Kotlin Multiplatform Collection of Benchmarks focused on Kotlin/Wasm performance.

Based on data from Compose Multiplatform Benchmarks

Description

These benchmarks are based on are-we-fast-yet benchmarks collection and JetBrains/Kotlin micro-benchmarks (work-in-progress). To perform benchmarks it uses kotlinx-benchmarks library.

Build and Run

Specify Kotlin version in gradle.properties file or use additional gradle argument -Pkotlin_version=2.4.20.

Run benchmark

The benchmarks can be run with the following parameters:

• Name: FastMacro, FastMicro, SlowMacro, SlowMicro, VolatileMicro
• Binaries: Js, Wasm
• Build: Prod, Dev
• Engine: D8, Jsc, JsShell, Wasmtime, WasmEdge

To run the benchmarks you need to specify those parameters in the following command and run it:

./gradlew wasmJs[Name]_[Binaries]_[Build]_[Engine]Benchmark

For example to have FastMicro benchmarks compiled to Wasm in Prod mode with Wasmtime one need to run:

./gradlew wasmJsFastMicro_Wasm_Prod_WasmtimeBenchmark

To see all tasks:

./gradlew tasks

Result output is located in build/reports directory.

License

See LICENSE.md file for details.

Generic rules writing the benchmarks

Some rules considering a benchmark writing:

1. Whenever is possible, consume some benchmark result into Blackhole parameter.
2. Do not use function return value as a blackhole consumer.
3. Always prepare benchmark data in @Setup method.
4. Never mutate shared data in the benchmark.
5. If the benchmark uses other functions, try make them top-level to be sure that function call does not use VTable.
6. Prefer const val's when possible.
7. Call blackhole consume as rarely as possible but consumed value should cover as much data-flow as possible.
8. If you write a micro benchmark, write as less code as possible to be sure that this benchmark is still being micro.
9. Do not use kotlinx.benchmark annotations to tune your benchmark, use gradle file.
10. Benchmark iterations could be also tuned with a loop on the BENCHMARK_SIZE value. If you need to use small value, it is also fine.
11. Do not consume into blackhole instances of external classes or external interfaces.
12. Do not have multiple access to non-local properties in the benchmark. Read it once into a local variable and use that variable instead.
13. Use microBenchmarks.Random random generator to generate random data.