Research On Synthesis Technology For Embedded Software Of MPSoC

Along with the semiconductor technology into the nanometer era, embedded SoC can be integrated more and more processor cores. It has pushed processor technology into the many-core era. A many-core hardware platform has great potential parallel-computing capability due to its numerous cores. However, many-core systems’ performance cannot be fully played because the matching infrastructure software techniques, like many-core software synthesis tools, etc., lag behind that of hardware. Therefore, it has great significance to study many-core software synthesis technology, and break through the bottleneck of many-core technology.Tasks binding and scheduling algorithm is the key for many-core software synthesis technology. Researching on the task binding and scheduling algorithm with fast convergence and high accuracy is the basic way to elevate the efficiency for many-core software synthesis. Moreover, it is complex and hard to get a better solution for many-core tasks binding and scheduling, which is an instance of NP-hard problem. Problem solving needs to construct a better algorithm according to the particularity of the problem itself. So this paper focuses on the 2D-Torus many-core processor platform, and does some studies on its tasks binding and scheduling problem as follows:Firstly, the BAMSE algorithm is improved according to the features of 2D-Torus NoC. Meanwhile,the modified BAMSE algorithm implements the binding of non-independent tasks with communication cost on 2D-Torus physical processor. The algorithm’s efficiency is explored by experiments.Secondly, a scheme of tasks binding and scheduling based on ant colony algorithm is presented by introducing convergence degree and information weight, as well as combining with new mechanism of pheromone updating. Experiments show that it can get better results than the improved BAMSE algorithm in different CCR.Finally, in order to solve the convergence speed problem of ant colony algorithm, a scheme of tasks binding and scheduling based on genetic ant colony algorithm is putted forward by extracting the advantage of both genetic algorithm and ant colony algorithm. Experiments show the effectiveness of the proposed algorithm.In this paper, we have tried to get some optimal tasks binding and scheduling algorithms with fast convergence and high accuracy. All our work give a foundation for the design of software synthesis tools based on 2D-Torus homogeneous many-core processor platform.