#今日看了什么
#学习
Victima: Drastically Increasing Address Translation Reach by Leveraging Underutilized Cache Resources
https://arxiv.org/pdf/2310.04158
#学习
Victima: Drastically Increasing Address Translation Reach by Leveraging Underutilized Cache Resources
https://arxiv.org/pdf/2310.04158
#学习
《An Alternative TAGE-like Conditional Branch Predictor》
提出一套自适应的 Allocation Throttling 方法,可以消除 Cold entry 问题和 Allocation Rate 不合适的问题
https://dl.acm.org/doi/10.1145/3226098
《An Alternative TAGE-like Conditional Branch Predictor》
提出一套自适应的 Allocation Throttling 方法,可以消除 Cold entry 问题和 Allocation Rate 不合适的问题
https://dl.acm.org/doi/10.1145/3226098
#学习 #BranchPrediction #分支预测 #强化学习
Branch Prediction as a Reinforcement Learning Problem: Why, How and Case Studies
https://arxiv.org/pdf/2106.13429
Branch Prediction as a Reinforcement Learning Problem: Why, How and Case Studies
https://arxiv.org/pdf/2106.13429
#学习 #Prefetcher #数据预取 #强化学习
Pythia: A Customizable Hardware Prefetching Framework Using Online Reinforcement Learning
https://arxiv.org/pdf/2109.12021
Pythia: A Customizable Hardware Prefetching Framework Using Online Reinforcement Learning
https://arxiv.org/pdf/2109.12021
#学习
今日思考了一个问题。
对于 顺序一致性内存模型(Sequential Consistency),简单的load值预测可能会导致违例。
考虑以下事件:
1. Core#1 load [x] -> predict A
2. Core#1 执行load [x] -> value B
3. Core#2 store A -> [x]
4. Core#1 commit load, 重发load -> value A
提出几个问题:
1. load的值预测是否应该认为是正确的?
2. 如果不正确,哪里会出现问题?
3. 如何解决?何种情况才有必要replay/recover?
今日思考了一个问题。
对于 顺序一致性内存模型(Sequential Consistency),简单的load值预测可能会导致违例。
考虑以下事件:
1. Core#1 load [x] -> predict A
2. Core#1 执行load [x] -> value B
3. Core#2 store A -> [x]
4. Core#1 commit load, 重发load -> value A
提出几个问题:
1. load的值预测是否应该认为是正确的?
2. 如果不正确,哪里会出现问题?
3. 如何解决?何种情况才有必要replay/recover?
#学习
#阅读推荐
A Primer on Memory Consistency and Cache Coherence
https://pages.cs.wisc.edu/~markhill/papers/primer2020_2nd_edition.pdf
#阅读推荐
A Primer on Memory Consistency and Cache Coherence
https://pages.cs.wisc.edu/~markhill/papers/primer2020_2nd_edition.pdf
像FDIP喵
Fetch Directed Instruction Prefetching
https://web.eecs.umich.edu/~taustin/papers/MICRO32-fdp.pdf
#学习
大开眼界喵
Draco(MICRO’20) Architectural and Operating System Support for System Call Security 阅读笔记
https://blog.cyyself.name/draco-notes/
大开眼界喵
Draco(MICRO’20) Architectural and Operating System Support for System Call Security 阅读笔记
https://blog.cyyself.name/draco-notes/
#学习
解决解耦前端误预测恢复延迟高的问题
提出了一种由IFU充当紧耦合前端减少分支较为稀疏情况下的误预测惩罚
Elastic Instruction Fetching
https://ieeexplore.ieee.org/document/8675212/
解决解耦前端误预测恢复延迟高的问题
提出了一种由IFU充当紧耦合前端减少分支较为稀疏情况下的误预测惩罚
Elastic Instruction Fetching
https://ieeexplore.ieee.org/document/8675212/
#学习
主要提出了H2P(Hard to Predict)分支的一些特点和为什么SOTA预测器无法有效预测的原因
还有这些分支对宽后端IPC的影响
Branch Prediction Is Not a Solved Problem: Measurements, Opportunities, and Future Directions
https://arxiv.org/abs/1906.08170
主要提出了H2P(Hard to Predict)分支的一些特点和为什么SOTA预测器无法有效预测的原因
还有这些分支对宽后端IPC的影响
Branch Prediction Is Not a Solved Problem: Measurements, Opportunities, and Future Directions
https://arxiv.org/abs/1906.08170