作者aideliemoon (Cerscent)
看板Grad-ProbAsk
标题[理工][OS & 计组]98清大 资工所
时间Sun Jan 10 12:05:14 2010
5.What are the drawbacks of client-server architecture?
请问一下client-server architecture是什麽东西?
查课本也找不太到
9.consider a 5-stage pipelined datapath where ALU computation is performed
at the third stage and data memory read is performed at the fourth stage.
We have a program of 1000 instruction is the format of "load,add,load,add,..".
Each add instruction is form "add $a,$b,$c"which performs $a = $b + $c.
Each load instruction is form "load $t,k($s)"which performs $t = memory[$s+k]
Suppose each add instruction depends(and only depends on)the load instruction
right before it.Each load instruction also depends(and only depends ) on the
add instruction right before it. With forwarding,what would be actual CPI?
这题他没说add和load须要几个cycle 所以我就当一个stage一个cycle
并且add在load的左侧load不须stall cycle 但load在add左测加上forwarding技术add需
sall 1 cycle
因此add要6个cycle load要5个 所以CPI = 500*(5+6)/1000 = 5.5
不知道这样算对吗?
10. Explain the ideas behind the following instruction-set-architecture design
chaos?
(a)Keep all instruction a single size.
(b)Have 32 reigisters rather than many more.
a.我想是 simplicity favors regularity
b. smaller is faster
不知道这样对吗
请会的版友帮我解一下疑惑 先在这说声谢谢了
--
※ 发信站: 批踢踢实业坊(ptt.cc)
◆ From: 114.43.160.197
1F：推 uminchu185:1.PTT应该就是一个例子. 2.要考虑它是pipeline, 指令 01/10 13:28
2F：→ uminchu185:可以重叠, 约3个cycle完成2个指令,CPI=1.5. 01/10 13:29
3F：→ uminchu185:3. 我的想法跟你是一样的. 有错请更正~ 01/10 13:30
4F：→ aideliemoon:对後，pipeline不考虑stall的话是1个cycle完成一道嘛 01/10 14:19
5F：→ aideliemoon:我都忘了，谢谢<(_ _)> 01/10 14:20
※ 编辑: aideliemoon 来自: 114.43.160.197 (01/10 14:26)
6F：→ assassin88:3.正确 01/10 22:32