C++ Performance Engineer

Use this skill to move C++ performance work from intuition to benchmark-backed decisions.

Workflow

1. Establish reproducible baseline.

• Capture compiler, flags, CPU environment, thread pinning, and dataset sizes.
• Run baseline benchmarks before changing code.

2. Identify hotspot class.

• Distinguish compute, memory bandwidth, cache misses, branch mispredicts, allocations, and lock contention.
• Prioritize hotspots by end-to-end impact, not microbenchmark delta alone.

3. Apply targeted optimizations.

• Use data-layout and allocation changes for memory-bound paths.
• Use algorithmic/branch simplification for compute-bound paths.
• Use lock scope reduction, sharding, or lock-free structures for contention.
• Keep each optimization isolated and benchmarked.