MurMurHash3 .Net (C#) implementation

MurMurHash3.cs

/*
This code is public domain.

The MurmurHash3 algorithm was created by Austin Appleby and put into the public domain.  See http://code.google.com/p/smhasher/

This C# variant was authored by
Elliott B. Edwards and was placed into the public domain as a gist
Status...Working on verification (Test Suite)
Set up to run as a LinqPad (linqpad.net) script (thus the ".Dump()" call)
*/
public static class MurMurHash3
{
	//Change to suit your needs
	const uint seed = 144;

	public static int Hash(Stream stream)
	{
		const uint c1 = 0xcc9e2d51;
		const uint c2 = 0x1b873593;

		uint h1 = seed;
		uint k1 = 0;
		uint streamLength = 0;

		using (BinaryReader reader = new BinaryReader(stream))
		{
			byte[] chunk = reader.ReadBytes(4);
			while (chunk.Length > 0)
			{
				streamLength += (uint)chunk.Length;
				switch(chunk.Length)
				{
					case 4:
						/* Get four bytes from the input into an uint */
						k1 = (uint)
						   (chunk[0]
						  | chunk[1] << 8
						  | chunk[2] << 16
						  | chunk[3] << 24);

						/* bitmagic hash */
						k1 *= c1;
						k1 = rotl32(k1, 15);
						k1 *= c2;

						h1 ^= k1;
						h1 = rotl32(h1, 13);
						h1 = h1 * 5 + 0xe6546b64;
						break;
					case 3:
						k1 = (uint)
						   (chunk[0]
						  | chunk[1] << 8
						  | chunk[2] << 16);
						k1 *= c1;
						k1 = rotl32(k1, 15);
						k1 *= c2;
						h1 ^= k1;
						break;
					case 2:
						k1 = (uint)
						   (chunk[0]
						  | chunk[1] << 8);
						k1 *= c1;
						k1 = rotl32(k1, 15);
						k1 *= c2;
						h1 ^= k1;
						break;
					case 1:
						k1 = (uint)(chunk[0]);
						k1 *= c1;
						k1 = rotl32(k1, 15);
						k1 *= c2;
						h1 ^= k1;
						break;
					
				}
				chunk = reader.ReadBytes(4);
			}
		}
				
		// finalization, magic chants to wrap it all up
		h1 ^= streamLength;
		h1 = fmix(h1);

		unchecked //ignore overflow
		{
			return (int)h1;
		}
	}
	
	private static uint rotl32(uint x, byte r)
	{
		return (x << r) | (x >> (32 - r));
	}

	private static uint fmix(uint h)
	{
		h ^= h >> 16;
		h *= 0x85ebca6b;
		h ^= h >> 13;
		h *= 0xc2b2ae35;
		h ^= h >> 16;
		return h;
	}
}

void Main()
{
	Encoding encoding = new UTF8Encoding();
	byte[] input = encoding.GetBytes("The quick brown fox jumps over the lazy dog");
	using (MemoryStream stream = new MemoryStream(input))
	{
		int hash = MurMurHash3.Hash(stream);
		new {Hash=hash, Bytes = BitConverter.GetBytes(hash) }.Dump("Result");
	}
}

Just confirmed with @sapbucket, same result when using with UTF8. DONT USE THIS IMPLEMENTATION

I ended up using this one https://github.com/jitbit/MurmurHash.net

Hmm... MurMurHash3 looks fine to me. I can't reproduce @sapbucket's problem (with .NET 4.5.2, Windows 10, 64-bit) using any default encodings from System.Text.Encoding or new UTF8Encoding()

public static int GetHash(string uniqueString) // "0,0,0,15,7" and "0,3,13,25,12" return different values
{
  byte[] input = Encoding.UTF8.GetBytes(uniqueString); // Values stay different for System.Text.Encoding.Unicode, UTF8, etc.
  using (var stream = new MemoryStream(input))
  return MurMurHash3.Hash(stream);
}

@sapbucket

It was my understanding that each unique byte[] would produce a unique hash.

This is incorrect understanding of hash as a concept. What you see is "hash collision" and it is NORMAL.

Hash is a function that maps all possible byte sequences of arbitrary length into a hash space - 32 bit for this. There are WAY more (countable but infinite) number of possible combinations of bytes that you can feed into hash functions but hash can output only 2^32 numbers. By pidgin hole principle - some inputs will be mapped to the same hash value.
if hash(x) = hash(y) => it doesn't mean that x=y.
Since this hash is a deterministic algorithm (for same input it always gives same output) - The only guarantee that hash give you is that:
if hash(x) <> hash(y) => it means that x <> y

You can think of hash as a guy that assigns a bucket number to the given byte string input.
Where hash space (all possible values, 2^32 here) - is total number of buckets

If you feed small number of input samples and your hash function is "good" (regarding random uniform distribution) - you can hope that all your inputs will get unique hash values assigned. But there is a chance that some buckets (hash values) will be associated with two or even more inputs. The "better" the hash - the lower such chances. Even with SHA1 and other crypto hashes this "hash collisions" are possible.

Imagine that your test set has more than 2^32 input samples - how hash function would be able to give you unique values for each of the input sample?