• 转载自：http://igoro.com/archive/gallery-of-processor-cache-effects/
• 原作者：Igor Ostrovsky

这是少有的将缓存讲解地如此透彻的文章，全文转载如下

Most of my readers will understand that cache is a fast but small type of memory that stores recently accessed memory locations. This description is reasonably accurate, but the “boring” details of how processor caches work can help a lot when trying to understand program performance.

In this blog post, I will use code samples to illustrate various aspects of how caches work, and what is the impact on the performance of real-world programs.

The examples are in C#, but the language choice has little impact on the performance scores and the conclusions they lead to.

Example 1: Memory accesses and performance

How much faster do you expect Loop 2 to run, compared Loop 1?

几十年来，传统的储存模型几乎没有什么变化。如下图所示，操作系统负责与储存介质交互，对用户程序提供open/close，read/write等基本API。

除了这样直接访问，Windows和Linux均支持内存映射，这在旧的编程模型中可能并不常用，但它却是NVM编程模型的核心。

环境准备

• OS: Windows 10
• JDK: 1.8.0_191
• soot: 3.0.1

步骤

下载Soot

Soot的版本要刚刚好（3.0.1）。如果版本过低不支持JDK1.8，过高会提示某些库找不到。

下载jar包最好下载***-jar-with-dependencies.jar，参考下载地址：链接。

手动编译java源文件

Soot既可以读取*.java，也可以读取编译好的*.class。我选择先手动将java源文件编译成class文件，因为这样可以准确地暴露出代码中的错误。手动编译使用命令：

在Apollo 3.5中，百度自研了运行时计算框架Cyber RT来代替ROS，官方FAQ中提到其性能、延迟以及吞吐量都要优于ROS。这篇博客记录了我依据Issue #7220来验证性能是否提高的过程。

我们测试的是发送和接收消息的延时。在Cyber RT和ROS中分别有自带的demo来完成这一功能，我们只需将这两个demo中发送的数据统一一下，并在接受时输出latency即可。

测试Cyber RT

Clone apollo最新代码

1

git clone https://github.com/ApolloAuto/apollo.git -b master

运行docker环境

1
2

./docker/scripts/dev_start.sh
./docker/scripts/dev_into.sh

apollo提供了docker环境，已经预装了各种依赖的环境，官方也建议在docker中运行。

build项目

1

bash apollo.sh build

build后的可执行文件在bazel_bin/目录下。

我们要测试的两个可执行文件即为talker和listener，它们在bazel_bin/cyber/examples/目录下，此时可以运行这两个文件测试功能是否正常。

一些快捷键

移动光标

修改文本

剪切和粘贴

@在python中是函数的修饰符，为函数提供了包装的功能，为函数提供了更细粒度的扩展控制。详见官方文档。

本文仅从使用的角度来探究一下@的特性。

例1

1
2
3
4
5
6
7
8
9

def wrapper(fn):
    print("This is a wrapper!")

@wrapper
def func():
    print("func")
    
if __name__ == "__main__":
    func()

输出为：

1
2
3
4
5

This is a wrapper!
Traceback (most recent call last):
  File "***.py", line 9, in <module>
    func()
TypeError: 'NoneType' object is not callable

在执行line 9前这里并没有任何函数调用，但是wrapper已经被执行了。这是因为在解析到@wrapper的时候，等效的代码为：

Bazel is an open-source build and test tool similar to Make, Maven, and Gradle. It uses a human-readable, high-level build language. Bazel supports projects in multiple languages and builds outputs for multiple platforms. Bazel supports large codebases across multiple repositories, and large numbers of users.

Installing on Ubuntu

It’s easy and recommended to install bazel via the binary installer, which can be downloaded from Bazel’s Github releases page

Run the binary directly, and it should be done gracefully. Note that some libraries must be installed for Bazel to work. Run command:

1

sudo apt install pkg-config zip g++ zlib1g-dev unzip python

Official examples

Bazel can be used to build various projects, while we just focus on C++ in this blog. The codes can be found here.

The directory structure looks like this:

1
2
3
4
5
6
7
8

root-dir
|-------main
        |-------BUILD
        |-------hello-world.cc
        |-------[OTHER FILES]
        |
|-------WORKSPACE
|

The WORKSPACE file is necessary for Bazel to work, although it can be remained empty for now.