SMA:前瞻性多线程体系结构

SMA: A SPECULATIVE MULTITHREADED ARCHITECTURE

提出了一种新的ＩＬＰ处理器体系结构——前瞻性多线程体系结构，简称ＳＭＡ．它结合了前瞻性执行机制和多线程执行机制，以整个线程为步长进行前瞻性执行，多个线程并行执行并且共享处理器硬件资源．这样，处理器既通过组合每个线程的指令窗口形成一个大的动态指令窗口，开发出程序中更大的ＩＬＰ，又利用多线程执行机制屏蔽各种长延迟操作，达到较高的资源利用率；介绍了ＳＭＡ执行模型，并讨论了ＳＭＡ处理器的实现和其中的关键技术．这些关键技术包括：线程控制、分布式寄存器重命名、指令发射执行、中断处理和处理器对编译优化的要求．性能分析结果表明ＳＭＡ处理器可以适应各种不同类型的程序，以达到高性能，如果采用较好的编译支持和线程预测算法。

Most processors improve their performance by exploiting instruction level parallelism in program, called ILP processors. This paper puts forward a new architecture for ILP processor, speculative multithreaded architecture, shortly SMA. It combines the speculative execution and multithreading. Every time, a whole thread is predicted and speculatively executed. Multiple threads execute in parallel to share the processor resources. By this way, more ILP can be exploited and the utilization of the function units in the processor increases. High performance can be expected in SMA. This paper presents the philosophy of the SMA architecture, discusses the implementation and key technologies of the SMA processor. These technologies include thread control, distributed register renaming, instruction issue and execution, exception handling and compiler requirement. Experimental results about the performance of the SMA are also presented. The results show that it could reach high ILP performance for different program structures. If optimized compiler and better thread predicator are used, performance can reach higher.