vfxreview/RenderWorker/VFXReviewWorker/XxHash64.cs

namespace VFXReviewWorker;

/// <summary>
/// xxHash64 (§8.2 `checksumAlgo: "xxh64"`). Implemented here rather than taken
/// from System.IO.Hashing so the worker stays a dependency-free single-file
/// publish for offline workstation installs. Verified against the reference
/// test vectors in XxHash64Tests.
/// </summary>
public static class XxHash64
{
    private const ulong Prime1 = 11400714785074694791UL;
    private const ulong Prime2 = 14029467366897019727UL;
    private const ulong Prime3 = 1609587929392839161UL;
    private const ulong Prime4 = 9650029242287828579UL;
    private const ulong Prime5 = 2870177450012600261UL;

    public static ulong Hash(ReadOnlySpan<byte> data, ulong seed = 0)
    {
        var state = new State(seed);
        state.Update(data);
        return state.Digest();
    }

    public static string HashFile(string path, int bufferSize = 1 << 20)
    {
        var state = new State(0);
        using var fs = new FileStream(path, FileMode.Open, FileAccess.Read, FileShare.ReadWrite, bufferSize, FileOptions.SequentialScan);
        var buffer = new byte[bufferSize];
        int read;
        while ((read = fs.Read(buffer, 0, buffer.Length)) > 0)
        {
            state.Update(new ReadOnlySpan<byte>(buffer, 0, read));
        }
        return state.Digest().ToString("x16");
    }

    /// <summary>Streaming state so multi-GB sequences hash without being held in memory.</summary>
    public struct State
    {
        private ulong _v1, _v2, _v3, _v4;
        private readonly ulong _seed;
        private ulong _length;
        private readonly byte[] _buffer;
        private int _bufferLen;

        public State(ulong seed)
        {
            _seed = seed;
            _v1 = seed + Prime1 + Prime2;
            _v2 = seed + Prime2;
            _v3 = seed;
            _v4 = seed - Prime1;
            _length = 0;
            _buffer = new byte[32];
            _bufferLen = 0;
        }

        public void Update(ReadOnlySpan<byte> data)
        {
            _length += (ulong)data.Length;

            if (_bufferLen > 0)
            {
                var need = Math.Min(32 - _bufferLen, data.Length);
                data[..need].CopyTo(_buffer.AsSpan(_bufferLen));
                _bufferLen += need;
                data = data[need..];
                if (_bufferLen < 32) return;
                ProcessStripe(_buffer);
                _bufferLen = 0;
            }

            while (data.Length >= 32)
            {
                ProcessStripe(data[..32]);
                data = data[32..];
            }

            if (data.Length > 0)
            {
                data.CopyTo(_buffer.AsSpan(0));
                _bufferLen = data.Length;
            }
        }

        private void ProcessStripe(ReadOnlySpan<byte> stripe)
        {
            _v1 = Round(_v1, BitConverter.ToUInt64(stripe[..8]));
            _v2 = Round(_v2, BitConverter.ToUInt64(stripe.Slice(8, 8)));
            _v3 = Round(_v3, BitConverter.ToUInt64(stripe.Slice(16, 8)));
            _v4 = Round(_v4, BitConverter.ToUInt64(stripe.Slice(24, 8)));
        }

        public ulong Digest()
        {
            ulong h;
            if (_length >= 32)
            {
                h = RotateLeft(_v1, 1) + RotateLeft(_v2, 7) + RotateLeft(_v3, 12) + RotateLeft(_v4, 18);
                h = MergeRound(h, _v1);
                h = MergeRound(h, _v2);
                h = MergeRound(h, _v3);
                h = MergeRound(h, _v4);
            }
            else
            {
                h = _seed + Prime5;
            }

            h += _length;

            var tail = new ReadOnlySpan<byte>(_buffer, 0, _bufferLen);
            int i = 0;
            while (i + 8 <= tail.Length)
            {
                h ^= Round(0, BitConverter.ToUInt64(tail.Slice(i, 8)));
                h = RotateLeft(h, 27) * Prime1 + Prime4;
                i += 8;
            }
            if (i + 4 <= tail.Length)
            {
                h ^= BitConverter.ToUInt32(tail.Slice(i, 4)) * Prime1;
                h = RotateLeft(h, 23) * Prime2 + Prime3;
                i += 4;
            }
            while (i < tail.Length)
            {
                h ^= tail[i] * Prime5;
                h = RotateLeft(h, 11) * Prime1;
                i++;
            }

            h ^= h >> 33;
            h *= Prime2;
            h ^= h >> 29;
            h *= Prime3;
            h ^= h >> 32;
            return h;
        }

        private static ulong Round(ulong acc, ulong input)
        {
            acc += input * Prime2;
            acc = RotateLeft(acc, 31);
            acc *= Prime1;
            return acc;
        }

        private static ulong MergeRound(ulong acc, ulong val)
        {
            acc ^= Round(0, val);
            acc = acc * Prime1 + Prime4;
            return acc;
        }

        private static ulong RotateLeft(ulong value, int bits) => (value << bits) | (value >> (64 - bits));
    }
}