private static uint QueueRound(uint hash, uint queuedValue)
{
unchecked
{
return(BitOperations.RotateLeft(hash + queuedValue * Prime3, 17) * Prime4);
}
}
private static uint Round(uint hash, uint input)
{
unchecked
{
return(BitOperations.RotateLeft(hash + input * Prime2, 13) * Prime1);
}
}
private static uint MixState(uint v1, uint v2, uint v3, uint v4)
{
unchecked
{
return(BitOperations.RotateLeft(v1, 1) + BitOperations.RotateLeft(v2, 7) + BitOperations.RotateLeft(v3, 12) + BitOperations.RotateLeft(v4, 18));
}
}
// Use this if and only if 'Denial of Service' attacks are not a concern (i.e. never used for free-form user input),
// or are otherwise mitigated
internal unsafe int GetNonRandomizedHashCode()
{
fixed(char *src = &_firstChar)
{
Debug.Assert(src[this.Length] == '\0', "src[this.Length] == '\\0'");
Debug.Assert(((int)src) % 4 == 0, "Managed string should start at 4 bytes boundary");
uint hash1 = (5381 << 16) + 5381;
uint hash2 = hash1;
uint *ptr = (uint *)src;
int length = this.Length;
while (length > 2)
{
length -= 4;
// Where length is 4n-1 (e.g. 3,7,11,15,19) this additionally consumes the null terminator
hash1 = (BitOperations.RotateLeft(hash1, 5) + hash1) ^ ptr[0];
hash2 = (BitOperations.RotateLeft(hash2, 5) + hash2) ^ ptr[1];
ptr += 2;
}
if (length > 0)
{
// Where length is 4n-3 (e.g. 1,5,9,13,17) this additionally consumes the null terminator
hash2 = (BitOperations.RotateLeft(hash2, 5) + hash2) ^ ptr[0];
}
return((int)(hash1 + (hash2 * 1566083941)));
}
}
//public override void NextBytes(byte[] buffer) => NextBytes((Span<byte>)buffer);
public override unsafe void NextBytes(byte[] buffer)
{
ulong s0 = _s0, s1 = _s1, s2 = _s2, s3 = _s3;
int pos = 0;
int ulongs = buffer.Length / sizeof(ulong) * sizeof(ulong);
//while (buffer.Length >= sizeof(ulong))
for (; pos < ulongs; pos += sizeof(ulong))
{
/*Unsafe.WriteUnaligned(
* ref MemoryMarshal.GetReference(buffer),
* BitOperations.RotateLeft(s1 * 5, 7) * 9);*/
Unsafe.As <byte, ulong>(ref buffer[pos]) =
BitOperations.RotateLeft(s1 * 5, 7) * 9;
// Update PRNG state.
ulong t = s1 << 17;
s2 ^= s0;
s3 ^= s1;
s1 ^= s2;
s0 ^= s3;
s2 ^= t;
s3 = BitOperations.RotateLeft(s3, 45);
//buffer = buffer.Slice(sizeof(ulong));
}
//if (!buffer.IsEmpty)
if (pos < buffer.Length)
{
ulong next = BitOperations.RotateLeft(s1 * 5, 7) * 9;
byte *remainingBytes = (byte *)&next;
Debug.Assert(buffer.Length < sizeof(ulong));
//for (int i = 0; i < buffer.Length; i++)
for (int i = 0; pos < buffer.Length; i++, pos++)
{
//buffer[i] = remainingBytes[i];
buffer[pos] = remainingBytes[i];
}
// Update PRNG state.
ulong t = s1 << 17;
s2 ^= s0;
s3 ^= s1;
s1 ^= s2;
s0 ^= s3;
s2 ^= t;
s3 = BitOperations.RotateLeft(s3, 45);
}
_s0 = s0;
_s1 = s1;
_s2 = s2;
_s3 = s3;
}
internal unsafe int GetNonRandomizedHashCodeOrdinalIgnoreCase()
{
uint hash1 = (5381 << 16) + 5381;
uint hash2 = hash1;
fixed(char *src = &_firstChar)
{
Debug.Assert(src[this.Length] == '\0', "src[this.Length] == '\\0'");
Debug.Assert(((int)src) % 4 == 0, "Managed string should start at 4 bytes boundary");
uint *ptr = (uint *)src;
int length = this.Length;
// We "normalize to lowercase" every char by ORing with 0x0020. This casts
// a very wide net because it will change, e.g., '^' to '~'. But that should
// be ok because we expect this to be very rare in practice.
const uint NormalizeToLowercase = 0x0020_0020u; // valid both for big-endian and for little-endian
while (length > 2)
{
uint p0 = ptr[0];
uint p1 = ptr[1];
if (!Utf16Utility.AllCharsInUInt32AreAscii(p0 | p1))
{
goto NotAscii;
}
length -= 4;
// Where length is 4n-1 (e.g. 3,7,11,15,19) this additionally consumes the null terminator
hash1 = (BitOperations.RotateLeft(hash1, 5) + hash1) ^ (p0 | NormalizeToLowercase);
hash2 = (BitOperations.RotateLeft(hash2, 5) + hash2) ^ (p1 | NormalizeToLowercase);
ptr += 2;
}
if (length > 0)
{
uint p0 = ptr[0];
if (!Utf16Utility.AllCharsInUInt32AreAscii(p0))
{
goto NotAscii;
}
// Where length is 4n-3 (e.g. 1,5,9,13,17) this additionally consumes the null terminator
hash2 = (BitOperations.RotateLeft(hash2, 5) + hash2) ^ (p0 | NormalizeToLowercase);
}
}
return((int)(hash1 + (hash2 * 1566083941)));
NotAscii:
return(GetNonRandomizedHashCodeOrdinalIgnoreCaseSlow(this));
/// <summary>
/// Called when pos reaches 64.
/// </summary>
private void Drain()
{
for (int i = 16; i != 80; i++)
{
_w[i] = BitOperations.RotateLeft(_w[i - 3] ^ _w[i - 8] ^ _w[i - 14] ^ _w[i - 16], 1);
}
uint a = _w[80];
uint b = _w[81];
uint c = _w[82];
uint d = _w[83];
uint e = _w[84];
for (int i = 0; i != 20; i++)
{
const uint k = 0x5A827999;
uint f = (b & c) | ((~b) & d);
uint temp = BitOperations.RotateLeft(a, 5) + f + e + k + _w[i]; e = d; d = c; c = BitOperations.RotateLeft(b, 30); b = a; a = temp;
}
for (int i = 20; i != 40; i++)
{
uint f = b ^ c ^ d;
const uint k = 0x6ED9EBA1;
uint temp = BitOperations.RotateLeft(a, 5) + f + e + k + _w[i]; e = d; d = c; c = BitOperations.RotateLeft(b, 30); b = a; a = temp;
}
for (int i = 40; i != 60; i++)
{
uint f = (b & c) | (b & d) | (c & d);
const uint k = 0x8F1BBCDC;
uint temp = BitOperations.RotateLeft(a, 5) + f + e + k + _w[i]; e = d; d = c; c = BitOperations.RotateLeft(b, 30); b = a; a = temp;
}
for (int i = 60; i != 80; i++)
{
uint f = b ^ c ^ d;
const uint k = 0xCA62C1D6;
uint temp = BitOperations.RotateLeft(a, 5) + f + e + k + _w[i]; e = d; d = c; c = BitOperations.RotateLeft(b, 30); b = a; a = temp;
}
_w[80] += a;
_w[81] += b;
_w[82] += c;
_w[83] += d;
_w[84] += e;
_length += 512; // 64 bytes == 512 bits
_pos = 0;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)] // small-ish hot path used by a handful of "next" methods
internal uint NextUInt32()
{
uint result = BitOperations.RotateLeft(_s1 * 5, 7) * 9;
uint t = _s1 << 9;
_s2 ^= _s0;
_s3 ^= _s1;
_s1 ^= _s2;
_s0 ^= _s3;
_s2 ^= t;
_s3 = BitOperations.RotateLeft(_s3, 11);
return(result);
}
[MethodImpl(MethodImplOptions.AggressiveInlining)] // small-ish hot path used by a handful of "next" methods
internal ulong NextUInt64()
{
ulong result = BitOperations.RotateLeft(_s1 * 5, 7) * 9;
ulong t = _s1 << 17;
_s2 ^= _s0;
_s3 ^= _s1;
_s1 ^= _s2;
_s0 ^= _s3;
_s2 ^= t;
_s3 = BitOperations.RotateLeft(_s3, 45);
return(result);
}
public override unsafe void NextBytes(Span <byte> buffer)
{
ulong s0 = _s0, s1 = _s1, s2 = _s2, s3 = _s3;
while (buffer.Length >= sizeof(ulong))
{
Unsafe.WriteUnaligned(
ref MemoryMarshal.GetReference(buffer),
BitOperations.RotateLeft(s1 * 5, 7) * 9);
// Update PRNG state.
ulong t = s1 << 17;
s2 ^= s0;
s3 ^= s1;
s1 ^= s2;
s0 ^= s3;
s2 ^= t;
s3 = BitOperations.RotateLeft(s3, 45);
buffer = buffer.Slice(sizeof(ulong));
}
if (!buffer.IsEmpty)
{
ulong next = BitOperations.RotateLeft(s1 * 5, 7) * 9;
byte *remainingBytes = (byte *)&next;
Debug.Assert(buffer.Length < sizeof(ulong));
for (int i = 0; i < buffer.Length; i++)
{
buffer[i] = remainingBytes[i];
}
// Update PRNG state.
ulong t = s1 << 17;
s2 ^= s0;
s3 ^= s1;
s1 ^= s2;
s0 ^= s3;
s2 ^= t;
s3 = BitOperations.RotateLeft(s3, 45);
}
_s0 = s0;
_s1 = s1;
_s2 = s2;
_s3 = s3;
}
static int GetNonRandomizedHashCodeOrdinalIgnoreCaseSlow(string str)
{
int length = str.Length;
char[]? borrowedArr = null;
// Important: leave an additional space for '\0'
Span <char> scratch = (uint)length < 64 ?
stackalloc char[64] : (borrowedArr = ArrayPool <char> .Shared.Rent(length + 1));
int charsWritten = System.Globalization.Ordinal.ToUpperOrdinal(str, scratch);
Debug.Assert(charsWritten == length);
scratch[length] = '\0';
const uint NormalizeToLowercase = 0x0020_0020u;
uint hash1 = (5381 << 16) + 5381;
uint hash2 = hash1;
// Duplicate the main loop, can be removed once JIT gets "Loop Unswitching" optimization
fixed(char *src = scratch)
{
uint *ptr = (uint *)src;
while (length > 2)
{
length -= 4;
hash1 = (BitOperations.RotateLeft(hash1, 5) + hash1) ^ (ptr[0] | NormalizeToLowercase);
hash2 = (BitOperations.RotateLeft(hash2, 5) + hash2) ^ (ptr[1] | NormalizeToLowercase);
ptr += 2;
}
if (length > 0)
{
hash2 = (BitOperations.RotateLeft(hash2, 5) + hash2) ^ (ptr[0] | NormalizeToLowercase);
}
}
if (borrowedArr != null)
{
ArrayPool <char> .Shared.Return(borrowedArr);
}
return((int)(hash1 + (hash2 * 1566083941)));
}
private static void Block(ref uint rp0, ref uint rp1)
{
uint p0 = rp0;
uint p1 = rp1;
p1 ^= p0;
p0 = BitOperations.RotateLeft(p0, 20);
p0 += p1;
p1 = BitOperations.RotateLeft(p1, 9);
p1 ^= p0;
p0 = BitOperations.RotateLeft(p0, 27);
p0 += p1;
p1 = BitOperations.RotateLeft(p1, 19);
rp0 = p0;
rp1 = p1;
}
private static void Block(ref uint rp0, ref uint rp1)
{
// Intrinsified in mono interpreter
uint p0 = rp0;
uint p1 = rp1;
p1 ^= p0;
p0 = BitOperations.RotateLeft(p0, 20);
p0 += p1;
p1 = BitOperations.RotateLeft(p1, 9);
p1 ^= p0;
p0 = BitOperations.RotateLeft(p0, 27);
p0 += p1;
p1 = BitOperations.RotateLeft(p1, 19);
rp0 = p0;
rp1 = p1;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)] // small-ish hot path used by a handful of "next" methods
internal uint NextUInt32()
{
uint s0 = _s0, s1 = _s1, s2 = _s2, s3 = _s3;
uint result = BitOperations.RotateLeft(s1 * 5, 7) * 9;
uint t = s1 << 9;
s2 ^= s0;
s3 ^= s1;
s1 ^= s2;
s0 ^= s3;
s2 ^= t;
s3 = BitOperations.RotateLeft(s3, 11);
_s0 = s0;
_s1 = s1;
_s2 = s2;
_s3 = s3;
return(result);
}
[MethodImpl(MethodImplOptions.AggressiveInlining)] // small-ish hot path used by a handful of "next" methods
internal ulong NextUInt64()
{
ulong s0 = _s0, s1 = _s1, s2 = _s2, s3 = _s3;
ulong result = BitOperations.RotateLeft(s1 * 5, 7) * 9;
ulong t = s1 << 17;
s2 ^= s0;
s3 ^= s1;
s1 ^= s2;
s0 ^= s3;
s2 ^= t;
s3 = BitOperations.RotateLeft(s3, 45);
_s0 = s0;
_s1 = s1;
_s2 = s2;
_s3 = s3;
return(result);
}
static int IBinaryInteger <int> .RotateLeft(int value, int rotateAmount) => (int)BitOperations.RotateLeft((uint)value, rotateAmount);
static ulong IBinaryInteger <ulong> .RotateLeft(ulong value, int rotateAmount) => BitOperations.RotateLeft(value, rotateAmount);
/// <inheritdoc cref="IBinaryInteger{TSelf}.RotateLeft(TSelf, int)" />
public static long RotateLeft(long value, int rotateAmount) => (long)BitOperations.RotateLeft((ulong)value, rotateAmount);
static uint IBinaryInteger <uint> .RotateLeft(uint value, int rotateAmount) => BitOperations.RotateLeft(value, rotateAmount);
private static uint QueueRound(uint hash, uint queuedValue)
{
return(BitOperations.RotateLeft(hash + (queuedValue * _prime3), 17) * _prime4);
}
public static int ComputeHash32(ref byte data, uint count, uint p0, uint p1)
{
// Control flow of this method generally flows top-to-bottom, trying to
// minimize the number of branches taken for large (>= 8 bytes, 4 chars) inputs.
// If small inputs (< 8 bytes, 4 chars) are given, this jumps to a "small inputs"
// handler at the end of the method.
if (count < 8)
{
// We can't run the main loop, but we might still have 4 or more bytes available to us.
// If so, jump to the 4 .. 7 bytes logic immediately after the main loop.
if (count >= 4)
{
goto Between4And7BytesRemain;
}
else
{
goto InputTooSmallToEnterMainLoop;
}
}
// Main loop - read 8 bytes at a time.
// The block function is unrolled 2x in this loop.
uint loopCount = count / 8;
Debug.Assert(loopCount > 0, "Shouldn't reach this code path for small inputs.");
do
{
// Most x86 processors have two dispatch ports for reads, so we can read 2x 32-bit
// values in parallel. We opt for this instead of a single 64-bit read since the
// typical use case for Marvin32 is computing String hash codes, and the particular
// layout of String instances means the starting data is never 8-byte aligned when
// running in a 64-bit process.
p0 += Unsafe.ReadUnaligned <uint>(ref data);
uint nextUInt32 = Unsafe.ReadUnaligned <uint>(ref Unsafe.AddByteOffset(ref data, 4));
// One block round for each of the 32-bit integers we just read, 2x rounds total.
Block(ref p0, ref p1);
p0 += nextUInt32;
Block(ref p0, ref p1);
// Bump the data reference pointer and decrement the loop count.
// Decrementing by 1 every time and comparing against zero allows the JIT to produce
// better codegen compared to a standard 'for' loop with an incrementing counter.
// Requires https://github.com/dotnet/coreclr/issues/7566 to be addressed first
// before we can realize the full benefits of this.
data = ref Unsafe.AddByteOffset(ref data, 8);
} while (--loopCount > 0);
// n.b. We've not been updating the original 'count' parameter, so its actual value is
// still the original data length. However, we can still rely on its least significant
// 3 bits to tell us how much data remains (0 .. 7 bytes) after the loop above is
// completed.
if ((count & 0b_0100) == 0)
{
goto DoFinalPartialRead;
}
Between4And7BytesRemain:
// If after finishing the main loop we still have 4 or more leftover bytes, or if we had
// 4 .. 7 bytes to begin with and couldn't enter the loop in the first place, we need to
// consume 4 bytes immediately and send them through one round of the block function.
Debug.Assert(count >= 4, "Only should've gotten here if the original count was >= 4.");
p0 += Unsafe.ReadUnaligned <uint>(ref data);
Block(ref p0, ref p1);
DoFinalPartialRead:
// Finally, we have 0 .. 3 bytes leftover. Since we know the original data length was at
// least 4 bytes (smaller lengths are handled at the end of this routine), we can safely
// read the 4 bytes at the end of the buffer without reading past the beginning of the
// original buffer. This necessarily means the data we're about to read will overlap with
// some data we've already processed, but we can handle that below.
Debug.Assert(count >= 4, "Only should've gotten here if the original count was >= 4.");
// Read the last 4 bytes of the buffer.
uint partialResult = Unsafe.ReadUnaligned <uint>(ref Unsafe.Add(ref Unsafe.AddByteOffset(ref data, (nuint)count & 7), -4));
// The 'partialResult' local above contains any data we have yet to read, plus some number
// of bytes which we've already read from the buffer. An example of this is given below
// for little-endian architectures. In this table, AA BB CC are the bytes which we still
// need to consume, and ## are bytes which we want to throw away since we've already
// consumed them as part of a previous read.
//
// (partialResult contains) (we want it to contain)
// count mod 4 = 0 -> [ ## ## ## ## | ] -> 0x####_#### -> 0x0000_0080
// count mod 4 = 1 -> [ ## ## ## ## | AA ] -> 0xAA##_#### -> 0x0000_80AA
// count mod 4 = 2 -> [ ## ## ## ## | AA BB ] -> 0xBBAA_#### -> 0x0080_BBAA
// count mod 4 = 3 -> [ ## ## ## ## | AA BB CC ] -> 0xCCBB_AA## -> 0x80CC_BBAA
count = ~count << 3;
if (BitConverter.IsLittleEndian)
{
partialResult >>= 8; // make some room for the 0x80 byte
partialResult |= 0x8000_0000u; // put the 0x80 byte at the beginning
partialResult >>= (int)count & 0x1F; // shift out all previously consumed bytes
}
else
{
partialResult <<= 8; // make some room for the 0x80 byte
partialResult |= 0x80u; // put the 0x80 byte at the end
partialResult <<= (int)count & 0x1F; // shift out all previously consumed bytes
}
DoFinalRoundsAndReturn:
// Now that we've computed the final partial result, merge it in and run two rounds of
// the block function to finish out the Marvin algorithm.
p0 += partialResult;
Block(ref p0, ref p1);
Block(ref p0, ref p1);
return((int)(p1 ^ p0));
InputTooSmallToEnterMainLoop:
// We had only 0 .. 3 bytes to begin with, so we can't perform any 32-bit reads.
// This means that we're going to be building up the final result right away and
// will only ever run two rounds total of the block function. Let's initialize
// the partial result to "no data".
if (BitConverter.IsLittleEndian)
{
partialResult = 0x80u;
}
else
{
partialResult = 0x80000000u;
}
if ((count & 0b_0001) != 0)
{
// If the buffer is 1 or 3 bytes in length, let's read a single byte now
// and merge it into our partial result. This will result in partialResult
// having one of the two values below, where AA BB CC are the buffer bytes.
//
// (little-endian / big-endian)
// [ AA ] -> 0x0000_80AA / 0xAA80_0000
// [ AA BB CC ] -> 0x0000_80CC / 0xCC80_0000
partialResult = Unsafe.AddByteOffset(ref data, (nuint)count & 2);
if (BitConverter.IsLittleEndian)
{
partialResult |= 0x8000;
}
else
{
partialResult <<= 24;
partialResult |= 0x800000u;
}
}
if ((count & 0b_0010) != 0)
{
// If the buffer is 2 or 3 bytes in length, let's read a single ushort now
// and merge it into the partial result. This will result in partialResult
// having one of the two values below, where AA BB CC are the buffer bytes.
//
// (little-endian / big-endian)
// [ AA BB ] -> 0x0080_BBAA / 0xAABB_8000
// [ AA BB CC ] -> 0x80CC_BBAA / 0xAABB_CC80 (carried over from above)
if (BitConverter.IsLittleEndian)
{
partialResult <<= 16;
partialResult |= (uint)Unsafe.ReadUnaligned <ushort>(ref data);
}
else
{
partialResult |= (uint)Unsafe.ReadUnaligned <ushort>(ref data);
partialResult = BitOperations.RotateLeft(partialResult, 16);
}
}
// Everything is consumed! Go perform the final rounds and return.
goto DoFinalRoundsAndReturn;
}
/// <inheritdoc cref="IBinaryInteger{TSelf}.RotateLeft(TSelf, int)" />
public static uint RotateLeft(uint value, int rotateAmount) => BitOperations.RotateLeft(value, rotateAmount);
private static uint Round(uint hash, uint input)
{
return(BitOperations.RotateLeft(hash + (input * _prime2), 13) * _prime1);
}
/// <inheritdoc cref="IBinaryInteger{TSelf}.RotateLeft(TSelf, int)" />
public static ulong RotateLeft(ulong value, int rotateAmount) => BitOperations.RotateLeft(value, rotateAmount);
/// <inheritdoc cref="IBinaryInteger{TSelf}.RotateLeft(TSelf, int)" />
public static int RotateLeft(int value, int rotateAmount) => (int)BitOperations.RotateLeft((uint)value, rotateAmount);
static long IBinaryInteger <long> .RotateLeft(long value, int rotateAmount) => (long)BitOperations.RotateLeft((ulong)value, rotateAmount);