Multi-Threaded Programming 1: Basics

Original Author: Joseph Simons

This will be a series of posts in which I will cover a good portion about what I have learned in terms of multi-threaded development in the realm of video games. I am writing this since I have been inspired by informative series of posts here such as Alex Darby’s Low Level C/C++ Curriculum. So my goal is to explain how multi-threading works at a low and high level, as well as looking at a few common scenarios that can benefit from these techniques. I might even touch on a few situations you can get yourself into and some techniques for debugging them. I hope you will enjoy reading about it as much as I enjoy discussing it.

 

I am not the first person to make this joke…

 

What is Multi-Threading?

Lets start at the very beginning and cover exactly what a thread is and why having more of them (up to a certain point) is a good thing.

Every program that runs on a modern computer (such as your PC, game console, or smartphone) is a threads associated with them. These threads are what execute the actual code for the running program. Threads run code independently but share memory. This allows them to operate easily on the same data but perform different calculations. The fact that they share memory is the powerful, double-edged sword of their functionality.

A modern CPU is typically composed of several cores. Each core can run one thread at a time (though context switch, and while it isn’t exorbitantly expensive, you still want to avoid it as much as you can to achieve optimal speed.

The main parts of a thread are the local data storage. The program counter keeps track of what part of the code it is currently executing. The registers keep track of the current values of the executing code. The stack holds any values that don’t fit in the registers. And the local data is program specific, but generally keeps track of data that is local to that thread. This local storage is accomplished via a lookup table, managed by the OS, that will have each thread look up into its specified index to see where its data is kept. In short, this allows you to have variables that you can access like a global, but are in fact unique to each thread.

So, taking into account all of this, multi-threaded programming is when you use multiple threads in your program. Typically, you are doing this so that you can execute your program faster than you would using a single thread. Another common usage is to separate out your UI from the rest of the program, so that it always feels responsive. The difficulty in doing so lies in designing the systems in such a way that they can make use of multiple threads in these manners.

 

Show me your cores!

Why do we need it?

In short, we need this in order to make the most use of the available resources. But the real question is more likely, “Why do we need it, now?” and this is due to the trends in computing.

Since the dawn of the CPU, as years have gone by hardware manufacturers have been able to increase the speed of computers by increasing their 5GHz for consumer use, that left ten years where chip designers had to think laterally. And this means that one of the key innovations was putting more CPU cores onto a single chip. So no longer could you expect your games to be faster just by putting in new hardware, you had to design the game so that it could work well using multiple threads running on multiple cores.

One of the first areas where multi-threaded computing required games to adapt was with the release of the Microsoft Xbox 360 and Playstation 3, and for two similar but different reasons. The 360 had three cores with Hyper Threading technology (HTT), meaning that it could conceivably run six threads in parallel. The PS3 had a single core with HTT, and seven Synergistic Processing Elements (SPE), in a new architecture called the Cell Broadband Engine. The 360 had unified memory, so that all the threads could access all the memory. This made it easy to implement multi-threaded applications since you didn’t have to worry about who could access what, unlike with the PS3. Each SPE there had only a limited amount of memory that it could make use of, meaning that as a developer you had to carefully design your algorithms to take this into account.

Also, from a pure performance standpoint, the CPUs in the consoles were slower than their traditional desktop counterparts. While they were clocked similarly in speed, they were simpler chips that didn’t have a lot of the fancier technologies such as Out-of-order execution that made the most of the available power. So this forced developers to cope with multi-threaded strategies in order to make their games stand out amongst the rest.

In the desktop world, multi-core CPUs were quickly becoming the standard, but it appeared that developers were slow to adopt their programs to make use of the additional cores. A lot of this was likely due to two main factors. First is that the desktop world has a much broader range of hardware to support, and so they tend to design with the lowest common denominator in mind. Meaning that they wanted to support older systems that likely only had a single core available. Second is that multi-threaded programming is more difficult, and it takes a different approach to handling it. Couple this with the fact that a lot of game teams tend to reuse technologies, so in order to take advantage of multiple cores a lot of things would have to be rewritten. Also, graphics tended to be a big bottleneck in games, and submitting all the stuff for the GPU to draw was restricted to being done in a serial manner by the APIs (DirectX or OpenGL, typically). Only with the latest versions of the APIs (released only a few years ago) has it really been possible to make the most use of all the available cores in modern CPUs. And now with the latest generation of consoles upon us, developers have no excuse to not use heavily multi-threaded game engines for the top tier games.

 

Next Time…

For the next post in this series, I will cover how communicating between threads works on a hardware and software level. This is important to understand what is exactly going on with the CPU in order to avoid many of the potential pitfalls that are inherent in multi-threaded programming.

If you have any interest in other areas dealing with multi-threaded programming, please let me know in the comments and I will see if I can make a future post answering any questions or topics you would like covered. Thanks for reading!

Next Post – Multi-Threaded Programming 2: Communication