Benchmarks & profiling (repository)¶

This page is for contributors working on Incan’s performance: running the repository benchmark suite, adding new benchmarks, and profiling the compiler.

If you want to speed up your program, see Performance (how-to).

Benchmark suite¶

The repository includes a benchmark suite that compares generated binaries against Rust and Python baselines.

Available benchmarks¶

Category	Benchmark	Description
Compute	`fib`	Iterative Fibonacci (N=1,000,000)
Compute	`collatz`	Collatz sequence (1,000,000 numbers)
Compute	`gcd`	GCD of 10,000,000 pairs
Compute	`mandelbrot`	2000×2000 escape iterations
Compute	`nbody`	N-body simulation (500,000 steps)
Compute	`primes`	Sieve up to 50,000,000
Sorting	`quicksort`	In-place sort (1M integers)
Sorting	`mergesort`	Merge sort (1M integers)

Running benchmarks¶

# Prerequisites (macOS)
brew install hyperfine jq bc

# Build the compiler
cargo build --release

# Run all benchmarks
make benchmarks

# Or run the benchmark runner directly
./benchmarks/run_all.sh

Running an individual benchmark¶

cd benchmarks/compute/fib

# Build Incan version
../../../target/release/incan build fib.incn
cp ../../../target/incan/fib/target/release/fib ./fib_incan

# Build Rust baseline
rustc -O fib.rs -o fib_rust

# Compare
hyperfine --warmup 2 --min-runs 5 \
  './fib_incan' \
  './fib_rust' \
  'python3 fib.py'

Results¶

For current measured results, see benchmarks/results/results.md in the repository.

Profiling¶

Profiling generated code (your benchmark/program)¶

# Build (incan always builds the generated project in release mode today)
incan build myprogram.incn

# Profile with Instruments (macOS)
xcrun xctrace record --template "Time Profiler" \
  --launch ./target/incan/myprogram/target/release/myprogram

# Profile with perf (Linux)
perf record ./target/incan/myprogram/target/release/myprogram
perf report

Profiling the compiler¶

# Profile compilation itself
cargo flamegraph --bin incan -- build examples/advanced/async_await.incn

# Or with samply (macOS)
samply record ./target/release/incan build large_program.incn

Adding a new benchmark¶

Create a directory under benchmarks/compute/ or benchmarks/sorting/.
Add three implementations: name.incn, name.rs, name.py.
Each should print a single result line for verification.
Run ./benchmarks/run_all.sh to include it in the suite.