1. 程式人生 > 實用技巧 >Exploiting Choice: Instruction Fetch and Issue on an Implementable Simultaneous Multithreading Processor(2)

Exploiting Choice: Instruction Fetch and Issue on an Implementable Simultaneous Multithreading Processor(2)

Time

2020.10.31

Summary

Research Objective

Problem Statement

Method(s)

The methodology in this paper closely follows the simulation and measurement methodology of[27]

Evaluation

Conclusion

Notes

RR.4.2 has four output buses, each two instructions wide.
RR.4.2有四個輸出匯流排,每一個是兩指令寬度。

The efficiency of the entire processor is affected by the quality of instructions fetched.

https://blog.csdn.net/shuiliusheng/article/details/84255514

Words

interleaving 交織
decoupled 解耦的
branch target buffer (BTB) 分支目標快取???
pattern history table (PHT) 圖形歷史表???
implementation 實現
emulation 模擬
Multiflow 多流程
scheduling 排程
differentiate 鑑別
reap benefits 收穫利益
appear to 似乎
culprit 罪魁禍首
simultaneous multi-threading(SMT) 同時多執行緒
定義:在超標量結構的基礎上增加執行緒上下文控制,允許在同一個時鐘週期內發射來自不同執行緒的多條指令,以充分利用功能部件的一種處理器結構。
exploit 利用
misalignment 錯位
fragmentation 碎片
fetch 讀取
fetch block fragmentation(our term for the various factors that prevent us from fetching the
maximum number of instructions)
term 術語
indistinguishable 難以區分
finely 精細地
multiplexer 多路選擇器
multiplexed 多路複用
circuitry 電路
replicated 複製的
additions 附加物
MUX 複用器
bank conflict 儲存體衝突
當被訪問的儲存體沒有恢復時又出現對該儲存體新訪問的現象
negligible 微不足道
extensive 廣泛的
rename register 換名暫存器
暫存器重新命名,由於x86指令的限制,可以使用的暫存器就那麼幾個,但現在CPU的內部實際的暫存器數量是高於x86指令中的暫存器數量的,所以就要在執行指令的時候把X86暫存器重新對映到內部實際暫存器
,暫存器重新命名可以提高指令執行的並行度。
dependency chains 依賴鏈
assume 假設
desirable 可取的
attack 破壞
clog 阻塞
outstanding 優秀的
heuristics 啟發式
saturated 飽和的
Cache miss 快取記憶體缺失
gain 實現
metrics 指標
Intelligent fetching 智慧獲取
weighted 加權的
blockked 被阻塞
lookups 查詢

Sentence

fetch efficiency, by partitioning the fetch unit among threads
通過線上程之間劃分獲取單元來獲取效率
fetch effectiveness, by improving the quality of the instructions fetched
通過提高所提取指令的質量來獲取有效性
four each from two threads
兩個執行緒各四個
reduced by a factor of 2 and 4
減少了2到4倍
splitting the fetch over multiple threads.
將獲取指令分為多個執行緒
we stand to lose
我們承受的損失
we can fetch around cache misses:
我們可以獲取快取未命中

TimtLine

Using a combination of partitioning the fetch unit, intelligent
fetching, and early I cache tag Iookups