README.md

Benchmarks

This compares 4 projects:

• pure javascript
• pure rust
• wasm generated from rust
• rust with node using neon

There are be some differences in the datastructures between javascript and the port to rust, but I think these differences are idiomatic. Overall the algorithms are implemented in a similar way and behave equally, the tests were also ported. Everything is singlethreaded.

Test

for (var i = 0; i < 1000; i++) {
    wanakana.toKana('aiueosashisusesonaninunenokakikukeko')
    wanakana.toKana('AIUEOSASHISUSESONANINUNENOKAKIKUKEKO')
    wanakana.toHiragana('aiueosashisusesonaninunenokakikukeko')
    wanakana.toHiragana('アイウエオサシスセソナニヌネノカキクケコ')
    wanakana.toKatakana('aiueosashisusesonaninunenokakikukeko')
    wanakana.toKatakana('あいうえおさしすせそなにぬねのかきくけこ')
    wanakana.toRomaji('あいうえおさしすせそなにぬねのかきくけこ')
    wanakana.toRomaji('アイウエオサシスセソナニヌネノカキクケコ')
}

Results

Num Loops		javascript	rust	wasm (rust)	node with rust bindings
1_000	Wall Time[ms]	253	9	72	18
	Wall Time Relative	28x	1x	8x	2x
100_000	Wall Time[ms]	16149	1959	3783	3131
	Wall Time Relative	8,2x	1x	1,93x	1,59x
1_000_000	Wall Time[ms]	181879	21956	36885	31294
	Wall Time Relative	8,2x	1x	1,67x	1,42x

Conclusion

It can make sense to port performance sensitive code to wasm if total compatiblity is required (e.g. browser and node support, or no rust compiler feasible). More gains are possible when using native rust code or calling rust from nodejs.

Open

No analysis regarding memory analysis has been done yet.