mzHash32

[sergeevabc]

My JavaScript port of mzHash32 v0.5 (2026-01-05).
It works even with pre-ES6 engines like JScript.
Original author is Matteo Zapparoli aka @matteo65.

function mzHash32(bytes, seed) {
    if (!bytes || typeof bytes === "string" || bytes.length == null)
        throw Error("Byte array expected.");
    var M = 0x514430D1;
    var Z = 0x003C50DB;
    var hash = (M ^ seed) >>> 0;
    for (var i = 0, len = bytes.length; i < len; i++) {
        // Convert unsigned byte (0-255) to signed (-128 to 127)
        var signedByte = (bytes[i] << 24) >> 24;
        // Logical shift >> in C is >>> in JS
        var shiftCombo = ((hash << 1) ^ (hash >>> 1));
        hash = (Z * (signedByte ^ shiftCombo)) >>> 0;
    }
    // Output: uint32
    return hash;
}

[matteo65]

I'm testing a further version of mzHash32 (6 - the final one) that further improves the collision count, getting very close to the ideal universal hash function. It's only slightly different, but I'll report it in the issues.

[hjerabek]

That 3byte back and forth shift is a really creative way to handle the sign bit. Yet, assuming the input bytes are randomly distributed you have to perform 2 unnecessary shifts in 50% of the cases. And in the other 50% you could do a single bitwise OR with a bit mask that has all bits set to 1 in the places where the signed right shift would put them. Maybe I'm wrong, but I think you could save 75% time (minus the cost of performing a <128 check) when calculating the signed byte.