• Introduction • Memory • Host vs Device Functions • Return on Investment • Concluding remarks Outline 2 |• I work the RiskLab team at CSIRO on applied mathematics for Financial Risk. • The aim of from introductory CUDA examples to an existing project that has a meaningful amount of non-trivial code. • Provide some guidance to people about to embark on using CUDA to speed up existing software separated by the PCI-Express bus.” -- https://developer.nvidia.com/blog/unified-memory-in-cuda- 6/ CPU vs GPU Memory System Memory GPU Memory 12 |“Unified Memory creates a pool of managed memory that is

experience for Visual Studio Code Our agenda today Visual Studio Code Session coming up! #1 most used code editor [StackOverflow Developer Surveys] Free, open-source code editor Runs on Windows, macOS and Linux C++ IntelliSense, debug & code browsing CMake, vcpkg and git integrations SSH/container/WSL remote experiences and much much more… https://aka.ms/cpp/code Thu 10/28 – 2pm An Editor Can Do That That? Debugging Assembly Language and GPU Kernels in Visual Studio Code Julia Reid – _3 Visual Studio CppCon 2020 Visual Studio 2019 Preview 3 Preview 4 Preview 5 Preview 6 Version 16

monetization • App store discovery • … Hybrid "Promise" 100% Web 100% Nativ e Hybri d Reach Code/Skill Reuse Richness Premium experience Device APIs Best of both? Compromises. Mobile Web - PWA Native UI driven by JavaScript Native UI JavaScript-to-Native bridge JavaScript Engine (Your app code runs here) Native APIs NativeScri pt (by Progress) React Native (by Facebook) JavaScript-Driven JavaScript-to-Native runtime JavaScript Engine (Your app code runs here) Native APIs Native App Native UI JavaScript Engine (Your app code runs here) Native APIs API Wrapper API Wrapper Plugin

C++20 features • Majority from our amazing contributors • Extensively reviewed and tested • Released VS 2019 16.5, 16.6, 16.7, 16.8 Preview 3 • GitHub migration ongoing • Build system (native desktop) vector13 constexpr In Everyday Programming • constexpr has steadily increased in power • It's just code that runs on your machine instead of the user's • Example: lookup tables • Everyone uses lookup print(v); // dog cat ox dog erase(v, "dog"); print(v); // cat ox }18 atomic_ref19 atomic vs. atomic_ref • atomic: atomic access and storage for T • atomic_ref: atomic access for separate

use a pure eBPF service? • Not mature enough eBPF brief • Write C • Compile into eBPF assembly code • Inject to kernel • Attach to network tc hooks • Triggered by ingress/egress packets IPVS bypass map id is passed to IPVS module • Ip_vs_new_conn() inserts eBPF map • Key: (protocol, cip:cport , rsip:rsport) • Value: (protocol, lip:lport, rsip:rsport) • Ip_vs_conn_unlink() deletes entries in eBPF https://careers.tencent.com/home.html Bugs solved – 1/2 • IPVS conn_reuse_mode=1 low cps Ip_vs_conn nf_conn New ip_vs_conn Bugs solved – 2/2 • DNS resolution delays for 5s Iptables SNAT Conntrack insert

compare_exchange(old_ccb , new_ccb)) if (old_ccb.ctrl != prev_ccb.ctrl) { ??? } } Disclaimer: Slide code omits some nullptr checks to save space. atomic> What happens when a concurrent Protect then increment Q1: Does this prevent the use- after-free on the control block? Q2: Is this code correct? This might increment the reference count from zero!!! Correct: Protect then increment Benchmarking multithreaded/lock-free code is hard… • There are many factors to consider: • Measurement: Throughput vs latency? • Workload: Proportion of reads vs writes • Hotness: Does the data fit

Web ODA vs Web2 app Web2 App Open Data App Centralization Centralized Decentralized Performance High High, less than Web2 Query Capability Powerful Powerful, less than Web2 Open Source Code Open or No Yes ODA vs Web3 DApp Web3 DApp Open Data App Centralization Decentralized Decentralized Underlying Nodes Many Few Performance Low High Query Capability Low High Open Source Code Open Open Hot

MySQL MySQL 优化方式 优化方式  MySQL MySQL 技巧分享 技巧分享  Q Q & & AA MyISAM MyISAM 特点 特点 MyISAM vs MyISAM vs InnoDB InnoDB • 数据存储方式简单，使用 数据存储方式简单，使用 B+ Tree B+ Tree 进行索引 进行索引 • 使用三个文件定义一个表： 使用三个文件定义一个表： 开发的把 开发的把 Slow Log Slow Log 输出到屏幕，功能简单 输出到屏幕，功能简单  mysql-log-filter - Google code - Google code 上一个开源产品，报表 上一个开源产品，报表 简洁 简洁

cpp 源文件用于测试 CMake 中添加一个可执行文件作为构建目标 另一种方式：先创建目标，稍后再添加源文件 如果有多个源文件呢？ 逐个添加即可 使用变量来存储 建议把头文件也加上，这样在 VS 里可以出现在“ Header Files” 一栏 使用 GLOB 自动查找当前目录下指定扩展名的文件，实现批量添加源文件 启用 CONFIGURE_DEPENDS 选项，当添加新文件时，自动更新变量 1749 • CMAKE_BUILD_TOOL: 执行构建过程的工具。该变量设置为 CMake 构建时输出所需的程序。对于 VS 6 ， CMAKE_BUILD_TOOL 设置为 msdev ， 对于 Unix ，它被设置为 make 或 gmake 。 对于 VS 7 ， 它被设置为 devenv. 对于 Nmake 构建文件，它的值为 nmake 。 • CMAKE_DL_LIBS: 解决：让静态库编译时也生成位置无关的代码 (PIC) ，这样才能装在动态库 里 也可以只针对一个库，只对他启用位置无关的代码 (PIC) 第 4 章：对象的属性 除了 POSITION_INDEPENDENT_CODE 还有哪些这样的属性？ 另一种方式： set_target_properties 批量设置多个属性 另一种方式：通过全局的变量，让之后创建的所有对象都享有同样的属性 相当于改变了各个属性的初始默认值。

DevOps 方案对比 来自客户的评价： 2 Zadig 产品特性 Zadig 核心特性 面向开发者的云原生环境 灵活易用的高并发工作流 高效协同的测试管理 云原生 IDE 插件（ VS CODE) 客观精准的效能洞察 强大免运维的模版库 • 自动生成面向开发、测试、运维角 色的工作流 • 多个微服务并行构建、部署、测 试，代码验证效率 UP • 自定义工作流，灵活编排发布、自 高开发效率 可重用降低运维成本 • 一些组织可能过度依赖高级工程 师管理发布流程和基础设施，导 致资源浪费和效率低下 Zadig 平台工程模式 工程规模数据： • 1500+ 产研工程师 Vs. 2 人运维 • 50 + 个 Kubernetes 全球集群 • 300+ 个数字产品（资产沉淀） • 1600+ 条自动化工作流（全域） • 900+ 个服务化运行环境 通过整