Wyn Benchmarks

March 2026 — Updated with honest numbers

All benchmarks run on Apple M4, wyn build (clang -O2). Run them yourself:

cd wyn
./wyn build tests/benchmarks/bench_suite.wyn && ./tests/benchmarks/bench_suite
./wyn build tests/benchmarks/spawn_benchmark.wyn && ./tests/benchmarks/spawn_benchmark

CPU: Fibonacci(35)

Backend	Time
wyn build (clang -O2)	~120ms
wyn run (TCC)	~32ms

Wyn emits C and compiles with the system C compiler. CPU-bound performance is determined by the C compiler's optimizer, not Wyn.

Concurrency

Metric	Value
Sequential spawn+await (1K)	~20 μs/op
Concurrent batch (100 + await_all)	~15 μs/op
Max concurrent spawns	~5-10K
Throughput via batching (10×1K)	~10K/s

Limitations

Each spawn allocates an 8MB virtual stack via mmap. Only ~4-8KB of physical memory is used for simple tasks, but the virtual address space limits concurrent spawns to ~5-10K on macOS. Higher concurrency requires smaller stacks (WYN_CORO_STACK=262144).

The pool allocator recycles destroyed coroutine memory via MAP_FIXED remap, making sequential spawn+await patterns efficient.

Build Speed

Command	Time
wyn build (clang)	~1.1s
wyn run (TCC)	~1.9s

Binary Size

A typical Wyn program compiles to ~50KB (statically linked runtime).

What We Don't Benchmark

We don't compare against Go, Rust, or Python because:

1. Wyn's runtime is much younger — comparisons would be misleading
2. CPU-bound code performance depends on the C compiler, not Wyn
3. Concurrency overhead (~20μs/spawn) is adequate for real workloads but not competitive with Go (~1μs/goroutine)

When Wyn's runtime matures, we'll publish honest cross-language comparisons with reproducible methodology.