摘要 |
<p><P>PROBLEM TO BE SOLVED: To improve the cost-to-performance ratio of a multiprocessor for parallel-processing so as to achieve performance enhancement scalable to increasing semiconductor integration. <P>SOLUTION: The multiprocessor includes a plurality of processing elements each having a CPU, a distributed shared memory and a local data memory; and a centralized shared memory connected to each of the processing elements. Before data shared among tasks assigned to the processing elements is needed for each of the tasks, the data is transferred to the distributed shared memory of the processing element where the data is to be consumed. In dynamic scheduling used for conditional branching in coarse grain parallel processing, the centralized shared memory stores data unassigned to any CPU until execution of a program. <P>COPYRIGHT: (C)2010,JPO&INPIT</p> |