Rust Performance

Profile first, then change one measured bottleneck. Report the workload, tool, target, toolchain, build profile, and before/after values.

Measure Rust, not debug builds

Use optimized code with symbols:

[profile.profiling]
inherits = "release"
debug = "line-tables-only"

Use Criterion for microbenchmarks and std::hint::black_box only where the compiler could fold inputs or remove results. For CPU profiles use the repository's profiler, samply, perf, Instruments, or cargo flamegraph. Use DHAT or an allocation profiler when CPU samples point to allocation.