Profiling Uno Platform Applications

Profiling .NET 6 Android applications

As of Preview 7, .NET 6 provides the ability to do CPU profiling through dotnet-trace for android applications.

Profiling Xamarin.Android and Xamarin.iOS applications


Run the following commands

  • dotnet tool update -g dotnet-dsrouter --add-source=
  • dotnet tool update -g dotnet-trace --add-source=

Adjust your application to enable profiling

Profiling has to first be enabled in the application. Some additional properties need to be added in the MyApp.Mobile project :

    <RuntimeIdentifier Condition="'$(TargetFramework)' == 'net6.0-android'">android-x64</RuntimeIdentifier>

<PropertyGroup Condition="'$(AndroidEnableProfiler)'=='true'">
    <IsEmulator Condition="'$(IsEmulator)' == ''">true</IsEmulator>

<ItemGroup Condition="'$(AndroidEnableProfiler)'=='true'">
    <AndroidEnvironment Condition="'$(IsEmulator)' == 'true'" Include="Android\environment.emulator.txt" />
    <AndroidEnvironment Condition="'$(IsEmulator)' != 'true'" Include="Android\environment.device.txt" />

Then in the Android application folder, add the following two files:

  • environment.device.txt
  • environment.emulator.txt

Note that the suspend directive means that if dotnet-trace is not running, the application waits for it to start.

Profiling the application

  • Start the diagnostics router, in any folder:
    dotnet-dsrouter client-server -tcps -ipcc /tmp/uno-app --verbose debug
  • Start dotnet-trace, in the app folder or where you want your traces to be stored:
    dotnet-trace collect --diagnostic-port /tmp/uno-app --format speedscope -o uno-app-trace
  • Start an x86-64 emulator or arm64 (armv8) 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 -f net6.0-android -t:run -c Release -p:IsEmulator=true /p:RunAOTCompilation=true /p:AndroidEnableProfiler=true
  • The app will start and the dotnet-trace will display a MB number counting up
  • Use the app and once done, stop dotnet-trace using the specified method (Likely Enter or Ctr+C)
  • Open a browser at and drop the uno-app-trace.speedscope.json file on it

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 finding 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 the strict minimum, where impactful.

Profiling Skia/GTK and Skia/WPF 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:

  • Enable AOT compilation:

  • Build and deploy the application

  • Open the Performance tab in your browser

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


  • 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.