基于OpenMP的事务存储同步语义研究

Transactional Memory Synchronization Semantics Research Based on OpenMP

多核处理器环境下必须解决多核处理器的并行编程问题,才能够充分发挥多核处理器的性能。事务存储(Transactional Memory)机制提供了一种在多核环境下程序并行执行和同步的方法。已有的工作已将事务存储扩展到了OpenMP,为程序员提供满足事务原子性、一致性和隔离性的共享存储访问。但当前事务存储的语义并不完善,事务间不能交换中间结果,不能实现锁的部分语义。提出并实现了一种基于开放嵌套的事务存储的同步语义,从而解决了事务间不能交换中间结果的问题,增强了扩展事务存储后OpenMP的并行编程能力。

A good prarallel programming environment is the key to explore effectively Multi-core processor＇ s performance. Transactional Memory provides a way to execute program parallell and synchronization for Multi-core Processors programming, and it has been extended to OpenMP which supports atomic, consistent and isolated shared memory access. However, the semantics of transactional memory is not perfect. Intermediate results can＇t be exchanged among transactions, and some semantic of locks can＇t be realized by transactional memory. Based on open nested transactions, this paper presented and realized a kind of transactional memory synchronization semantics, and solved the problem that the exchange of intermediate results among transactions is infeasible, and enhanced the ability of its parallel programing after extending transaction memory to OpenMP.