目前，片内集成大量轻量级处理器核的众核处理器已经成为研究热点。类似于MPP系统，众核处理器的合理使用模式为多任务共享，每个任务分区（Partition）使用处理器资源。 但是，一方面，不同任务具有不同的资源需求，同一任务的资源需求在执行过程中也体现出明显的阶段性特征。另一方面，由于Cache一致性的协议开销，在众核处理器中实现全局Cache一致性是不可行的。 基于以上原因，本课题研究支持动态可扩展Cache一致性分区（SCCP）的众核处理器关键技术，实现任务资源的按需分配和分区隔离，达到提高资源利用率并有效降低Cache一致性开销的目的。本课题的主要研究内容包括：硬件动态可重构子网、动态可扩展Cache一致性协议，以及支持SCCP的软件动态资源调度机制。 本课题拟采用软件模拟器和关键模块FPGA实现相结合的方法进行仿真验证，并通过典型应用和benchmark进行性能评测。

Manycore processors integrating a large number of lightweight cores have become current research focus. Similar to MPP systems, multi-task sharing is a reasonable way to take full advantage of manycore's resources, by means of partitioning processor resources for each task. However, different tasks have different requirements of processor resources, and even for a single task, its resource requirement has obvious characteristic of phase during the task execution. At the same time, implementing the global cache coherence in manycore processor is not feasible due to the huge overhead of the Cache coherence protocols. Because of the reasons mentioned above, the key techniques supporting dynamically Scalable Cache Coherence Partition (SCCP) in manycore processor are proposed, which can allocate processor resources according to the demands of tasks and isolate tasks' communication from each other. These key techniques can improve the utilization of processor resources and reduce the cost of cache coherence effectively. The main research contents of this project are: hardware-based dynamically reconfigurable subnetting, dynamically scalable cache coherence protocol and software-based SCCP-oriented dynamic scheduling algorithm. In order to verify the proposed techniques, a software-based simulator will be implemented to simulate the full manycore system, and some key modules will be implemented using FPGA. In addition, typical applications and benchmarks will be used to evaluate the overall performance.