基于推理多线程技术的一种改进模型

Improved model based on speculative multithreading

为了解决多线程处理器不同线程之间并行性低、相互之间数据依赖性高的问题,在推理多线程技术基础上提出了一种新的多线程技术模型(分级多线程Hierarchical Multithreading HMT).该模型采用两种等级处理元的方法,低等级使用指令级并行和细粒度线程级并行;高等级更多地使用间隔并行机制.通过详细的模拟研究,证明分级多线程技术通过对线程的不同粒度采用并行机制能够切实可行地提高线程之间的并行性.

In order to solve the problem caused by low parallelism among different threads and strong data interdependence, a new multithreading model was presented based on speculative multith/eading. The model, called Hierarchical Multithreading （HMT）, used a two-layered hierarchical method for processing elements. The lower level of the hierarchy exploited instruction-level parallelism and fine-granularity thread-level parallelism, whereas the upper level exploited more distant parallelism. Detailed simulation studies have shown that Hierarchical Multithreading can effectively improve the parallelism among threads by exploiting parallelism among different granularities.