Skip to content

benchmarking.md

Latest commit

 

History

History
234 lines (177 loc) · 8.28 KB

benchmarking.md

File metadata and controls

234 lines (177 loc) · 8.28 KB

Benchmarking in AndroidX

[TOC]

The public documentation at d.android.com/benchmark explains how to use the library - this page focuses on specifics to writing libraries in the AndroidX repo, and our continuous testing / triage process.

This page is for MICRO benchmarks measuring CPU performance of small sections of code. If you're looking for measuring startup or jank, see the guide for MACRObenchmarks here.

Writing the benchmark

Benchmarks are just regular instrumentation tests! Just use the BenchmarkRule provided by the library:

Kotlin {.new-tab}

@RunWith(AndroidJUnit4::class)
class ViewBenchmark {
    @get:Rule
    val benchmarkRule = BenchmarkRule()

    @Test
    fun simpleViewInflate() {
        val context = InstrumentationRegistry
                .getInstrumentation().targetContext
        val inflater = LayoutInflater.from(context)
        val root = FrameLayout(context)

        benchmarkRule.measure {
            inflater.inflate(R.layout.test_simple_view, root, false)
        }
    }
}

Java {.new-tab}

@RunWith(AndroidJUnit4.class)
public class ViewBenchmark {
    @Rule
    public BenchmarkRule mBenchmarkRule = new BenchmarkRule();

    @Test
    public void simpleViewInflate() {
        Context context = InstrumentationRegistry
                .getInstrumentation().getTargetContext();
        final BenchmarkState state = mBenchmarkRule.getState();
        LayoutInflater inflater = LayoutInflater.from(context);
        FrameLayout root = new FrameLayout(context);

        while (state.keepRunning()) {
            inflater.inflate(R.layout.test_simple_view, root, false);
        }
    }
}

Project structure

As in the public documentation, benchmarks in the AndroidX repo are test-only library modules. Differences for AndroidX repo:

  1. Module must apply id("androidx.benchmark") in the plugin block
  2. Module should live in integration-tests group directory to follow convention
  3. Module name should end with -benchmark in settings.gradle to follow convention
  4. Module should not contain non-benchmark tests to avoid wasting resources in benchmark postsubmit

Applying the benchmark plugin give you benefits from the AndroidX plugin:

  • Inclusion in microbenchmark CI runs
  • AOT Compilation of module (local and CI) for stability
  • Disable ANR avoidance in local runs (so you always get method traces)

But note that these can be detrimental for non-benchmark code.

I'm lazy and want to start quickly

Start by copying one of the following non-Compose projects:

Many Compose libraries already have benchmark modules:

Profiling

See the public profiling guide for more details.

Jetpack benchmark supports capturing profiling information by setting instrumentation arguments. Stack sampling and method tracing can be performed either from CLI or Studio invocation.

Set Arguments in Gradle

Args can be set in your benchmark's build.gradle, which will affect both Studio / command-line gradlew runs. Runs from Studio will link result traces that can be opened directly from the IDE.

android {
    defaultConfig {
        // must be one of: 'None', 'StackSampling', or 'MethodTracing'
        testInstrumentationRunnerArgument 'androidx.benchmark.profiling.mode', 'StackSampling'
    }
}

Set Arguments on Command Line

Args can also be passed from CLI. Here's an example which runs the androidx.compose.material.benchmark.CheckboxesInRowsBenchmark#draw method with StackSampling profiling:

./gradlew compose:material:material-benchmark:cC \
    -P android.testInstrumentationRunnerArguments.androidx.benchmark.profiling.mode=StackSampling \
    -P android.testInstrumentationRunnerArguments.class=androidx.compose.material.benchmark.CheckboxesInRowsBenchmark#draw

The command output will tell you where to look for the file on your host machine:

04:33:49 I/Benchmark: Benchmark report files generated at
/androidx-main/out/ui/ui/integration-tests/benchmark/build/outputs/connected_android_test_additional_output

To inspect the captured trace, open the appropriate *.trace file in that directory with Android Studio, using File > Open.

NOTE For stack sampling, it's recommended to profile on Android Q(API 29) or higher, as this enables the benchmark library to use Simpleperf when capturing samples.

For more information on the StackSampling and MethodTracing profiling modes, see the Studio Profiler recording configuration docs, specifically "Sample C/C++ Functions" (called "Callstack sample" in recent versions), and Java Method Tracing.

Sample flame chart

Advanced: Connected Studio Profiler

Profiling for allocations requires Studio to capture, and a debuggable build. Do not commit the following changes.

First, set your benchmark to be debuggable in your benchmark module's androidTest/AndroidManifest.xml:

  <application
    ...
    android:debuggable="false"
    tools:ignore="HardcodedDebugMode"/>

Note that switching to the debug variant will likely not work, as Studio will fail to find the benchmark as a test source.

Next select ConnectedAllocation in your benchmark module's build.gradle:

android {
    defaultConfig {
        // --- Local only, don't commit this! ---
        // pause for manual profiler connection before/after a single run of
        // the benchmark loop, after warmup
        testInstrumentationRunnerArgument 'androidx.benchmark.profiling.mode', 'ConnectedAllocation'
    }
}

Run File > Sync Project with Gradle Files, or sync if Studio asks you. Now any benchmark runs in that project will permit debuggable, and pause before and after the test, to allow you to connect a profiler and start recording, and then stop recording.

Running and Profiling

After the benchmark test starts, you have about 20 seconds to connect the profiler:

  1. Click the profiler tab at the bottom
  2. Click the plus button in the top left, <device name>, <process name>
  3. Click the memory section, and right click the window, and select Record allocations.
  4. Approximately 20 seconds later, right click again and select Stop recording.

If timed correctly, you'll have started and stopped collection around the single run of your benchmark loop, and see all allocations in detail with call stacks in Studio.

Minification / R8

As many Android apps don't yet enable R8, the default for microbenchmarks in AndroidX is to run with R8 disabled to measure worst-case performance. It may still be useful to run your microbenchmarks with R8 enabled locally however, and that is supported experimentally. To do this in your microbench module, set the androidTest minification property:

android {
    buildTypes.release.androidTest.enableMinification = true
}

Then, if you see any errors from classes not found at runtime, you can add proguard rules here, or in a similar place for your module.