Research On Key Technology Of Multi - Core Processor Mapping

With the fast development of multi-core technic, the cost of manual partition becomes higher and higher. How to realize high-performance program with limited resource on multi-core platform is a very hard problem. We intend to solve this problem with automatic parallelism tool. With task partition algorithm, we get program with higher performance.Task partition on multicore platform is a NP-Hard problem. But when we intend to achieve more parallelism, the necessity is to find more parallelism between program language items. One result is that grain of the task-dependence graph becomes smaller. This result greatly increases the challenge of algorithm design.Cluster Analysis is a common unsupervised learning method. It aims at classifying objects with no enough knowledge. With various source codes of program, task partition is also a classification problem that could not get enough priori knowledge. This thsis analyses cluster algorithm, and applys it into task partition problem. Under new thought of task grain, distance function, cluster progress, quantization of performance, we designs cluster algorithm and achieves the tool with C programming language. Our tool reads the quantization formation of task graph, outputs data and graphs that represents all kinds of performance targets and partition result. Owning to the simulation function in this tool, the time-consuming hardware simulation could be replaced in some way, so we can not only partition the graph, but also do research more effectively.