On Apple M4, Silverfir-nano beats Wasmtime's Cranelift on multiple benchmarks and goes head-to-head with V8 TurboFan, while staying ultra-compact and no_std-compatible.

All benchmarks on Apple M4 (MacBook Air, macOS 26). Silverfir vs Wasmtime Cranelift (optimizing JIT) and V8 TurboFan (Node.js 25.4).

See the charts above and full benchmark results for numbers.

• Competes with optimizing JITs — beats Cranelift on CoreMark and Lua benchmarks, beats V8 on STREAM and floating-point workloads
• Ultra-compact — the minimal no_std JIT binary is just 277 KB stripped, with zero runtime dependencies
• Full WebAssembly 2.0 — multi-value, reference types, bulk memory, multiple tables
• no_std — core library requires only alloc; runs on embedded and bare-metal

Silverfir-nano uses a micro-JIT that translates WebAssembly to native ARM64 machine code at function granularity. Key techniques:

• Native code generation — direct ARM64 emission, no interpreter dispatch overhead
• Register allocation — maps WebAssembly locals and stack to hardware registers
• Inline operations — arithmetic, comparisons, and memory access compiled to native instructions

If you need a pure interpreter without JIT (e.g., for platforms without executable memory), check out the interp branch:

git checkout interp

This branch includes the fast interpreter with instruction fusion and register caching, but no native code generation.

# Default build (micro-JIT)
cargo build --release

# Run a WASI program
./target/release/sf-nano-cli program.wasm [args...]

# Run benchmarks
python3 benchmarks/wasi/run_tests.py

# Run benchmarks in V8 (Node.js)
node benchmarks/wasi/run_v8.mjs

# Minimal no_std build (277 KB, no WASI, JIT only)
# Must be built standalone (excluded from workspace due to no_std)
cd sf-nano-cli-minimal && cargo run --release

Tested against the official WebAssembly spec testsuite:

• Multi-value returns
• Reference types (funcref, externref)
• Bulk memory operations
• Multiple tables
• Mutable globals import/export

MIT / Apache-2.0