-core processors was established by lognormal function of fault distribution.The impacts of different software mapping methods on the lifetime reliability of many-core processors were revealed by the simulation experiments.That provided a basis for the later algorithm design.(4)The dual-optimization problem model of reliability and throughput for many-core software mapping was formulated.Furthermore,the linear weighting method and the two-stage method of dual-objectives optimization mapping algorithm were designed and implemented by the framework of simulated annealing algorithm.The experiments indicated that both algorithms can improve the system throughput and ensure the system reliability.(5)A user-friendly many-core software mapping tool was implemented after the above algorithms were integrated together.The tool provides multiple alternatives for many-core software mapping so that the system can not only exploit parallel computing ability,but also guarantee the lifetime reliability.Some exploratory work has been done for the later development of a more powerful manycore software synthesis tool in this thesis.It is helpful for China's high-end industrial software in the field of many-core techniques to break the technology blockade of foreign countries.