Benchmarks

Performance characteristics of Keystone operations measured with Criterion.

Arithmetic Operations

Rounding Operations

Financial Calculations

Risk Metrics

Running Benchmarks

cd keystone
cargo bench

Specific Benchmark

cargo bench --bench arithmetic
cargo bench --bench financial
cargo bench --bench risk

With Baseline Comparison

# Save baseline
cargo bench -- --save-baseline main

# Compare to baseline
cargo bench -- --baseline main

Benchmark Code

Located in crates/precision-core/benches/:

#![allow(unused)]
fn main() {
use criterion::{black_box, criterion_group, criterion_main, Criterion};
use precision_core::Decimal;

fn bench_addition(c: &mut Criterion) {
    let a = Decimal::from(12345i64);
    let b = Decimal::from(67890i64);

    c.bench_function("decimal_add", |bencher| {
        bencher.iter(|| black_box(a).checked_add(black_box(b)))
    });
}

criterion_group!(benches, bench_addition);
criterion_main!(benches);
}

WASM Performance

WASM operations include JS-WASM boundary overhead:

WASM Binary Size

Optimization Settings

From Cargo.toml:

[profile.release]
opt-level = "z"      # Optimize for size
lto = "fat"          # Full link-time optimization
codegen-units = 1    # Better optimization
panic = "abort"      # Smaller binary
strip = true         # Remove symbols
overflow-checks = true  # Keep overflow checking

Memory Usage

Stack-only allocation for all operations.

Comparison with Alternatives

vs JavaScript Number

// JavaScript floating-point
0.1 + 0.2  // 0.30000000000000004

// Keystone
keystone.add("0.1", "0.2")  // "0.3"

Keystone is slower but guarantees correctness for financial calculations.

vs BigInt

JavaScript BigInt handles integers only. Keystone provides:

• Decimal precision up to 28 digits
• Native rounding modes
• Financial functions

vs decimal.js

Similar precision, but Keystone provides:

• Deterministic cross-platform results
• Smaller WASM bundle
• ZK-proof compatibility
• Rust native performance

Profiling

CPU Profiling

cargo bench --bench arithmetic -- --profile-time 10

Memory Profiling

cargo bench --bench arithmetic -- --plotting-backend disabled

CI Benchmarks

Benchmarks run on every PR to detect regressions:

- name: Run benchmarks
  run: cargo bench -- --noplot

- name: Compare with main
  run: |
    cargo bench -- --save-baseline pr
    git checkout main
    cargo bench -- --baseline pr