qmlbench is a tool for benchmarking Qt, QML and QtQuick as a whole stack rather than in isolation. The benchmarks it provides exercise a very large part of Quick, QML, Gui, Core, and as a result can be considered a decent metric for overall Qt performance.
As with any benchmarking, if you want reliable results, the first thing you must attend to is ensuring that the system under test is doing the absolute minimum possible. "Typical" operating systems (OS X, Ubuntu, ...) all have a lot of "background noise" in the form of file indexing, automatic updates. All of this should be disabled if possible. You may want to consider disabling networking for the duration of the benchmarking run to help keep things "quiet".
In terms of running applications, you should have the bare minimum if at all possible. No email clients, no IRC, no media players. Only your terminal, and qmlbench should be running. If possible (e.g. on Linux), you may want to even look into setting up a custom desktop environment session that doesn't launch anything else (so no gnome-session etc).
Remember: Any background process that might pop up will cause instability in the results!
qmlbench provides a number of "shells", which are the containers for the benchmarking logic. Different shells work in different ways, and might measure different things.
The frame-count shell constantly performs a set number of operations each frame, for a given amount of (real world) time, and counts the number of frames the window is able to generate. If the GL drivers & OS support it, vsync will be disabled.
The frame-count shell is used for automated benchmark runs, as it is less susceptible to variance than the sustained-fps shell.
The sustained-fps shell provides a measure of how many times a certain operation can be performed per frame while sustaining a velvet smooth framerate matching the host's screen refresh rate.
Summarized, it can be looked at as a test of: How much stuff can I do before things start to fall apart..
Note that the sustained-fps shell requires a stable frame rate (see the prerequisites section), which is often hard for systems to provide, so this shell may not be a good choice for you, and may not be appropriate for automated regression testing.
As already mentioned, the goal of the sustained-fps shell is to find how much is possible while hitting a perfect framerate. So the first goal is to verify that your system is capable of hitting a perfect framerate at all. This is called the swap test. Test it by running the following, and observing the screen:
If you see a pulsating purple rectangle you are in good shape. If you see flashes of red and/or blue or if the purple rectangle has horizontal regions of red and blue, your system is not able to hit a perfect framerate.
Fix that first, or you will not get stable/sensible results from the sustained-fps shell.
Lastly, check if your system is running with the 'basic' render loop (QSG_INFO=1 in the environment will tell you). If you are, you will most likely not be able to get velvet FPS because the animations are timer driven and will skip once or twice per frame by default. If threaded is not an option for whatever reason, try with QSG_RENDER_LOOP=windows (which uses vsync driven animations).
The static-count shell constantly performs a set number of operations (the staticCount property in benchmarks) each frame. It is useful for profiling.
You can override the staticCount value by passing a
--count N command line
argument, where N is the staticCount you want to use instead.
The static-count shell is most useful for profiling, as it will repeat the same operation continuously.
the qmlbench tool
It comes with a number of default settings which aim to help give stable numbers:
--fps-override: This one may be very important when using the sustained-fps shell. When you ran decidefps, it told you a fairly accurate FPS. If QScreen reports something different through refreshRate() the calibration won't work and the resulting numbers won't mean anything. On OSX, we should be mostly good. On Linux, it is not uncommon that the QScreen value is wrong so this option is needed as an override to make the tests stable. The tool has not been tested on Windows at the time when this is written.
--delay [ms]: defaults to 2 seconds. An idle time after showing the window before starting the benchmark. Especially useful on OSX where the system specific to-fullscreen animation takes a while. And that to-fullscreen will severly damage the benchmarks.
--help: Tells you the other options.
the actual benchmarks
Benchmarks are divided into two types, automated, and manual. The automated tests are useful for regression testing, and the manual tests are useful for determining the capabilities of a new piece of hardware.
To run all the automated tests, simply run this, and go have lunch (it is not a fast process):
You can also run a subset of them by providing that directory instead:
Or even individual tests, by passing the filenames directly.
Here's an example of running
src/qmlbench --shell frame-count /Users/mitch/dev/qmlbench/benchmarks/auto/creation/quick.controls2/delegates_tumbler.qml on macOS:
ID: OS: macOS High Sierra (10.13) QPA: cocoa GL_VENDOR: ATI Technologies Inc. GL_RENDERER: AMD Radeon Pro 460 OpenGL Engine GL_VERSION: 2.1 ATI-1.60.24 running: /Users/mitch/dev/qmlbench/benchmarks/auto/creation/quick.controls2/delegates_tumbler.qml 153 frames 152 frames 154 frames 154 frames 153 frames Average: 153.2 frames; MedianAll=153; StdDev=0.83666, CoV=0.00546123 All done...
The test was repeated 5 times, and for each run, the frames are printed. For the frame-count shell, the higher the number of frames is, the better.
- "Average: 153.2 frames" - this is the average of each run.
- "MedianAll=153" - the median of all of the runs: (152, 153, 153, 154, 154) ^
- "StdDev=0.83666" - the standard deviation. The lower the better.
- "CoV=0.00546123" - the coefficient of variation. The lower the better; a general guide is to aim for less than 5% CoV.
Running tests on Android
To run qmlbench on Android, you need to use the subprocess mode. To pass arguments to the Android app, they need to be base64-encoded. Here's an example:
~/Android/Sdk/platform-tools/adb shell am start -n org.qtproject.example.qmlbench/org.qtproject.qt.android.bindings.QtActivity -e extraappparams `echo -n "--subprocess-mode --json-file :benchmarks/auto/creation/quick.text/delegates_longtext_arabic.qml" | base64 -w 0`
So, what we're doing here is
- launch the right app with the right activity:
- pass it the parameters, --subprocess-mode --json-file
- the command line is created with echo -n so that we don't insert a superfluous newline into the data
- and then we encode with base64 -w 0 to disable base64 line wrapping, so that that doesn't insert superfluous newlines either.
Since we're passing --json-file, that tells qmlbench to output the benchmark result into a file in JSON format. The file is qmlbench.txt in the application's data directory, so the usual incantation to get it out from a phone is
~/Android/Sdk/platform-tools/adb pull sdcard/Android/data/org.qtproject.example.qmlbench/files/Documents/qmlbench.txt
The benchmark app will not recreate the file, it always appends to it. If you need to start from a clean slate, just adb rm that same file.