saffron-crypto

SIMD-accelerated hashing routines over Vec4f lanes. Four parallel digests per instruction, no heap traffic in the hot loop.

Lane mixing

Vectors cross function boundaries as params and returns on every backend:

.rnx

import { Vec4f } from "@std/simd";


let block = new Vec4f(1.0, 2.0, 3.0, 4.0);
let key = Vec4f.splat(0.5);
let scaled = block * key;
print(scaled.dot(key));

Key schedule preview

.rnx

import { Vec4f } from "@std/simd";


let k = Vec4f.splat(1.0);
print(k.get(0), k.get(3));