推 mqazz1:謝啦! 不然這兩篇可能都石沉大海了XD 08/22 20:52
→ blackZ2:是因為我也看不懂,然後google到這一篇,但是沒有答案Q.Q 08/23 00:46
→ blackZ2:所以與同學討論出來的... 08/23 00:46
※ 引述《mqazz1 (無法顯示)》之銘言：
assume that the breakdown of dynamic instructions into various instruction
categories is as follows:
R-type 50%, beq 15%, jump 10%, lw 15%, sw 10%
branch predictor accuracies: always-taken 40%, always not-taken 60%, 2-bit 80%
(1) with the 2-bit prdictor, what speed-up would be achieved if we could
convert half of the branch instructions in a way that replaces a branch
instruction with an ALU instruction?
assume that correctly and incorrectly predicted instructions have the same
chance of being replaced
1 + 3*(1-0.8)*0.15
--------------------------------
1 + 3*(1-0.8)*0.15*0.5
請問這題為什麼要加1?
因為理想CPI為1，然後加上branches penalty 就是實際CPI
(2)同上，把黃字改成two ALU
1 + 3*(1-0.8))*0.15
--------------------------------
1 + (1 + 3*(1-0.8))*0.15*0.5
請問為什麼還要再加1?
因為這各式子化簡了，所以看起來很怪
請想成 1 + 0.15*0.5*0.2*3 + 0.15*0.5*1
理想CPI 沒有轉換的一半branch 一半branch轉成兩個ALU,指令會變多
化簡之後就是 1+(1+0.2*3)*0.15*0.5
這是我自己的想法，希望是對的
--
其實我是來賺P幣的
--
※ 發信站: 批踢踢實業坊(ptt.cc)
◆ From: 118.168.81.69