一种有效的同时多线程处理器取指控制机制  

An Effective Instruction Fetch Control Mechanism for SMT Processors

作　　者：何立强[1] 刘志勇[2] 

机构地区：[1]内蒙古大学计算机学院 [2]国家自然科学基金委员会信息学部,北京100085

出　　处：《计算机学报》2006年第4期535-543,共9页Chinese Journal of Computers

基　　金：国家自然科学基金(60325205;60303033;60373043)资助

摘　　要：同时多线程处理器通过每时钟周期从多个运行的线程取指令执行,极大地提高了处理器的性能.分支预测器的预测精度和取指策略的效率是影响同时多线程处理器性能的关键.通过将一个基于值的分支预测器和一个基于线程推进速度的取指策略相结合,提出一种新的取指控制机制.该结构的硬件开销较小,实现复杂度较低.实验结果表明,该取指控制机制有效地提高了处理器的性能,其相对于传统取指控制机制的性能加速比为28%且该加速比也高于目前基于流缓冲区和基于分支分类器的取指控制机制.Simultaneous Multithreaded （SMT） processors improve the instruction throughput by allowing fetching and running instructions from several threads simultaneously at a single cycle. Accurate branch prediction and effective instruction fetching are crucial to an SMT processor. In this paper, the authors propose an effective instruction fetch control mechanism for SMT processors through combining a value based branch predictor and an instruction throughput based fetch policy. The implementation of the mechanism is easy, and the hardware budget is less than that of the traditional SMT fetch control schemes. Execution-driven simulation results show that the mechanism improves the performance of an SMT processor dramatically. The average speedup of instruction throughput is up to 28% over that of the traditional SMT fetch mechanism, which is also higher than that of stream buffer based scheme and the branch classifier based scheme in the literature.

关 键 词：同时多线程处理器 取指策略 分支预测器 分支分类器 取指控制机制 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]