Awesome
Melatonin Perfetto
Sounds like an ice cream flavor (albeit a sleepy one).
However, it's just a way to use the amazing Perfetto performance tracing in JUCE.
Perfetto lets you accurately measure different parts of your code and visualize it over time. It's the successor to chrome://tracing
.
Why would I use Perfetto instead of the profiler?
✨
✨✨
✨✨✨
Read the blog post!
✨✨✨
✨✨
✨
Requirements
- JUCE, any version after JUCE 5 should be happy
- C++17 (a Perfetto requirement since v31.0)
Installing with CMake
We worked hard so you don't have to.
Not only do we handle building Perfetto (which is fairly annoying on Windows), but we've made it very easy to add this module into your CMake projects.
You have several options. In all cases, the exported target you should link against is: Melatonin::Perfetto
.
Option #1: FetchContent
Example usage:
include (FetchContent)
FetchContent_Declare (melatonin_perfetto
GIT_REPOSITORY https://github.com/sudara/melatonin_perfetto.git
GIT_TAG origin/main)
FetchContent_MakeAvailable (melatonin_perfetto)
target_link_libraries (yourTarget PRIVATE Melatonin::Perfetto)
Option #2: submodules and add_subdirectory
If you are a git submodule aficionado, add this repository as a git submodule to your project:
git submodule add -b main https://github.com/sudara/melatonin_perfetto.git modules/melatonin_perfetto
and then simply call add_subdirectory
in your CMakeLists.txt:
add_subdirectory (modules/melatonin_perfetto)
target_link_libraries (yourTarget PRIVATE Melatonin::Perfetto)
Option #3: Installing and using find_package
Install the module to your system by cloning the code and then running the following commands:
cmake -B Builds
cmake --build Builds
cmake --install Builds
The --install
command will write to system directories, so it may require sudo
.
Once this module is installed to your system, you can simply add to your CMake project:
find_package (MelatoninPerfetto)
target_link_libraries (yourTarget PRIVATE Melatonin::Perfetto)
Installing with Projucer
Step 1: Download the Perfetto SDK
It can go anywhere. You'll actually need to use git to grab it though, there's no way to download it otherwise. Paste this into your macOS terminal (or download git on windows and use git bash):
git clone https://android.googlesource.com/platform/external/perfetto -b v31.0
Step 2: Download this module and add to your project
Use git to add it as a submodule if you'd like stay up to date with any changes:
git submodule add -b main https://github.com/sudara/melatonin_perfetto.git modules/melatonin_perfetto
Or just download it and stick it somewhere.
Be sure to add it in Projucer under "Modules".
Step 3: Add the perfetto headers in File Explorer
This is necessary to actually compile the perfetto tracing sdk from source.
In the File Explorer, hit the +
, Add Existing Files
and make sure the following two are added:
sdk/perfetto.h
sdk/perfetto.cc
Step 4: Add to your project's Header Search Paths
In the Project Settings (gear at the top right of the sidebar), tell the Projucer where to find perfetto/sdk
folder.
For example, if you downloaded it as a sibling folder to the project, you would add the following to Header Search Paths
:
../perfetto/sdk
Step 5: Add read/write permissions where necessary (macOS)
If you have App Sandbox
enabled, you'll have to enable the following:
`File Access: Read/Write: Download Folder (Read/Write)`
This lets perfetto write out the trace files.
<img src="https://user-images.githubusercontent.com/472/213724719-be39512e-cda2-43cb-a589-0c3478625228.jpg" width="400"/>Step 5: Add compile flags and preprocessor definitions (Windows)
Windows Projucer builds require some extra love.
Go to the Settings page for the Visual Studio Exporter and add these to "Extra Compile Flags":
/bigobj
/Zc:__cplusplus
/permissive-
/Zc:externC-
In addition, you'll need the following "Extra Preprocessor Definitions" set on that same page:
NOMINMAX=1
WIN32_LEAN_AND_MEAN=1
Step 6: Enable/Disable via the Projucer
As you'll see in "How to use", you can toggle perfetto traces on/off by adding the following in the exporter's preprocessor definitions:
PERFETTO=1
If that's too clunky, you can also just toggle in the source (read more below).
How to use
Step 1: Add a few pieces of guarded code to your plugin
Include library:
#include <melatonin_perfetto/melatonin_perfetto.h>
Add a member of the plugin processor:
#if PERFETTO
std::unique_ptr<perfetto::TracingSession> tracingSession;
#endif
Put this in PluginProcessor's constructor:
#if PERFETTO
MelatoninPerfetto::get().beginSession();
#endif
and in the destructor:
#if PERFETTO
MelatoninPerfetto::get().endSession();
#endif
Step 2: Pepper around some sweet sweet trace macros
Perfetto will only measure functions you specifically tell it to.
Pepper around TRACE_DSP()
or TRACE_COMPONENT()
macros at the start of the functions you want to measure.
For example, in your PluginProcessor::processBlock
:
void SineMachineAudioProcessor::processBlock (juce::AudioBuffer<float>& buffer, juce::MidiBuffer& midiMessages)
{
TRACE_DSP();
// your dsp code here
}
or a component's paint
method:
void paint (juce::Graphics& g) override
{
TRACE_COMPONENT();
// your paint code here
}
By default, perfetto is disabled (PERFETTO=0
) and the macros will evaporate on compile, so feel free to leave them in your code, especially if you plan on profiling again in the future.
Step 3: Enable profiling
When you are ready to profile, you'll need to set PERFETTO=1
This module offers a CMake option, PERFETTO
, that when ON
, adds the PERFETTO
symbol to the module's exported
compile definitions. This CMake option
is an easy way for you to turn tracing on and off from the command line when building your project:
cmake -B build -D PERFETTO=ON
The value of PERFETTO
will be saved in the CMake cache, so you don't need to re-specify this every time you re-run
CMake configure.
To turn it off again, you can do:
cmake -B build -D PERFETTO=OFF
You can add this as a CMake option in your IDE build settings, for example in CLion:
<img width="254" alt="CLion - 2023-01-07 06@2x" src="https://user-images.githubusercontent.com/472/211118287-0a3cd6bb-84e3-430b-a7d4-34474e51e42d.png">Aaaaand if you are lazy like Sudara sometimes is, you can just edit melatonin_perfetto and change the default to PERFETTO=1
...
That'll actually include the google lib and will profile the code.
Step 4: Run your app
Reminder: do you want to profile Release? Probably! I love profiling Debug builds too, but if you are looking for real-world numbers, you'll want to use Release.
Start your app and perform the actions you want traced.
When you quit your app, a trace file will be dumped
(Note: don't just terminate it via your IDE, the file will be only dumped on a graceful quit).
Step 5: Drag the trace into Perfetto
Find the trace file and drag it into https://ui.perfetto.dev
You can keep the macros peppered around in your app during normal dev/release.
Just remember to set PERFETTO
back to 0
or OFF
so everything gets turned into a no-op.
Customizing
By default, there are two perfetto "categories" defined, dsp
and components
.
The current function name is passed as the name.
You can also add custom parameters that will show up in perfetto's UI:
TRACE_DSP("startSample", startSample, "numSamples", numSamples);
TRACE_EVENT
You can also use the built in TRACE_EVENT
which takes a name if you don't want it to derive a name based on the function.
This is also if you want to do things like have multiple traces in a function, for example in a loop.
TRACE_DSP(); // start the trace, use the function name
if (someCondition)
{
TRACE_EVENT("dsp", "someCondition");
// do something expensive, traced separately
}
moreFuctionCode(); // included in the main trace
TRACE_EVENT_BEGIN and TRACE_EVENT_END
Sometimes you want to go full granular and not just depend on scoping.
To do this, use TRACE_EVENT_BEGIN
and TRACE_EVENT_END
:
TRACE_EVENT_BEGIN ("dsp", "memset");
// CLEAR the temp buffer
temp.clear();
TRACE_EVENT_END ("dsp");
Dynamic Names
Perfetto is optimized to be low overhead, shipping in RELEASE production builds. By default usage is via compile-time strings.
Working with strings dynamically introduces runtime overhead — but hey, sometimes you just want to look at something real quick. For those cases, Perfetto provides a helper:
TRACE_EVENT ("dsp", perfetto::DynamicString{my_dynamic_string});
"Solo" the message or audio thread
If you are focusing on UI and want to temporarily rid of the audio thread in the trace, set PERFETTO_ENABLE_TRACE_DSP=0
in your preprocessor definitions (or just modify the header like I do) and it will be a no-op.
You can do the reverse and disable all UI component tracing on the message thread with PERFETTO_ENABLE_TRACE_COMPONENT=0
.
Go wild!
Assumptions / Caveats
- On Mac, the trace is dumped to your Downloads folder. On Windows, it's dumped to your Desktop (sorry not sorry).
- Traces are set to in memory, 80MB by default.
Troubleshooting
Perfetto gives me an error about C++17 on Windows
Make sure you are passing the /Zc:__cplusplus
flag so MSVC's version detection actually works, see this link
In CMake:
target_compile_options("${PROJECT_NAME}" PUBLIC /Zc:__cplusplus)
I don't see a trace file
Did you quit your app gracefully, such as with cmd-Q? If you instead just hit STOP on your IDE, you won't get a trace file.
My traces appear empty
You probably went over the memory size that Perfetto is set to use by default (80MB).
If you are doing intensive profiling with lots of functions being called many times, you'll probably want to increase this limit. You can do this by passing the number of kb
you want to beginSession
:
MelatoninPerfetto::get().beginSession(300000); # 300MB
I keep forgetting to turn perfetto off!
Get warned by your tests if someone left PERFETTO=1
on with a test like this (this example is Catch2):
TEST_CASE ("Perfetto not accidentally left enabled", "[perfetto]")
{
#if defined(PERFETTO) && PERFETTO
FAIL_CHECK ("PERFETTO IS ENABLED");
#else
SUCCEED ("PERFETTO DISABLED");
#endif
}
If you use perfetto regularly, you can also do what I do and check for PERFETTO
in your plugin editor and display something in the UI:
Running Melatonin::Perfetto's tests
melatonin_perfetto
includes a test suite using CTest. To run the tests, clone the code and run these commands:
cmake -B Builds
cmake --build Builds --config Debug
cd Builds
ctest -C Debug
The tests attempt to build two minimal CMake projects that depend on the melatonin_perfetto
module; one tests
finding an install tree using find_package()
and one tests calling add_subdirectory()
. These tests serve to
verify that this module's packaging and installation scripts are correct, and that it can be successfully imported
to other projects using the methods advertised above. Another test case verifies that attempting to configure a
project that adds melatonin_perfetto
before JUCE will fail with the proper error message.
Acknowledgements
- Thanks to @benthevining for extra CMake love and tests!
- Thanks to @dikadk for getting a Projucer version running on MacOS
- Thanks to stephenk for putting in the effort to getting the Projucer version working on Windows.