|
using System;
|
|
using System.Runtime.InteropServices;
|
|
|
|
namespace MiscUtil.Conversion
|
|
{
|
|
/// <summary>
|
|
/// Equivalent of System.BitConverter, but with either endianness.
|
|
/// </summary>
|
|
public abstract class EndianBitConverter
|
|
{
|
|
#region Endianness of this converter
|
|
/// <summary>
|
|
/// Indicates the byte order ("endianess") in which data is converted using this class.
|
|
/// </summary>
|
|
/// <remarks>
|
|
/// Different computer architectures store data using different byte orders. "Big-endian"
|
|
/// means the most significant byte is on the left end of a word. "Little-endian" means the
|
|
/// most significant byte is on the right end of a word.
|
|
/// </remarks>
|
|
/// <returns>true if this converter is little-endian, false otherwise.</returns>
|
|
public abstract bool IsLittleEndian();
|
|
|
|
/// <summary>
|
|
/// Indicates the byte order ("endianess") in which data is converted using this class.
|
|
/// </summary>
|
|
public abstract Endianness Endianness { get; }
|
|
#endregion
|
|
|
|
#region Factory properties
|
|
static LittleEndianBitConverter little = new LittleEndianBitConverter();
|
|
/// <summary>
|
|
/// Returns a little-endian bit converter instance. The same instance is
|
|
/// always returned.
|
|
/// </summary>
|
|
public static LittleEndianBitConverter Little
|
|
{
|
|
get { return little; }
|
|
}
|
|
|
|
static BigEndianBitConverter big = new BigEndianBitConverter();
|
|
/// <summary>
|
|
/// Returns a big-endian bit converter instance. The same instance is
|
|
/// always returned.
|
|
/// </summary>
|
|
public static BigEndianBitConverter Big
|
|
{
|
|
get { return big; }
|
|
}
|
|
#endregion
|
|
|
|
#region Double/primitive conversions
|
|
/// <summary>
|
|
/// Converts the specified double-precision floating point number to a
|
|
/// 64-bit signed integer. Note: the endianness of this converter does not
|
|
/// affect the returned value.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert. </param>
|
|
/// <returns>A 64-bit signed integer whose value is equivalent to value.</returns>
|
|
public long DoubleToInt64Bits(double value)
|
|
{
|
|
return BitConverter.DoubleToInt64Bits(value);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Converts the specified 64-bit signed integer to a double-precision
|
|
/// floating point number. Note: the endianness of this converter does not
|
|
/// affect the returned value.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert. </param>
|
|
/// <returns>A double-precision floating point number whose value is equivalent to value.</returns>
|
|
public double Int64BitsToDouble (long value)
|
|
{
|
|
return BitConverter.Int64BitsToDouble(value);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Converts the specified single-precision floating point number to a
|
|
/// 32-bit signed integer. Note: the endianness of this converter does not
|
|
/// affect the returned value.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert. </param>
|
|
/// <returns>A 32-bit signed integer whose value is equivalent to value.</returns>
|
|
public int SingleToInt32Bits(float value)
|
|
{
|
|
return new Int32SingleUnion(value).AsInt32;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Converts the specified 32-bit signed integer to a single-precision floating point
|
|
/// number. Note: the endianness of this converter does not
|
|
/// affect the returned value.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert. </param>
|
|
/// <returns>A single-precision floating point number whose value is equivalent to value.</returns>
|
|
public float Int32BitsToSingle (int value)
|
|
{
|
|
return new Int32SingleUnion(value).AsSingle;
|
|
}
|
|
#endregion
|
|
|
|
#region To(PrimitiveType) conversions
|
|
/// <summary>
|
|
/// Returns a Boolean value converted from one byte at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <returns>true if the byte at startIndex in value is nonzero; otherwise, false.</returns>
|
|
public bool ToBoolean (byte[] value, int startIndex)
|
|
{
|
|
CheckByteArgument(value, startIndex, 1);
|
|
return BitConverter.ToBoolean(value, startIndex);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a Unicode character converted from two bytes at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <returns>A character formed by two bytes beginning at startIndex.</returns>
|
|
public char ToChar (byte[] value, int startIndex)
|
|
{
|
|
return unchecked((char) (CheckedFromBytes(value, startIndex, 2)));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a double-precision floating point number converted from eight bytes
|
|
/// at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <returns>A double precision floating point number formed by eight bytes beginning at startIndex.</returns>
|
|
public double ToDouble (byte[] value, int startIndex)
|
|
{
|
|
return Int64BitsToDouble(ToInt64(value, startIndex));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a single-precision floating point number converted from four bytes
|
|
/// at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <returns>A single precision floating point number formed by four bytes beginning at startIndex.</returns>
|
|
public float ToSingle (byte[] value, int startIndex)
|
|
{
|
|
return Int32BitsToSingle(ToInt32(value, startIndex));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a 16-bit signed integer converted from two bytes at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <returns>A 16-bit signed integer formed by two bytes beginning at startIndex.</returns>
|
|
public short ToInt16 (byte[] value, int startIndex)
|
|
{
|
|
return unchecked((short) (CheckedFromBytes(value, startIndex, 2)));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a 32-bit signed integer converted from four bytes at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <returns>A 32-bit signed integer formed by four bytes beginning at startIndex.</returns>
|
|
public int ToInt32 (byte[] value, int startIndex)
|
|
{
|
|
return unchecked((int) (CheckedFromBytes(value, startIndex, 4)));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a 64-bit signed integer converted from eight bytes at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <returns>A 64-bit signed integer formed by eight bytes beginning at startIndex.</returns>
|
|
public long ToInt64 (byte[] value, int startIndex)
|
|
{
|
|
return CheckedFromBytes(value, startIndex, 8);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a 16-bit unsigned integer converted from two bytes at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <returns>A 16-bit unsigned integer formed by two bytes beginning at startIndex.</returns>
|
|
public ushort ToUInt16 (byte[] value, int startIndex)
|
|
{
|
|
return unchecked((ushort) (CheckedFromBytes(value, startIndex, 2)));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a 32-bit unsigned integer converted from four bytes at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <returns>A 32-bit unsigned integer formed by four bytes beginning at startIndex.</returns>
|
|
public uint ToUInt32 (byte[] value, int startIndex)
|
|
{
|
|
return unchecked((uint) (CheckedFromBytes(value, startIndex, 4)));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a 64-bit unsigned integer converted from eight bytes at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <returns>A 64-bit unsigned integer formed by eight bytes beginning at startIndex.</returns>
|
|
public ulong ToUInt64 (byte[] value, int startIndex)
|
|
{
|
|
return unchecked((ulong) (CheckedFromBytes(value, startIndex, 8)));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Checks the given argument for validity.
|
|
/// </summary>
|
|
/// <param name="value">The byte array passed in</param>
|
|
/// <param name="startIndex">The start index passed in</param>
|
|
/// <param name="bytesRequired">The number of bytes required</param>
|
|
/// <exception cref="ArgumentNullException">value is a null reference</exception>
|
|
/// <exception cref="ArgumentOutOfRangeException">
|
|
/// startIndex is less than zero or greater than the length of value minus bytesRequired.
|
|
/// </exception>
|
|
static void CheckByteArgument(byte[] value, int startIndex, int bytesRequired)
|
|
{
|
|
if (value==null)
|
|
{
|
|
throw new ArgumentNullException("value");
|
|
}
|
|
if (startIndex < 0 || startIndex > value.Length-bytesRequired)
|
|
{
|
|
throw new ArgumentOutOfRangeException("startIndex");
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Checks the arguments for validity before calling FromBytes
|
|
/// (which can therefore assume the arguments are valid).
|
|
/// </summary>
|
|
/// <param name="value">The bytes to convert after checking</param>
|
|
/// <param name="startIndex">The index of the first byte to convert</param>
|
|
/// <param name="bytesToConvert">The number of bytes to convert</param>
|
|
/// <returns></returns>
|
|
long CheckedFromBytes(byte[] value, int startIndex, int bytesToConvert)
|
|
{
|
|
CheckByteArgument(value, startIndex, bytesToConvert);
|
|
return FromBytes(value, startIndex, bytesToConvert);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Convert the given number of bytes from the given array, from the given start
|
|
/// position, into a long, using the bytes as the least significant part of the long.
|
|
/// By the time this is called, the arguments have been checked for validity.
|
|
/// </summary>
|
|
/// <param name="value">The bytes to convert</param>
|
|
/// <param name="startIndex">The index of the first byte to convert</param>
|
|
/// <param name="bytesToConvert">The number of bytes to use in the conversion</param>
|
|
/// <returns>The converted number</returns>
|
|
protected abstract long FromBytes(byte[] value, int startIndex, int bytesToConvert);
|
|
#endregion
|
|
|
|
#region ToString conversions
|
|
/// <summary>
|
|
/// Returns a String converted from the elements of a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <remarks>All the elements of value are converted.</remarks>
|
|
/// <returns>
|
|
/// A String of hexadecimal pairs separated by hyphens, where each pair
|
|
/// represents the corresponding element in value; for example, "7F-2C-4A".
|
|
/// </returns>
|
|
public static string ToString(byte[] value)
|
|
{
|
|
return BitConverter.ToString(value);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a String converted from the elements of a byte array starting at a specified array position.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <remarks>The elements from array position startIndex to the end of the array are converted.</remarks>
|
|
/// <returns>
|
|
/// A String of hexadecimal pairs separated by hyphens, where each pair
|
|
/// represents the corresponding element in value; for example, "7F-2C-4A".
|
|
/// </returns>
|
|
public static string ToString(byte[] value, int startIndex)
|
|
{
|
|
return BitConverter.ToString(value, startIndex);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a String converted from a specified number of bytes at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <param name="length">The number of bytes to convert.</param>
|
|
/// <remarks>The length elements from array position startIndex are converted.</remarks>
|
|
/// <returns>
|
|
/// A String of hexadecimal pairs separated by hyphens, where each pair
|
|
/// represents the corresponding element in value; for example, "7F-2C-4A".
|
|
/// </returns>
|
|
public static string ToString(byte[] value, int startIndex, int length)
|
|
{
|
|
return BitConverter.ToString(value, startIndex, length);
|
|
}
|
|
#endregion
|
|
|
|
#region Decimal conversions
|
|
/// <summary>
|
|
/// Returns a decimal value converted from sixteen bytes
|
|
/// at a specified position in a byte array.
|
|
/// </summary>
|
|
/// <param name="value">An array of bytes.</param>
|
|
/// <param name="startIndex">The starting position within value.</param>
|
|
/// <returns>A decimal formed by sixteen bytes beginning at startIndex.</returns>
|
|
public decimal ToDecimal (byte[] value, int startIndex)
|
|
{
|
|
// HACK: This always assumes four parts, each in their own endianness,
|
|
// starting with the first part at the start of the byte array.
|
|
// On the other hand, there's no real format specified...
|
|
int[] parts = new int[4];
|
|
for (int i=0; i < 4; i++)
|
|
{
|
|
parts[i] = ToInt32(value, startIndex+i*4);
|
|
}
|
|
return new Decimal(parts);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the specified decimal value as an array of bytes.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <returns>An array of bytes with length 16.</returns>
|
|
public byte[] GetBytes(decimal value)
|
|
{
|
|
byte[] bytes = new byte[16];
|
|
int[] parts = decimal.GetBits(value);
|
|
for (int i=0; i < 4; i++)
|
|
{
|
|
CopyBytesImpl(parts[i], 4, bytes, i*4);
|
|
}
|
|
return bytes;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Copies the specified decimal value into the specified byte array,
|
|
/// beginning at the specified index.
|
|
/// </summary>
|
|
/// <param name="value">A character to convert.</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
public void CopyBytes(decimal value, byte[] buffer, int index)
|
|
{
|
|
int[] parts = decimal.GetBits(value);
|
|
for (int i=0; i < 4; i++)
|
|
{
|
|
CopyBytesImpl(parts[i], 4, buffer, i*4+index);
|
|
}
|
|
}
|
|
#endregion
|
|
|
|
#region GetBytes conversions
|
|
/// <summary>
|
|
/// Returns an array with the given number of bytes formed
|
|
/// from the least significant bytes of the specified value.
|
|
/// This is used to implement the other GetBytes methods.
|
|
/// </summary>
|
|
/// <param name="value">The value to get bytes for</param>
|
|
/// <param name="bytes">The number of significant bytes to return</param>
|
|
byte[] GetBytes(long value, int bytes)
|
|
{
|
|
byte[] buffer = new byte[bytes];
|
|
CopyBytes(value, bytes, buffer, 0);
|
|
return buffer;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the specified Boolean value as an array of bytes.
|
|
/// </summary>
|
|
/// <param name="value">A Boolean value.</param>
|
|
/// <returns>An array of bytes with length 1.</returns>
|
|
public byte[] GetBytes(bool value)
|
|
{
|
|
return BitConverter.GetBytes(value);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the specified Unicode character value as an array of bytes.
|
|
/// </summary>
|
|
/// <param name="value">A character to convert.</param>
|
|
/// <returns>An array of bytes with length 2.</returns>
|
|
public byte[] GetBytes(char value)
|
|
{
|
|
return GetBytes(value, 2);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the specified double-precision floating point value as an array of bytes.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <returns>An array of bytes with length 8.</returns>
|
|
public byte[] GetBytes(double value)
|
|
{
|
|
return GetBytes(DoubleToInt64Bits(value), 8);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the specified 16-bit signed integer value as an array of bytes.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <returns>An array of bytes with length 2.</returns>
|
|
public byte[] GetBytes(short value)
|
|
{
|
|
return GetBytes(value, 2);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the specified 32-bit signed integer value as an array of bytes.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <returns>An array of bytes with length 4.</returns>
|
|
public byte[] GetBytes(int value)
|
|
{
|
|
return GetBytes(value, 4);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the specified 64-bit signed integer value as an array of bytes.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <returns>An array of bytes with length 8.</returns>
|
|
public byte[] GetBytes(long value)
|
|
{
|
|
return GetBytes(value, 8);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the specified single-precision floating point value as an array of bytes.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <returns>An array of bytes with length 4.</returns>
|
|
public byte[] GetBytes(float value)
|
|
{
|
|
return GetBytes(SingleToInt32Bits(value), 4);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the specified 16-bit unsigned integer value as an array of bytes.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <returns>An array of bytes with length 2.</returns>
|
|
public byte[] GetBytes(ushort value)
|
|
{
|
|
return GetBytes(value, 2);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the specified 32-bit unsigned integer value as an array of bytes.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <returns>An array of bytes with length 4.</returns>
|
|
public byte[] GetBytes(uint value)
|
|
{
|
|
return GetBytes(value, 4);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the specified 64-bit unsigned integer value as an array of bytes.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <returns>An array of bytes with length 8.</returns>
|
|
public byte[] GetBytes(ulong value)
|
|
{
|
|
return GetBytes(unchecked((long)value), 8);
|
|
}
|
|
|
|
#endregion
|
|
|
|
#region CopyBytes conversions
|
|
/// <summary>
|
|
/// Copies the given number of bytes from the least-specific
|
|
/// end of the specified value into the specified byte array, beginning
|
|
/// at the specified index.
|
|
/// This is used to implement the other CopyBytes methods.
|
|
/// </summary>
|
|
/// <param name="value">The value to copy bytes for</param>
|
|
/// <param name="bytes">The number of significant bytes to copy</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
void CopyBytes(long value, int bytes, byte[] buffer, int index)
|
|
{
|
|
if (buffer==null)
|
|
{
|
|
throw new ArgumentNullException("buffer", "Byte array must not be null");
|
|
}
|
|
if (buffer.Length < index+bytes)
|
|
{
|
|
throw new ArgumentOutOfRangeException("Buffer not big enough for value");
|
|
}
|
|
CopyBytesImpl(value, bytes, buffer, index);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Copies the given number of bytes from the least-specific
|
|
/// end of the specified value into the specified byte array, beginning
|
|
/// at the specified index.
|
|
/// This must be implemented in concrete derived classes, but the implementation
|
|
/// may assume that the value will fit into the buffer.
|
|
/// </summary>
|
|
/// <param name="value">The value to copy bytes for</param>
|
|
/// <param name="bytes">The number of significant bytes to copy</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
protected abstract void CopyBytesImpl(long value, int bytes, byte[] buffer, int index);
|
|
|
|
/// <summary>
|
|
/// Copies the specified Boolean value into the specified byte array,
|
|
/// beginning at the specified index.
|
|
/// </summary>
|
|
/// <param name="value">A Boolean value.</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
public void CopyBytes(bool value, byte[] buffer, int index)
|
|
{
|
|
CopyBytes(value ? 1 : 0, 1, buffer, index);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Copies the specified Unicode character value into the specified byte array,
|
|
/// beginning at the specified index.
|
|
/// </summary>
|
|
/// <param name="value">A character to convert.</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
public void CopyBytes(char value, byte[] buffer, int index)
|
|
{
|
|
CopyBytes(value, 2, buffer, index);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Copies the specified double-precision floating point value into the specified byte array,
|
|
/// beginning at the specified index.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
public void CopyBytes(double value, byte[] buffer, int index)
|
|
{
|
|
CopyBytes(DoubleToInt64Bits(value), 8, buffer, index);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Copies the specified 16-bit signed integer value into the specified byte array,
|
|
/// beginning at the specified index.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
public void CopyBytes(short value, byte[] buffer, int index)
|
|
{
|
|
CopyBytes(value, 2, buffer, index);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Copies the specified 32-bit signed integer value into the specified byte array,
|
|
/// beginning at the specified index.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
public void CopyBytes(int value, byte[] buffer, int index)
|
|
{
|
|
CopyBytes(value, 4, buffer, index);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Copies the specified 64-bit signed integer value into the specified byte array,
|
|
/// beginning at the specified index.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
public void CopyBytes(long value, byte[] buffer, int index)
|
|
{
|
|
CopyBytes(value, 8, buffer, index);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Copies the specified single-precision floating point value into the specified byte array,
|
|
/// beginning at the specified index.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
public void CopyBytes(float value, byte[] buffer, int index)
|
|
{
|
|
CopyBytes(SingleToInt32Bits(value), 4, buffer, index);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Copies the specified 16-bit unsigned integer value into the specified byte array,
|
|
/// beginning at the specified index.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
public void CopyBytes(ushort value, byte[] buffer, int index)
|
|
{
|
|
CopyBytes(value, 2, buffer, index);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Copies the specified 32-bit unsigned integer value into the specified byte array,
|
|
/// beginning at the specified index.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
public void CopyBytes(uint value, byte[] buffer, int index)
|
|
{
|
|
CopyBytes(value, 4, buffer, index);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Copies the specified 64-bit unsigned integer value into the specified byte array,
|
|
/// beginning at the specified index.
|
|
/// </summary>
|
|
/// <param name="value">The number to convert.</param>
|
|
/// <param name="buffer">The byte array to copy the bytes into</param>
|
|
/// <param name="index">The first index into the array to copy the bytes into</param>
|
|
public void CopyBytes(ulong value, byte[] buffer, int index)
|
|
{
|
|
CopyBytes(unchecked((long)value), 8, buffer, index);
|
|
}
|
|
|
|
#endregion
|
|
|
|
#region Private struct used for Single/Int32 conversions
|
|
/// <summary>
|
|
/// Union used solely for the equivalent of DoubleToInt64Bits and vice versa.
|
|
/// </summary>
|
|
[StructLayout(LayoutKind.Explicit)]
|
|
struct Int32SingleUnion
|
|
{
|
|
/// <summary>
|
|
/// Int32 version of the value.
|
|
/// </summary>
|
|
[FieldOffset(0)]
|
|
int i;
|
|
/// <summary>
|
|
/// Single version of the value.
|
|
/// </summary>
|
|
[FieldOffset(0)]
|
|
float f;
|
|
|
|
/// <summary>
|
|
/// Creates an instance representing the given integer.
|
|
/// </summary>
|
|
/// <param name="i">The integer value of the new instance.</param>
|
|
internal Int32SingleUnion(int i)
|
|
{
|
|
this.f = 0; // Just to keep the compiler happy
|
|
this.i = i;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Creates an instance representing the given floating point number.
|
|
/// </summary>
|
|
/// <param name="f">The floating point value of the new instance.</param>
|
|
internal Int32SingleUnion(float f)
|
|
{
|
|
this.i = 0; // Just to keep the compiler happy
|
|
this.f = f;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the value of the instance as an integer.
|
|
/// </summary>
|
|
internal int AsInt32
|
|
{
|
|
get { return i; }
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns the value of the instance as a floating point number.
|
|
/// </summary>
|
|
internal float AsSingle
|
|
{
|
|
get { return f; }
|
|
}
|
|
}
|
|
#endregion
|
|
}
|
|
}
|