SMA处理器中的持续优化技术

CONTINUOUS OPTIMIZATION OF AN SMA PROCESSOR

前瞻多线程结构 (SMA)是在超标量前瞻执行技术和多线程技术的基础上结合了二者的优点而发展起来的 .首先研究了 SMA模型的特点 ,指出了 3个关键性能要素 :现场负载不均衡、线程间控制前瞻失效与线程间数据前瞻失效 .为了有效地开发 SMA结构的潜能 ,引入了若干启发式规则 ,设计了基于线程的动态轮廓采样机制 ,并在此基础上实现了一个持续优化框架原型 .对上述优化规则的模拟表明 ,该优化原型能够较好地完成线程优化任务 ,有效地开发 SMA结构的性能潜力 .

Speculative multithread architecture (SMA) combines both benefits of speculative execution and multithreading architecture. Performance factors are discussed in this paper. In the SMA model, there are three key factors: contexts load imbalance, inter-thread control speculation failure, and inter-thread data speculation failure. To achieve high instruction level parallelism, some optimization heuristic rules are introduced with emphasis on the design of per-thread dynamic profiling mechanism. An optimization framework prototype is built on the basis of those heuristic rules and mechanisms. Benchmark simulation of SMA optimization framework shows the satisfying results.