Research On Multi-task Parallel Online Scheduling Technology For Heterogeneous Multicore System

With the continuous development of VLSI technology,the integration of the resources on the processor chip will be greatly increased,using these resources effectively play the key to the performance of the processor.The performance of traditional processor mainly improved by developing instruction level parallelism,but it becomes more and more difficult to improve processor performance,which is limited by hardware complexity,application inherent ILP and storage latency.Developing a higher level of parallelism and thread-level parallelism becomes the primary choice for a new generation of processor architectures.Therefore,based on the application of superscalar technology and multithread technology,several technologies are proposed for designing and optimizing the target system.Superscalar technology is a parallel operation of instruction level parallelism in the processor,which can achieve higher instruction throughput in the same working environment with space in exchange for time.Out-of-order execution is a common technology of superscalar technology,allowing instructions to be executed in a different order in the program,which avoid the instruction delay of the processor operation,thus shortening the execution time of the program.Multithreading technology improves thread-level parallelism and instruction-level parallelism by executing instructions from multiple threads to achieve higher program efficiency.Based on the technical background,this thesis presents a multithreaded processor for heterogeneous multi-core systems,the main work of the thesis is as follows:Firstly,the technical background and research status of parallel technology,coarsegrained reconfigurable technology and multi-core technology are introduced.Based on the analysis of the model of coarse-grained heterogeneous multi-core system,the task of the target system is instructed,and a two-layer programming framework is proposed,which is designed for the target system,and the programming method is standardized.Secondly,the key techniques of multitask parallelism are studied,based on the function of multithreading instruction and resource strategy,a strategy of fetch and resource allocation for target system is proposed,which aims at using processor resources rationally and speeding up instruction execution time.This thesis analyzes the multithread instruction scheduling technology at the same time.Based on the Out-of-order execution technology,some key technologies are proposed to optimize the thread scheduling.Thirdly,the main controller is designed and implemented,with analyzing the working principle and implementation scheme of each key module.At the same time,we monitor the resource of the system in real time by integrating them into the management of the main controller,and the resources are allocated dynamically according to the instruction demand.Finally,the design is integrated and tested,the function and correctness of the main controller are validated by loading different task instructions,and the direction of further optimization is put forward.