NetMurmurHash3 1.0.0

NetMurmurHash3

MurmurHash3 for .NET 8 and .NET 10, with an API modeled on System.IO.Hashing (XxHash32, XxHash128).

Both derive from NonCryptographicHashAlgorithm, take an optional uint seed (default 0), and match the reference C++ output byte for byte. Not for cryptographic use.

Usage

using NetMurmurHash3;

// One-shot
byte[] hash = MurmurHash3x64_128.Hash(data);
uint h32 = MurmurHash3x86_32.HashToUInt32(data, seed: 42);
UInt128 h128 = MurmurHash3x64_128.HashToUInt128(data);

// Into a span, without allocating
bool ok = MurmurHash3x64_128.TryHash(data, destination, out int written);

// Incremental
MurmurHash3x86_32 hasher = new MurmurHash3x86_32();
hasher.Append(chunk1);
hasher.Append(chunk2);
uint result = hasher.GetCurrentHashAsUInt32();

Notes

• The total length is mixed in as 64 bits. The reference takes an int, so results match it for inputs up to int.MaxValue bytes; longer incremental inputs have no reference counterpart.
• 128-bit output is h1 then h2, each little-endian, as in the reference. HashToUInt128 returns h2 as the upper and h1 as the lower 64 bits.

Benchmarks

BenchmarkDotNet v0.15.8, Windows 11 (10.0.26200.9550/25H2/2025Update/HudsonValley2)
Intel Core i9-10885H CPU 2.40GHz, 1 CPU, 16 logical and 8 physical cores
.NET SDK 10.0.401
  [Host]     : .NET 10.0.12 (10.0.12, 10.0.1226.42308), X64 RyuJIT x86-64-v3
  DefaultJob : .NET 10.0.12 (10.0.12, 10.0.1226.42308), X64 RyuJIT x86-64-v3

Inputs are random bytes (fixed seed 42). Ratio is relative to NetMurmurHash3 in each group. XxHash32 and XxHash128 from System.IO.Hashing are not MurmurHash3; they are included as a reference point.

x86_32

JeremyEspresso/MurmurHash (v0.0.1) and darrenkopp/murmurhash-net as DarrenKopp. DarrenKopp goes through HashAlgorithm.ComputeHash and returns a byte[].

JeremyEspresso's code is compiled into the benchmark project rather than referenced as a package. Its NuGet package ships MurmurHash.dll, the same assembly name as darrenkopp's package, so the two can't be loaded in one process.

x64_128

History

This project is born because Claude accidentally created an implementation of the x64_128 version of the algorithm while I was working for another project, because the base class implementation NonCryptographicHashAlgorithm was needed.

I then decided to expand it and create a separated project for that, which turned out to be super fast.