C++11: Multi-core Programming – PPL Parallel Aggregation Explained

In the first part of this series we looked at general multi-threading and multi-core programming concepts without getting into the meat of any real problems, and in the second part we looked at the theory and application of the parallel aggregation pattern using either C++11’s standard thread library or Boost using classical parallel programming techniques. In this part we shall solve exactly the same summation problem from part 2 (adding all the numbers from 0 to 1 billion) but use the new Parallel Patterns Library (PPL) to do so. We will first look at a direct port using classical parallel programming techniques, then find out how to leverage new language features in C++11 which are employed by PPL to simplify the code and write parallel algorithms using an entirely new paradigm. We shall also look at some fundamental differences between PPL and other multi-threading/multi-core libraries, the pros and cons, and some gotchas to be aware of.

Advertisement

C++11 / Boost: Multi-threading – The Parallel Aggregation Pattern

In the first part of this series we looked at general multi-threading and multi-core programming concepts without getting into the meat of any real problems. Tutorials on how to spin up worker threads in C(++) using POSIX/Pthreads, Windows or Boost.Thread are a dime a dozen so I won’t spend too much time on that here; instead I’ll look at a much less-documented and more complicated real-world multi-threading problem, namely that of parallel aggregation, and how to implement it both using new C++11 standard library functions, and with Boost.Thread for those who don’t have access to C++11 at present.

In the example developed in this article, we shall learn how to add all of the numbers from 0 to 1,000,000,000 (1 US billion) exclusive, using multi-threading on a multi-core system to speed up execution and make all the available cores work for us, rather than just one.

(do note that this particular problem can be solved very simply with the sum of a series formula: S = n(a₁ + a_n) / 2 where S is the sum of the series, a_x is the x‘th term in the series and n is the number of terms in the series; we are using the brute-force approach here purely for illustration)

Introduction to Multi-Threaded, Multi-Core and Parallel Programming concepts

In this article I’m going to present a gentle and modernized introduction to multi-threaded and parallel programming. While there are no concrete examples in this overview, I’m going to cover the general concepts and terminology, as well as an overview of the tools available to you as a developer to leverage multi-threaded techniques in our modern era of programming.

What is multi-threaded programming?

Single-threaded programs execute one line of code at a time, then move onto the next line in sequential order (except for branches, function calls etc.). This is generally the default behaviour when you write code. Multi-threaded programs are executing from two or more locations in your program at the same time (or at least, with the illusion of running at the same time). For example, suppose you want to perform a long file download from the internet, but don’t want to keep the user waiting while this happens. Imagine how inconvenient it would be if we couldn’t browse other web pages while waiting for files to download! So, we create a new thread (in the browser program for example) to do the download. In the meantime, the main original thread keeps processing mouse clicks and keyboard input so you can continue using the browser. When the file download completes, the main thread is signaled so it knows about it and can notify the user visually, and the thread performing the download closes down.