Skip to main content

crossbench

Running benchmarks with Crossbench.

Datos de origen

Repositorio
nodejs/node
Última actividad en el origen
4 de octubre de 2026 a las 21:04
Idioma detectado de SKILL.md
inglés
Estrellas
122.301
Forks
39.450

Opciones de instalación

De forma predeterminada está seleccionado el prompt que primero revisa el origen. Puedes cambiar a un comando directo o descargar una copia local.

Revisa los archivos de origen

Lee SKILL.md y los archivos complementarios que muestra SkillsMP antes de decidir si quieres instalarlo.

Mostrando SKILL.md

SKILL.md
Instrucciones de origen · Vista previa de solo lectura
name
crossbench
description
Running benchmarks with Crossbench.
# Crossbench Crossbench is the central chrome runner harness for press benchmarks like JetStream, Speedometer or Motionmark, as well as for loading live pages. ## Location and Setup - **Location**: It is often installed in `~/crossbench` or in `CHROMIUM_CHECKOUT/src/tools/perf/cb`, but could be anywhere. If not found in standard locations, ask the user for the path. - **Update**: Always ensure you are running the latest version. ## Basic Usage Navigate to the crossbench directory and use `./cb.py` directly (or via `poetry` if vpython3 is not available). ```bash # Run the latest stable JetStream with a specific d8 binary ./cb.py jetstream --browser=/path/to/d8 --env-validation=warn ``` And for chrome: ```bash ./cb.py jetstream --browser=out/x64.release/chrome --env-validation=warn ``` ## Filtering - You can run benchmark workload separate with `--separate` for instance to generate individual profiles for each sub test - Use `--story=XXX` to only run a specific test ## Crossbench MCP Use `./cb.py mcp` to launch a local MCP to interact with benchmarks. ## Profiling with Probes - Use `./cb.py describe probes` to list all probes. - Use `./cb.py describe probe v8.log` to show detailed help for a specific probe. Important probes for v8 investigation: - `--probe='v8.log'` to generate a `v8.log` file (see v8-log skill for more information on how to analyse the log file) - `--probe='profiling'` to generate a Linux `perf` trace. - `--probe=perfetto` to collected perfetto traces Results: - Crossbench prints the results directory and provides a summary `cb.results.json` file listing all probe results ## Best Practices - Use `--env-validation=warn` to bypass environment input prompts. - Use `./cb.py describe` to understand subcommands and probes. - Prefer creating JSON files instead of HJSON for configurations to minimize quoting errors. - Validate generated configs with `poetry run cb_validate_hjson -- file.hjson`. ## Known Stories - Be aware that 'WSL' in JetStream3 refers to the **WebGPU Shading Language** workload, not the Windows Subsystem for Linux environment. ## Environment Fallbacks - Suggest install depot_toools if vpython3 is not available - Use poetry as backup if vpython3 is not available. - The JetStream benchmark can be run directly with d8
Ver en GitHub