Research On MPSoC Thread Partitioning And Migration Method For Communication Optimization

Automatic multi-threaded code generation technology can generate multi-threaded software code, adapting to different hardware architectures and communication protocols and shortening concurrent software development cycle based on MPSoC and improving development efficiency.In this study, an automatic multi-threaded code generation platform based on Simulink model, LESCEA, is introduced. Based on functional and system modelling, LESCEA takes CAAM, which describes application and system architecture accurately, as input and generates multi-threaded software code for each processor.This study is focused on task mapping, which is the key to automatic multi-threaded code generation. Based on static mapping, a thread migration strategy for MPSoC is proposed, which is applied to LESCEA. In the strategy, two factors, execution efficiency and migration overhead, are considered. As a result, reasonable thread migration opportunities can be selected, maintaining load balance of the processor and avoiding performance loss caused by excessive thread migration.In order to adapt to the characteristics of the thread migration system, in this study, a thread partition method based on ILP is also proposed. Different with mapping functional modules in Simulink model on different processors above all, which is common in conventional methods, in this method, application is divided into units of threads firstly to facilitate thread migration between processors. Aimed at minimizing inter-threads communication, utilizing this method can reduce inter-processor communication and synchronization overhead effectively.Experimental results demonstrate that multi-threaded software code generated by applying the thread migration strategy and thread partition method to LESCEA can increase processor utilization and reduce system synchronization and communication overhead, providing an effective method for improving system performance.