一种动态共享数据结构的并发访问控制分析方法

An analysis approach to concurrent access control for shared mutable data

共享数据的并发访问控制是并行编程的关键之一,而对动态共享数据结构的细粒度并发访问控制更是其中的难点.针对操作动态数据结构的指针程序,提出一种并发访问控制的分析方法,该方法采用形状图对程序运行时的动态数据结构和指针变量关系建立抽象,通过形状图推导完成对共享单元的访问控制分析.程序员只需要声明指针及指向单元的共享性和数据结构形状特征,编译器利用本文中的方法分析得到共享单元的访问控制点,并添加使用互斥锁进行访问控制的代码.

Concurrent access control for shared data is extremely important for parallel programming. And it is challenging to perform fine-grained concurrent access control for mutable-shared data. To address these problems, a novel approach was presented to achieve concurrent access control for concurrent programs with pointers. A shape gi＇aph was used to establish the abstract relationship between dynamic data structures and pointer variables in the run time, inferring the synchronization points. Our solution only requires programmers to provide some necessary sharing and shape information of the pointers and the memory cells for program analysis, which produces the synchronized points for each shared memory cell. Then, the compiler generates the mutual-lock-based code in terms of these synchronized points.