/// <summary>Merges two colors for DXT decompression.</summary>
/// <param name="result">The storage to be used for the result.</param>
/// <param name="color1">A color.</param>
/// <param name="color2">A color.</param>
internal static unsafe void DxtMergeHalves(ArgbColor* result, ArgbColor* color1, ArgbColor* color2)
{
result->B = (byte)((color1->B + color2->B) >> 1);
result->G = (byte)((color1->G + color2->G) >> 1);
result->R = (byte)((color1->R + color2->R) >> 1);
result->A = 255;
}
/// <summary>Copies an <see cref="ArgbColor"/> into another, with forced opaque alpha.</summary>
/// <param name="destination">The destination color.</param>
/// <param name="source">The source color.</param>
internal static unsafe void CopyWithAlpha(ArgbColor* destination, ArgbColor* source, byte alpha)
{
destination->B = source->B;
destination->G = source->G;
destination->R = source->R;
destination->A = alpha;
}
/// <summary>Copies an <see cref="ArgbColor"/> into another, with forced opaque alpha.</summary>
/// <param name="destination">The destination color.</param>
/// <param name="source">The source color.</param>
internal static unsafe void CopyOpaque(ArgbColor* destination, ArgbColor* source)
{
destination->B = source->B;
destination->G = source->G;
destination->R = source->R;
destination->A = 255;
}
/// <summary>Reads a 256 color palette from a stream</summary>
/// <param name="reader">The BinaryReader used for reading data in the stream</param>
/// <param name="alphaOperation">The operation to apply on each palette entry's alpha component</param>
/// <returns>An array of bytes containing the palette entries</returns>
private ArgbColor[] ReadPalette(BinaryReader reader, AlphaOperation alphaOperation)
{
var palette = new ArgbColor[256];
for (int i = 0; i < palette.Length; i++)
{
byte b = reader.ReadByte();
byte g = reader.ReadByte();
byte r = reader.ReadByte();
byte a = reader.ReadByte();
palette[i] = new ArgbColor(r, g, b, alphaOperation != AlphaOperation.None ? alphaOperation == AlphaOperation.SetAlpha ? (byte)255 : (byte)~a : a);
}
return(palette);
}
public PaletteSurface(ArgbColor[] palette, int width, int height, bool opaque = true, byte separateAlphaBitCount = 0, bool alphaPremultiplied = false, bool sharePalette = false)
: base(width, height, GetAlphaBitCount(opaque, separateAlphaBitCount), alphaPremultiplied)
{
if (palette == null) throw new ArgumentNullException("palette");
if (palette.Length != 256) throw new ArgumentException();
int length = Width * Height;
if (this.separateAlpha = separateAlphaBitCount != 0)
{
int divisor = 8 / separateAlphaBitCount;
int remainder = length % divisor;
length = length / divisor;
if (remainder != 0) length++;
}
this.data = new byte[length];
this.palette = sharePalette ? palette : palette.Clone() as ArgbColor[];
}
private unsafe void CopyToArgbAlpha8(SurfaceData surfaceData)
{
fixed(byte *dataPointer = data)
fixed(ArgbColor * palettePointer = palette)
{
byte *destinationRowPointer = (byte *)surfaceData.DataPointer;
byte *sourceColorPointer = dataPointer;
byte *sourceAlphaPointer = dataPointer + Width * Height;
for (int i = Height; i-- != 0; destinationRowPointer += surfaceData.Stride)
{
ArgbColor *destinationPointer = (ArgbColor *)destinationRowPointer;
for (int j = Width; j-- != 0;)
{
ArgbColor.CopyWithAlpha(destinationPointer++, palettePointer + *sourceColorPointer++, *sourceAlphaPointer++);
}
}
}
}
private unsafe void CopyToArgbOpaque(SurfaceData surfaceData)
{
int rowLength = Width;
fixed(byte *dataPointer = data)
fixed(ArgbColor * palettePointer = palette)
{
byte *destinationRowPointer = (byte *)surfaceData.DataPointer;
byte *sourcePointer = dataPointer;
for (int i = Height; i-- != 0; destinationRowPointer += surfaceData.Stride)
{
ArgbColor *destinationPointer = (ArgbColor *)destinationRowPointer;
for (int j = Width; j-- != 0;)
{
ArgbColor.CopyOpaque(destinationPointer++, palettePointer + *sourcePointer++);
}
}
}
}
protected unsafe override void CopyToArgbInternal(SurfaceData surfaceData)
{
var colors = stackalloc ArgbColor[4];
fixed(byte *dataPointer = data)
{
var destinationRowPointer = (byte *)surfaceData.DataPointer;
var sourcePointer = dataPointer;
int rowBlockStride = surfaceData.Stride << 2;
for (int i = Height; i > 0; i -= 4, destinationRowPointer += rowBlockStride)
{
var destinationPointer = destinationRowPointer;
for (int j = Width; j > 0; j -= 4)
{
var alphaPointer = (ushort *)sourcePointer; // Save the alpha block pointer for later
sourcePointer += 8; // Get to the color block
ushort color0 = (ushort)(*sourcePointer++ | *sourcePointer++ << 8);
ushort color1 = (ushort)(*sourcePointer++ | *sourcePointer++ << 8);
colors[0] = new ArgbColor(color0);
colors[1] = new ArgbColor(color1);
ArgbColor.DxtMergeThirds(colors + 2, colors + 1, colors);
ArgbColor.DxtMergeThirds(colors + 3, colors, colors + 1);
// Handle the case where the surface's width is not a multiple of 4.
int inverseBlockWidth = j > 4 ? 0 : 4 - j;
var blockRowDestinationPointer = destinationPointer;
for (int k = 4; k-- != 0; blockRowDestinationPointer += surfaceData.Stride)
{
byte rowData = *sourcePointer++;
if (i + k < 4)
{
continue; // Handle the case where the surface's height is not a multiple of 4.
}
var blockDestinationPointer = (ArgbColor *)blockRowDestinationPointer;
// The small loop here has been unrolled, which should be well worth it:
// - No loop variable is needed.
// - No useless shift and incrementation for the last step.
// - Only one conditional jump.
// - The loop will be executed very often, as the blocks are very small.
switch (inverseBlockWidth)
{
case 0:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], (byte)(*alphaPointer << 4));
rowData >>= 2;
goto case 1;
case 1:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], (byte)(*alphaPointer & 0xF0));
rowData >>= 2;
goto case 2;
case 2:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], (byte)((*alphaPointer >> 4) & 0xF0));
rowData >>= 2;
goto case 3;
case 3:
ArgbColor.CopyWithAlpha(blockDestinationPointer, &colors[rowData & 3], (byte)((*alphaPointer++ >> 8) & 0xF0));
break;
}
}
destinationPointer += 4 * sizeof(uint); // Skip the 4 processed pixels
}
}
}
}
protected unsafe override void CopyToArgbInternal(SurfaceData surfaceData)
{
var colors = stackalloc ArgbColor[4];
var alpha = stackalloc byte[8];
fixed (byte* dataPointer = data)
{
var destinationRowPointer = (byte*)surfaceData.DataPointer;
var sourcePointer = dataPointer;
int rowBlockStride = surfaceData.Stride << 2;
for (int i = Height; i > 0; i -= 4, destinationRowPointer += rowBlockStride)
{
var destinationPointer = destinationRowPointer;
for (int j = Width; j > 0; j -= 4)
{
alpha[0] = *sourcePointer++;
alpha[1] = *sourcePointer++;
// The divisions here have been optimized with multiply/shift
// Maybe there is way for some optimization in the multiplications, (using shifts and additions/substractions where applicable)
// but for now I just hope the .NET JIT knows how to do those optimizations when they are possible…
if (alpha[0] <= alpha[1])
{
alpha[2] = (byte)((4 * alpha[0] + alpha[1]) * 1639 >> 13);
alpha[3] = (byte)((3 * alpha[0] + 2 * alpha[1]) * 1639 >> 13);
alpha[4] = (byte)((2 * alpha[0] + 3 * alpha[1]) * 1639 >> 13);
alpha[5] = (byte)((alpha[0] + 4 * alpha[1]) * 1639 >> 13);
alpha[6] = 0;
alpha[7] = 255;
}
else
{
alpha[2] = (byte)((6 * alpha[0] + alpha[1]) * 2341 >> 14);
alpha[3] = (byte)((5 * alpha[0] + 2 * alpha[1]) * 2341 >> 14);
alpha[4] = (byte)((4 * alpha[0] + 3 * alpha[1]) * 2341 >> 14);
alpha[5] = (byte)((3 * alpha[0] + 4 * alpha[1]) * 2341 >> 14);
alpha[6] = (byte)((2 * alpha[0] + 5 * alpha[1]) * 2341 >> 14);
alpha[7] = (byte)((alpha[0] + 6 * alpha[1]) * 2341 >> 14);
}
// Store the block's alpha data in a 64 bit integer. This will probably be a bit slower on 32-bit CPUs, but who cares… :p
ulong blockAlphaData = (ulong)(*sourcePointer++ | (uint)*sourcePointer++ << 8 | (uint)*sourcePointer++ << 16 | (uint)*sourcePointer++ << 24) | (ulong)(*sourcePointer++ | (uint)*sourcePointer++ << 8) << 32;
ushort color0 = (ushort)(*sourcePointer++ | *sourcePointer++ << 8);
ushort color1 = (ushort)(*sourcePointer++ | *sourcePointer++ << 8);
colors[0] = new ArgbColor(color0);
colors[1] = new ArgbColor(color1);
ArgbColor.DxtMergeThirds(colors + 2, colors + 1, colors);
ArgbColor.DxtMergeThirds(colors + 3, colors, colors + 1);
// Handle the case where the surface's width is not a multiple of 4.
int inverseBlockWidth = j > 4 ? 0 : 4 - j;
var blockRowDestinationPointer = destinationPointer;
for (int k = 4; k-- != 0; blockRowDestinationPointer += surfaceData.Stride)
{
byte rowData = *sourcePointer++;
if (i + k < 4) continue; // Handle the case where the surface's height is not a multiple of 4.
var blockDestinationPointer = (ArgbColor*)blockRowDestinationPointer;
if (inverseBlockWidth != 0) blockAlphaData >>= 3 * inverseBlockWidth;
// The small loop here has been unrolled, which shoudl be well worth it:
// - No loop variable is needed.
// - No useless shift and incrementation for the last step.
// - Only one conditional jump.
// - The loop will be executed very often, as the blocks are very small.
switch (inverseBlockWidth)
{
case 0:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], alpha[blockAlphaData & 7]);
rowData >>= 2;
blockAlphaData >>= 3;
goto case 1;
case 1:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], alpha[blockAlphaData & 7]);
rowData >>= 2;
blockAlphaData >>= 3;
goto case 2;
case 2:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], alpha[blockAlphaData & 7]);
rowData >>= 2;
blockAlphaData >>= 3;
goto case 3;
case 3:
ArgbColor.CopyWithAlpha(blockDestinationPointer, &colors[rowData & 3], alpha[blockAlphaData & 7]);
blockAlphaData >>= 3;
break;
}
}
destinationPointer += 4 * sizeof(uint); // Skip the 4 processed pixels
}
}
}
}
protected unsafe override void CopyToArgbInternal(SurfaceData surfaceData)
{
var colors = stackalloc ArgbColor[4];
fixed (byte* dataPointer = data)
{
var destinationRowPointer = (byte*)surfaceData.DataPointer;
var sourcePointer = dataPointer;
int rowBlockStride = surfaceData.Stride << 2;
for (int i = Height; i > 0; i -= 4, destinationRowPointer += rowBlockStride)
{
var destinationPointer = destinationRowPointer;
for (int j = Width; j > 0; j -= 4)
{
var alphaPointer = (ushort*)sourcePointer; // Save the alpha block pointer for later
sourcePointer += 8; // Get to the color block
ushort color0 = (ushort)(*sourcePointer++ | *sourcePointer++ << 8);
ushort color1 = (ushort)(*sourcePointer++ | *sourcePointer++ << 8);
colors[0] = new ArgbColor(color0);
colors[1] = new ArgbColor(color1);
ArgbColor.DxtMergeThirds(colors + 2, colors + 1, colors);
ArgbColor.DxtMergeThirds(colors + 3, colors, colors + 1);
// Handle the case where the surface's width is not a multiple of 4.
int inverseBlockWidth = j > 4 ? 0 : 4 - j;
var blockRowDestinationPointer = destinationPointer;
for (int k = 4; k-- != 0; blockRowDestinationPointer += surfaceData.Stride)
{
byte rowData = *sourcePointer++;
if (i + k < 4) continue; // Handle the case where the surface's height is not a multiple of 4.
var blockDestinationPointer = (ArgbColor*)blockRowDestinationPointer;
// The small loop here has been unrolled, which should be well worth it:
// - No loop variable is needed.
// - No useless shift and incrementation for the last step.
// - Only one conditional jump.
// - The loop will be executed very often, as the blocks are very small.
switch (inverseBlockWidth)
{
case 0:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], (byte)(*alphaPointer << 4));
rowData >>= 2;
goto case 1;
case 1:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], (byte)(*alphaPointer & 0xF0));
rowData >>= 2;
goto case 2;
case 2:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], (byte)((*alphaPointer >> 4) & 0xF0));
rowData >>= 2;
goto case 3;
case 3:
ArgbColor.CopyWithAlpha(blockDestinationPointer, &colors[rowData & 3], (byte)((*alphaPointer++ >> 8) & 0xF0));
break;
}
}
destinationPointer += 4 * sizeof(uint); // Skip the 4 processed pixels
}
}
}
}
protected unsafe override void CopyToArgbInternal(SurfaceData surfaceData)
{
var colors = stackalloc ArgbColor[4];
var alpha = stackalloc byte[8];
fixed(byte *dataPointer = data)
{
var destinationRowPointer = (byte *)surfaceData.DataPointer;
var sourcePointer = dataPointer;
int rowBlockStride = surfaceData.Stride << 2;
for (int i = Height; i > 0; i -= 4, destinationRowPointer += rowBlockStride)
{
var destinationPointer = destinationRowPointer;
for (int j = Width; j > 0; j -= 4)
{
alpha[0] = *sourcePointer++;
alpha[1] = *sourcePointer++;
// The divisions here have been optimized with multiply/shift
// Maybe there is way for some optimization in the multiplications, (using shifts and additions/substractions where applicable)
// but for now I just hope the .NET JIT knows how to do those optimizations when they are possible…
if (alpha[0] <= alpha[1])
{
alpha[2] = (byte)((4 * alpha[0] + alpha[1]) * 1639 >> 13);
alpha[3] = (byte)((3 * alpha[0] + 2 * alpha[1]) * 1639 >> 13);
alpha[4] = (byte)((2 * alpha[0] + 3 * alpha[1]) * 1639 >> 13);
alpha[5] = (byte)((alpha[0] + 4 * alpha[1]) * 1639 >> 13);
alpha[6] = 0;
alpha[7] = 255;
}
else
{
alpha[2] = (byte)((6 * alpha[0] + alpha[1]) * 2341 >> 14);
alpha[3] = (byte)((5 * alpha[0] + 2 * alpha[1]) * 2341 >> 14);
alpha[4] = (byte)((4 * alpha[0] + 3 * alpha[1]) * 2341 >> 14);
alpha[5] = (byte)((3 * alpha[0] + 4 * alpha[1]) * 2341 >> 14);
alpha[6] = (byte)((2 * alpha[0] + 5 * alpha[1]) * 2341 >> 14);
alpha[7] = (byte)((alpha[0] + 6 * alpha[1]) * 2341 >> 14);
}
// Store the block's alpha data in a 64 bit integer. This will probably be a bit slower on 32-bit CPUs, but who cares… :p
ulong blockAlphaData = (ulong)(*sourcePointer++ | (uint)*sourcePointer++ << 8 | (uint)*sourcePointer++ << 16 | (uint)*sourcePointer++ << 24) | (ulong)(*sourcePointer++ | (uint)*sourcePointer++ << 8) << 32;
ushort color0 = (ushort)(*sourcePointer++ | *sourcePointer++ << 8);
ushort color1 = (ushort)(*sourcePointer++ | *sourcePointer++ << 8);
colors[0] = new ArgbColor(color0);
colors[1] = new ArgbColor(color1);
ArgbColor.DxtMergeThirds(colors + 2, colors + 1, colors);
ArgbColor.DxtMergeThirds(colors + 3, colors, colors + 1);
// Handle the case where the surface's width is not a multiple of 4.
int inverseBlockWidth = j > 4 ? 0 : 4 - j;
var blockRowDestinationPointer = destinationPointer;
for (int k = 4; k-- != 0; blockRowDestinationPointer += surfaceData.Stride)
{
byte rowData = *sourcePointer++;
if (i + k < 4)
{
continue; // Handle the case where the surface's height is not a multiple of 4.
}
var blockDestinationPointer = (ArgbColor *)blockRowDestinationPointer;
if (inverseBlockWidth != 0)
{
blockAlphaData >>= 3 * inverseBlockWidth;
}
// The small loop here has been unrolled, which shoudl be well worth it:
// - No loop variable is needed.
// - No useless shift and incrementation for the last step.
// - Only one conditional jump.
// - The loop will be executed very often, as the blocks are very small.
switch (inverseBlockWidth)
{
case 0:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], alpha[blockAlphaData & 7]);
rowData >>= 2;
blockAlphaData >>= 3;
goto case 1;
case 1:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], alpha[blockAlphaData & 7]);
rowData >>= 2;
blockAlphaData >>= 3;
goto case 2;
case 2:
ArgbColor.CopyWithAlpha(blockDestinationPointer++, &colors[rowData & 3], alpha[blockAlphaData & 7]);
rowData >>= 2;
blockAlphaData >>= 3;
goto case 3;
case 3:
ArgbColor.CopyWithAlpha(blockDestinationPointer, &colors[rowData & 3], alpha[blockAlphaData & 7]);
blockAlphaData >>= 3;
break;
}
}
destinationPointer += 4 * sizeof(uint); // Skip the 4 processed pixels
}
}
}
}
/// <summary>Merges two colors for DXT decompression.</summary>
/// <param name="result">The storage to be used for the result.</param>
/// <param name="minColor">The color whose weight will be 1/3.</param>
/// <param name="maxColor">The color whose weight will be 2/3.</param>
internal static unsafe void DxtMergeThirds(ArgbColor* result, ArgbColor* minColor, ArgbColor* maxColor)
{
// Formula used here:
// x / 3 = x * 683 >> 11 (for 0 ≤ x ≤ 3 * 255)
// Need to verify that this is indeed faster, but it'll do the work for now.
result->B = (byte)((minColor->B + maxColor->B + maxColor->B) * 683 >> 11);
result->G = (byte)((minColor->G + maxColor->G + maxColor->G) * 683 >> 11);
result->R = (byte)((minColor->R + maxColor->R + maxColor->R) * 683 >> 11);
result->A = 255;
}
