// This method has different signature for x64 and other platforms and is done for performance reasons.
private static void Memmove(ref byte dest, ref byte src, nuint len)
{
// P/Invoke into the native version when the buffers are overlapping.
if (((nuint)Unsafe.ByteOffset(ref src, ref dest) < len) || ((nuint)Unsafe.ByteOffset(ref dest, ref src) < len))
{
goto BuffersOverlap;
}
// Use "(IntPtr)(nint)len" to avoid overflow checking on the explicit cast to IntPtr
ref byte srcEnd = ref Unsafe.Add(ref src, (IntPtr)(nint)len);
public unsafe static int GetIndexOfFirstInvalidUtf8Sequence(ReadOnlySpan <byte> utf8Data, out bool isAscii)
{
fixed(byte *pUtf8Data = &MemoryMarshal.GetReference(utf8Data))
{
byte *pFirstInvalidByte = GetPointerToFirstInvalidByte(pUtf8Data, utf8Data.Length, out int utf16CodeUnitCountAdjustment, out _);
int index = (int)(void *)Unsafe.ByteOffset(ref *pUtf8Data, ref *pFirstInvalidByte);
isAscii = (utf16CodeUnitCountAdjustment == 0); // If UTF-16 char count == UTF-8 byte count, it's ASCII.
return((index < utf8Data.Length) ? index : -1);
}
}
private static IntPtr MeasureStringAdjustment()
{
string sampleString = "a";
unsafe
{
fixed(char *pSampleString = sampleString)
{
return(Unsafe.ByteOffset <char>(ref Unsafe.As <Pinnable <char> >(sampleString).Data, ref Unsafe.AsRef <char>(pSampleString)));
}
}
}
public static void ByteOffsetStackByte4()
{
var byte4 = new Byte4();
Assert.Equal(new IntPtr(0), Unsafe.ByteOffset(ref byte4.B0, ref byte4.B0));
Assert.Equal(new IntPtr(1), Unsafe.ByteOffset(ref byte4.B0, ref byte4.B1));
Assert.Equal(new IntPtr(-1), Unsafe.ByteOffset(ref byte4.B1, ref byte4.B0));
Assert.Equal(new IntPtr(2), Unsafe.ByteOffset(ref byte4.B0, ref byte4.B2));
Assert.Equal(new IntPtr(-2), Unsafe.ByteOffset(ref byte4.B2, ref byte4.B0));
Assert.Equal(new IntPtr(3), Unsafe.ByteOffset(ref byte4.B0, ref byte4.B3));
Assert.Equal(new IntPtr(-3), Unsafe.ByteOffset(ref byte4.B3, ref byte4.B0));
}
public static void Run()
{
var z1 = (1, 2L, 'Z', "test");
var zz = "new mark test";
var z2 = (10, 20L, 'Z', "0test");
Console.WriteLine("SizeOf(...) = {0}", Unsafe.SizeOf <ValueTuple <int, int, UIntPtr> >());
Console.WriteLine("SizeOf(...) = {0}", Unsafe.SizeOf <ValueTuple <string> >());
Console.WriteLine("Offset = {0}", Unsafe.ByteOffset(ref z2.Item4, ref z1.Item4));
Console.WriteLine(Unsafe.AddByteOffset(ref z1.Item4, new IntPtr(8)));
}
// Copies the 8 byte args between mem and corrects offsets dest
private static void CopyArgs(
ref ulong *dest,
ulong *destLimit,
ulong *source,
uint argsLength
)
{
var diff = (uint)Unsafe.ByteOffset(ref *dest, ref *destLimit);
Debug.Assert((int)diff > 0);
Buffer.MemoryCopy(source, dest, diff, argsLength * sizeof(ulong));
dest += argsLength;
}
/// <summary>
/// Determines whether two sequences overlap in memory and outputs the element offset.
/// </summary>
public static bool Overlaps <T>(this ReadOnlySpan <T> first, ReadOnlySpan <T> second, out int elementOffset)
{
if (first.IsEmpty || second.IsEmpty)
{
elementOffset = 0;
return(false);
}
IntPtr byteOffset = Unsafe.ByteOffset(
ref MemoryMarshal.GetReference(first),
ref MemoryMarshal.GetReference(second));
if (Unsafe.SizeOf <IntPtr>() == sizeof(int))
{
if ((uint)byteOffset < (uint)(first.Length * Unsafe.SizeOf <T>()) ||
(uint)byteOffset > (uint)-(second.Length * Unsafe.SizeOf <T>()))
{
if ((int)byteOffset % Unsafe.SizeOf <T>() != 0)
{
ThrowHelper.ThrowArgumentException_OverlapAlignmentMismatch();
}
elementOffset = (int)byteOffset / Unsafe.SizeOf <T>();
return(true);
}
else
{
elementOffset = 0;
return(false);
}
}
else
{
if ((ulong)byteOffset < (ulong)((long)first.Length * Unsafe.SizeOf <T>()) ||
(ulong)byteOffset > (ulong)-((long)second.Length * Unsafe.SizeOf <T>()))
{
if ((long)byteOffset % Unsafe.SizeOf <T>() != 0)
{
ThrowHelper.ThrowArgumentException_OverlapAlignmentMismatch();
}
elementOffset = (int)((long)byteOffset / Unsafe.SizeOf <T>());
return(true);
}
else
{
elementOffset = 0;
return(false);
}
}
}
public static unsafe bool Overlap(
ReadOnlySpan <byte> first,
ReadOnlySpan <byte> second)
{
if (second.IsEmpty)
{
return(false);
}
IntPtr diff = Unsafe.ByteOffset(ref first.DangerousGetPinnableReference(), ref second.DangerousGetPinnableReference());
return((sizeof(IntPtr) == sizeof(int))
? unchecked ((uint)diff < (uint)first.Length || (uint)diff > (uint)-second.Length)
: unchecked ((ulong)diff < (ulong)first.Length || (ulong)diff > (ulong)-second.Length));
}
public static void ByteOffsetArray()
{
var a = new byte[] { 0, 1, 2, 3, 4, 5, 6, 7 };
Assert.Equal(new IntPtr(0), Unsafe.ByteOffset(ref a[0], ref a[0]));
Assert.Equal(new IntPtr(1), Unsafe.ByteOffset(ref a[0], ref a[1]));
Assert.Equal(new IntPtr(-1), Unsafe.ByteOffset(ref a[1], ref a[0]));
Assert.Equal(new IntPtr(2), Unsafe.ByteOffset(ref a[0], ref a[2]));
Assert.Equal(new IntPtr(-2), Unsafe.ByteOffset(ref a[2], ref a[0]));
Assert.Equal(new IntPtr(3), Unsafe.ByteOffset(ref a[0], ref a[3]));
Assert.Equal(new IntPtr(4), Unsafe.ByteOffset(ref a[0], ref a[4]));
Assert.Equal(new IntPtr(5), Unsafe.ByteOffset(ref a[0], ref a[5]));
Assert.Equal(new IntPtr(6), Unsafe.ByteOffset(ref a[0], ref a[6]));
Assert.Equal(new IntPtr(7), Unsafe.ByteOffset(ref a[0], ref a[7]));
}
public static Span <T> DangerousCreate(object obj, ref T objectData, int length)
{
if (obj == null)
{
ThrowHelper.ThrowArgumentNullException(ExceptionArgument.obj);
}
if (length < 0)
{
ThrowHelper.ThrowArgumentOutOfRangeException(ExceptionArgument.length);
}
Pinnable <T> pinnable = Unsafe.As <Pinnable <T> >(obj);
IntPtr byteOffset = Unsafe.ByteOffset <T>(ref pinnable.Data, ref objectData);
return(new Span <T>(pinnable, byteOffset, length));
}
private static void _BulkMoveWithWriteBarrier(
ref byte destination,
ref byte source,
nuint byteCount
)
{
Debug.Assert(byteCount > BulkMoveWithWriteBarrierChunk);
if (Unsafe.AreSame(ref source, ref destination))
{
return;
}
// This is equivalent to: (destination - source) >= byteCount || (destination - source) < 0
if ((nuint)(nint)Unsafe.ByteOffset(ref source, ref destination) >= byteCount)
{
// Copy forwards
do
{
byteCount -= BulkMoveWithWriteBarrierChunk;
__BulkMoveWithWriteBarrier(
ref destination,
ref source,
BulkMoveWithWriteBarrierChunk
);
destination = ref Unsafe.AddByteOffset(
ref destination,
BulkMoveWithWriteBarrierChunk
);
source = ref Unsafe.AddByteOffset(ref source, BulkMoveWithWriteBarrierChunk);
} while (byteCount > BulkMoveWithWriteBarrierChunk);
}
else
{
// Copy backwards
do
{
byteCount -= BulkMoveWithWriteBarrierChunk;
__BulkMoveWithWriteBarrier(
ref Unsafe.AddByteOffset(ref destination, byteCount),
ref Unsafe.AddByteOffset(ref source, byteCount),
BulkMoveWithWriteBarrierChunk
);
} while (byteCount > BulkMoveWithWriteBarrierChunk);
}
__BulkMoveWithWriteBarrier(ref destination, ref source, byteCount);
}
// This method has different signature for x64 and other platforms and is done for performance reasons.
private static void Memmove(ref byte dest, ref byte src, nuint len)
{
#if AMD64 || (BIT32 && !ARM)
const nuint CopyThreshold = 2048;
#else
const nuint CopyThreshold = 512;
#endif // AMD64 || (BIT32 && !ARM)
// P/Invoke into the native version when the buffers are overlapping.
if (((nuint)Unsafe.ByteOffset(ref src, ref dest) < len) || ((nuint)Unsafe.ByteOffset(ref dest, ref src) < len))
{
goto BuffersOverlap;
}
// Use "(IntPtr)(nint)len" to avoid overflow checking on the explicit cast to IntPtr
ref byte srcEnd = ref Unsafe.Add(ref src, (IntPtr)(nint)len);
/// <summary>
/// Implements the copy functionality used by Span and ReadOnlySpan.
///
/// NOTE: Fast span implements TryCopyTo in corelib and therefore this implementation
/// is only used by portable span. The code must live in code that only compiles
/// for portable span which means either each individual span implementation
/// of this shared code file. Other shared SpanHelper.X.cs files are compiled
/// for both portable and fast span implementations.
/// </summary>
public static unsafe void CopyTo <T>(ref T dst, int dstLength, ref T src, int srcLength)
{
Debug.Assert(dstLength != 0);
IntPtr srcByteCount = Unsafe.ByteOffset(ref src, ref Unsafe.Add(ref src, srcLength));
IntPtr dstByteCount = Unsafe.ByteOffset(ref dst, ref Unsafe.Add(ref dst, dstLength));
IntPtr diff = Unsafe.ByteOffset(ref src, ref dst);
bool isOverlapped = (sizeof(IntPtr) == sizeof(int))
? (uint)diff <(uint)srcByteCount || (uint)diff> (uint) - (int)dstByteCount
: (ulong)diff <(ulong)srcByteCount || (ulong)diff> (ulong) - (long)dstByteCount;
if (!isOverlapped && !SpanHelpers.IsReferenceOrContainsReferences <T>())
{
ref byte dstBytes = ref Unsafe.As <T, byte>(ref dst);
ref byte srcBytes = ref Unsafe.As <T, byte>(ref src);
/// <summary>
/// Determines whether two sequences overlap in memory.
/// </summary>
public static bool Overlaps <T>(this ReadOnlySpan <T> first, ReadOnlySpan <T> second)
{
if (first.IsEmpty || second.IsEmpty)
{
return(false);
}
IntPtr byteOffset = Unsafe.ByteOffset(
ref MemoryMarshal.GetReference(first),
ref MemoryMarshal.GetReference(second));
if (Unsafe.SizeOf <IntPtr>() == sizeof(int))
{
return((uint)byteOffset < (uint)(first.Length * Unsafe.SizeOf <T>()) ||
(uint)byteOffset > (uint)-(second.Length * Unsafe.SizeOf <T>()));
}
else
{
return((ulong)byteOffset < (ulong)((long)first.Length * Unsafe.SizeOf <T>()) ||
(ulong)byteOffset > (ulong)-((long)second.Length * Unsafe.SizeOf <T>()));
}
}
// This method has different signature for x64 and other platforms and is done for performance reasons.
private static void Memmove(ByReference <byte> dest, ByReference <byte> src, nuint len)
{
#if AMD64 || (BIT32 && !ARM)
const nuint CopyThreshold = 2048;
#elif ARM64
#if PLATFORM_WINDOWS
// Determined optimal value for Windows.
// https://github.com/dotnet/coreclr/issues/13843
const nuint CopyThreshold = UInt64.MaxValue;
#else // PLATFORM_WINDOWS
// Managed code is currently faster than glibc unoptimized memmove
// TODO-ARM64-UNIX-OPT revisit when glibc optimized memmove is in Linux distros
// https://github.com/dotnet/coreclr/issues/13844
const nuint CopyThreshold = UInt64.MaxValue;
#endif // PLATFORM_WINDOWS
#else
const nuint CopyThreshold = 512;
#endif // AMD64 || (BIT32 && !ARM)
// P/Invoke into the native version when the buffers are overlapping.
if (((nuint)Unsafe.ByteOffset(ref src.Value, ref dest.Value) < len) || ((nuint)Unsafe.ByteOffset(ref dest.Value, ref src.Value) < len))
{
goto BuffersOverlap;
}
// Use "(IntPtr)(nint)len" to avoid overflow checking on the explicit cast to IntPtr
ref byte srcEnd = ref Unsafe.Add(ref src.Value, (IntPtr)(nint)len);
public static int TestStructWithNonBlittableTypesOffset(ref StructWithNonBlittableTypes a)
{
return(Unsafe.ByteOffset(ref a.a0, ref a.a1).ToInt32());
}
public static int LoadWav(string path)
#endif
{
try
{
bool swapped = false;
// First: we open the file and copy it into a single large memory buffer for processing.
byte[] buffer;
using (var stream = File.OpenRead(path))
{
buffer = new byte[stream.Length];
stream.Read(buffer, 0, buffer.Length);
}
Span <byte> bufferSpan = buffer;
// Second: find the RIFF chunk. Note that by searching for RIFF both normal
// and reversed, we can automatically determine the endian swap situation for
// this file regardless of what machine we are on.
var riff = FindChunk(bufferSpan, RiffId, false);
if (riff.IsEmpty)
{
riff = FindChunk(bufferSpan, RiffId, true);
if (riff.IsEmpty)
{
XPlane.Trace.WriteLine("[OpenAL Sample] Could not find RIFF chunk in wave file.");
return(0);
}
swapped = true;
}
// The wave chunk isn't really a chunk at all. :-( It's just a "WAVE" tag
// followed by more chunks. This strikes me as totally inconsistent, but
// anyway, confirm the WAVE ID and move on.
if (riff[0] != (byte)'W' ||
riff[1] != (byte)'A' ||
riff[2] != (byte)'V' ||
riff[3] != (byte)'E')
{
XPlane.Trace.WriteLine("[OpenAL Sample] Could not find WAVE signature in wave file.");
return(0);
}
bufferSpan = bufferSpan.Slice(4 + (int)Unsafe.ByteOffset(ref bufferSpan[0], ref riff[0]));
var formatChunk = FindChunk(bufferSpan, FmtId, swapped);
if (formatChunk.IsEmpty)
{
XPlane.Trace.WriteLine("[OpenAL Sample] Could not find FMT chunk in wave file.");
}
var format = MemoryMarshal.Read <FormatInfo>(formatChunk);
if (swapped)
{
format.Format = BinaryPrimitives.ReverseEndianness(format.Format);
format.ChannelCount = BinaryPrimitives.ReverseEndianness(format.ChannelCount);
format.SampleRate = BinaryPrimitives.ReverseEndianness(format.SampleRate);
format.ByteRate = BinaryPrimitives.ReverseEndianness(format.ByteRate);
format.BlockAlign = BinaryPrimitives.ReverseEndianness(format.BlockAlign);
format.BitsPerSample = BinaryPrimitives.ReverseEndianness(format.BitsPerSample);
}
// Reject things we don't understand...expand this code to support weirder audio formats.
if (format.Format != 1)
{
XPlane.Trace.WriteLine("[OpenAL Sample] Wave file is not PCM format data.");
return(0);
}
if (format.ChannelCount != 1 && format.ChannelCount != 2)
{
XPlane.Trace.WriteLine("[OpenAL Sample] Must have mono or stereo sound.");
return(0);
}
if (format.BitsPerSample != 8 && format.BitsPerSample != 16)
{
XPlane.Trace.WriteLine("[OpenAL Sample] Must have 8 or 16 bit sounds.");
return(0);
}
var data = FindChunk(bufferSpan, DataId, swapped);
if (data.IsEmpty)
{
XPlane.Trace.WriteLine("[OpenAL Sample] I could not find the DATA chunk.");
return(0);
}
var sampleSize = format.ChannelCount * format.BitsPerSample / 8;
var dataSamples = data.Length / sampleSize;
// If the file is swapped and we have 16-bit audio, we need to endian-swap the audio too or we'll
// get something that sounds just astoundingly bad!
if (format.BitsPerSample == 16 && swapped)
{
var wordSpan = MemoryMarshal.Cast <byte, ushort>(data);
foreach (ref uint word in wordSpan)
{
word = BinaryPrimitives.ReverseEndianness(word);
}
}
return(BufferData(data));
#if SILK_NET
unsafe uint BufferData(in ReadOnlySpan <byte> audioData)
{
var bufferId = al.GenBuffer();
if (bufferId == 0)
{
XPlane.Trace.WriteLine("[OpenAL Sample] Could not generate buffer id.");
return(0);
}
fixed(void *pData = audioData)
{
al.BufferData(
bufferId,
format.BitsPerSample == 16
? (format.ChannelCount == 2 ? BufferFormat.Stereo16 : BufferFormat.Mono16)
: (format.ChannelCount == 2 ? BufferFormat.Stereo8 : BufferFormat.Mono8),
pData,
audioData.Length,
format.SampleRate);
}
return(bufferId);
}
#elif OPEN_TOOLKIT
int BufferData(in Span <byte> audioData)
{
AL.GenBuffer(out var bufferId);
if (bufferId == 0)
{
XPlane.Trace.WriteLine("[OpenAL Sample] Could not generate buffer id.");
return(0);
}
AL.BufferData(
bufferId,
format.BitsPerSample == 16
? (format.ChannelCount == 2 ? ALFormat.Stereo16 : ALFormat.Mono16)
: (format.ChannelCount == 2 ? ALFormat.Stereo8 : ALFormat.Mono8),
audioData,
format.SampleRate);
return(bufferId);
}
#endif
}
catch (Exception ex)
{
XPlane.Trace.WriteLine(ex.ToString());
return(0);
}
}
/// <summary>
/// Calculates a byte offset of a given field within a struct.
/// </summary>
/// <typeparam name="T">Struct type</typeparam>
/// <typeparam name="T2">Field type</typeparam>
/// <param name="storage">Parent struct</param>
/// <param name="target">Field</param>
/// <returns>The byte offset of the given field in the given struct</returns>
public static int OffsetOf <T, T2>(ref T2 storage, ref T target)
{
return((int)Unsafe.ByteOffset(ref Unsafe.As <T2, T>(ref storage), ref target));
}
// Array header sizes are a runtime implementation detail and aren't the same across all runtimes. (The CLR made a tweak after 4.5, and Mono has an extra Bounds pointer.)
private static IntPtr MeasureArrayAdjustment()
{
T[] sampleArray = new T[1];
return(Unsafe.ByteOffset <T>(ref Unsafe.As <Pinnable <T> >(sampleArray).Data, ref sampleArray[0]));
}
public static nint ByteOffset <T>(T *tgt, T *cur) where T : unmanaged => Unsafe.ByteOffset(ref *tgt, ref *cur);
public static unsafe int UnsafeByteOffset(ref byte start, ReadOnlySpan <byte> end) => (int)(void *)Unsafe.ByteOffset(ref start, ref MemoryMarshal.GetReference(end));
private static int StorageOffset <T>(ref NativeCtxStorage storage, ref T target)
{
return((int)Unsafe.ByteOffset(ref Unsafe.As <NativeCtxStorage, T>(ref storage), ref target));
}
public static Span <LbrEntry64> Entries(ref LbrTraceEventData64 data, int totalSize)
{
IntPtr entriesOffset = Unsafe.ByteOffset(ref Unsafe.As <LbrTraceEventData64, byte>(ref data), ref Unsafe.As <LbrEntry64, byte>(ref data._entries));
return(MemoryMarshal.CreateSpan(ref data._entries, (totalSize - (int)entriesOffset) / sizeof(LbrEntry64)));
}
private static int OffsetOf <T>(ref SupportBuffer storage, ref T target)
{
return((int)Unsafe.ByteOffset(ref Unsafe.As <SupportBuffer, T>(ref storage), ref target));
}
private static unsafe SafeCertStoreHandle SelectFromStore(SafeCertStoreHandle safeSourceStoreHandle, string?title, string?message, X509SelectionFlag selectionFlags, IntPtr hwndParent)
{
int dwErrorCode = ERROR_SUCCESS;
SafeCertStoreHandle safeCertStoreHandle = Interop.Crypt32.CertOpenStore(
(IntPtr)Interop.Crypt32.CERT_STORE_PROV_MEMORY,
Interop.Crypt32.X509_ASN_ENCODING | Interop.Crypt32.PKCS_7_ASN_ENCODING,
IntPtr.Zero,
0,
IntPtr.Zero);
if (safeCertStoreHandle == null || safeCertStoreHandle.IsInvalid)
{
throw new CryptographicException(Marshal.GetLastWin32Error());
}
Interop.CryptUI.CRYPTUI_SELECTCERTIFICATE_STRUCTW csc = default;
// Older versions of CRYPTUI do not check the size correctly,
// so always force it to the oldest version of the structure.
#if NET7_0_OR_GREATER
// Declare a local for Native to enable us to get the managed byte offset
// without having a null check cause a failure.
Interop.CryptUI.CRYPTUI_SELECTCERTIFICATE_STRUCTW.Marshaller.Native native;
Unsafe.SkipInit(out native);
csc.dwSize = (uint)Unsafe.ByteOffset(ref Unsafe.As <Interop.CryptUI.CRYPTUI_SELECTCERTIFICATE_STRUCTW.Marshaller.Native, byte>(ref native), ref Unsafe.As <IntPtr, byte>(ref native.hSelectedCertStore));
#else
csc.dwSize = (uint)Marshal.OffsetOf(typeof(Interop.CryptUI.CRYPTUI_SELECTCERTIFICATE_STRUCTW), "hSelectedCertStore");
#endif
csc.hwndParent = hwndParent;
csc.dwFlags = (uint)selectionFlags;
csc.szTitle = title;
csc.dwDontUseColumn = 0;
csc.szDisplayString = message;
csc.pFilterCallback = IntPtr.Zero;
csc.pDisplayCallback = IntPtr.Zero;
csc.pvCallbackData = IntPtr.Zero;
csc.cDisplayStores = 1;
IntPtr hSourceCertStore = safeSourceStoreHandle.DangerousGetHandle();
csc.rghDisplayStores = new IntPtr(&hSourceCertStore);
csc.cStores = 0;
csc.rghStores = IntPtr.Zero;
csc.cPropSheetPages = 0;
csc.rgPropSheetPages = IntPtr.Zero;
csc.hSelectedCertStore = safeCertStoreHandle.DangerousGetHandle();
SafeCertContextHandle safeCertContextHandle = Interop.CryptUI.CryptUIDlgSelectCertificateW(ref csc);
if (safeCertContextHandle != null && !safeCertContextHandle.IsInvalid)
{
// Single select, so add it to our hCertStore
SafeCertContextHandle ppStoreContext = SafeCertContextHandle.InvalidHandle;
if (!Interop.Crypt32.CertAddCertificateLinkToStore(safeCertStoreHandle,
safeCertContextHandle,
Interop.Crypt32.CERT_STORE_ADD_ALWAYS,
ppStoreContext))
{
dwErrorCode = Marshal.GetLastWin32Error();
}
}
if (dwErrorCode != ERROR_SUCCESS)
{
throw new CryptographicException(dwErrorCode);
}
return(safeCertStoreHandle);
}
// Returns &inputBuffer[inputLength] if the input buffer is valid.
/// <summary>
/// Given an input buffer <paramref name="pInputBuffer"/> of byte length <paramref name="inputLength"/>,
/// returns a pointer to where the first invalid data appears in <paramref name="pInputBuffer"/>.
/// </summary>
/// <remarks>
/// Returns a pointer to the end of <paramref name="pInputBuffer"/> if the buffer is well-formed.
/// </remarks>
public static byte *GetPointerToFirstInvalidByte(byte *pInputBuffer, int inputLength, out int utf16CodeUnitCountAdjustment, out int scalarCountAdjustment)
{
Debug.Assert(inputLength >= 0, "Input length must not be negative.");
Debug.Assert(pInputBuffer != null || inputLength == 0, "Input length must be zero if input buffer pointer is null.");
// First, try to drain off as many ASCII bytes as we can from the beginning.
{
nuint numAsciiBytesCounted = ASCIIUtility.GetIndexOfFirstNonAsciiByte(pInputBuffer, (uint)inputLength);
pInputBuffer += numAsciiBytesCounted;
// Quick check - did we just end up consuming the entire input buffer?
// If so, short-circuit the remainder of the method.
inputLength -= (int)numAsciiBytesCounted;
if (inputLength == 0)
{
utf16CodeUnitCountAdjustment = 0;
scalarCountAdjustment = 0;
return(pInputBuffer);
}
}
#if DEBUG
// Keep these around for final validation at the end of the method.
byte *pOriginalInputBuffer = pInputBuffer;
int originalInputLength = inputLength;
#endif
// Enregistered locals that we'll eventually out to our caller.
int tempUtf16CodeUnitCountAdjustment = 0;
int tempScalarCountAdjustment = 0;
if (inputLength < sizeof(uint))
{
goto ProcessInputOfLessThanDWordSize;
}
byte *pFinalPosWhereCanReadDWordFromInputBuffer = pInputBuffer + (uint)inputLength - sizeof(uint);
// Begin the main loop.
#if DEBUG
byte *pLastBufferPosProcessed = null; // used for invariant checking in debug builds
#endif
while (pInputBuffer <= pFinalPosWhereCanReadDWordFromInputBuffer)
{
// Read 32 bits at a time. This is enough to hold any possible UTF8-encoded scalar.
uint thisDWord = Unsafe.ReadUnaligned <uint>(pInputBuffer);
AfterReadDWord:
#if DEBUG
Debug.Assert(pLastBufferPosProcessed < pInputBuffer, "Algorithm should've made forward progress since last read.");
pLastBufferPosProcessed = pInputBuffer;
#endif
// First, check for the common case of all-ASCII bytes.
if (ASCIIUtility.AllBytesInUInt32AreAscii(thisDWord))
{
// We read an all-ASCII sequence.
pInputBuffer += sizeof(uint);
// If we saw a sequence of all ASCII, there's a good chance a significant amount of following data is also ASCII.
// Below is basically unrolled loops with poor man's vectorization.
// Below check is "can I read at least five DWORDs from the input stream?"
// n.b. Since we incremented pInputBuffer above the below subtraction may result in a negative value,
// hence using nint instead of nuint.
if ((nint)(void *)Unsafe.ByteOffset(ref *pInputBuffer, ref *pFinalPosWhereCanReadDWordFromInputBuffer) >= 4 * sizeof(uint))
{
// We want reads in the inner loop to be aligned. So let's perform a quick
// ASCII check of the next 32 bits (4 bytes) now, and if that succeeds bump
// the read pointer up to the next aligned address.
thisDWord = Unsafe.ReadUnaligned <uint>(pInputBuffer);
if (!ASCIIUtility.AllBytesInUInt32AreAscii(thisDWord))
{
goto AfterReadDWordSkipAllBytesAsciiCheck;
}
pInputBuffer = (byte *)((nuint)(pInputBuffer + 4) & ~(nuint)3);
// At this point, the input buffer offset points to an aligned DWORD. We also know that there's
// enough room to read at least four DWORDs from the buffer. (Heed the comment a few lines above:
// the original 'if' check confirmed that there were 5 DWORDs before the alignment check, and
// the alignment check consumes at most a single DWORD.)
byte *pInputBufferFinalPosAtWhichCanSafelyLoop = pFinalPosWhereCanReadDWordFromInputBuffer - 3 * sizeof(uint); // can safely read 4 DWORDs here
uint mask;
do
{
if (Sse2.IsSupported && Bmi1.IsSupported)
{
// pInputBuffer is 32-bit aligned but not necessary 128-bit aligned, so we're
// going to perform an unaligned load. We don't necessarily care about aligning
// this because we pessimistically assume we'll encounter non-ASCII data at some
// point in the not-too-distant future (otherwise we would've stayed entirely
// within the all-ASCII vectorized code at the entry to this method).
mask = (uint)Sse2.MoveMask(Sse2.LoadVector128((byte *)pInputBuffer));
if (mask != 0)
{
goto Sse2LoopTerminatedEarlyDueToNonAsciiData;
}
}
else
{
if (!ASCIIUtility.AllBytesInUInt32AreAscii(((uint *)pInputBuffer)[0] | ((uint *)pInputBuffer)[1]))
{
goto LoopTerminatedEarlyDueToNonAsciiDataInFirstPair;
}
if (!ASCIIUtility.AllBytesInUInt32AreAscii(((uint *)pInputBuffer)[2] | ((uint *)pInputBuffer)[3]))
{
goto LoopTerminatedEarlyDueToNonAsciiDataInSecondPair;
}
}
pInputBuffer += 4 * sizeof(uint); // consumed 4 DWORDs
} while (pInputBuffer <= pInputBufferFinalPosAtWhichCanSafelyLoop);
continue; // need to perform a bounds check because we might be running out of data
Sse2LoopTerminatedEarlyDueToNonAsciiData:
Debug.Assert(BitConverter.IsLittleEndian);
Debug.Assert(Sse2.IsSupported);
Debug.Assert(Bmi1.IsSupported);
// The 'mask' value will have a 0 bit for each ASCII byte we saw and a 1 bit
// for each non-ASCII byte we saw. We can count the number of ASCII bytes,
// bump our input counter by that amount, and resume processing from the
// "the first byte is no longer ASCII" portion of the main loop.
Debug.Assert(mask != 0);
pInputBuffer += Bmi1.TrailingZeroCount(mask);
if (pInputBuffer > pFinalPosWhereCanReadDWordFromInputBuffer)
{
goto ProcessRemainingBytesSlow;
}
thisDWord = Unsafe.ReadUnaligned <uint>(pInputBuffer); // no longer guaranteed to be aligned
goto BeforeProcessTwoByteSequence;
LoopTerminatedEarlyDueToNonAsciiDataInSecondPair:
pInputBuffer += 2 * sizeof(uint); // consumed 2 DWORDs
LoopTerminatedEarlyDueToNonAsciiDataInFirstPair:
// We know that there's *at least* two DWORDs of data remaining in the buffer.
// We also know that one of them (or both of them) contains non-ASCII data somewhere.
// Let's perform a quick check here to bypass the logic at the beginning of the main loop.
thisDWord = *(uint *)pInputBuffer; // still aligned here
if (ASCIIUtility.AllBytesInUInt32AreAscii(thisDWord))
{
pInputBuffer += sizeof(uint); // consumed 1 more DWORD
thisDWord = *(uint *)pInputBuffer; // still aligned here
}
goto AfterReadDWordSkipAllBytesAsciiCheck;
}
continue; // not enough data remaining to unroll loop - go back to beginning with bounds checks
}
AfterReadDWordSkipAllBytesAsciiCheck:
Debug.Assert(!ASCIIUtility.AllBytesInUInt32AreAscii(thisDWord)); // this should have been handled earlier
// Next, try stripping off ASCII bytes one at a time.
// We only handle up to three ASCII bytes here since we handled the four ASCII byte case above.
{
uint numLeadingAsciiBytes = ASCIIUtility.CountNumberOfLeadingAsciiBytesFromUInt32WithSomeNonAsciiData(thisDWord);
pInputBuffer += numLeadingAsciiBytes;
if (pFinalPosWhereCanReadDWordFromInputBuffer < pInputBuffer)
{
goto ProcessRemainingBytesSlow; // Input buffer doesn't contain enough data to read a DWORD
}
else
{
// The input buffer at the current offset contains a non-ASCII byte.
// Read an entire DWORD and fall through to multi-byte consumption logic.
thisDWord = Unsafe.ReadUnaligned <uint>(pInputBuffer);
}
}
BeforeProcessTwoByteSequence:
// At this point, we suspect we're working with a multi-byte code unit sequence,
// but we haven't yet validated it for well-formedness.
// The masks and comparands are derived from the Unicode Standard, Table 3-6.
// Additionally, we need to check for valid byte sequences per Table 3-7.
// Check the 2-byte case.
thisDWord -= (BitConverter.IsLittleEndian) ? 0x0000_80C0u : 0xC080_0000u;
if ((thisDWord & (BitConverter.IsLittleEndian ? 0x0000_C0E0u : 0xE0C0_0000u)) == 0)
{
// Per Table 3-7, valid sequences are:
// [ C2..DF ] [ 80..BF ]
//
// Due to our modification of 'thisDWord' above, this becomes:
// [ 02..1F ] [ 00..3F ]
//
// We've already checked that the leading byte was originally in the range [ C0..DF ]
// and that the trailing byte was originally in the range [ 80..BF ], so now we only need
// to check that the modified leading byte is >= [ 02 ].
if ((BitConverter.IsLittleEndian && (byte)thisDWord < 0x02u) ||
(!BitConverter.IsLittleEndian && thisDWord < 0x0200_0000u))
{
goto Error; // overlong form - leading byte was [ C0 ] or [ C1 ]
}
ProcessTwoByteSequenceSkipOverlongFormCheck:
// Optimization: If this is a two-byte-per-character language like Cyrillic or Hebrew,
// there's a good chance that if we see one two-byte run then there's another two-byte
// run immediately after. Let's check that now.
// On little-endian platforms, we can check for the two-byte UTF8 mask *and* validate that
// the value isn't overlong using a single comparison. On big-endian platforms, we'll need
// to validate the mask and validate that the sequence isn't overlong as two separate comparisons.
if ((BitConverter.IsLittleEndian && UInt32EndsWithValidUtf8TwoByteSequenceLittleEndian(thisDWord)) ||
(!BitConverter.IsLittleEndian && (UInt32EndsWithUtf8TwoByteMask(thisDWord) && !UInt32EndsWithOverlongUtf8TwoByteSequence(thisDWord))))
{
// We have two runs of two bytes each.
pInputBuffer += 4;
tempUtf16CodeUnitCountAdjustment -= 2; // 4 UTF-8 code units -> 2 UTF-16 code units (and 2 scalars)
if (pInputBuffer <= pFinalPosWhereCanReadDWordFromInputBuffer)
{
// Optimization: If we read a long run of two-byte sequences, the next sequence is probably
// also two bytes. Check for that first before going back to the beginning of the loop.
thisDWord = Unsafe.ReadUnaligned <uint>(pInputBuffer);
if (BitConverter.IsLittleEndian)
{
if (UInt32BeginsWithValidUtf8TwoByteSequenceLittleEndian(thisDWord))
{
// The next sequence is a valid two-byte sequence.
goto ProcessTwoByteSequenceSkipOverlongFormCheck;
}
}
else
{
if (UInt32BeginsWithUtf8TwoByteMask(thisDWord))
{
if (UInt32BeginsWithOverlongUtf8TwoByteSequence(thisDWord))
{
goto Error; // The next sequence purports to be a 2-byte sequence but is overlong.
}
goto ProcessTwoByteSequenceSkipOverlongFormCheck;
}
}
// If we reached this point, the next sequence is something other than a valid
// two-byte sequence, so go back to the beginning of the loop.
goto AfterReadDWord;
}
else
{
goto ProcessRemainingBytesSlow; // Running out of data - go down slow path
}
}
// The buffer contains a 2-byte sequence followed by 2 bytes that aren't a 2-byte sequence.
// Unlikely that a 3-byte sequence would follow a 2-byte sequence, so perhaps remaining
// bytes are ASCII?
tempUtf16CodeUnitCountAdjustment--; // 2-byte sequence + (some number of ASCII bytes) -> 1 UTF-16 code units (and 1 scalar) [+ trailing]
if (UInt32ThirdByteIsAscii(thisDWord))
{
if (UInt32FourthByteIsAscii(thisDWord))
{
pInputBuffer += 4;
}
else
{
pInputBuffer += 3;
// A two-byte sequence followed by an ASCII byte followed by a non-ASCII byte.
// Read in the next DWORD and jump directly to the start of the multi-byte processing block.
if (pInputBuffer <= pFinalPosWhereCanReadDWordFromInputBuffer)
{
thisDWord = Unsafe.ReadUnaligned <uint>(pInputBuffer);
goto BeforeProcessTwoByteSequence;
}
}
}
else
{
pInputBuffer += 2;
}
continue;
}
// Check the 3-byte case.
// We need to restore the C0 leading byte we stripped out earlier, then we can strip out the expected E0 byte.
thisDWord -= (BitConverter.IsLittleEndian) ? (0x0080_00E0u - 0x0000_00C0u) : (0xE000_8000u - 0xC000_0000u);
if ((thisDWord & (BitConverter.IsLittleEndian ? 0x00C0_C0F0u : 0xF0C0_C000u)) == 0)
{
ProcessThreeByteSequenceWithCheck:
// We assume the caller has confirmed that the bit pattern is representative of a three-byte
// sequence, but it may still be overlong or surrogate. We need to check for these possibilities.
//
// Per Table 3-7, valid sequences are:
// [ E0 ] [ A0..BF ] [ 80..BF ]
// [ E1..EC ] [ 80..BF ] [ 80..BF ]
// [ ED ] [ 80..9F ] [ 80..BF ]
// [ EE..EF ] [ 80..BF ] [ 80..BF ]
//
// Big-endian examples of using the above validation table:
// E0A0 = 1110 0000 1010 0000 => invalid (overlong ) patterns are 1110 0000 100# ####
// ED9F = 1110 1101 1001 1111 => invalid (surrogate) patterns are 1110 1101 101# ####
// If using the bitmask ......................................... 0000 1111 0010 0000 (=0F20),
// Then invalid (overlong) patterns match the comparand ......... 0000 0000 0000 0000 (=0000),
// And invalid (surrogate) patterns match the comparand ......... 0000 1101 0010 0000 (=0D20).
//
// It's ok if the caller has manipulated 'thisDWord' (e.g., by subtracting 0xE0 or 0x80)
// as long as they haven't touched the bits we're about to use in our mask checking below.
if (BitConverter.IsLittleEndian)
{
// The "overlong or surrogate" check can be implemented using a single jump, but there's
// some overhead to moving the bits into the correct locations in order to perform the
// correct comparison, and in practice the processor's branch prediction capability is
// good enough that we shouldn't bother. So we'll use two jumps instead.
// Can't extract this check into its own helper method because JITter produces suboptimal
// assembly, even with aggressive inlining.
// Code below becomes 5 instructions: test, jz, lea, test, jz
if (((thisDWord & 0x0000_200Fu) == 0) || (((thisDWord - 0x0000_200Du) & 0x0000_200Fu) == 0))
{
goto Error; // overlong or surrogate
}
}
else
{
if (((thisDWord & 0x0F20_0000u) == 0) || (((thisDWord - 0x0D20_0000u) & 0x0F20_0000u) == 0))
{
goto Error; // overlong or surrogate
}
}
ProcessSingleThreeByteSequenceSkipOverlongAndSurrogateChecks:
// Occasionally one-off ASCII characters like spaces, periods, or newlines will make their way
// in to the text. If this happens strip it off now before seeing if the next character
// consists of three code units.
// Branchless: consume a 3-byte UTF-8 sequence and optionally an extra ASCII byte hanging off the end
nint asciiAdjustment;
if (BitConverter.IsLittleEndian)
{
asciiAdjustment = (int)thisDWord >> 31; // smear most significant bit across entire value
}
else
{
asciiAdjustment = (nint)(sbyte)thisDWord >> 7; // smear most significant bit of least significant byte across entire value
}
// asciiAdjustment = 0 if fourth byte is ASCII; -1 otherwise
// Please *DO NOT* reorder the below two lines. It provides extra defense in depth in case this method
// is ever changed such that pInputBuffer becomes a 'ref byte' instead of a simple 'byte*'. It's valid
// to add 4 before backing up since we already checked previously that the input buffer contains at
// least a DWORD's worth of data, so we're not going to run past the end of the buffer where the GC can
// no longer track the reference. However, we can't back up before adding 4, since we might back up to
// before the start of the buffer, and the GC isn't guaranteed to be able to track this.
pInputBuffer += 4; // optimistically, assume consumed a 3-byte UTF-8 sequence plus an extra ASCII byte
pInputBuffer += asciiAdjustment; // back up if we didn't actually consume an ASCII byte
tempUtf16CodeUnitCountAdjustment -= 2; // 3 (or 4) UTF-8 bytes -> 1 (or 2) UTF-16 code unit (and 1 [or 2] scalar)
SuccessfullyProcessedThreeByteSequence:
if (IntPtr.Size >= 8 && BitConverter.IsLittleEndian)
{
// x64 little-endian optimization: A three-byte character could indicate CJK text,
// which makes it likely that the character following this one is also CJK.
// We'll try to process several three-byte sequences at a time.
// The check below is really "can we read 9 bytes from the input buffer?" since 'pFinalPos...' is already offset
// n.b. The subtraction below could result in a negative value (since we advanced pInputBuffer above), so
// use nint instead of nuint.
if ((nint)(pFinalPosWhereCanReadDWordFromInputBuffer - pInputBuffer) >= 5)
{
ulong thisQWord = Unsafe.ReadUnaligned <ulong>(pInputBuffer);
// Stage the next 32 bits into 'thisDWord' so that it's ready for us in case we need to jump backward
// to a previous location in the loop. This offers defense against reading main memory again (which may
// have been modified and could lead to a race condition).
thisDWord = (uint)thisQWord;
// Is this three 3-byte sequences in a row?
// thisQWord = [ 10yyyyyy 1110zzzz | 10xxxxxx 10yyyyyy 1110zzzz | 10xxxxxx 10yyyyyy 1110zzzz ] [ 10xxxxxx ]
// ---- CHAR 3 ---- --------- CHAR 2 --------- --------- CHAR 1 --------- -CHAR 3-
if ((thisQWord & 0xC0F0_C0C0_F0C0_C0F0ul) == 0x80E0_8080_E080_80E0ul && IsUtf8ContinuationByte(in pInputBuffer[8]))
{
// Saw a proper bitmask for three incoming 3-byte sequences, perform the
// overlong and surrogate sequence checking now.
// Check the first character.
// If the first character is overlong or a surrogate, fail immediately.
if ((((uint)thisQWord & 0x200Fu) == 0) || ((((uint)thisQWord - 0x200Du) & 0x200Fu) == 0))
{
goto Error;
}
// Check the second character.
// At this point, we now know the first three bytes represent a well-formed sequence.
// If there's an error beyond here, we'll jump back to the "process three known good bytes"
// logic.
thisQWord >>= 24;
if ((((uint)thisQWord & 0x200Fu) == 0) || ((((uint)thisQWord - 0x200Du) & 0x200Fu) == 0))
{
goto ProcessSingleThreeByteSequenceSkipOverlongAndSurrogateChecks;
}
// Check the third character (we already checked that it's followed by a continuation byte).
thisQWord >>= 24;
if ((((uint)thisQWord & 0x200Fu) == 0) || ((((uint)thisQWord - 0x200Du) & 0x200Fu) == 0))
{
goto ProcessSingleThreeByteSequenceSkipOverlongAndSurrogateChecks;
}
pInputBuffer += 9;
tempUtf16CodeUnitCountAdjustment -= 6; // 9 UTF-8 bytes -> 3 UTF-16 code units (and 3 scalars)
goto SuccessfullyProcessedThreeByteSequence;
}
// Is this two 3-byte sequences in a row?
// thisQWord = [ ######## ######## | 10xxxxxx 10yyyyyy 1110zzzz | 10xxxxxx 10yyyyyy 1110zzzz ]
// --------- CHAR 2 --------- --------- CHAR 1 ---------
if ((thisQWord & 0xC0C0_F0C0_C0F0ul) == 0x8080_E080_80E0ul)
{
// Saw a proper bitmask for two incoming 3-byte sequences, perform the
// overlong and surrogate sequence checking now.
// Check the first character.
// If the first character is overlong or a surrogate, fail immediately.
if ((((uint)thisQWord & 0x200Fu) == 0) || ((((uint)thisQWord - 0x200Du) & 0x200Fu) == 0))
{
goto Error;
}
// Check the second character.
// At this point, we now know the first three bytes represent a well-formed sequence.
// If there's an error beyond here, we'll jump back to the "process three known good bytes"
// logic.
thisQWord >>= 24;
if ((((uint)thisQWord & 0x200Fu) == 0) || ((((uint)thisQWord - 0x200Du) & 0x200Fu) == 0))
{
goto ProcessSingleThreeByteSequenceSkipOverlongAndSurrogateChecks;
}
pInputBuffer += 6;
tempUtf16CodeUnitCountAdjustment -= 4; // 6 UTF-8 bytes -> 2 UTF-16 code units (and 2 scalars)
// The next byte in the sequence didn't have a 3-byte marker, so it's probably
// an ASCII character. Jump back to the beginning of loop processing.
continue;
}
if (UInt32BeginsWithUtf8ThreeByteMask(thisDWord))
{
// A single three-byte sequence.
goto ProcessThreeByteSequenceWithCheck;
}
else
{
// Not a three-byte sequence; perhaps ASCII?
goto AfterReadDWord;
}
}
}
if (pInputBuffer <= pFinalPosWhereCanReadDWordFromInputBuffer)
{
thisDWord = Unsafe.ReadUnaligned <uint>(pInputBuffer);
// Optimization: A three-byte character could indicate CJK text, which makes it likely
// that the character following this one is also CJK. We'll check for a three-byte sequence
// marker now and jump directly to three-byte sequence processing if we see one, skipping
// all of the logic at the beginning of the loop.
if (UInt32BeginsWithUtf8ThreeByteMask(thisDWord))
{
goto ProcessThreeByteSequenceWithCheck; // Found another [not yet validated] three-byte sequence; process
}
else
{
goto AfterReadDWord; // Probably ASCII punctuation or whitespace; go back to start of loop
}
}
else
{
goto ProcessRemainingBytesSlow; // Running out of data
}
}
// Assume the 4-byte case, but we need to validate.
if (BitConverter.IsLittleEndian)
{
thisDWord &= 0xC0C0_FFFFu;
// After the above modifications earlier in this method, we expect 'thisDWord'
// to have the structure [ 10000000 00000000 00uuzzzz 00010uuu ]. We'll now
// perform two checks to confirm this. The first will verify the
// [ 10000000 00000000 00###### ######## ] structure by taking advantage of two's
// complement representation to perform a single *signed* integer check.
if ((int)thisDWord > unchecked ((int)0x8000_3FFF))
{
goto Error; // didn't have three trailing bytes
}
// Now we want to confirm that 0x01 <= uuuuu (otherwise this is an overlong encoding)
// and that uuuuu <= 0x10 (otherwise this is an out-of-range encoding).
thisDWord = BitOperations.RotateRight(thisDWord, 8);
// Now, thisDWord = [ 00010uuu 10000000 00000000 00uuzzzz ].
// The check is now a simple add / cmp / jcc combo.
if (!UnicodeUtility.IsInRangeInclusive(thisDWord, 0x1080_0010u, 0x1480_000Fu))
{
goto Error; // overlong or out-of-range
}
}
else
{
thisDWord -= 0x80u;
// After the above modifications earlier in this method, we expect 'thisDWord'
// to have the structure [ 00010uuu 00uuzzzz 00yyyyyy 00xxxxxx ]. We'll now
// perform two checks to confirm this. The first will verify the
// [ ######## 00###### 00###### 00###### ] structure.
if ((thisDWord & 0x00C0_C0C0u) != 0)
{
goto Error; // didn't have three trailing bytes
}
// Now we want to confirm that 0x01 <= uuuuu (otherwise this is an overlong encoding)
// and that uuuuu <= 0x10 (otherwise this is an out-of-range encoding).
// This is a simple range check. (We don't care about the low two bytes.)
if (!UnicodeUtility.IsInRangeInclusive(thisDWord, 0x1010_0000u, 0x140F_FFFFu))
{
goto Error; // overlong or out-of-range
}
}
// Validation of 4-byte case complete.
pInputBuffer += 4;
tempUtf16CodeUnitCountAdjustment -= 2; // 4 UTF-8 bytes -> 2 UTF-16 code units
tempScalarCountAdjustment--; // 2 UTF-16 code units -> 1 scalar
continue; // go back to beginning of loop for processing
}
goto ProcessRemainingBytesSlow;
ProcessInputOfLessThanDWordSize:
Debug.Assert(inputLength < 4);
nuint inputBufferRemainingBytes = (uint)inputLength;
goto ProcessSmallBufferCommon;
ProcessRemainingBytesSlow:
inputBufferRemainingBytes = (nuint)(void *)Unsafe.ByteOffset(ref *pInputBuffer, ref *pFinalPosWhereCanReadDWordFromInputBuffer) + 4;
ProcessSmallBufferCommon:
Debug.Assert(inputBufferRemainingBytes < 4);
while (inputBufferRemainingBytes > 0)
{
uint firstByte = pInputBuffer[0];
if ((byte)firstByte < 0x80u)
{
// 1-byte (ASCII) case
pInputBuffer++;
inputBufferRemainingBytes--;
continue;
}
else if (inputBufferRemainingBytes >= 2)
{
uint secondByte = pInputBuffer[1]; // typed as 32-bit since we perform arithmetic (not just comparisons) on this value
if ((byte)firstByte < 0xE0u)
{
// 2-byte case
if ((byte)firstByte >= 0xC2u && IsLowByteUtf8ContinuationByte(secondByte))
{
pInputBuffer += 2;
tempUtf16CodeUnitCountAdjustment--; // 2 UTF-8 bytes -> 1 UTF-16 code unit (and 1 scalar)
inputBufferRemainingBytes -= 2;
continue;
}
}
else if (inputBufferRemainingBytes >= 3)
{
if ((byte)firstByte < 0xF0u)
{
if ((byte)firstByte == 0xE0u)
{
if (!UnicodeUtility.IsInRangeInclusive(secondByte, 0xA0u, 0xBFu))
{
goto Error; // overlong encoding
}
}
else if ((byte)firstByte == 0xEDu)
{
if (!UnicodeUtility.IsInRangeInclusive(secondByte, 0x80u, 0x9Fu))
{
goto Error; // would be a UTF-16 surrogate code point
}
}
else
{
if (!IsLowByteUtf8ContinuationByte(secondByte))
{
goto Error; // first trailing byte doesn't have proper continuation marker
}
}
if (IsUtf8ContinuationByte(in pInputBuffer[2]))
{
pInputBuffer += 3;
tempUtf16CodeUnitCountAdjustment -= 2; // 3 UTF-8 bytes -> 2 UTF-16 code units (and 2 scalars)
inputBufferRemainingBytes -= 3;
continue;
}
}
}
}
// Error - no match.
goto Error;
}
// If we reached this point, we're out of data, and we saw no bad UTF8 sequence.
#if DEBUG
// Quick check that for the success case we're going to fulfill our contract of returning &inputBuffer[inputLength].
Debug.Assert(pOriginalInputBuffer + originalInputLength == pInputBuffer, "About to return an unexpected value.");
#endif
Error:
// Report back to our caller how far we got before seeing invalid data.
// (Also used for normal termination when falling out of the loop above.)
utf16CodeUnitCountAdjustment = tempUtf16CodeUnitCountAdjustment;
scalarCountAdjustment = tempScalarCountAdjustment;
return(pInputBuffer);
}
public Span <LbrEntry32> Entries(int totalSize)
{
IntPtr entriesOffset = Unsafe.ByteOffset(ref Unsafe.As <LbrTraceEventData32, byte>(ref this), ref Unsafe.As <LbrEntry32, byte>(ref _entries));
return(MemoryMarshal.CreateSpan(ref _entries, (totalSize - (int)entriesOffset) / sizeof(LbrEntry32)));
}
