作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 体系结构与软件技术 • 上一篇    下一篇

浮点与整数资源区别分配的SMT处理器取指策略

蒋生健,胡向东,杨剑新   

  1. (国家高性能集成电路(上海)设计中心,上海 201204)
  • 收稿日期:2016-04-12 出版日期:2017-04-15 发布日期:2017-04-14
  • 作者简介:蒋生健(1989—),男,硕士研究生,主研方向为计算机体系结构、高性能微处理器设计;胡向东、杨剑新,高级工程师。
  • 基金资助:
    “核高基”重大专项“超级计算机处理器研发”(2013ZX01028-001-001-001)。

Instruction Fetch Policy for SMT Processors with Different Allocations of Floating-point and Integer Resources

JIANG Shengjian,HU Xiangdong,YANG Jianxin   

  1. (National High Performance Integrated Circuit Design Center,Shanghai 201204,China)
  • Received:2016-04-12 Online:2017-04-15 Published:2017-04-14

摘要: 在同时多线程处理器中,各线程对于浮点和整数资源需求不同,合理分配线程的共享资源是提升处理器整体性能的重要因素。为此,提出一种浮点与整数资源区别分配的取指策略,合理分配各个线程对于浮点和整数资源的使用情况。实验结果表明,与ICOUNT,STALL等策略相比,该策略在算术平均IPC和调和平均IPC方面均取得一定的性能提升,同时其在处理浮点和整数混合型程序时也具有优势。

关键词: 同时多线程, 取指策略, 资源分配, 线程级并行, 浮点, 整数

Abstract: In Simultaneous Multithreading(SMT) processors,different threads have different demands for floating-point and integer resources.How to allocate shared resources among threads is the key point to improve the whole performance for SMT processors.Aiming at this problem,this paper proposes an instruction fetch policy with different allocations of floating-point and integer resources,which reasonably allocates the usage of each thread for floating-point and integer resources.Experimental results show that,compared with ICOUNT,STALL policy,etc.,the proposed policy improves performance when using average IPC and harmonic average IPC as a metric.Meanwhile,it also has advantages when processing the programs mixing floating-point and integer.

Key words: Simultaneous Multithreading(SMT), instruction fetch policy, resource allocation, thread-level parallelism, floating-point, integer

中图分类号: