/* This Source Code Form is subject to the terms of the Mozilla Public * License, v. 2.0. If a copy of the MPL was not distributed with this * file, You can obtain one at https://mozilla.org/MPL/2.0/. */ // This comment contains the central documentation for how we configured Jetpack Benchmark. Currently: // - microbenchmark: configured differently than recommended (see inline notes below) // - macrobenchmark: not configured // // To run our benchmarks, you need to set the "benchmark" gradle property. You can: // - (preferred) Run via the command line (change the class you run on): // ./gradlew -Pbenchmark app:connectedCheck -Pandroid.testInstrumentationRunnerArguments.class=net.waterfox.android.perf.SampleBenchmark // - Use the IDE. Temporarily set the "benchmark" property in app/build.gradle with "ext.benchmark=true" // near the top of the file. DO NOT COMMIT THIS. // - (note: I was unable to get IDE run configurations working) // // To get the results, look at this file (we recommend using the median; results are in nanoseconds): // app/build/outputs/connected_android_test_additional_output/nightlyAndroidTest/connected//net.waterfox.android-benchmarkData.json // // I was unable to get the results to print directly in Android Studio (perhaps it's my device). // // The official documentation suggests configuring microbenchmark in a separate module. This would // require any benchmarked code to be in a library module, not the :app module (see below). To avoid // this requirement, we created the "benchmark" gradle property. // // For the most accurate results, the documentation recommends running tests on rooted devices with // the CPU clock locked. // // See https://developer.android.com/studio/profile/benchmark#what-to-benchmark for when writing a // jetpack microbenchmark is a good fit. // I think `android` represents this object: // https://google.github.io/android-gradle-dsl/3.3/com.android.build.gradle.AppExtension.html ext.maybeConfigForJetpackBenchmark = { android -> if (!project.hasProperty("benchmark")) { return } // The official documentation https://developer.android.com/studio/profile/benchmark#full-setup // recommends setting up the Microbenchmark library in a separate module from your app: AFAICT, // the reason for this is to prevent the benchmarks from being configured against debug // builds. We chose not to do this because it's a lot of work to pull code out into a // separate module just to benchmark it. We were able to replicate the outcome by setting // this testBuildType property. android.testBuildType "release" // WARNING: our proguard configuration for androidTest is not set up correctly so the tests // fail if we don't disable minification. DISABLING MINIFICATION PRODUCES BENCHMARKS THAT ARE // LESS REPRESENTATIVE TO THE USER EXPERIENCE, however, so we made this tradeoff to reduce // implementation time. project.ext.disableOptimization = true android.defaultConfig { // WARNING: the benchmark framework warns you if you're running the test in a configuration // that will compromise the accuracy of the results. Unfortunately, I couldn't get everything // working so I had to suppress some things. testInstrumentationRunnerArguments = [ // - ACTIVITY-MISSING: we're supposed to use the test instrumentation runner, // "androidx.benchmark.junit4.AndroidBenchmarkRunner". However, when I do so, I get an error // that we're unable to launch the activity. My understanding is that this runner will use an // "IsolationActivity" to reduce the impact of other work on the device from affecting the benchmark // and to opt into a lower-max CPU frequency on unrooted devices that support it // - UNLOCKED: ./gradlew lockClocks, which locks the CPU frequency, fails on my device. See // https://issuetracker.google.com/issues/176836267 for potential workarounds. 'androidx.benchmark.suppressErrors' : 'ACTIVITY-MISSING,UNLOCKED', // The tests don't always output a JSON file with the data. To make sure it does, we have to // set androidx.benchmark.output.enable to true. 'androidx.benchmark.output.enable' : 'true', // We set the the output directory simply for simplicity since the benchmark_runner.py script // can't know the name of the phone in the /build/outputs/ directory. The system defaults to // {phone_name} which can be troublesome finding in some case. // // NOTE: Jetpack Benchmark outputs to Logcat too. However, the output in the logcat is // the min of the several repeats, for more statistics. Therefore, to get more stats, // we refer to the JSON file. additionalTestOutputDir : '/storage/emulated/0/benchmark' ] } }