没有合适的资源?快使用搜索试试~ 我知道了~
Argobots: A Lightweight Threading Framework for Massive Fine-Gra
需积分: 1 0 下载量 61 浏览量
2023-09-30
17:08:40
上传
评论
收藏 1.63MB PDF 举报
温馨提示
试读
33页
Argobots: A Lightweight Threading Framework for Massive Fine-Gra
资源推荐
资源详情
资源评论
Hardware Trend: More and More Cores
▪ Increase of cores in a processor.
▪ Finer-grained parallelism is important to utilize all cores.
2
1.0E-1
1.0E+0
1.0E+1
1.0E+2
1.0E+3
1.0E+4
1.0E+5
1968 1982 1995 2009 2023
CPU Frequency [MHz]
Year
1
10
100
1968 1982 1995 2009 2023
# of Cores per Processor
Year
Frequency # of cores
[Right] Intel Xeon Phi (Knights Landing) 72 cores, 288 HWTs
(https://software.intel.com/en-us/articles/what-disclosures-has-intel-made-
about-knights-landing)
[Left] ARM ThunderX2 up to 32 cores, 128 HWTs
(https://www.servethehome.com/cavium-thunderx2-review-benchmarks-real-
arm-server-option/)
CPU DB: (http://cpudb.stanford.edu/)
Software Trend: Irregular and Complex Applications
▪ Dynamic and irregular parallel algorithms.
– Adaptive mesh refinement, FMM, graph processing, …
▪ Difficult parallelism control over complicated software stacks.
– Hard to reduce parallelism in an optimal way. Can’t we utilize all?
3
Regular Parallelism Irregular Parallelism
High-Level
Runtime System
Threading System
Scientific Library
Math Library A
User Application
Math Library B
Parallel code
Parallel code
Parallel Code
Parallel Code
nested!
nested!
CPU
Lightweight Threads for Scalability
▪ In-node parallel units are called threads.
– Runnable by multiple cores.
▪ For scalability, threads become finer-grained.
▪ Demand of extremely lightweight threads to utilize both
hardware and software parallelism.
4
Core
Core Core
Core
Irregular algorithm
and complicated input
More cores
CPU
Core
Core Core
Core
Core
Core Core
Core
Core
Core Core
Core
Each block: thread.
Core
Core Core
Core
Core
Core Core
Core
Core
Core Core
Core
Argobots: A Low-Level Lightweight Threading Library
▪ Argobots is an open-source project
– URL: https://argobots.org/
– Maintained by Argonne National Laboratory
• Collaborators: UIUC, Univ. of Tennessee, PNNL,
Intel, UTokyo, Riken, …
• Supported by DOE
▪ Unleashes user-level threads (ULTs)
▪ Provides for the future scalable systems.
– Extremely lightweight “thread” implementation
– Rich and powerful threading capabilities
– Interoperability with other programming models
5
Argobots
CPU
Core Core Core
OS-level thread OS-level thread OS-level thread
Pool Pool Pool
Execution
Stream
Execution
Stream
Execution
Stream
Scheduler Scheduler Scheduler
ULT ULT ULT
ULT ULT
Pool
ULT ULT
剩余32页未读,继续阅读
资源评论
bandaoyu
- 粉丝: 16w+
- 资源: 56
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- 基于matlab实现用有限元法计算电磁场的Matlab工具 .rar
- 基于matlab实现有限元算法 计算电磁场问题 边界条件包括第一类边界和第二类边界.rar
- 基于matlab实现用于计算不同车重下的电动汽车动力性和经济性.rar
- 基于matlab实现遗传算法求解多车场车辆路径问题 有多组算例可以用.rar
- 浏览器.apk
- 基于matlab实现是一个matlab中的power system 中搭建的一个模型
- 基于JSP毕业设计-教学管理系统(源代码+论文).zip
- 基于JSP毕业设计-家政管理系统-毕业设计.zip
- 基于Python实现淘宝商品评论采集(含逆向)源代码
- 基于matlab实现多目标进化算法NSGAⅡ&Matlab讲解.rar
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功