The Game Loop
The game loop is the overall flow control for the entire game program. It’s a loop because the game keeps doing a series of actions over and over again until the user quits. Each iteration of the game loop is known as a frame. Most real-time games update several times per second: 30 and 60 are the two most common intervals. If a game runs at 60 FPS (frames per second), this means that the game loop completes 60 iterations every second.
There can be many variations of the game loop, depending on a number of factors, most notably the target hardware. Let’s first discuss the traditional game loop before exploring a more advanced formulation that’s designed for more modern hardware.
Traditional Game Loop
A traditional game loop is broken up into three distinct phases: processing inputs, updating the game world, and generating outputs. At a high level, a basic game loop might look like this:
whilegame is running process inputs update game world generate outputs
Each of these three phases has more depth than might be apparent at first glance. For instance, processing inputs clearly involves detecting any inputs from devices such as a keyboard, mouse, or controller. But those aren’t the only inputs to be considered; any external input must be processed during this phase of the game loop.
As one example, consider a game that supports online multiplayer. An important input for such a game is any data received over the Internet, because the state of the game world will directly be affected by this information. Or take the case of a sports game that supports instant replay. When a previous play is being viewed in replay mode, one of the inputs is the saved replay information. In certain types of mobile games, another input might be what’s visible by the camera, or perhaps GPS information. So there are quite a few potential input options, depending on the particular game and hardware it’s running on.
Updating the game world involves going through everything that is active in the game and updating it as appropriate. This could be hundreds or even thousands of objects. Later in this chapter, we will cover exactly how we might represent said game objects.
As for generating outputs, the most computationally expensive output is typically the graphics, which may be 2D or 3D. But there are other outputs as well—for example, audio, including sound effects, music, and dialogue, is just as important as visual outputs. Furthermore, most console games have “rumble” effects, where the controller begins to shake when something exciting happens in the game. The technical term for this is force feedback, and it, too, is another output that must be generated. And, of course, for an online multiplayer game, an additional output would be data sent to the other players over the Internet.
We’ll fill in these main parts of the game loop further as this chapter continues. But first, let’s look at how this style of game loop applies to the classic Namco arcade game Pac-Man.
The primary input device in the arcade version of Pac-Man is a quad-directional joystick, which enables the player to control Pac-Man’s movement. However, there are other inputs to consider: the coin slot that accepts quarters and the Start button. When a Pac-Man arcade cabinet is not being played, it simply loops in a demo mode that tries to attract potential players. Once a quarter is inserted into the machine, it then asks the user to press Start to commence the actual game.
When in a maze level, there are only a handful of objects to update in Pac-Man—the main character and the four ghosts. Pac-Man’s position gets updated based on the processed joystick input. The game then needs to check if Pac-Man has run into any ghosts, which could either kill him or the ghosts, depending on whether or not Pac-Man has eaten a power pellet. The other thing Pac-Man can do is eat any pellets or fruits he moves over, so the update portion of the loop also needs to check for this. Because the ghosts are fully AI controlled, they also must update their logic.
Finally, in classic Pac-Man the only outputs are the audio and video. There isn’t any force feedback, networking, or anything else necessary to output. A high-level version of the Pac-Man game loop during the gameplay state would look something like what is shown in Listing 1.1.
Listing 1.1 Theoretical Pac-Man Game Loop
whileplayer.lives > 0
// Process Inputs
JoystickDataj = grab raw data from joystick
// Update Game Worldupdate player.position based on j
Ghostg in world
ifplayer collides with g kill either player or g
elseupdate AI for g based on player.position
// Pac-Man eats any pellets...
// Generate Outputsdraw graphics update audio
Note that the actual code for Pac-Man does have several different states, including the aforementioned attract mode, so these states would have to be accounted for in the full game’s code. However, for simplicity the preceding pseudo-code gives a representation of what the main game loop might look like if there were only one state.
Multithreaded Game Loops
Although many mobile and independent titles still use a variant of the traditional game loop, most AAA console and PC titles do not. That’s because newer hardware features CPUs that have multiple cores. This means the CPU is physically capable of running multiple lines of execution, or threads, at the same time.
All of the major consoles, most new PCs, and even some mobile devices now feature multicore CPUs. In order to achieve maximum performance on such systems, the game loop should be designed to harness all available cores. Several different methods take advantage of multiple cores, but most are well beyond the scope of this book. However, multithreaded programming is something that has become prevalent in video games, so it bears mentioning at least one basic multithreaded game loop technique.
Rendering graphics is an extremely time-consuming operation for AAA games. There are numerous steps in the rendering pipeline, and the amount of data that needs to be processed is rather massive; some console games now render well over a million polygons per frame. Suppose it takes 30 milliseconds to render the entire scene for a particular game. It also takes an additional 20 milliseconds to perform the game world update. If this is all running on a single thread, it will take a total of 50 milliseconds to complete a frame, which will result in an unacceptably low 20 FPS. But if the rendering and world update could be completed in parallel, it would only take 30 milliseconds to complete a frame, which means that a 30 FPS target can be achieved.
In order for this to work, the main game loop thread must be changed so it processes all inputs, updates the game world, and outputs anything other than the graphics. It then must hand off any relevant data to a secondary rendering thread, which can then draw all the graphics.
But there is a catch to this: What should the main thread do while the rendering thread is drawing? We don’t want it to simply wait for the drawing to finish, because that would be no faster than performing all the actions on one thread. The way this problem is solved is by having the rendering thread always lag one frame behind the main thread. So every frame, the main thread is updating the world while the rendering thread draws the results of the last main thread update.
One big drawback of this delay is an increase of input lag, or how long it takes for a player’s action to be visible onscreen. Suppose the “jump” button is pressed during frame 2. In the multithreaded loop, the input will not get processed until the beginning of frame 3, and the graphics will not be visible until the end of frame 4. This is illustrated in Figure 1.1.
Figure 1.1 The jump is delayed a couple of frames due to input lag.
If a particular game relies on extremely quick response time, including fighting games such as Street Fighter, this increased input lag may be deemed unacceptable. But for most other genres, the increased lag may not be particularly noticeable. Several other factors increase input lag. The game loop can be one of these factors, and some, like the display lag most LCD panels have, might be out of the programmer’s control. For more information on this topic, check the references at the end of this chapter, which includes an interesting set of articles on the topic of measuring and solving input lag in games.