WASM promises faster code execution than Javascript, but it has a weakness: moving data between WASM-land and Javascript-land is not free. Every time we generate a path, flow field, etc., we incur that boundary cost.
I ran some tests to find the relative CPU usage for passing these data types across the WASM boundary:
CPU usage per 1,000 iterations
| Length: | 1 | 100 | 10,000 |
|---|---|---|---|
| Integer | ~0.03 CPU | - | - |
| Array | ~1 CPU | ~1 CPU | ~23 CPU |
| String | ~0.5 CPU | ~1.75 CPU | ~110 CPU |
For reference, a typical cost matrix is an array of 2,500 items (50x50).
Sometimes, we don’t need to pass all the data back and forth: we can return a Rust struct, and wasm-bindgen will generate a class that tracks the reference for us. This will be ideal for most scenarios. However, users may need to clean up these objects manually.