针对投机并行化中如何权衡策略并确定合适的执行模型来获取理想性能的问题,提出了一种基于交互信息的投机并行化方法,利用交互信息来确定投机并行化的执行模型,建立相关评价模型,并着重从线程抽取创建角度提出了相应的策略及对应的性能评价。通过实验表明,基于交互信息进行"按需"并行化,可以达到所需的性能要求。
Aiming at the problem about how to trade off between different policies and execution models to get ideal perfor-mance in the exploration of speculative thread-level parallelism, this paper proposed an interaction-based method to exploit speculative thread level parallelism. This solution used interactive information to set up the execution model and evaluation model, especially from the view of thread extraction and spawning. The experimental results show that on demand parallelization based on interactive information can achieve the performance goal.