Mapping of Nested Loops to Multiprocessors
嵌套循环到多处理机的映射

Two new methods for partitioning and mapping nested loops with non-constant dependencies into distributed memory multiprocessors are presented. By partitioning the dependencies vectors or using direction vectors, the methods can partition the loops with non-constant dependencies into independent parts without any mutual dependencies. These parts can be processed independently so as to be mapped into multiprocessors and be executed in parallel.