English | 简体中文
Evidence-driven Go performance engineering with documentation, Agent Skill, and reproducible proofs.
This repository focuses on understanding and improving Go performance through evidence.
It is not a collection of isolated optimization tricks.
The project covers:
- compiler and SSA optimization;
- CPU cache and memory layout;
- concurrency and synchronization;
- allocation and garbage collection;
- unsafe and runtime boundaries;
- benchmarking and profiling methodology.
.
├── docs/
│ Complete performance engineering documentation
│
├── skill/
│ Agent Skill package
│
└── proofs/
Minimal reproducible experiments
The docs/ directory contains the complete knowledge base.
Start here:
The skill/ directory contains the AI-oriented performance workflow.
It provides:
- diagnostic routing;
- evidence requirements;
- optimization risk rules;
- maintainability checks;
- validation workflow.
Start here:
The proofs/ directory contains minimal experiments for specific performance behaviors.
A proof demonstrates:
This mechanism exists and can be reproduced under documented conditions.
It does not claim:
This optimization is always faster.
See:
Articles published on Gopher Atlas connect Go performance concepts with reproducible experiments and documented evidence. Both English and Simplified Chinese editions are available.
| Study | Articles | Reproducible evidence |
|---|---|---|
| Slice Retention — How a 100-byte slice can keep a 100 MiB backing array alive | English · 简体中文 | Proof · Windows report · Linux report |
| Heap Ballast (Historical) — The mechanism and evolution of a classic Go GC tuning technique | English · 简体中文 | Proof · Windows report · Linux report |
| Bounds Check Elimination — How the compiler proves redundant slice bounds checks | English · 简体中文 | Proof · Windows report · Linux report |
The experiments establish specific mechanisms under recorded conditions, not universal performance gains. Historical techniques are preserved for study and reproducibility, not as default recommendations for modern Go applications.
Measure
↓
Understand the cost model
↓
Make the smallest justified change
↓
Benchmark
↓
Validate system impact
↓
Preserve the reason for the optimization
Apache License 2.0