论文部分内容阅读
六十年代以来,由于迫切需要速度更快、成本较低的系统,因而流水线的计算机结构得到了很大重视。流水技术的优点在于:它能使各子系统的速度相匹配,而又不会使整个系统的成本成倍增加。随着技术的发展,不断提供更快速、更便宜的大规模集成电路,这样采用不论是简单形式或复杂形式的流水线结构就更有希望了。本文考察了有关流水技术的许多理论上的考虑,评价和比较了各种以时序方式或向量方式进行操作的有代表性的流水线机器、已采用的解决有关问题的实际方法以及对各种因素的折中考虑。分析了简单流水线的性能、实际的速度限制和花费的控制设备代价。解决了系统设计者面临的各种问题,包括缓冲、总线结构、转移和中断处理。研究了时序处理和向量处理的概念、向量处理的基本优点、它所需要的附加成本以及如何在时序和向量流水线间进行权衡。最后介绍了二台最新机器(CRAY-1和Amdahl 470V/6),用以分别说明可采用的流水线技术的复杂性以及可利用的流水线概念的简单而先进的程度。
Since the 1960s, due to the urgent need for faster, lower cost systems, the pipeline computer architecture has received a great deal of attention. The advantage of streaming technology is that it matches the speeds of the subsystems without doubling the cost of the overall system. As technology evolves, ever-faster, cheaper, larger scale integrated circuits continue to be offered, and even more hope is the adoption of pipelined architectures, whether simple or complex. This paper examines many theoretical considerations regarding pipelining techniques, assesses and compares various representative pipelining machines that operate in a time-series or vector-based fashion, practical approaches that have been adopted to address these issues, Compromise The performance of simple pipelines, the actual speed limits, and the cost of control equipment were analyzed. Solve various problems faced by system designers, including buffering, bus structure, transfer and interrupt handling. The concept of sequential processing and vector processing is studied, the basic advantages of vector processing, the additional costs it requires, and how to trade off between timing and vector pipelining. Finally, two state-of-the-art machines (CRAY-1 and Amdahl 470V / 6) are introduced to illustrate the sophistication of the pipeline techniques available and the simplicity and sophistication of the available pipeline concepts.