Concurrent Programming in C++

From Gridkaschool

Introduction

This tutorial consists of two sections:

  • C++11 Concurrency
  • Intel Threaded Building Blocks

C++11 Concurrency

C++11 introduced direct support in the language for concurrency, with the ability to launch and manage different threads. Support for mutexes and other synchronisation mechanisms are available.

Threaded Building Blocks

Threaded Building Blocks is a higher level C++ that supports many common design patterns for applications to take advantage of multi-threading capabilities. Basic patterns like parallel loops and reductions are available. We'll also look at the TBB graph constructs.

TBB includes useful containers, which are thread safe, as well as a high performance multi-threaded implementation of malloc.

Talk

The introductory talk to this tutorial is here.

Setup

General

It's important to setup the Redhat devtoolset to have access to a modern version of gcc for all of the tutorial exercises.

To do this login and execute this command:

   $ scl enable devtoolset-2 bash

You can verify that all is well by asking for the gcc version, which should be 4.8.2:

   $ g++ --version
   g++ (GCC) 4.8.2 20140120 (Red Hat 4.8.2-15)


C++11 Compiler Options

To compile concurrent C++11 programs you'll need two flags for g++:

  • -std=c++11 - Use the new C++11 standards
  • -pthread - Enable posix thread support, which is the underlying thread library used by libstdc++ on linux platforms

You might well find this Makefile useful. It will compile any .cc file into an executable with the correct compiler flags.

A good reference for C++ concurrency libraries is http://www.cplusplus.com/ (use the reference section, under multi-threading).

Threaded Building Blocks

Intel Threaded Building Blocks has been compiled and installed into /usr/local (specifically /usr/local/include for header files and /usr/local/lib64 for libraries). Thus you shouldn't need any special extra setup to use is during the school. (Actually TBB is very easy to install from source.)

Compiling against TBB should just work, but linking requires -ltbb and -lrt.

The Makefile here will add all the flags you need, and enable -std=c++11 so it's recommended.

Documentation for TBB is installed into /usr/local/share/tbb/html, but using the Intel documentation is more convenient.

Writing Code

You will have been given access to a machine on which to complete the tutorial. There are a suite of editors available and you can pick whichever you like best:

  • nano - A very simple text editor (but lacks helpful things such as automatic indentation and code colourisation)
  • emacs - The classic Stallman text editor
  • vi - The other classic editor from the editor wars
  • geany - A nice standalone GUI editor with indentation and code highlighting, but you will need to switch on X forwarding to use it (so ssh -Y -p 24 multithr@gks-XXX.scc.kit.edu)

We have also setup working VNC on these machines, if you would prefer to use a VNC client from your laptop.

If you want us to really recommend something then we pick geany ;-)

Exercises

Introductions to each section of the tutorial are here:

https://github.com/graeme-a-stewart/cpp-concurrency/tree/gridka14/doc

and exercises are at the end of each document.

The file

https://github.com/graeme-a-stewart/cpp-concurrency/blob/gridka14/doc/XX-Answers.md

lists the source code that is the solution to each of the exercises. Feel free to look at that if you need a hint or want to check your answer.

You can clone the git repository like this:

git clone https://github.com/graeme-a-stewart/cpp-concurrency.git -b gridka14

N.B. This will clone the GridKA 2014 version of the repository, if you want to clone the master branch then just do

git clone https://github.com/graeme-a-stewart/cpp-concurrency.git

References

https://github.com/graeme-a-stewart/cpp-concurrency/blob/master/doc/00-TipsReferences.md

Is the current 'master' copy of the introduction, which provides references.