This is an implementation of Bernard Widynski's counter-based PRNG as an ES6 module. The sampling is done in WASM. Relative to a pure-javascript version the WASM version is 5x-6x faster.
We use a cache and sample in blocks to reduce call overhead. When using
the cache, sampling values is on par with calls to Math.random
. It's
faster in v8, slower in webkit. Not because webkit runs WASM slowly, but
because webkit's Math.Random is 2x faster than v8. The WASM runs about the
same in all engines.
TODO: perf table
Squares uses a key and a counter. The key needs an appropriate bit pattern to ensure quality output. We provide a utility method to generate a suitable key from an integer seed.
This is a low-effort key generator. There are likely to be collisions, but probably not from nearby seed values. We found no collisions within groups of 4096 sequential values for the first 1e7 integers.
import { SquaresRNG, GenerateKey } from 'squares-rng';
const key = GenerateKey(123);
console.info(key.toString(16));
/**
bc8f2e15d5c8ef82
**/
const rng = new SquaresRNG(key)
for (let i = 0; i< 8; i++) {
console.info(rng.Next().toFixed(5));
}
/**
0.80028
0.46273
0.33456
0.44465
0.71814
0.54016
0.39462
0.66978
**/
TODO: API
https://arxiv.org/pdf/2004.06278
https://squaresrng.wixsite.com/rand