Profile before optimizing (C++)

Steps

1. 1Define target metric: p95 latency? memory ceiling? throughput?
2. 2Build benchmark using PRODUCTION-shaped data volumes
3. 3Profile CPU + memory + allocations; identify top three offenders
4. 4Optimize one offender at a time; re-measure after each change
5. 5Check second-order effects: cache misses, GC pressure, lock contention
6. 6Record before/after numbers in the PR for future archaeologists
7. 7Sanitizer builds (ASan/UBSan) in CI before perf tuning
8. 8Profile with perf/VTune; watch cache lines and allocations

Common Pitfalls

• ▲Optimizing intuition while profile shows I/O waits
• ▲Micro-benchmarks on toy data lying about production

Commands

$ npx skills add aniruddhaadak80/skills --skill code-quality-performance-profiling-cpp

$ npx skills add aniruddhaadak80/skills --skill code-quality-performance-profiling-cpp -g

Tags