This is a copy of the Java documentation for libGDX. It is not finished being ported.
This article describes the little helpers and utilities that might come in handy in case you are running into performance problems and need to start profiling your game.
FPSLogger
The FPSLogger
is a simple helper class to log the frames per seconds achieved. Just invoke the log()
method in your rendering method. The output will be logged once per second.
PerformanceCounter
The PerformanceCounter
keeps track of the time and load (percentage of total time) a specific task takes. Call start()
just before starting the task and stop()
right after. You can do this multiple times if required. Every render or update call tick()
to update the values. The time
FloatCounter provides access to the minimum, maximum, average, total and current time the task takes. Likewise for the load
value, which is the percentage of the total time.
OpenGL
Profiling
Profiling the actual OpenGL calls that happen while your game is running is often not very easy to do, since libGDX tries to abstract all those low-level things away. In order to enable the collection of this information, there is the GLProfiler
.
To enable it you need to instantiate a new GLProfiler object, then call the method glProfiler.enable()
. Behind the scenes this will replace the original GL20
and GL30
instances (Gdx.gl
etc.) with the profilers.
Now those will be active and start to monitor the actual GL calls (and GL errors, see below) for you. One information you might be interested in, could be the amount of texture bindings that happen, which are costly and might slow down your game. To optimize this, you might start to use a TextureAtlas
. To prove with actual numbers that the texture bindings become less, you can call glProfiler.getTextureBindings()
.
You might also implement something like view frustum culling to render only those things that are visible on the screen. The method glProfiler.getDrawCalls()
will return the total number of draw calls made.
The following information is provided by the profiler:
- Amount of total OpenGL calls
- Amount of draw calls
- Amount of texture bindings
- Amount of shader switches
- Amount of used vertices
glProfiler.getVertexCount()
returns a FloatCounter
. From the FloatCounter you can access fields which provide more information (count, total, min, max, average, latest, and value).
In order to reset all these numbers once you have read and displayed them (probably once per frame), you have to call the glProfiler.reset()
method. To completely disable the profiling and replace the profilers with the original GL20
and GL30
instances, use glProfiler.disable()
.
Note that in case you are using Gdx.graphics.getGL20()
or Gdx.graphics.getGL30()
you are bypassing the profiler and that’s why you should use Gdx.gl20
or Gdx.gl30
directly.
To see how to use this you can have a look at the Benchmark3DTest
Error checking (since 1.6.5)
GLProfiler
has one more useful feature and that is error checking.
Almost all GL calls can in some circumstances produce errors. These errors are not thrown or logged like Java errors, but they have to be explicitly checked for (Gdx.gl.getError()
), so they can be hard to find. Enabling GLProfiler
(see above on how to enable) will automatically check for GL errors after every GL call and report it, so you don’t have to.
By default, encountered errors will be printed to the console (using Gdx.app.error
). However, this can be customized (for example for your own logging/crash reporting system) by setting up a different error listener with glProfiler.setListener()
.
If you want to know where exactly the error happened in your code, you may want to use the GLErrorListener.THROWING_LISTENER
which throws an exception on any GL error. Error listener callback is called inside GL call, so the stack trace will reveal where exactly things went wrong. (Throwing an exception on GL error will most likely crash your application in case of errors, so it is not used by default.)
For example use and testing, there is a GLProfilerErrorTest
apitrace
apitrace is an open source cross platform debugger and profiler for OpenGL. You run the tracer to record the state and each call (including contents of buffers). You then run the viewing tool will read the trace and playback any state at any point in time. It breaks down the cpu/gpu time, every single OpenGL call. You can see the contents of the framebuffer and each texture bound at any point you choose.
Running: For linux, do: apitrace trace java -cp /home/me/my-app/desktop/build/libs/*.jar -Dorg.lwjgl.opengl.libname=/usr/lib/apitrace/wrappers/glxtrace.so com.my.app.desktop.DesktopLauncher
Then just exit your app (if you want), run qapitrace
and open that trace file.