数据并行语言中的任务并行

Task Parallelism in a Data Parallel Language

数据并行虽然已经获得了广泛的应用，但是，仍然有一些应用程序不适于数据并行语言的并行模式，如树结构算法。数据并行与任务并行的结合可以很好地解决这些问题。该文主要讨论了在数据并行中引入任务并行时，遇到的共享变量、代码生成和处理器分配等问题，比较和分析了基于编译、基于语言和基于协作库的方法。

Data parallelism has been used in a wide field, but many applications can't be whtten in the way of data parallelism, for example, data structure algorithm. Integraing data parallel and task parallel can solve these problems.This paper discusses share variables, code generation and processor distribution, compares the compiler-based, language-based and coordination libop method when task parallelism is implemented in a data parallel langUage.