Start with a function you already have.

1. Install

   npm install jz

2. Compile a function

   import jz from 'jz'
   
   const { exports } = jz`
     export const dist = (x, y) =>
       (x*x + y*y) ** 0.5
   `
   
   console.log(exports.dist(3, 4)) // 5

3. Pass whole arrays

   import jz from 'jz'
   
   const { exports, memory } = jz`
     export const sum = a => {
       let total = 0
       for (const x of a) total += x
       return total
     }
   `
   
   const values = new Float64Array([1, 2, 3])
   console.log(exports.sum(values)) // 6
   memory.reset()

   Value	Crosses
   number	directly
   string, array, typed array	copied
   plain object	by reference
   memory.Float64Array(n)	shared, no copy
   heap allocations	freed on return or by memory.reset()

4. Ship the .wasm

   npx jz kernel.js               # → kernel.wasm
   npx jz kernel.js -O3           # fastest; -Os smallest
   npx jz kernel.js --host wasi   # standalone WASI module
   npx jz kernel.js --why         # what kept a loop scalar

   import { instantiate } from 'jz/interop' // 19 kB gzip
   
   const wasm = await fetch('kernel.wasm').then(r => r.arrayBuffer())
   const { exports } = instantiate(wasm)

5. Check both

   Tests	run them against kernel.js and kernel.wasm
   Speed	time the whole call, transfer included
   Scalar loops	--why names the operation that blocked SIMD

What compiles

No GC, ASCII case, UTC dates: where it differs from JS.