I’ve just published mojo-benchsuite, a Mojo benchmarking package focused on practical project workflows (not only one-off “microbenchmarks”).
Mojo’s stdlib benchmarking is excellent at the low level.
This package aims to complement it at the suite/workflow level:
• organised bench_*.mojo suite runs
• adaptive sampling with richer stats (mean/min/max + p50/p95/p99)
• baseline save/compare for regression tracking
• machine-readable summary output for CI
• report artefacts (console + Markdown + CSV)
It also includes practical benchmark suites and examples:
• config workloads
• data-transform workloads
• startup-cost workloads
• longer-running examples (plus a GPU example with CPU reference path)
Quick start:
• pixi install
• pixi run bench-all
• pixi run run-example-complex
• pixi run run-example-startup
• pixi run run-example-gpu
I’d really value feedback on:
• API ergonomics
• baseline/regression workflow
• CI expectations for benchmark projects
• additional real-world workload patterns people want included