docs/macrobenchmarking.md
[TOC]
<!-- Copied from macrobenchmark docs --> <table> <tr> <td><strong>Macrobenchmark</strong></td> <td><strong>Benchmark</strong></td> </tr> <tr> <td>Measure high-level entry points (Activity launch / Scrolling a list)</td> <td>Measure individual functions</td> </tr> <tr> <td>Out-of-process test of full app</td> <td>In-process test of CPU work</td> </tr> <tr> <td>Slow iteration speed (Often more than a minute)</td> <td>Fast iteration speed (Often less than 10 seconds)</td> </tr> <tr> <td>Configure compilation with <a href="https://developer.android.com/reference/androidx/benchmark/macro/CompilationMode">CompilationMode</a></td> <td>Always fully AOT (<code>speed</code>) compiled.</td> </tr> <tr> <td>Relatively lower stability measurements</td> <td>Higher stability measurements</td> </tr> </table>The public documentation for macrobenchmark explains how to use the library. This page focuses on specifics to writing library macrobenchmarks in the AndroidX repo. If you're looking for measuring CPU perf of individual functions, see the guide for MICRObenchmarks here.
Benchmarks are just regular instrumentation tests! Just use the
MacrobenchmarkRule
provided by the library:
@get:Rule
val benchmarkRule = MacrobenchmarkRule()
@Test
fun startup() = benchmarkRule.measureRepeated(
packageName = "mypackage.myapp",
metrics = listOf(StartupTimingMetric()),
startupMode = StartupMode.COLD,
iterations = 10
) { // this = MacrobenchmarkScope
pressHome()
val intent = Intent()
intent.setPackage("mypackage.myapp")
intent.setAction("mypackage.myapp.myaction")
startActivityAndWait(intent)
}
@Rule
public MacrobenchmarkRule mBenchmarkRule = MacrobenchmarkRule()
@Test
public void startup() {
mBenchmarkRule.measureRepeated(
"mypackage.myapp",
Collections.singletonList(new StartupTimingMetric()),
StartupMode.COLD,
/* iterations = */ 10,
scope -> {
scope.pressHome();
Intent intent = Intent();
intent.setPackage("mypackage.myapp");
intent.setAction("mypackage.myapp.myaction");
scope.startActivityAndWait(intent);
return Unit.INSTANCE;
}
);
}
As in the public documentation, macrobenchmarks in the AndroidX repo are comprised of an app, and a separate macrobenchmark test module. In the AndroidX repository, there are additional requirements:
Macrobenchmark test module path in settings.gradle must end with
macrobenchmark to run in CI.
Macrobenchmark target module path in settings.gradle should end with
macrobenchmark-target to follow convention.
Macrobenchmark modules must use project dependencies where available (so
implementation(project(":activity:activity-ktx")) rather than
implementation("androidx.activity:activity-ktx:1.5.0")). This prevents
accidentally testing against out-of-date versions, and increases coverage of
lower level libraries.
Each library group must declare its own in-group macrobenchmark test and
app module, with out using these modules for anything else (e.g. samples).
We want to be intentional about which changes affect measurements. More than
one of either is allowed, which is sometimes necessary to compare different
startup behaviors, see e.g.
:emoji2:integration-tests:init-<disabled/enabled>-macrobenchmark-target.
Note that comparing multiple app variants are not currently supported by CI.
Compose Macrobenchmark Examples:
Note: Compose macrobenchmarks are ideally duplicated with View system
counterparts, defined in :benchmark:integration-tests:macrobenchmark-target.
This is how we compare performance of the two systems.