Profiling Uno Platform Applications

Profiling .NET Android/iOS applications

.NET 7 and later provides the ability to do CPU profiling through dotnet-trace for Android applications.

Pre-requisites

Run the following commands

  • dotnet tool update -g dotnet-dsrouter
  • dotnet tool update -g dotnet-trace
  • dotnet tool update -g dotnet-gcdump

Profiling .NET iOS applications

Note

This documentation is based on .NET iOS profiling documentation.

Profiling iOS apps needs to be done on a mac machine.

First, create an alias to mlaunch:

alias mlaunch=/Library/Frameworks/Xamarin.iOS.framework/Versions/Current/bin/mlaunch

Profiling on an iOS Simulator

  1. The first step is to launch the tool that provides a connection between the app and the .NET tracing tools:

    dotnet-dsrouter client-server -ipcc ~/my-sim-port -tcps 127.0.0.1:9000

  2. Launch the app and make it suspend upon launch (waiting for the .NET tooling to connect):

    mlaunch --launchsim bin/Debug/net*/*/*.app --device :v2:runtime=com.apple.CoreSimulator.SimRuntime.iOS-15-4,devicetype=com.CoreSimulator.SimDeviceType.iPhone-11 --wait-for-exit --stdout=$(tty) --stderr=$(tty) --argument --connection-mode --argument none '--setenv:DOTNET_DiagnosticPorts=127.0.0.1:9000,suspend'

  3. At this point it's necessary to wait until the following line shows up in the terminal:

    The runtime has been configured to pause during startup and is awaiting a Diagnostics IPC ResumeStartup command from a Diagnostic Port

  4. Once that's printed, go ahead and start profiling:

    dotnet-trace collect --diagnostic-port ~/my-sim-port --format speedscope

To find which device to use, use:

xcrun simctl list devices

Then reference the UDID of the simulator in the mlaunch command:

mlaunch ... --device :v2:udid=50BCC90D-7E56-4AFB-89C5-3688BF345998 ...

Profiling on a physical iOS device

Launch the tool that bridges the app and the .NET tracing tools:

dotnet-dsrouter server-client -ipcs ~/my-dev-port -tcpc 127.0.0.1:9001 --forward-port iOS

Install & launch the app and make it suspended upon launch:

mlaunch --installdev bin/Debug/net*/*/*.app --devname ... 
mlaunch --launchdev bin/Debug/net*/*/*.app --devname ... --wait-for-exit --argument --connection-mode --argument none '--setenv:DOTNET_DiagnosticPorts=127.0.0.1:9001,suspend,listen'

At this point, it's necessary to wait until the following line shows up in the terminal:

The runtime has been configured to pause during startup and is awaiting a Diagnostics IPC ResumeStartup command from a Diagnostic Port

Once that's printed, go ahead and start profiling:

dotnet-trace collect --diagnostic-port ~/my-dev-port,connect --format speedscope

Profiling Catalyst apps

  1. Launch the executable, passing the DOTNET_DiagnosticPorts variable directly:

    DOTNET_DiagnosticPorts=~/my-desktop-port,suspend ./bin/Debug/net7.0-*/*/MyTestApp.app/Contents/MacOS/MyTestApp

  2. At this point it's necessary to wait until the following line shows up in the terminal:

    The runtime has been configured to pause during startup and is awaiting a Diagnostics IPC ResumeStartup command from a Diagnostic Port

  3. Once that's printed, go ahead and start profiling:

    dotnet-trace collect --diagnostic-port ~/my-desktop-port --format speedscope

Profiling .NET Android applications (.NET 8)

Enable profiling in your application

In Platforms/Android/environment.conf, add one of the following lines:

  • For devices:

    DOTNET_DiagnosticPorts=127.0.0.1:9000,suspend,connect

  • For emulators:

    DOTNET_DiagnosticPorts=10.0.2.2:9000,suspend,connect

The suspend directive means that the application will wait for dotnet-trace connections before starting, nosuspend may also be used.

Profiling the application

  • Start the diagnostics router:

    • For devices, run adb reverse tcp:9000 tcp:9001 then dotnet-dsrouter android -v debug

    • For emulators, run dotnet-dsrouter android-emu -v debug

  • Run dotnet-trace, in the folder where you want your traces to be stored, using the PID provided by the dotnet-dsrouter output:

    dotnet-trace collect -p PID --format speedscope

  • Start the x64 emulator or the arm64 device

    Running on a 32 bits device is not supported and will generate unusable traces in SpeedScope

  • Build the application with profiling enabled

    dotnet build -c Release -f net9.0-android -r android-arm64 -t:Run -p:AndroidEnableProfiler=true

    Use -r android-x64 for emulators instead.

  • The app will start and dotnet-trace will display a MB number counting up

  • Use the app, once done, stop dotnet-trace by pressing Enter or Ctrl+C

  • Open a browser at https://speedscope.app and drop the *.speedscope.json file in it

Getting GC memory dumps

To take a GC memory dump of a running android app, follow the same steps above, but instead of dotnet-trace collect -p <port>, use dotnet-gcdump collect -p <port>. It will create a .gcdump file that can be viewed in Visual Studio and Perfview on Windows and heapview on non-Windows platforms.

See complete documentation for more details.

Analyzing the trace data

This section provides insights into what to look for when analyzing flame charts.

  • When building without AOT, a lot of the startup traces will show time spent in System.Private.CoreLib!System.Runtime.CompilerServices.RuntimeHelpers.CompileMethod(object), indicating that that the JIT is doing a lot of work. This can make performance improvements harder to find.
  • When building with AOT, most of the IL is compiled to native code with some exceptions. You may still find RuntimeHelpers.CompileMethod invocations. In such cases, you may need to find what is causing the AOT compiler to skip IL portions. If the JIT still impacts cold paths of your application, you may still need to adjust your code to avoid the JIT. For instance, some generics constructs force the AOT compiler to still use JITing. In other cases, it could be accessing static-type members. The JIT conditions are runtime version dependent, and looking at the runtime code can help to find out which ones.
  • Some of the time is spent in the .NET Android binding framework (e.g. Android.Runtime.JNIEnv or Java.Interop.TypeManager), operations that cannot be adjusted by the application. One change to consider is to reduce the native code invocations to a strict minimum, where impactful.

Profiling Skia Desktop applications

Profiling Skia-based Uno Platform targets can be done on Windows in Visual Studio 2019 and 2022 using time and memory profilers.

Profiling WebAssembly applications

Profiling WebAssembly applications can be done through the use of AOT compilation, and browsers' performance tab.

Setup the WebAssembly application for profiling

  • Enable emcc profiling:

    <PropertyGroup>
    <WasmShellEnableEmccProfiling>true</WasmShellEnableEmccProfiling>
</PropertyGroup>

  • Enable AOT compilation:

    <PropertyGroup>
    <WasmShellMonoRuntimeExecutionMode>InterpreterAndAOT</WasmShellMonoRuntimeExecutionMode>
</PropertyGroup>

  • Build and deploy the application

  • Open the Performance tab in your browser

  • Use your application or restart your application while recording the trace

Troubleshooting

  • Deep traces found in large async code patterns or complex UI trees may hit this chromium issue. This generally makes traces very long to load; you'll need to be patient.