Time
2020.11.1
Summary
Research Objective
Problem Statement
Method(s)
Evaluation
Conclusion
This paper presents a simultaneous multithreading architecture that：
● borrows heavily from conventional superscalar design, requiring little additional hardware support,
● minimizes the impact on single-thread performance, running
only 2％ slower in that scenario, and
● achieves significant throughput improvements over the superscalar when many threads are running: a 2.5 throughput gain at 8 threads, achieving 5.4 IPC
Notes
BRANCHYIRST：issues branches as early as possible in order to identify tnispredicted branches quickly.
Perfect branch prediction boosts throughput by 25% at 1 thread,15% at 4 threads, and 9% at 8 threads.
Words
speculative 投机的
side effect 副作用
proposing 提出
alignment 校准
dual 双重
interfere 干扰
saturate 饱和
excess 过量
access 存取
the dispatch stack 调度堆栈
Multiscalar 多标量
Sentence
Much as the fetch unit can choose which threads to fetch
就像存取单元可以选择要存取的线程一样
the amount of improvement
改善程度
benefits much less from
从...获得的收益少得多