OPTIMAL PIPELINING

被引:13
作者
DUBEY, PK [1 ]
FLYNN, MJ [1 ]
机构
[1] STANFORD UNIV,STANFORD,CA 94305
基金
美国国家航空航天局;
关键词
D O I
10.1016/0743-7315(90)90064-V
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
Pipelining is one of the most attractive and widely used design alternatives in high-speed computer systems as it offers a potential speedup of N when N pipeline stages are used. This study is an attempt to understand the trade-offs and overhead that limit this theoretical speedup. A mathematical model has been developed to provide insight into the effective roles played by different parameters involved. Optimization of a pipeline refers to partitioning the pipeline into an optimum number of segments such that overall throughput is maximized. The first section of this paper summarizes previous related work in this area. The next section provides the details of the model. The following section derives certain inferences from the model and illustrates its close agreement with corresponding results obtained via CRAY-IS simulations by Kunkel and Smith [Proc. 13th Annual Symposium on Computer Architecture, 1986, pp. 404-411]. The last section discusses some potential improvements in the model. © 1990.
引用
收藏
页码:10 / 19
页数:10
相关论文
共 6 条
[1]  
COTTON LW, 1965, FAL P JOINT COMP C, P489
[2]  
FAWCETT BK, 1975, THESIS U ILLINOIS UR
[3]   PIPELINING OF ARITHMETIC FUNCTIONS [J].
HALLIN, TG ;
FLYNN, MJ .
IEEE TRANSACTIONS ON COMPUTERS, 1972, C 21 (08) :880-&
[4]  
Kunkel S. R., 1986, 13th Annual International Symposium on Computer Architecture (Cat. No.86CH2291-3), P404
[5]  
SHAPIRO HD, 1977, 1977 INT C PAR PROC, P67
[6]   DETECTION AND PARALLEL EXECUTION OF INDEPENDENT INSTRUCTIONS [J].
TJADEN, GS ;
FLYNN, MJ .
IEEE TRANSACTIONS ON COMPUTERS, 1970, C 19 (10) :889-&