The Dart project benchmark harness is the recommended starting point when building a benchmark for Dart.
Interpreting Results
By default, the reported runtime in BenchmarkBase is not for a single call to
run(), but for the average time it takes to call run() 10 times for
legacy reasons. The benchmark harness executes a 10-call timing loop repeatedly
until 2 seconds have elapsed; the reported result is the average of the runtimes
for each loop. This behavior will change in a future major version.
Benchmarks extending BenchmarkBase can opt into the reporting the average time
to call run() once by overriding the exercise method:
@override
void exercise() => run();
AsyncBenchmarkBase already reports the average time to call run() once.
Comparing Results
If you are running the same benchmark, on the same machine, running the same OS, the reported run times can be carefully compared across runs. Carefully because there are a variety of factors which could cause error in the run time, for example, the load from other applications running on your machine could alter the result.
Comparing the run time of different benchmarks is not recommended. In other words, don't compare apples with oranges.
Features
BenchmarkBaseclass that all new benchmarks shouldextend.AsyncBenchmarkBasefor asynchronous benchmarks.- Template benchmark that you can copy and paste when building new benchmarks.
Getting Started
- Add the following to your project's pubspec.yaml
dependencies:
benchmark_harness: any
- Install pub packages
dart pub install
- Add the following import:
import 'package:benchmark_harness/benchmark_harness.dart';
- Create a benchmark class which inherits from
BenchmarkBaseorAsyncBenchmarkBase.
Example
Create a dart file in the
benchmark/
folder of your package.
// Import BenchmarkBase class.
import 'package:benchmark_harness/benchmark_harness.dart';
// Create a new benchmark by extending BenchmarkBase
class TemplateBenchmark extends BenchmarkBase {
const TemplateBenchmark() : super('Template');
static void main() {
const TemplateBenchmark().report();
}
// The benchmark code.
@override
void run() {}
// Not measured setup code executed prior to the benchmark runs.
@override
void setup() {}
// Not measured teardown code executed after the benchmark runs.
@override
void teardown() {}
// To opt into the reporting the time per run() instead of per 10 run() calls.
//@override
//void exercise() => run();
}
void main() {
// Run TemplateBenchmark
TemplateBenchmark.main();
}
Output
Template(RunTime): 0.1568472448997197 us.
This is the average amount of time it takes to run run() 10 times for
BenchmarkBase and once for AsyncBenchmarkBase.
usis an abbreviation for microseconds.
bench command
A convenience command available in package:benchmark_harness.
If a package depends on benchmark_harness, invoke the command by running
dart run benchmark_harness:bench
If not, you can use this command by activating it.
dart pub global activate benchmark_harness
dart pub global run benchmark_harness:bench
Output from dart run benchmark_harness:bench --help
Runs a dart script in a number of runtimes.
Meant to make it easy to run a benchmark executable across runtimes to validate
performance impacts.
-f, --flavor
[aot] Compile and run as a native binary.
[jit] Run as-is without compilation, using the just-in-time (JIT) runtime.
[js] Compile to JavaScript and run on node.
[wasm] Compile to WebAssembly and run on node.
--target The target script to compile and run.
(defaults to "benchmark/benchmark.dart")
-h, --help Print usage information and quit.
-v, --verbose Print the full stack trace if an exception is thrown.
Example usage:
dart run benchmark_harness:bench --flavor aot --target example/template.dart
AOT - COMPILE
/dart_installation/dart-sdk/bin/dart compile exe example/template.dart -o /temp_dir/bench_1747680526905_GtfAeM/out.exe
Generated: /temp_dir/bench_1747680526905_GtfAeM/out.exe
AOT - RUN
/temp_dir/bench_1747680526905_GtfAeM/out.exe
Template(RunTime): 0.005620051244379949 us.